Deep neural network for automatic volumetric segmentation of whole-body CT images for body composition assessment

KEVS: enhancing segmentation of visceral adipose tissue in pre-cystectomy CT with Gaussian kernel density estimation

PURPOSE

The distribution of visceral adipose tissue (VAT) in cystectomy patients is indicative of the incidence of postoperative complications. Existing VAT segmentation methods for computed tomography (CT) employing intensity thresholding have limitations relating to inter-observer variability. Moreover, the difficulty in creating ground-truth masks limits the development of deep learning (DL) models for this task. This paper introduces a novel method for VAT prediction in pre-cystectomy CT, which is fully automated and does not require ground-truth VAT masks for training, overcoming aforementioned limitations.

METHODS

We introduce the kernel density-enhanced VAT segmentator (KEVS), combining a DL semantic segmentation model, for multi-body feature prediction, with Gaussian kernel density estimation analysis of predicted subcutaneous adipose tissue to achieve accurate scan-specific predictions of VAT in the abdominal cavity. Uniquely for a DL pipeline, KEVS does not require ground-truth VAT masks.

RESULTS

We verify the ability of KEVS to accurately segment abdominal organs in unseen CT data and compare KEVS VAT segmentation predictions to existing state-of-the-art (SOTA) approaches in a dataset of 20 pre-cystectomy CT scans, collected from University College London Hospital (UCLH-Cyst), with expert ground-truth annotations. KEVS presents a 4.80 % and 6.02 % improvement in Dice coefficient over the second best DL and thresholding-based VAT segmentation techniques respectively when evaluated on UCLH-Cyst.

CONCLUSION

This research introduces KEVS, an automated, SOTA method for the prediction of VAT in pre-cystectomy CT which eliminates inter-observer variability and is trained entirely on open-source CT datasets which do not contain ground-truth VAT masks.

Deep learning for automatic volumetric bowel segmentation on body CT images

OBJECTIVES

To develop a deep neural network for automatic bowel segmentation and assess its applicability for estimating large bowel length (LBL) in individuals with constipation.

MATERIALS AND METHODS

We utilized contrast-enhanced and non-enhanced abdominal, chest, and whole-body CT images for model development. External testing involved paired pre- and post-contrast abdominal CT images from another hospital. We developed 3D nnU-Net models to segment the gastrointestinal tract and separate it into the esophagus, stomach, small bowel, and large bowel. Segmentation accuracy was evaluated using the Dice similarity coefficient (DSC) based on radiologists' segmentation. We employed the network to estimate LBL in individuals having abdominal CT for health check-ups, and the height-corrected LBL was compared between groups with and without constipation.

RESULTS

One hundred thirty-three CT scans (88 patients; age, 63.6 ± 10.6 years; 39 men) were used for model development, and 60 for external testing (30 patients; age, 48.9 ± 15.8 years; 16 men). In the external dataset, the mean DSC for the entire gastrointestinal tract was 0.985 ± 0.008. The mean DSCs for four-part separation exceeded 0.95, outperforming TotalSegmentator, except for the esophagus (DSC, 0.807 ± 0.173). For LBL measurements, 100 CT scans from 51 patients were used (age, 67.0 ± 6.9 years; 59 scans from men; 59 with constipation). The height-corrected LBL were significantly longer in the constipation group on both per-exam (79.1 ± 12.4 vs 88.8 ± 15.8 cm/m, p = 0.001) and per-subject basis (77.6 ± 13.6 vs 86.9 ± 17.1 cm/m, p = 0.04).

CONCLUSION

Our model accurately segmented the entire gastrointestinal tract and its major compartments from CT scans and enabled the noninvasive estimation of LBL in individuals with constipation.

KEY POINTS

Questions Automated bowel segmentation is a first step for algorithms, including bowel tracing and length measurement, but the complexity of the gastrointestinal tract limits its accuracy. Findings Our 3D nnU-Net model showed high performance in segmentation and four-part separation of the GI tract (DSC > 0.95), except for the esophagus. Clinical relevance Our model accurately segments the gastrointestinal tract and separates it into major compartments. Our model potentially has use in various clinical applications, including semi-automated measurement of LBL in individuals with constipation.

Association of Deep Learning-based Chest CT-derived Respiratory Parameters with Disease Progression in Amyotrophic Lateral Sclerosis

Background Forced vital capacity (FVC) is a standard measure of respiratory function in patients with amyotrophic lateral sclerosis (ALS) but has limitations, particularly for patients with bulbar impairment. Purpose To determine the value of deep learning-based chest CT-derived respiratory parameters in predicting ALS progression and survival. Materials and Methods This retrospective study included patients with ALS diagnosed between January 2010 and July 2023 who underwent chest CT at a tertiary hospital. Deep learning-based software was used to measure lung and respiratory muscle volume, normalized for height as the lung volume index (LVI) and respiratory muscle index (RMI). Differences in these parameters across King clinical stages were assessed using ordinal logistic regression. Tracheostomy-free survival was evaluated using Cox regression and time-dependent receiver operating characteristic analysis. Subgroup analysis was conducted for patients with bulbar impairment. In addition, a Gaussian process regressor model was developed to estimate FVC based on lung volume, respiratory muscle volume, age, and sex. Results A total of 261 patients were included in the study (mean age, 65.2 years ± 11.9 [SD]; 156 male patients). LVI and RMI decreased with increasing King stage (both P < .001). The high LVI and high RMI groups had better survival (both P < .001). After adjustment, LVI (hazard ratio [HR] = 0.998 [95% CI: 0.996, 1.000]; P = .021) and RMI (HR = 0.992 [95% CI: 0.988, 0.996]; P < .001) remained independent prognostic factors. In patients with bulbar impairment, LVI (HR = 0.998 [95% CI: 0.996, 1.000]; P = .029) and RMI (HR = 0.991 [95% CI: 0.987, 0.996]; P < .001) were independent prognostic factors. Time-dependent receiver operating characteristic curve analysis revealed no significant differences in survival prediction performance among LVI, RMI, and FVC. The Gaussian process regressor model estimated FVC with approximately 8% error. Conclusion The deep learning-derived CT metrics LVI and RMI reflected ALS stage, enabled FVC prediction, and supported assessment in patients with limited respiratory function. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Rahsepar and Bedayat in this issue.

Development of Multicenter Deep Learning Models for Predicting Surgical Complexity and Surgical Site Infection in Abdominal Wall Reconstruction, a Pilot Study

Objective

Hernia recurrence and surgical site infection (SSI) are grave complications in Abdominal Wall Reconstruction (AWR). This study aimed to develop multicenter deep learning models (DLMs) developed for predicting surgical complexity, using Component Separation Technique (CST) as a surrogate, and the risk of surgical site infections (SSI) in AWR, using preoperative computed tomography (CT) images.

Methods

Multicenter models were created using deidentified CT images from two tertiary AWR centers. The models were developed with ResNet-18 architecture. Model performance was reported as accuracy and AUC.

Results

The CST model underperformed with an AUC of 0.569, while the SSI model exhibited strong performance with an AUC of 0.898.

Conclusion

The study demonstrated the successful development of a multicenter DLM for SSI prediction in AWR, highlighting the impact of patient factors over surgical practice variability in predicting SSIs with DLMs. The CST model's prediction remained challenging, which we hypothesize reflects the subjective nature of surgical decisions and varying institutional practices. Our findings underscore the potential of AI-enhanced surgical risk calculators to risk stratify patients and potentially improve patient outcomes.

Validation of body composition parameters extracted via deep learning-based segmentation from routine computed tomographies

Sarcopenia and body composition metrics are strongly associated with patient outcomes. In this study, we developed and validated a flexible, open-access pipeline integrating available deep learning-based segmentation models with pre- and postprocessing steps to extract body composition measures from routine computed tomography (CT) scans. In 337 surgical oncology patients, total skeletal muscle tissue (SMtotal), psoas muscle tissue (SMpsoas), visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT) were quantified both manually and using the pipeline. Automated and manual measurements showed strong correlations (SMpsoas: r = 0.776, VAT: r = 0.993, SAT: r = 0.984; all P < 0.001). Measurement discrepancies primarily resulted from segmentation errors, anatomical anomalies or image irregularities. SMpsoas measurements showed substantial variability depending on slice selection, whereas SMtotal, averaged across all L3 levels, provided greater measurement stability. Overall, SMtotal performed comparably to SMpsoas in predicting overall survival (OS). In summary, body composition measures derived from the pipeline strongly correlated with manual measurements and were prognostic for OS. The increased stability of SMtotal across vertebral levels suggests it may serve as a more reliable alternative to psoas-based assessments. Future studies should address the identified areas of improvement to enhance the accuracy of automated segmentation models.

Prognostic impact of low muscle mass on clinical outcomes in patients who undergo lung transplant

BACKGROUND

Low muscle mass (LMM) is recognized as a poor prognostic factor in various chronic lung diseases. However, its prognostic impact on recipients of lung transplants remains inconclusive.

METHODS

We retrospectively analyzed patients who underwent lung transplantation at a tertiary referral center in South Korea. Pretransplant skeletal muscle mass was quantified at the L1 vertebral level by computed tomography scans of the chest using a commercially available body composition analysis software. Patients were classified into LMM and non-LMM group using a threshold for LMM that had been previously validated in the South Korean population. We then evaluated the prognostic impact of preoperative LMM on clinical outcomes after lung transplantation.

RESULTS

A total of 107 patients were included in this analysis, of whom 44 (41.1%) were classified into the LMM group. The median follow-up duration was 958 days posttransplantation. A preoperative LMM was identified as an independent factor associated with a greater risk of overall mortality (adjusted hazard ratio, 2.15; 95% confidence interval, 1.07-4.34). In addition, patients with LMM had a greater risk of developing primary graft dysfunction (adjusted odds ratio, 3.56; 95% confidence interval, 1.25-10.18). At the 1-year follow-up, 37.5% of the patients with baseline LMM had recovered and were reclassified into the non-LMM group, and this improvement was found to mitigate the negative impact of preoperative LMM.

CONCLUSIONS

Pretransplant LMM was significantly associated with poor clinical outcomes in recipients of lung transplants. These findings highlight the importance of maintaining adequate muscle mass during the waiting period for lung transplantation.

Application of Artificial Intelligence in Thoracic Radiology: A Narrative Review

Thoracic radiology has emerged as a primary field in which artificial intelligence (AI) is extensively researched. Recent advancements highlight the potential to enhance radiologists' performance through AI. AI aids in detecting and classifying abnormalities, and in quantifying both normal and abnormal anatomical structures. Additionally, it facilitates prognostication by leveraging these quantitative values. This review article will discuss the recent achievements of AI in thoracic radiology, focusing primarily on deep learning, and explore the current limitations and future directions of this cutting-edge technique.

Automated abdominal organ segmentation algorithms for non-enhanced CT for volumetry and 3D radiomics analysis

PURPOSE

To develop fully-automated abdominal organ segmentation algorithms from non-enhanced abdominal CT and low-dose chest CT and assess their feasibility for automated CT volumetry and 3D radiomics analysis of abdominal solid organs.

METHODS

Fully-automated nnU-Net-based models were developed to segment the liver, spleen, and both kidneys in non-enhanced abdominal CT, and the liver and spleen in low-dose chest CT. 105 abdominal CTs and 60 low-dose chest CTs were used for model development, and 55 abdominal CTs and 10 low-dose chest CTs for external testing. The segmentation performance for each organ was assessed using the Dice similarity coefficients, with manual segmentation results serving as the ground truth. Agreements between ground-truth measurements and model estimates of organ volume and 3D radiomics features were assessed using the Bland-Altman analysis and intraclass correlation coefficients (ICC).

RESULTS

The models accurately segmented the liver, spleen, right kidney, and left kidney in abdominal CT and the liver and spleen in low-dose chest CT, showing mean Dice similarity coefficients in the external dataset of 0.968, 0.960, 0.952, and 0.958, respectively, in abdominal CT, and 0.969 and 0.960, respectively, in low-dose chest CT. The model-estimated and ground truth volumes of these organs exhibited mean differences between - 0.7% and 2.2%, with excellent agreements. The automatically extracted mean and median Hounsfield units (ICCs, 0.970-0.999 and 0.994-0.999, respectively), uniformity (ICCs, 0.985-0.998), entropy (ICCs, 0.931-0.993), elongation (ICCs, 0.978-0.992), and flatness (ICCs, 0.973-0.997) showed excellent agreement with ground truth measurements for each organ; however, skewness (ICCs, 0.210-0.831), kurtosis (ICCs, 0.053-0.933), and sphericity (ICCs, 0.368-0.819) displayed relatively low and inconsistent agreement.

CONCLUSION

Our nnU-Net-based models accurately segmented abdominal solid organs in non-enhanced abdominal and low-dose chest CT, enabling reliable automated measurements of organ volume and specific 3D radiomics features.

Harnessing radiomics and nutritional metrics to predict long-term survival in Fournier's gangrene patients

PURPOSE

To evaluate the association of traditional and novel nutritional measurements with survival in Fournier's gangrene (FG) patients.

METHODS

We reviewed records of FG patients from our tertiary center (Jan 2013-Jan 2022). Radiomic sarcopenia parameters (Psoas Muscle Area [PMA], Roundness, Solidity, and calculated PMA-Index) were measured from admission CT scans at the L3 level using ImageJ software. We assessed sarcopenia's impact on survival through three analyses: Model 1 used a PMI below the sex-adjusted median; Models 2 and 3 used published cutoffs. Kaplan-Meier curves were used to compare survival between sarcopenic and non-sarcopenic patients. Multivariable Cox and logistic regression analyses adjusted for age and the Charlson Comorbidity Index (CCI) to assess mortality risk.

RESULTS

Of 130 men and 31 women (82% white), 60 patients (37.3%) had died after a median follow-up of 2.2 years (IQR 0.9-4.4). Survival rates were 94% at 30 days, 92% at 90 days, 80% at 1 year, 77% at 2 years, and 56% at 5 years. Non-survivors were older (median age 63 vs 55.1 years, p < 0.001) and had higher median CCI (4.8 vs 3; p < 0.001).In Model 1, sarcopenic patients had a non-significant increased mortality risk with hazard ratio (HR 1.47, 95% CI 0.82-2.64, p = 0.196). Models 2 and 3 showed similar results (HR 1.41, 95% CI 0.70-2.84, p = 0.325; HR 1.35, 95% CI 0.70-2.61, p = 0.364). None of the models were significant when adjusting for CCI and age. Survivors had better traditional metabolic profiles, including higher albumin (3.1vs 2.7 g/dL), hemoglobin (12.4vs 11.4 g/dL), and lower creatinine (1.39 vs 2.1 mg/dL); however, none of these were significant when adjusting for age and CCI.

CONCLUSIONS

Despite a mild trend, none of the sarcopenia models were able to predict long-term mortality in FG patients in our cohort. This well-known, cost-effective nutritional predictor still requires further research to optimize its utilization in the FG patient population.

CT-Less Whole-Body Bone Segmentation of PET Images Using a Multimodal Deep Learning Network

In bone cancer imaging, positron emission tomography (PET) is ideal for the diagnosis and staging of bone cancers due to its high sensitivity to malignant tumors. The diagnosis of bone cancer requires tumor analysis and localization, where accurate and automated wholebody bone segmentation (WBBS) is often needed. Current WBBS for PET imaging is based on paired Computed Tomography (CT) images. However, mismatches between CT and PET images often occur due to patient motion, which leads to erroneous bone segmentation and thus, to inaccurate tumor analysis. Furthermore, there are some instances where CT images are unavailable for WBBS. In this work, we propose a novel multimodal fusion network (MMF-Net) for WBBS of PET images, without the need for CT images. Specifically, the tracer activity ($\lambda$-MLAA), attenuation map ($\mu$-MLAA), and synthetic attenuation map ($\mu$-DL) images are introduced into the training data. We first design a multi-encoder structure employed to fully learn modalityspecific encoding representations of the three PET modality images through independent encoding branches. Then, we propose a multimodal fusion module in the decoder to further integrate the complementary information across the three modalities. Additionally, we introduce revised convolution units, SE (Squeeze-and-Excitation) Normalization and deep supervision to improve segmentation performance. Extensive comparisons and ablation experiments, using 130 whole-body PET image datasets, show promising results. We conclude that the proposed method can achieve WBBS with moderate to high accuracy using PET information only, which potentially can be used to overcome the current limitations of CT-based approaches, while minimizing exposure to ionizing radiation.

Low Skeletal Muscle Radiodensity Predicts Response to CDK4/6 Inhibitors Plus Aromatase Inhibitors in Advanced Breast Cancer

BACKGROUND

Recent evidence indicates that a dysregulated host metabolism influences treatment outcomes in patients with breast cancer. We investigated the association of computed tomography (CT)-derived body composition indices with therapeutic responses in patients with hormone receptor-positive, HER2-negative advanced breast cancer (ABC) on endocrine plus CDK4/6 inhibitor (CDK4/6i) treatment.

METHODS

The study involved a retrospective cohort of patients with ABC at the Yonsei Cancer Center who received CDK4/6i and aromatase inhibitors as first-line therapy between January 2017 and October 2020. Body composition parameters were estimated from the non-enhanced CT images of the third lumbar spine by commercialized deep learning software. Patients with low skeletal muscle radiodensity (SMD) were defined as patients with SMD of low tertile (≤ 28.7 Hounsfield Units). The primary outcome was progression-free survival (PFS).

RESULTS

Among the 247 female participants (median age, 53 years; mean body mass index [BMI], 23.7 kg/m2), 45.7% had disease progression or death during a median follow-up of 36.4 months. After adjusting for age and visceral metastasis, SMD was the only independent predictor among body composition parameters for worse PFS (adjusted hazard ratio [HR] = 1.20 per standard deviation decrement, 95% CI: 1.01-1.42, p = 0.041), whereas BMI, muscle area, and fat area were not. Participants with low SMD had a higher risk of progression than those without (PFS, 27.2 vs. 51.1 months, p = 0.009; adjusted HR 1.84, 95% CI: 1.22-2.76, p = 0.003). Strong associations between low SMD and poor PFS were observed in groups with pre-menopause status (HR, 3.04 vs. 1.19 in post-menopause; 95% CI: 1.54-5.99, p for interaction < 0.05) and without visceral metastases (HR, 2.95 vs. 1.19 in with visceral metastases; 95% CI: 1.59-5.49, p for interaction < 0.05).

CONCLUSIONS

CT-defined low SMD predicts poor treatment outcomes in patients with ABC undergoing first-line treatment with aromatase inhibitors and CDK4/6i.

Body composition and cardiometabolic risks of patients with adrenal tumours in relation to hormonal activity: a large cross-sectional single-centre study

OBJECTIVE

We aimed to examine how different types of adrenal hormone excess influence body composition.

DESIGN

A retrospective, cross-sectional, single-centre study.

METHODS

We retrospectively enrolled 2971 consecutive adults with adrenal tumours and age-, sex-, and body mass index-matched controls at a 1:3 ratio. The area and attenuation of skeletal muscle and fat at the L3 vertebrae were measured using computed tomography-based analysis software. Prevalence ratios of cardiometabolic outcomes were calculated using the Poisson regression.

RESULTS

Patients with non-functioning adenoma (n = 1354) and mild autonomous cortisol secretion (MACS; n = 786) showed similar body compositions. Patients with overt Cushing's syndrome (CS) had the highest visceral fat (VF) area to skeletal muscle area ratio (1.14), while pheochromocytoma (PHEO) patients had the lowest (0.52). Muscle attenuation was lowest in CS and highest in PHEO (32.6 vs 41.5 Hounsfield units, P < .001). Mild autonomous cortisol secretion patients had higher risks of hypertension and dyslipidaemia than non-functioning adenoma patients. Non-functioning adenoma and MACS patients had higher VF area and lower muscle/fat attenuation compared with controls, while primary aldosteronism patients had body compositions similar to controls, except for higher fat attenuation.

CONCLUSIONS

Adrenal tumours are associated with altered body composition. Even patients with non-functioning adenoma and MACS had increased VF area and lower muscle and fat attenuation compared with controls, indicating potential cardiometabolic risks.

Artificial intelligence-based body composition analysis using computed tomography images predicts both prevalence and incidence of diabetes mellitus

AIM/INTRODUCTION

We assess the efficacy of artificial intelligence (AI)-based, fully automated, volumetric body composition metrics in predicting the risk of diabetes.

MATERIALS AND METHODS

This was a cross-sectional and 10-year retrospective longitudinal study. The cross-sectional analysis included health check-up data of 15,330 subjects with abdominal computed tomography (CT) images between January 1, 2011, and September 30, 2012. Of these, 10,570 subjects with available follow-up data were included in the longitudinal analyses. The volume of each body segment included in the abdominal CT images was measured using AI-based image analysis software.

RESULTS

Visceral fat (VF) proportion and VF/subcutaneous fat (SF) ratio increased with age, and both strongly predicted the presence and risk of developing diabetes. Optimal cut-offs for VF proportion were 24% for men and 16% for women, while VF/SF ratio values were 1.2 for men and 0.5 for women. The subjects with higher VF/SF ratio and VF proportion were associated with a greater risk of having diabetes (adjusted OR 2.0 [95% CI 1.7-2.4] in men; 2.9 [2.2-3.9] in women). In subjects with normal glucose tolerance, higher VF proportion and VF/SF ratio were associated with higher risk of developing prediabetes or diabetes (adjusted HR 1.3 [95% CI 1.1-1.4] in men; 1.4 [1.2-1.7] in women). These trends were consistently observed across each specified cut-off value.

CONCLUSIONS

AI-based volumetric analysis of abdominal CT images can be useful in obtaining body composition data and predicting the risk of diabetes.

Effects of respiratory sarcopenia on the postoperative course in elderly lung cancer patient: a retrospective study

OBJECTIVES

Recently, sarcopenia has been linked to unfavorable outcomes in various surgical procedures, including lung cancer surgery. This study aimed to investigate the impact of respiratory sarcopenia (RS) on postoperative and long-term outcomes in elderly patients undergoing lung cancer surgery.

METHODS

This retrospective study included patients aged 70 years and older who underwent lobectomy with curative intent for lung cancer between 2017 and 2019. RS was defined as having values below the median for both the L3 skeletal muscle index, measured from preoperative PET-CT images, and peak expiratory flow (PEF). An inverse probability of treatment weighting (IPTW) approach was applied to balance covariates between the RS and non-RS groups. Baseline characteristics and postoperative outcomes were compared between groups using t-tests and chi-square tests. Kaplan-Meier curves and log-rank tests were used to compare overall and recurrence-free survival. Multivariable logistic regression analysis incorporating IPTW weights was performed to assess the impact of RS on respiratory complications.

RESULTS

A total of 509 patients were included, of whom 123 (24.2%) had RS. After IPTW adjustment, baseline characteristics, including pulmonary function, were similar between the RS and non-RS groups. All patients underwent lobectomy, with 78.8% of the RS group and 80.9% of the non-RS group undergoing minimally invasive surgery. The RS group had a significantly higher rate of respiratory complications compared to the non-RS group (14.5% vs. 7.7%, p = 0.041). Multivariable logistic regression analysis showed that male sex (odds ratio = 15.2, p < 0.01) and lower DLCO (odds ratio = 0.96, p < 0.01) were significantly associated with respiratory complications, whereas RS did not show a significant association (p = 0.05). No significant differences were found in overall survival (p = 0.11) or recurrence-free survival (p = 0.51) between the groups.

CONCLUSIONS

In this study, RS had a limited impact on both postoperative and long-term outcomes in elderly patients undergoing lung cancer surgery. These findings suggest that other factors, such as DLCO and male sex, may play a more prominent role in predicting respiratory complications.

Association of Obesity and Skeletal Muscle with Postoperative Survival in Non-Small Cell Lung Cancer

Background A comprehensive assessment of skeletal muscle health is crucial to understanding the association between improved clinical outcomes and obesity as defined by body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) in lung cancer, but limited studies have been conducted on this topic. Purpose To investigate the association between BMI-defined obesity and survival in patients with non-small cell lung cancer who underwent curative resection, with a specific focus on the status of skeletal muscle assessed at CT. Materials and Methods This retrospective study investigated Korean patients with non-small cell lung cancer who underwent curative resection between January 2008 and December 2019. Patients were classified into nonobese (BMI <25) or obese (BMI ≥25) groups. Skeletal muscle status was assessed at CT at the level of the third lumbar vertebrae. Low skeletal muscle mass (LSMM) was defined as the sex-specific lowest quartile. Cox regression analysis was used to evaluate the associations of BMI and muscle status with overall survival. Results A total of 7076 patients (mean age, 62.5 years ± 9.7 [SD]; 4081 male) were included, of whom 2512 (35.5%) had a BMI greater than or equal to 25 (obese group). In the setting of absent LSMM and myosteatosis, patients in the obese group had longer overall survival compared with patients in the nonobese group (hazard ratio [HR], 0.77; 95% CI: 0.66, 0.90; P = .001). The associations between obesity and lower mortality were observed only in male patients (HR, 0.72; 95% CI: 0.60, 0.85; P < .001) and patients who had ever smoked (HR, 0.71; 95% CI: 0.60, 0.85; P < .001) who were without LSMM and myosteatosis, with effect differing according to sex and smoking status (P value range, <.001 to .02 for interaction). Conclusion Obesity is associated with improved overall survival in patients with non-small cell lung cancer after curative resection when skeletal muscle mass and radiodensity are preserved. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Vannier in this issue.

Evaluation of a fully automated computed tomography image segmentation method for fast and accurate body composition measurements

INTRODUCTION

Body composition evaluation can be used to assess patients' nutritional status to predict clinical outcomes. To facilitate reliable and time-efficient body composition measurements eligible for clinical practice, fully automated computed tomography segmentation methods were developed. The aim of this study was to evaluate automated segmentation by Data Analysis Facilitation Suite in an independent dataset.

MATERIALS AND METHODS

Preoperative computed tomography images were used of 165 patients undergoing cytoreductive surgery with hyperthermic intraperitoneal chemotherapy from 2014 to 2019. Manual and automated measurements of skeletal muscle mass (SMM), visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), and intramuscular adipose tissue (IMAT) were performed at the third lumbar vertebra. Segmentation accuracy of automated measurements was assessed using the Jaccard index and intra-class correlation coefficients.

RESULTS

Automatic segmentation provided accurate measurements compared to manual analysis, resulting in Jaccard score coefficients of 94.9 for SMM, 98.4 for VAT, 99.1 for SAT, and 79.4 for IMAT. Intra-class correlation coefficients ranged from 0.98 to 1.00. Automated measurements on average overestimated SMM and SAT areas compared to manual analysis, with mean differences (±2 standard deviations) of 1.10 (-1.91 to 4.11) and 1.61 (-2.26 to 5.48) respectively. For VAT and IMAT, automated measurements on average underestimated the areas with mean differences of -1.24 (-3.35 to 0.87) and -0.93 (-5.20 to 3.35), respectively.

CONCLUSIONS

Commercially available Data Analysis Facilitation Suite provides similar results compared to manual measurements of body composition at the level of third lumbar vertebra. This software provides accurate and time-efficient body composition measurements, which is necessary for implementation in clinical practice.

AI model using CT-based imaging biomarkers to predict hepatocellular carcinoma in patients with chronic hepatitis B

BACKGROUND & AIMS

Various hepatocellular carcinoma (HCC) prediction models have been proposed for patients with chronic hepatitis B (CHB) using clinical variables. We aimed to develop an artificial intelligence (AI)-based HCC prediction model by incorporating imaging biomarkers derived from abdominal computed tomography (CT) images along with clinical variables.

METHODS

An AI prediction model employing a gradient-boosting machine algorithm was developed utilizing imaging biomarkers extracted by DeepFore, a deep learning-based CT auto-segmentation software. The derivation cohort (n = 5,585) was randomly divided into the training and internal validation sets at a 3:1 ratio. The external validation cohort included 2,883 patients. Six imaging biomarkers (i.e. abdominal visceral fat-total fat volume ratio, total fat-trunk volume ratio, spleen volume, liver volume, liver-spleen Hounsfield unit ratio, and muscle Hounsfield unit) and eight clinical variables were selected as the main variables of our model, PLAN-B-DF.

RESULTS

In the internal validation set (median follow-up duration = 7.4 years), PLAN-B-DF demonstrated an excellent predictive performance with a c-index of 0.91 and good calibration function (p = 0.78 by the Hosmer-Lemeshow test). In the external validation cohort (median follow-up duration = 4.6 years), PLAN-B-DF showed a significantly better discrimination function compared to previous models, including PLAN-B, PAGE-B, modified PAGE-B, and CU-HCC (c-index, 0.89 vs. 0.65-0.78; all p <0.001), and maintained a good calibration function (p = 0.42 by the Hosmer-Lemeshow test). When patients were classified into four groups according to the risk probability calculated by PLAN-B-DF, the 10-year cumulative HCC incidence was 0.0%, 0.4%, 16.0%, and 46.2% in the minimal-, low-, intermediate-, and high-risk groups, respectively.

CONCLUSION

This AI prediction model, integrating deep learning-based auto-segmentation of CT images, offers improved performance in predicting HCC risk among patients with CHB compared to previous models.

IMPACT AND IMPLICATIONS

The novel predictive model PLAN-B-DF, employing an automated computed tomography segmentation algorithm, significantly improves predictive accuracy and risk stratification for hepatocellular carcinoma in patients with chronic hepatitis B (CHB). Using a gradient-boosting algorithm and computed tomography metrics, such as visceral fat volume and myosteatosis, PLAN-B-DF outperforms previous models based solely on clinical and demographic data. This model not only shows a higher c-index compared to previous models, but also effectively classifies patients with CHB into different risk groups. This model uses machine learning to analyze the complex relationships among various risk factors contributing to hepatocellular carcinoma occurrence, thereby enabling more personalized surveillance for patients with CHB.

Automated extraction of body composition metrics from abdominal CT or MR imaging: A scoping review

PURPOSE

To review methodological approaches for automated segmentation of subcutaneous adipose tissue, visceral adipose tissue, and skeletal muscle from abdominal cross-sectional imaging for body composition analysis.

METHOD

Four databases were searched for publications describing automated segmentation of subcutaneous adipose tissue, visceral adipose tissue, and/or skeletal muscle from abdominal CT or MR imaging between 2019 and 2023. Included reports were evaluated to assess how imaging modality, cohort size, vertebral level, model dimensionality, and use of a volume or single slice affected segmentation accuracy and/or clinical utility. Exclusion criteria included reports not in English language, manual or semi-automated segmentation methods, reports prior to 2019 or solely of paediatric patients, and those not describing the use of abdominal CT or MR.

RESULTS

After exclusions, 172 reports were included in the review. CT imaging was utilised approximately four times as often as MRI, and segmentation accuracy did not significantly differ between the two modalities. Cohort size had no significant effect on segmentation accuracy. There was little evidence to refute the current practice of extracting body composition metrics from the third lumbar vertebral level. There was no clear benefit of using a 3D model to perform segmentation over a 2D approach.

CONCLUSION

Automated segmentation of intra-abdominal soft tissues for body composition analysis is an intense area of research activity. Segmentation accuracy is not affected by cross-sectional imaging modality. Extracting metrics from a single slice at the third lumbar vertebral level is a common approach, however, extracting metrics from a volumetric slab surrounding this level may increase the resilience of the technique, which is important for clinical translation. A paucity of publicly available datasets led to most reports using different data sources, preventing direct comparison of segmentation techniques. Future efforts should prioritise creating a standardised dataset to facilitate benchmarking of different algorithms and subsequent clinical adoption.

Artificial Intelligence-Driven Volumetric Analysis of Muscle Mass as a Predictor of Tumor Response to Neoadjuvant Chemoradiotherapy in Patients with Rectal Cancer

Background/Objectives: Artificial intelligence (AI)-based volumetric measurements for assessing sarcopenia are expected to offer comprehensive insight into three-dimensional muscle volume and distribution. Therefore, we investigated the role of sarcopenia using computed tomography (CT)-based automated AI volumetric muscle measurements in predicting neoadjuvant chemoradiotherapy (nCRT) response and prognosis in patients with rectal cancer who underwent nCRT. Methods: We retrospectively analyzed the data of patients who underwent nCRT followed by curative resection between March 2010 and August 2021. Sarcopenia was defined using the Q1 cutoff value of the volumetric skeletal muscle index (SMI). The association between pre-nCRT volumetric sarcopenia and nCRT response was analyzed using logistic regression. A Cox proportional hazards model was used to identify the prognostic value of the pre- and post-nCRT volumetric SMIs. Results: Notably, 22 (25.6%) of the 86 patients had volumetric sarcopenia. The sarcopenia group showed a poorer nCRT response than the non-sarcopenia group. Pre-nCRT sarcopenia was a significant predictor of poor nCRT response (OR, 0.34 [95% CI, 0.12-0.96]; p = 0.041). Furthermore, an increased volumetric SMI during nCRT was a more significant prognostic factor on recurrence-free survival (aHR, 0.26 [95% CI, 0.08-0.83]; p = 0.023) and overall survival (aHR, 0.41 [95% CI, 0.17-0.99]; p = 0.049) than a decreased SMI. Conclusions: Volumetric sarcopenia can be used to predict poor nCRT response. A reduction in volumetric sarcopenia can be a poor prognostic factor in patients with rectal cancer who undergo nCRT.

Can Preoperative Hounsfield Unit Measurement Help Predict Mechanical Failure in Metastatic Spinal Tumor Surgery?

Background/Objectives: This study aimed to identify risk factors associated with mechanical failure in patients undergoing spinal instrumentation without fusion for metastatic spinal tumors. Methods: We retrospectively evaluated data from 220 patients with spinal tumors who underwent instrumentation without fusion. Propensity scores were used to match preoperative variables, resulting in the inclusion of 24 patients in the failure group (F group) and 72 in the non-failure group (non-F group). Demographic, surgical, and radiological characteristics were compared between the two groups. Logistic regression and Kaplan-Meier survival analyses were conducted to identify predictors of mechanical failure. Results: Propensity score matching resulted in a balanced distribution of covariates. Lower Hounsfield unit (HU) values at the lowest instrumented vertebra (LIV) were the only independent predictor of implant failure (p = 0.037). A cutoff value of 127.273 HUs was determined to predict mechanical failure, with a sensitivity of 59.1%, specificity of 73.4%, and area under the curve of 0.655 (95% confidence interval: 0.49-0.79). A significant difference in survival was observed between the groups with HU values above and below the cutoff (p = 0.0057). Cement-augmented screws were underutilized, with an average of only 0.2 screws per patient in the F group. Conclusions: Preoperative LIV HU values < 127.273 were strongly associated with an increased risk of mechanical failure following spinal instrumentation without fusion. Alternative surgical strategies including the use of cement-augmented screws are recommended for patients with low HU values.

Initial sarcopenia and body composition changes as prognostic factors in cervical cancer patients treated with concurrent chemoradiation: An artificial intelligence-based volumetric study

OBJECTIVES

This study aimed to investigate the influence of baseline sarcopenia and changes in body composition on survival during cervical cancer treatment.

METHODS

Patients diagnosed with stage IB1-IVB cervical cancer who underwent primary concurrent chemoradiation therapy (CCRT) between 2002 and 2022 were included. The exclusion criteria were prior radical hysterectomy, lack of pretreatment computed tomography (CT) imaging, or significant comorbidities. An artificial intelligence-based automatic segmentation program assessed body composition by analyzing CT images, defining L3 sarcopenia (L3 skeletal muscle index [SMI] <39cm2/m2) and volumetric sarcopenia (volumetric SMI <180.4 cm3/m3). Comparative and multivariate analyses identified the prognostic factors. The impact of body component changes during CCRT was explored.

RESULTS

Among 347 patients, there were 125 recurrences and 59 deaths (median follow-up, 50.5 months). Seven patients were excluded from the volumetric sarcopenia analysis because of incomplete baseline CT data, and 175 patients were included in the analysis of body composition changes. Patients with L3 sarcopenia had a lower 5-year progression-free survival (PFS) rate (55.6% vs. 66.2%, p = 0.027), while those with volumetric sarcopenia showed a poorer 5-year overall survival rate (76.5% vs. 85.1%, p = 0.036). Patients with total fat loss during CCRT had a worse 5-year PFS rate than those with total fat gain (61.9% vs. 73.8%, p = 0.029). Multivariate analyses revealed that total fat loss (adjusted hazard ratio [aHR], 2.172; 95% confidence interval [CI], 1.066-4.424; p = 0.033) was a significant factor for recurrence, whereas L3 sarcopenia was not. Volumetric sarcopenia increased the risk of death by 1.75-fold (aHR, 1.750; 95% CI, 1.012-3.025; p = 0.045).

CONCLUSIONS

Among patients with cervical cancer undergoing CCRT, initial volumetric sarcopenia and fat loss during treatment are survival risk factors. These findings suggest the potential importance of personalized supportive care, including tailored nutrition and exercise interventions.

Accelerated muscle mass estimation from CT images through transfer learning

BACKGROUND

The cost of labeling to collect training data sets using deep learning is especially high in medical applications compared to other fields. Furthermore, due to variances in images depending on the computed tomography (CT) devices, a deep learning based segmentation model trained with a certain device often does not work with images from a different device.

METHODS

In this study, we propose an efficient learning strategy for deep learning models in medical image segmentation. We aim to overcome the difficulties of segmentation in CT images by training a VNet segmentation model which enables rapid labeling of organs in CT images with the model obtained by transfer learning using a small number of manually labeled images, called SEED images. We established a process for generating SEED images and conducting transfer learning a model. We evaluate the performance of various segmentation models such as vanilla UNet, UNETR, Swin-UNETR and VNet. Furthermore, assuming a scenario that a model is repeatedly trained with CT images collected from multiple devices, in which is catastrophic forgetting often occurs, we examine if the performance of our model degrades.

RESULTS

We show that transfer learning can train a model that does a good job of segmenting muscles with a small number of images. In addition, it was confirmed that VNet shows better performance when comparing the performance of existing semi-automated segmentation tools and other deep learning networks to muscle and liver segmentation tasks. Additionally, we confirmed that VNet is the most robust model to deal with catastrophic forgetting problems.

CONCLUSION

In the 2D CT image segmentation task, we confirmed that the CNN-based network shows better performance than the existing semi-automatic segmentation tool or latest transformer-based networks.

Metabolic phenotyping with computed tomography deep learning for metabolic syndrome, osteoporosis and sarcopenia predicts mortality in adults

BACKGROUND

Computed tomography (CT) body compositions reflect age-related metabolic derangements. We aimed to develop a multi-outcome deep learning model using CT multi-level body composition parameters to detect metabolic syndrome (MS), osteoporosis and sarcopenia by identifying metabolic clusters simultaneously. We also investigated the prognostic value of metabolic phenotyping by CT model for long-term mortality.

METHODS

The derivation set (n = 516; 75% train set, 25% internal test set) was constructed using age- and sex-stratified random sampling from two community-based cohorts. Data from participants in the individual health assessment programme (n = 380) were used as the external test set 1. Semi-automatic quantification of body compositions at multiple levels of abdominal CT scans was performed to train a multi-layer perceptron (MLP)-based multi-label classification model. External test set 2 to test the prognostic value of the model output for mortality was built using data from individuals who underwent abdominal CT in a tertiary-level institution (n = 10 141).

RESULTS

The mean ages of the derivation and external sets were 62.8 and 59.7 years, respectively, without difference in sex distribution (women 50%) or body mass index (BMI; 23.9 kg/m2). Skeletal muscle density (SMD) and bone density (BD) showed a more linear decrement across age than skeletal muscle area. Alternatively, an increase in visceral fat area (VFA) was observed in both men and women. Hierarchical clustering based on multi-level CT body composition parameters revealed three distinctive phenotype clusters: normal, MS and osteosarcopenia clusters. The L3 CT-parameter-based model, with or without clinical variables (age, sex and BMI), outperformed clinical model predictions of all outcomes (area under the receiver operating characteristic curve: MS, 0.76 vs. 0.55; osteoporosis, 0.90 vs. 0.79; sarcopenia, 0.85 vs. 0.81 in external test set 1; P < 0.05 for all). VFA contributed the most to the MS predictions, whereas SMD, BD and subcutaneous fat area were features of high importance for detecting osteoporosis and sarcopenia. In external test set 2 (mean age 63.5 years, women 79%; median follow-up 4.9 years), a total of 907 individuals (8.9%) died during follow-up. Among model-predicted metabolic phenotypes, sarcopenia alone (adjusted hazard ratio [aHR] 1.55), MS + sarcopenia (aHR 1.65), osteoporosis + sarcopenia (aHR 1.83) and all three combined (aHR 1.87) remained robust predictors of mortality after adjustment for age, sex and comorbidities.

CONCLUSIONS

A CT body composition-based MLP model detected MS, osteoporosis and sarcopenia simultaneously in community-dwelling and hospitalized adults. Metabolic phenotypes predicted by the CT MLP model were associated with long-term mortality, independent of covariates.

Differential Impact of Subcutaneous and Visceral Fat on Bone Changes after Gastrectomy

BACKGRUOUND

Osteoporosis and fragility fractures are crucial musculoskeletal complications in long-term survivors of gastric cancer. However, the relationship between changes in body composition after gastrectomy and bone loss has not been investigated. Therefore, this study aimed to explore whether computed tomography (CT)-derived body composition parameters are associated with bone loss after gastrectomy in patients with gastric cancer.

METHODS

We retrospectively reviewed medical records and abdomen CT scans of patients who underwent gastrectomy at Yonsei University Severance Hospital between 2009 and 2018. Patients with non-metastatic gastric adenocarcinoma and preoperative and postoperative non-contrast CT scans were analyzed. Section area of skeletal muscle (SMA), visceral fat (VFA), and subcutaneous fat (SFA) were assessed using semi-automatic segmentation software. Changes in trabecular bone attenuation of L1 mid-vertebra level (L1 Hounsfield units [HU]) were measured.

RESULTS

Fifty-seven patients (mean age, 65.5±10.6; 70.2% males) were analyzed, and the median duration was 31 months. Fortyseven patients (82.5%) lost weight after gastrectomy. Baseline SMA and VFA did not differ between the bone loss and preserved groups; however, baseline SFA was significantly higher in the bone preserved group than in the bone loss group (P=0.020). In a multivariable linear regression model adjusted for confounding factors, one standard deviation higher VFA at baseline was associated with greater annualized L1 HU loss (%) (P=0.034). However, higher preoperative SFA was associated with protection against bone loss after gastrectomy (P=0.025).

CONCLUSION

Higher preoperative SFA exhibited a protective effect against bone loss after gastrectomy in patients with non-metastatic gastric cancer, whereas VFA exhibited a negative effect.

Automated Comprehensive CT Assessment of the Risk of Diabetes and Associated Cardiometabolic Conditions

Background CT performed for various clinical indications has the potential to predict cardiometabolic diseases. However, the predictive ability of individual CT parameters remains underexplored. Purpose To evaluate the ability of automated CT-derived markers to predict diabetes and associated cardiometabolic comorbidities. Materials and Methods This retrospective study included Korean adults (age ≥ 25 years) who underwent health screening with fluorine 18 fluorodeoxyglucose PET/CT between January 2012 and December 2015. Fully automated CT markers included visceral and subcutaneous fat, muscle, bone density, liver fat, all normalized to height (in meters squared), and aortic calcification. Predictive performance was assessed with area under the receiver operating characteristic curve (AUC) and Harrell C-index in the cross-sectional and survival analyses, respectively. Results The cross-sectional and cohort analyses included 32166 (mean age, 45 years ± 6 [SD], 28833 men) and 27 298 adults (mean age, 44 years ± 5 [SD], 24 820 men), respectively. Diabetes prevalence and incidence was 6% at baseline and 9% during the 7.3-year median follow-up, respectively. Visceral fat index showed the highest predictive performance for prevalent and incident diabetes, yielding AUC of 0.70 (95% CI: 0.68, 0.71) for men and 0.82 (95% CI: 0.78, 0.85) for women and C-index of 0.68 (95% CI: 0.67, 0.69) for men and 0.82 (95% CI: 0.77, 0.86) for women, respectively. Combining visceral fat, muscle area, liver fat fraction, and aortic calcification improved predictive performance, yielding C-indexes of 0.69 (95% CI: 0.68, 0.71) for men and 0.83 (95% CI: 0.78, 0.87) for women. The AUC for visceral fat index in identifying metabolic syndrome was 0.81 (95% CI: 0.80, 0.81) for men and 0.90 (95% CI: 0.88, 0.91) for women. CT-derived markers also identified US-diagnosed fatty liver, coronary artery calcium scores greater than 100, sarcopenia, and osteoporosis, with AUCs ranging from 0.80 to 0.95. Conclusion Automated multiorgan CT analysis identified individuals at high risk of diabetes and other cardiometabolic comorbidities. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Pickhardt in this issue.

Visceral Adipose Tissue Reduction Measured by Deep Neural Network Architecture Improved Reflux Esophagitis Endoscopic Grade

INTRODUCTION

Visceral obesity is a risk factor for reflux esophagitis (RE). We investigated the risk of RE according to visceral adipose tissue (VAT) measured by deep neural network architecture using computed tomography (CT) and evaluated the longitudinal association between abdominal adipose tissue changes and the disease course of RE.

METHODS

Individuals receiving health checkups who underwent esophagogastroduodenoscopy (EGD) and abdominal CT at Seoul National University Healthcare System Gangnam Center between 2015 and 2016 were included. Visceral and subcutaneous adipose tissue areas and volumes were measured using a deep neural network architecture and CT. The association between the abdominal adipose tissue area and volume and the risk of RE was evaluated. Participants who underwent follow-up EGD and abdominal CT were selected; the effects of changes in abdominal adipose tissue area and volume on RE endoscopic grade were investigated using Cox proportional hazards regression.

RESULTS

We enrolled 6,570 patients who underwent EGD and abdominal CT on the same day. RE was associated with male sex, hypertension, diabetes, excessive alcohol intake, current smoking status, and levels of physical activity. The VAT area and volume increased the risk of RE dose-dependently. A decreasing VAT volume was significantly associated with improvement in RE endoscopic grade (hazard ratio: 3.22, 95% confidence interval: 1.82-5.71). Changes in subcutaneous adipose tissue volume and the disease course of RE were not significantly correlated.

DISCUSSION

Visceral obesity is strongly associated with RE. VAT volume reduction was prospectively associated with improvement in RE endoscopic grade dose-dependently. Visceral obesity is a potential target for RE treatment.

Radiopharmaceuticals for Skeletal Muscle PET Imaging

The skeletal muscles account for approximately 40% of the body weight and are crucial in movement, nutrient absorption, and energy metabolism. Muscle loss and decline in function cause a decrease in the quality of life of patients and the elderly, leading to complications that require early diagnosis. Positron emission tomography/computed tomography (PET/CT) offers non-invasive, high-resolution visualization of tissues. It has emerged as a promising alternative to invasive diagnostic methods and is attracting attention as a tool for assessing muscle function and imaging muscle diseases. Effective imaging of muscle function and pathology relies on appropriate radiopharmaceuticals that target key aspects of muscle metabolism, such as glucose uptake, adenosine triphosphate (ATP) production, and the oxidation of fat and carbohydrates. In this review, we describe how [18F]fluoro-2-deoxy-D-glucose ([18F]FDG), [18F]fluorocholine ([18F]FCH), [11C]acetate, and [15O]water ([15O]H2O) are suitable radiopharmaceuticals for diagnostic imaging of skeletal muscles.

Prognostic value of initial and longitudinal changes in body composition in metastatic pancreatic cancer

BACKGROUND

Sarcopenia or visceral adipose tissue has been reported to be related to pancreatic cancer prognosis. However, clinical relevance of the comprehensive analysis of body compositions and their longitudinal changes is lacking. This study analysed the association between body composition changes after chemotherapy and survival in patients with metastatic pancreatic cancer.

METHODS

We retrospectively included 456 patients (mean age ± standard deviation, 61.2 ± 10.0 years; 272 males and 184 females) with metastatic pancreatic cancer who received palliative chemotherapy from May 2011 to December 2019. Using deep learning-based, fully automated segmentation of contrast-enhanced computed tomography (CT) at the time of diagnosis, cross-sectional areas of muscle, subcutaneous adipose tissue and visceral adipose tissue were extracted from a single axial image of the portal venous phase at L3 level. Skeletal muscle index (SMI), visceral adipose tissue index (VATI), subcutaneous adipose tissue index (SATI) and mean skeletal muscle attenuation (MA) were calculated, and their effect on overall survival (OS) was analysed. Longitudinal changes in body composition and prognostic values were also analysed in a subgroup of patients with 2- and 6-month follow-up CT (n = 349).

RESULTS

A total of 452 deaths occurred during follow-up in the entire cohort. The survival rate was 49.3% (95% confidence interval [CI], 44.9-54.2) at 1 year and 3.7% (95% CI, 2.0-6.8) at 5 years. In multivariable analysis, higher MA (≥44.4 HU in males and ≥34.8 HU in females) at initial CT was significantly associated with better OS in both males and females (adjusted hazard ratio [HR], 0.706; 95% CI, 0.538-0.925; P = 0.012 for males, and HR, 0.656; 95% CI, 0.475-0.906; P = 0.010 for females), whereas higher SATI (≥42.8 cm2/m2 in males and ≥65.8 cm2/m2 in females) was significantly associated with better OS in female patients only (adjusted HR, 0.568; 95% CI, 0.388-0.830; P = 0.003). In longitudinal analysis, SMI, VATI and SATI significantly decreased between initial and 2-month follow-up CT, whereas mean MA significantly decreased between 2- and 6-month follow-up CT. In multivariable Cox regression analysis of longitudinal changes, which was stratified by disease control state, SATI change was significantly associated with OS in male patients (adjusted HR, 0.513; 95% CI, 0.354-0.745; P < 0.001), while other body composition parameters were not.

CONCLUSIONS

In patients with metastatic pancreatic cancer, body composition mostly changed during the first 2 months after starting chemotherapy, and the prognostic factors associated with OS differed between males and females. Initial and longitudinal changes of body composition are associated with OS of metastatic pancreatic cancer.

Advanced detection of coronary artery disease via deep learning analysis of plasma cytokine data

The 2017 World Health Organization Fact Sheet highlights that coronary artery disease is the leading cause of death globally, responsible for approximately 30% of all deaths. In this context, machine learning (ML) technology is crucial in identifying coronary artery disease, thereby saving lives. ML algorithms can potentially analyze complex patterns and correlations within medical data, enabling early detection and accurate diagnosis of CAD. By leveraging ML technology, healthcare professionals can make informed decisions and implement timely interventions, ultimately leading to improved outcomes and potentially reducing the mortality rate associated with coronary artery disease. Machine learning algorithms create non-invasive, quick, accurate, and economical diagnoses. As a result, machine learning algorithms can be employed to supplement existing approaches or as a forerunner to them. This study shows how to use the CNN classifier and RNN based on the LSTM classifier in deep learning to attain targeted "risk" CAD categorization utilizing an evolving set of 450 cytokine biomarkers that could be used as suggestive solid predictive variables for treatment. The two used classifiers are based on these "45" different cytokine prediction characteristics. The best Area Under the Receiver Operating Characteristic curve (AUROC) score achieved is (0.98) for a confidence interval (CI) of 95; the classifier RNN-LSTM used "450" cytokine biomarkers had a great (AUROC) score of 0.99 with a confidence interval of 0.95 the percentage 95, the CNN model containing cytokines received the second best AUROC score (0.92). The RNN-LSTM classifier considerably beats the CNN classifier regarding AUROC scores, as evidenced by a p-value smaller than 7.48 obtained via an independent t-test. As large-scale initiatives to achieve early, rapid, reliable, inexpensive, and accessible individual identification of CAD risk gain traction, robust machine learning algorithms can now augment older methods such as angiography. Incorporating 65 new sensitive cytokine biomarkers can increase early detection even more. Investigating the novel involvement of cytokines in CAD could lead to better risk detection, disease mechanism discovery, and new therapy options.

Association between body fat decrease during the first year after diagnosis and the prognosis of idiopathic pulmonary fibrosis: CT-based body composition analysis

BACKGROUND

The prognostic role of changes in body fat in patients with idiopathic pulmonary fibrosis (IPF) remains underexplored. We investigated the association between changes in body fat during the first year post-diagnosis and outcomes in patients with IPF.

METHODS

This single-center, retrospective study included IPF patients with chest CT scan and pulmonary function test (PFT) at diagnosis and a one-year follow-up between January 2010 and December 2020. The fat area (cm2, sum of subcutaneous and visceral fat) and muscle area (cm2) at the T12-L1 level were obtained from chest CT images using a fully automatic deep learning-based software. Changes in the body composition were dichotomized using thresholds dividing the lowest quartile and others, respectively (fat area: -52.3 cm2, muscle area: -7.4 cm2). Multivariable Cox regression analyses adjusted for PFT result and IPF extent on CT images and the log-rank test were performed to assess the association between the fat area change during the first year post-diagnosis and the composite outcome of death or lung transplantation.

RESULTS

In total, 307 IPF patients (69.3 ± 8.1 years; 238 men) were included. During the first year post-diagnosis, fat area, muscle area, and body mass index (BMI) changed by -15.4 cm2, -1 cm2, and - 0.4 kg/m2, respectively. During a median follow-up of 47 months, 146 patients had the composite outcome (47.6%). In Cox regression analyses, a change in the fat area < -52.3 cm2 was associated with composite outcome incidence in models adjusted with baseline clinical variables (hazard ratio [HR], 1.566, P = .022; HR, 1.503, P = .036 in a model including gender, age, and physiology [GAP] index). This prognostic value was consistent when adjusted with one-year changes in clinical variables (HR, 1.495; P = .030). However, the change in BMI during the first year was not a significant prognostic factor (P = .941). Patients with a change in fat area exceeding this threshold experienced the composite outcome more frequently than their counterparts (58.4% vs. 43.9%; P = .007).

CONCLUSION

A ≥ 52.3 cm2 decrease in fat area, automatically measured using deep learning technique, at T12-L1 in one year post-diagnosis was an independent poor prognostic factor in IPF patients.

Fully-automated multi-organ segmentation tool applicable to both non-contrast and post-contrast abdominal CT: deep learning algorithm developed using dual-energy CT images

A novel 3D nnU-Net-based of algorithm was developed for fully-automated multi-organ segmentation in abdominal CT, applicable to both non-contrast and post-contrast images. The algorithm was trained using dual-energy CT (DECT)-obtained portal venous phase (PVP) and spatiotemporally-matched virtual non-contrast images, and tested using a single-energy (SE) CT dataset comprising PVP and true non-contrast (TNC) images. The algorithm showed robust accuracy in segmenting the liver, spleen, right kidney (RK), and left kidney (LK), with mean dice similarity coefficients (DSCs) exceeding 0.94 for each organ, regardless of contrast enhancement. However, pancreas segmentation demonstrated slightly lower performance with mean DSCs of around 0.8. In organ volume estimation, the algorithm demonstrated excellent agreement with ground-truth measurements for the liver, spleen, RK, and LK (intraclass correlation coefficients [ICCs] > 0.95); while the pancreas showed good agreements (ICC = 0.792 in SE-PVP, 0.840 in TNC). Accurate volume estimation within a 10% deviation from ground-truth was achieved in over 90% of cases involving the liver, spleen, RK, and LK. These findings indicate the efficacy of our 3D nnU-Net-based algorithm, developed using DECT images, which provides precise segmentation of the liver, spleen, and RK and LK in both non-contrast and post-contrast CT images, enabling reliable organ volumetry, albeit with relatively reduced performance for the pancreas.

Three-dimensional body composition parameters using automatic volumetric segmentation allow accurate prediction of colorectal cancer outcomes

BACKGROUND

Parameters obtained from two-dimensional (2D) cross-sectional images have been used to determine body composition. However, data from three-dimensional (3D) volumetric body images reflect real body composition more accurately and may be better predictors of patient outcomes in cancer. This study aimed to assess the 3D parameters and determine the best predictive factors for patient prognosis.

METHODS

Patients who underwent surgery for colorectal cancer (CRC) between 2010 and 2016 were included in this study. Preoperative computed tomography images were analysed using an automatic segmentation program. Body composition parameters for muscle, muscle adiposity, subcutaneous fat (SF) and abdominal visceral fat (AVF) were assessed using 2D images at the third lumbar (L3) level and 3D images of the abdominal waist (L1-L5). The cut-off points for each parameter were determined using X-tile software. A Cox proportional hazards regression model was used to identify the association between the parameters and the treatment outcomes, and the relative influence of each parameter was compared using a gradient boosting model.

RESULTS

Overall, 499 patients were included in the study. At a median follow-up of 59 months, higher 3D parameters of the abdominal muscles and SF from the abdominal waist were found to be associated with longer overall survival (OS) and disease-free survival (all P < 0.001). Although the 3D parameters of AVF were not related to survival outcomes, patients with a high AVF volume and mass experienced higher rate of postoperative complications than those with low AVF volume (27.4% vs. 18.7%, P = 0.021, for mass; 27.1% vs. 19.0%, P = 0.028, for volume). Low muscle mass and volume (hazard ratio [HR] 1.959, P = 0.016; HR 2.093, P = 0.036, respectively) and low SF mass and volume (HR 1.968, P = 0.008; HR 2.561, P = 0.003, respectively), both in the abdominal waist, were identified as independent prognostic factors for worse OS. Along with muscle mass and volume, SF mass and volume in the abdominal waist were negatively correlated with mortality (all P < 0.001). Both AVF mass and volume in the abdominal waist were positively correlated with postoperative complications (P < 0.05); 3D muscle volume and SF at the abdominal waist were the most influential factors for OS.

CONCLUSIONS

3D volumetric parameters generated using an automatic segmentation program showed higher correlations with the short- and long-term outcomes of patients with CRC than conventional 2D parameters.

Abdominal adipose tissue and type 2 diabetic kidney disease: adipose radiology assessment, impact, and mechanisms

Diabetic kidney disease (DKD) is a significant healthcare burden worldwide that substantially increases the risk of kidney failure and cardiovascular events. To reduce the prevalence of DKD, extensive research is being conducted to determine the risk factors and consequently implement early interventions. Patients with type 2 diabetes mellitus (T2DM) are more likely to be obese. Abdominal adiposity is associated with a greater risk of kidney damage than general obesity. Abdominal adipose tissue can be divided into different fat depots according to the location and function, including visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), perirenal adipose tissue (PAT), and renal sinus adipose tissue (RSAT), which can be accurately measured by radiology techniques, such as computed tomography (CT) and magnetic resonance imaging (MRI). Abdominal fat depots may affect the development of DKD through different mechanisms, and radiologic abdominal adipose characteristics may serve as imaging indicators of DKD risk. This review will first describe the CT/MRI-based assessment of abdominal adipose depots and subsequently describe the current studies on abdominal adipose tissue and DKD development, as well as the underlying mechanisms in patients of T2DM with DKD.

Subtype-specific Body Composition and Metabolic Risk in Patients With Primary Aldosteronism

BACKGROUND

Primary aldosteronism (PA) is associated with increased metabolic risks. However, controversy exists as to which subtype of PA has a higher metabolic risk between bilateral and lateralized PA. This study aimed to assess the body composition of 2 PA subtypes, bilateral PA and lateralized PA, according to sex and autonomous cortisol secretion (ACS) and their contribution to comorbidities.

DESIGN AND METHODS

A total of 400 patients with PA (females, n = 210) and 1:10 age- and sex-matched healthy controls (n = 4000) were enrolled. The skeletal muscle area (SMA), subcutaneous fat area, and visceral fat area (VFA) at the third lumbar spine were calculated using abdominal computed tomography-based body composition analysis.

RESULTS

Patients with bilateral PA had higher body mass index (BMI) in both sexes (all P < .05). Hemoglobin A1c level and the prevalence of diabetes were higher in female patients with bilateral PA than in those with lateralized PA (all P < .05). The VFA/BMI ratio was significantly higher in bilateral PA patients than in lateralized PA patients (5.77 ± 2.69 vs 4.56 ± 2.35 in men; 4.03 ± 2.58 vs 2.53 ± 2.05 in women, all P < .001). PA patients with ACS showed decreased SMA compared to those without ACS. Compared with healthy controls, all patients with bilateral PA and female patients with lateralized PA showed significantly higher VFA and VFA/BMI.

CONCLUSIONS

Patients with bilateral PA were more obese and had higher VFA levels than those with lateralized PA. Despite a milder form of PA, this metabolically unfavorable visceral fat distribution may lead to a higher metabolic risk in patients with bilateral PA.

Development and Evaluation of Machine Learning in Whole-Body Magnetic Resonance Imaging for Detecting Metastases in Patients With Lung or Colon Cancer: A Diagnostic Test Accuracy Study

OBJECTIVES

Whole-body magnetic resonance imaging (WB-MRI) has been demonstrated to be efficient and cost-effective for cancer staging. The study aim was to develop a machine learning (ML) algorithm to improve radiologists' sensitivity and specificity for metastasis detection and reduce reading times.

MATERIALS AND METHODS

A retrospective analysis of 438 prospectively collected WB-MRI scans from multicenter Streamline studies (February 2013-September 2016) was undertaken. Disease sites were manually labeled using Streamline reference standard. Whole-body MRI scans were randomly allocated to training and testing sets. A model for malignant lesion detection was developed based on convolutional neural networks and a 2-stage training strategy. The final algorithm generated lesion probability heat maps. Using a concurrent reader paradigm, 25 radiologists (18 experienced, 7 inexperienced in WB-/MRI) were randomly allocated WB-MRI scans with or without ML support to detect malignant lesions over 2 or 3 reading rounds. Reads were undertaken in the setting of a diagnostic radiology reading room between November 2019 and March 2020. Reading times were recorded by a scribe. Prespecified analysis included sensitivity, specificity, interobserver agreement, and reading time of radiology readers to detect metastases with or without ML support. Reader performance for detection of the primary tumor was also evaluated.

RESULTS

Four hundred thirty-three evaluable WB-MRI scans were allocated to algorithm training (245) or radiology testing (50 patients with metastases, from primary 117 colon [n = 117] or lung [n = 71] cancer). Among a total 562 reads by experienced radiologists over 2 reading rounds, per-patient specificity was 86.2% (ML) and 87.7% (non-ML) (-1.5% difference; 95% confidence interval [CI], -6.4%, 3.5%; P = 0.39). Sensitivity was 66.0% (ML) and 70.0% (non-ML) (-4.0% difference; 95% CI, -13.5%, 5.5%; P = 0.344). Among 161 reads by inexperienced readers, per-patient specificity in both groups was 76.3% (0% difference; 95% CI, -15.0%, 15.0%; P = 0.613), with sensitivity of 73.3% (ML) and 60.0% (non-ML) (13.3% difference; 95% CI, -7.9%, 34.5%; P = 0.313). Per-site specificity was high (>90%) for all metastatic sites and experience levels. There was high sensitivity for the detection of primary tumors (lung cancer detection rate of 98.6% with and without ML [0.0% difference; 95% CI, -2.0%, 2.0%; P = 1.00], colon cancer detection rate of 89.0% with and 90.6% without ML [-1.7% difference; 95% CI, -5.6%, 2.2%; P = 0.65]). When combining all reads from rounds 1 and 2, reading times fell by 6.2% (95% CI, -22.8%, 10.0%) when using ML. Round 2 read-times fell by 32% (95% CI, 20.8%, 42.8%) compared with round 1. Within round 2, there was a significant decrease in read-time when using ML support, estimated as 286 seconds (or 11%) quicker ( P = 0.0281), using regression analysis to account for reader experience, read round, and tumor type. Interobserver variance suggests moderate agreement, Cohen κ = 0.64; 95% CI, 0.47, 0.81 (with ML), and Cohen κ = 0.66; 95% CI, 0.47, 0.81 (without ML).

CONCLUSIONS

There was no evidence of a significant difference in per-patient sensitivity and specificity for detecting metastases or the primary tumor using concurrent ML compared with standard WB-MRI. Radiology read-times with or without ML support fell for round 2 reads compared with round 1, suggesting that readers familiarized themselves with the study reading method. During the second reading round, there was a significant reduction in reading time when using ML support.

Body composition analysis using convolutional neural network in predicting postoperative pancreatic fistula and survival after pancreatoduodenectomy for pancreatic cancer

PURPOSE

To evaluate whether body composition measurements acquired using convolutional neural networks (CNNs) from preoperative CT images could predict postoperative pancreatic fistula (POPF) and overall survival (OS) after pancreaticoduodenectomy in patients with pancreatic ductal adenocarcinoma (PDAC).

METHODS

257 patients (160 men; median age [interquartile range], 67 [60-74]) who underwent pancreaticoduodenectomy for PDAC between January 2013 and December 2017 were included in this retrospective study. Body composition measurements were based on a CNN trained to segment CT images into skeletal muscle area, visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT). Skeletal muscle area, VAT, and SAT were normalized to height square and labeled as skeletal muscle, VAT, and SAT indices, respectively. The independent risk factors for clinically relevant POPF (grade B or C) were determined using a multivariate logistic regression model, and prognostic factors for OS were assessed using Cox proportional hazards regression analyses.

RESULTS

After pancreatioduodenectomy, 27 patients developed POPF grade B or C (10.5 %, 27/257). The VAT index (odds ratio [OR] = 7.43, p < 0.001) was the only independent prognostic factor for POPF grade B or C. During the median follow-up period of 23 months, 205 (79.8 % [205/257]) patients died. For prediction of OS, skeletal muscle index (hazard ratio [HR] = 0.58, p = 0.018) was a significant factor, along with vascular invasion (HR = 1.85, p < 0.001) and neoadjuvant therapy (HR = 0.58, p = 0.011).

CONCLUSIONS

A high VAT index and a low skeletal muscle index can be utilized in predicting the occurrence of POPF grade B or C and poor OS, respectively.

Association Between Fat Depletion and Prognosis of Amyotrophic Lateral Sclerosis: CT-Based Body Composition Analysis

OBJECTIVE

The purpose of this study was to present the results of our investigation of the prognostic value of adipopenia and sarcopenia in patients with amyotrophic lateral sclerosis (ALS).

METHODS

Consecutive patients with ALS with abdominal computed tomography (CT) were retrospectively identified at a single tertiary hospital between January 2010 and July 2021. Deep learning-based volumetric CT body composition analysis software was used to obtain abdominal waist fat volume, fat attenuation, and skeletal muscle area at the L3 level, then normalized to the fat volume index (FVI) and skeletal muscle index (SMI). Adipopenia and sarcopenia were defined as the sex-specific lowest quartile and SMI reference values, respectively. The associations of CT-derived body composition parameters with clinical variables, such as body mass index (BMI) and creatinine, were evaluated by Pearson correlation analyses, and associations with survival were assessed using the multivariable Cox regression analysis.

RESULTS

Eighty subjects (40 men, 65.5 ± 9.4 years of age) were investigated (median interval between disease onset and CT examination = 25 months). The mean BMI at the CT examination was 20.3 ± 4.3 kg/m2 . The BMI showed a positive correlation with both FVI (R = 0.70, p < 0.001) and SMI (R = 0.63, p < 0.001), and the serum creatinine level was associated with SMI (R = 0.68, p < 0.001). After adjusting for sex, age, King's stage, BMI, creatinine, progression rate, and sarcopenia, adipopenia was associated with shorter survival (hazard ratio [HR] = 5.94, 95% confidence interval [CI] = 1.01, 35.0, p = 0.049). In a subgroup analysis for subjects with nutritional failure (stage 4a), the HR of adipopenia was 15.1 (95% CI = 2.45, 93.4, p = 0.003).

INTERPRETATION

Deep learning-based CT-derived adipopenia in patients with ALS is an independent poor prognostic factor for survival. ANN NEUROL 2023;94:1116-1125.

Prognostic utility of body composition parameters based on computed tomography analysis of advanced non-small cell lung cancer treated with immune checkpoint inhibitors

OBJECTIVE

The purpose of this study was to evaluate the prognostic impact of body composition parameters based on computed tomography (CT) in patients with non-small cell lung cancer (NSCLC) who received ICI treatment.

METHODS

This retrospective study analyzed the data from advanced NSCLC patients treated with ICI therapy between 2013 and 2019. We included patients with NSCLC who underwent baseline CT scans. The exclusion criteria included patients who received three or more lines of chemotherapy, those with insufficient clinical information, or those without treatment response evaluation.

RESULTS

A total of 136 patients were enrolled. Among the volumetric body composition parameters, patients in the highest quartiles (Q2-4) of the visceral fat index (VFI) exhibited a higher response rate to ICI therapy than those in the lowest quartile (Q1) of VFI (Q1 vs. Q2-4: 18.2% vs. 43.1%, p = 0.012). Patients with a VFI in Q2-4 had significantly prolonged progression-free survival (PFS) and overall survival (OS) (PFS, Q1 vs. Q2-4: 3.0 months vs. 6.4 months, p = 0.043; OS, Q1 vs. Q2-4: 5.6 months vs. 16.3 months, p = 0.004). Kaplan-Meier analysis based on the VFI and visceral fat Hounsfield unit (HU) revealed that patients with VFI in Q1 and HU in Q2-4 had the worst prognosis.

CONCLUSIONS

Visceral fat volume is significantly associated with treatment outcomes in ICI-treated patients with NSCLC. Moreover, fat quality may impact the treatment outcomes. This finding underscores the potential significance of both fat compartments and fat quality as prognostic indicators.

CRITICAL RELEVANCE STATEMENT

Visceral fat volume is significantly associated with treatment outcomes in ICI-treated patients with non-small cell lung cancer. Moreover, fat quality may impact the treatment outcomes. This finding underscores the potential significance of both fat compartments and fat quality as prognostic indicators.

KEY POINTS

• We found that visceral fat volume positively correlated with treatment response and survival in patients with non-small cell lung cancer receiving immune checkpoint inhibitors. • Additionally, a trend toward a negative correlation between visceral fat attenuation and survival was observed. • The findings highlight the prognostic utility of fat compartments and fat quality.

Prognostic role of computed tomography analysis using deep learning algorithm in patients with chronic hepatitis B viral infection

BACKGROUND/AIMS

The prediction of clinical outcomes in patients with chronic hepatitis B (CHB) is paramount for effective management. This study aimed to evaluate the prognostic value of computed tomography (CT) analysis using deep learning algorithms in patients with CHB.

METHODS

This retrospective study included 2,169 patients with CHB without hepatic decompensation who underwent contrast-enhanced abdominal CT for hepatocellular carcinoma (HCC) surveillance between January 2005 and June 2016. Liver and spleen volumes and body composition measurements including subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), and skeletal muscle indices were acquired from CT images using deep learning-based fully automated organ segmentation algorithms. We assessed the significant predictors of HCC, hepatic decompensation, diabetes mellitus (DM), and overall survival (OS) using Cox proportional hazard analyses.

RESULTS

During a median follow-up period of 103.0 months, HCC (n=134, 6.2%), hepatic decompensation (n=103, 4.7%), DM (n=432, 19.9%), and death (n=120, 5.5%) occurred. According to the multivariate analysis, standardized spleen volume significantly predicted HCC development (hazard ratio [HR]=1.01, P=0.025), along with age, sex, albumin and platelet count. Standardized spleen volume (HR=1.01, P<0.001) and VAT index (HR=0.98, P=0.004) were significantly associated with hepatic decompensation along with age and albumin. Furthermore, VAT index (HR=1.01, P=0.001) and standardized spleen volume (HR=1.01, P=0.001) were significant predictors for DM, along with sex, age, and albumin. SAT index (HR=0.99, P=0.004) was significantly associated with OS, along with age, albumin, and MELD.

CONCLUSION

Deep learning-based automatically measured spleen volume, VAT, and SAT indices may provide various prognostic information in patients with CHB.

A systematic review of automated segmentation of 3D computed-tomography scans for volumetric body composition analysis

Automated computed tomography (CT) scan segmentation (labelling of pixels according to tissue type) is now possible. This technique is being adapted to achieve three-dimensional (3D) segmentation of CT scans, opposed to single L3-slice alone. This systematic review evaluates feasibility and accuracy of automated segmentation of 3D CT scans for volumetric body composition (BC) analysis, as well as current limitations and pitfalls clinicians and researchers should be aware of. OVID Medline, Embase and grey literature databases up to October 2021 were searched. Original studies investigating automated skeletal muscle, visceral and subcutaneous AT segmentation from CT were included. Seven of the 92 studies met inclusion criteria. Variation existed in expertise and numbers of humans performing ground-truth segmentations used to train algorithms. There was heterogeneity in patient characteristics, pathology and CT phases that segmentation algorithms were developed upon. Reporting of anatomical CT coverage varied, with confusing terminology. Six studies covered volumetric regional slabs rather than the whole body. One study stated the use of whole-body CT, but it was not clear whether this truly meant head-to-fingertip-to-toe. Two studies used conventional computer algorithms. The latter five used deep learning (DL), an artificial intelligence technique where algorithms are similarly organized to brain neuronal pathways. Six of seven reported excellent segmentation performance (Dice similarity coefficients > 0.9 per tissue). Internal testing on unseen scans was performed for only four of seven algorithms, whilst only three were tested externally. Trained DL algorithms achieved full CT segmentation in 12 to 75 s versus 25 min for non-DL techniques. DL enables opportunistic, rapid and automated volumetric BC analysis of CT performed for clinical indications. However, most CT scans do not cover head-to-fingertip-to-toe; further research must validate using common CT regions to estimate true whole-body BC, with direct comparison to single lumbar slice. Due to successes of DL, we expect progressive numbers of algorithms to materialize in addition to the seven discussed in this paper. Researchers and clinicians in the field of BC must therefore be aware of pitfalls. High Dice similarity coefficients do not inform the degree to which BC tissues may be under- or overestimated and nor does it inform on algorithm precision. Consensus is needed to define accuracy and precision standards for ground-truth labelling. Creation of a large international, multicentre common CT dataset with BC ground-truth labels from multiple experts could be a robust solution.

Pectoralis muscle predicts distant metastases in breast cancer by deep learning radiomics

BACKGROUND

Sarcopenia is associated with a poor prognosis in patients with breast cancer (BC). Currently, there are few quantitative assessments carried out between muscle biomarkers and distant metastasis using existing methods.

PURPOSE

To assess the predictive value of the pectoralis muscle for BC distant metastasis, we developed a deep learning radiomics nomogram model (DLR-N) in this study.

MATERIAL AND METHODS

A total of 493 patients with pathologically confirmed BC were registered. Image features were extracted from computed tomography (CT) images for each patient. Univariate and multivariate Cox regression analyses were performed to determine the independent prognostic factors for distant metastasis. The DLR-N was built based on independent prognostic factors and CT images to predict distant metastases. The model was assessed in terms of overall performance, discrimination, calibration, and clinical value. Finally, the predictive performance of the model was validated using the testing cohort.

RESULTS

The developed DLR-N combined multiple radiomic features and clinicopathological factors and demonstrated excellent predictive performance. The C-index of the training cohort was 0.983 (95% confidence interval [CI] = 0.969-0.998) and the C-index of the testing cohort was 0.948 (95% CI = 0.917-0.979). Decision curve analysis (DCA) showed that patients could benefit more from incorporating multimodal radiomic features into clinicopathological models.

CONCLUSIONS

DLR-N verified that there were biomarkers at the level of the fourth thoracic vertebra (T4) that affected distant metastasis. Multimodal prediction models based on deep learning could be a potential method to aid in the prediction of distant metastases in patients with BC.

Prognostic Value of Artificial Intelligence-Driven, Computed Tomography-Based, Volumetric Assessment of the Volume and Density of Muscle in Patients With Colon Cancer

OBJECTIVE

The prognostic value of the volume and density of skeletal muscles in the abdominal waist of patients with colon cancer remains unclear. This study aimed to investigate the association between the automated computed tomography (CT)-based volume and density of the muscle in the abdominal waist and survival outcomes in patients with colon cancer.

MATERIALS AND METHODS

We retrospectively evaluated 474 patients with colon cancer who underwent surgery with curative intent between January 2010 and October 2017. Volumetric skeletal muscle index and muscular density were measured at the abdominal waist using artificial intelligence (AI)-based volumetric segmentation of body composition on preoperative pre-contrast CT images. Patients were grouped based on their skeletal muscle index (sarcopenia vs. not) and muscular density (myosteatosis vs. not) values and combinations (normal, sarcopenia alone, myosteatosis alone, and combined sarcopenia and myosteatosis). Postsurgical disease-free survival (DFS) and overall survival (OS) were analyzed using univariable and multivariable analyses, including multivariable Cox proportional hazard regression.

RESULTS

Univariable analysis showed that DFS and OS were significantly worse for the sarcopenia group than for the non-sarcopenia group (P = 0.044 and P = 0.003, respectively, by log-rank test) and for the myosteatosis group than for the non-myosteatosis group (P < 0.001 by log-rank test for all). In the multivariable analysis, the myosteatotic muscle type was associated with worse DFS (adjusted hazard ratio [aHR], 1.89 [95% confidence interval, 1.25-2.86]; P = 0.003) and OS (aHR, 1.90 [95% confidence interval, 1.84-3.04]; P = 0.008) than the normal muscle type. The combined muscle type showed worse OS than the normal muscle type (aHR, 1.95 [95% confidence interval, 1.08-3.54]; P = 0.027).

CONCLUSION

Preoperative volumetric sarcopenia and myosteatosis, automatically assessed from pre-contrast CT scans using AI-based software, adversely affect survival outcomes in patients with colon cancer.

Deep learning of image-derived measures of body composition in pediatric, adolescent, and young adult lymphoma: association with late treatment effects

OBJECTIVES

The objective of this study was to translate a deep learning (DL) approach for semiautomated analysis of body composition (BC) measures from standard of care CT images to investigate the prognostic value of BC in pediatric, adolescent, and young adult (AYA) patients with lymphoma.

METHODS

This 10-year retrospective, single-site study of 110 pediatric and AYA patients with lymphoma involved manual segmentation of fat and muscle tissue from 260 CT imaging datasets obtained as part of routine imaging at initial staging and first therapeutic follow-up. A DL model was trained to perform semiautomated image segmentation of adipose and muscle tissue. The association between BC measures and the occurrence of 3-year late effects was evaluated using Cox proportional hazards regression analyses.

RESULTS

DL-guided measures of BC were in close agreement with those obtained by a human rater, as demonstrated by high Dice scores (≥ 0.95) and correlations (r > 0.99) for each tissue of interest. Cox proportional hazards regression analyses revealed that patients with elevated subcutaneous adipose tissue at baseline and first follow-up, along with patients who possessed lower volumes of skeletal muscle at first follow-up, have increased risk of late effects compared to their peers.

CONCLUSIONS

DL provides rapid and accurate quantification of image-derived measures of BC that are associated with risk for treatment-related late effects in pediatric and AYA patients with lymphoma. Image-based monitoring of BC measures may enhance future opportunities for personalized medicine for children with lymphoma by identifying patients at the highest risk for late effects of treatment.

KEY POINTS

• Deep learning-guided CT image analysis of body composition measures achieved high agreement level with manual image analysis. • Pediatric patients with more fat and less muscle during the course of cancer treatment were more likely to experience a serious adverse event compared to their clinical counterparts. • Deep learning of body composition may add value to routine CT imaging by offering real-time monitoring of pediatric, adolescent, and young adults at high risk for late effects of cancer treatment.

Synergic prognostic value of 3D CT scan subcutaneous fat and muscle masses for immunotherapy-treated cancer

BACKGROUND

Our aim was to explore the prognostic value of anthropometric parameters in a large population of patients treated with immunotherapy.

METHODS

We retrospectively included 623 patients with advanced non-small cell lung cancer (NSCLC) (n=318) or melanoma (n=305) treated by an immune-checkpoint-inhibitor having a pretreatment (thorax-)abdomen-pelvis CT scan. An external validation cohort of 55 patients with NSCLC was used. Anthropometric parameters were measured three-dimensionally (3D) by a deep learning software (Anthropometer3DNet) allowing an automatic multislice measurement of lean body mass, fat body mass (FBM), muscle body mass (MBM), visceral fat mass (VFM) and sub-cutaneous fat mass (SFM). Body mass index (BMI) and weight loss (WL) were also retrieved. Receiver operator characteristic (ROC) curve analysis was performed and overall survival was calculated using Kaplan-Meier (KM) curve and Cox regression analysis.

RESULTS

In the overall cohort, 1-year mortality rate was 0.496 (95% CI: 0.457 to 0.537) for 309 events and 5-year mortality rate was 0.196 (95% CI: 0.165 to 0.233) for 477 events. In the univariate Kaplan-Meier analysis, prognosis was worse (p<0.001) for patients with low SFM (<3.95 kg/m2), low FBM (<3.26 kg/m2), low VFM (<0.91 kg/m2), low MBM (<5.85 kg/m2) and low BMI (<24.97 kg/m2). The same parameters were significant in the Cox univariate analysis (p<0.001) and, in the multivariate stepwise Cox analysis, the significant parameters were MBM (p<0.0001), SFM (0.013) and WL (0.0003). In subanalyses according to the type of cancer, all body composition parameters were statistically significant for NSCLC in ROC, KM and Cox univariate analysis while, for melanoma, none of them, except MBM, was statistically significant. In multivariate Cox analysis, the significant parameters for NSCLC were MBM (HR=0.81, p=0.0002), SFM (HR=0.94, p=0.02) and WL (HR=1.06, p=0.004). For NSCLC, a KM analysis combining SFM and MBM was able to separate the population in three categories with the worse prognostic for the patients with both low SFM (<5.22 kg/m2) and MBM (<6.86 kg/m2) (p<0001). On the external validation cohort, combination of low SFM and low MBM was pejorative with 63% of mortality at 1 year versus 25% (p=0.0029).

CONCLUSIONS

3D measured low SFM and MBM are significant prognosis factors of NSCLC treated by immune checkpoint inhibitors and can be combined to improve the prognostic value.

Newly developed sarcopenia after liver transplantation, determined by a fully automated 3D muscle volume estimation on abdominal CT, can predict post-transplant diabetes mellitus and poor survival outcomes

BACKGROUND

Loss of muscle mass is the most common complication of end-stage liver disease and negatively affects outcomes for liver transplantation (LT) recipients. We aimed to determine the prognostic value of a fully automated three-dimensional (3D) muscle volume estimation using deep learning algorithms on abdominal CT in patients who underwent liver transplantation (LT).

METHODS

This retrospective study included 107 patients who underwent LT from 2014 to 2015. Serial CT scans, including pre-LT and 1- and 2-year follow-ups were performed. From the CT scans, deep learning-based automated body composition segmentation software was used to calculate muscle volumes in 3D. Sarcopenia was calculated by dividing average skeletal muscle area by height squared. Newly developed-(ND) sarcopenia was defined as the onset of sarcopenia 1 or 2 years after LT in patients without a history of sarcopenia before LT. Patients' clinical characteristics, including post-transplant diabetes mellitus (PTDM) and Model for end-stage liver disease score, were compared according to the presence or absence of sarcopenia after LT. A subgroup analysis was performed in the post-LT sarcopenic group. The Kaplan-Meier method was used for overall survival (OS).

RESULTS

Patients with ND-sarcopenia had poorer OS than those who did not (P = 0.04, hazard ratio [HR], 3.34; 95% confidence interval [CI] 1.05 - 10.7). In the subgroup analysis for post-LT sarcopenia (n = 94), 34 patients (36.2%) had ND-sarcopenia. Patients with ND-sarcopenia had significantly worse OS (P = 0.002, HR 7.12; 95% CI 2.00 - 25.32) and higher PTDM occurrence rates (P = 0.02, HR 4.93; 95% CI 1.18 - 20.54) than those with sarcopenia prior to LT.

CONCLUSION

ND-sarcopenia determined by muscle volume on abdominal CT can predict poor survival outcomes and the occurrence of PTDM for LT recipients.

Fat Hounsfield Unit Reference Interval Derived through an Indirect Method

BACKGROUND

In vivo Hounsfield Unit (HU) values have traditionally been determined using direct CT image measurements. These measurements are dependent on the window/level used to examine the CT image and the individual conducting the fat tissue tracing.

METHODS

Using an indirect method, a new reference interval (RI) is proposed. A total of 4000 samples of fat tissues were collected from routine abdominal CT examinations. A linear regression equation was then calculated using the linear part of the cumulative frequency plot of their average values.

RESULTS

The regression function for total abdominal fat was determined to be y = 35.376*x - 123.48, and a 95% confidence RI of -123 to -89 was computed. A significant difference of 3.82 was observed between the average fat HU values of visceral and subcutaneous areas.

CONCLUSIONS

Using statistical methods and the in vivo measurements of patient data, a series of RIs were determined for fat HU that is consistent with theoretical values.

CT analysis of thoracolumbar body composition for estimating whole-body composition

BACKGROUND

To evaluate the correlation between single- and multi-slice cross-sectional thoracolumbar and whole-body compositions.

METHODS

We retrospectively included patients who underwent whole-body PET-CT scans from January 2016 to December 2019 at multiple institutions. A priori-developed, deep learning-based commercially available 3D U-Net segmentation provided whole-body 3D reference volumes and 2D areas of muscle, visceral fat, and subcutaneous fat at the upper, middle, and lower endplate of the individual T1-L5 vertebrae. In the derivation set, we analyzed the Pearson correlation coefficients of single-slice and multi-slice averaged 2D areas (waist and T12-L1) with the reference values. We then built prediction models using the top three correlated levels and tested the models in the validation set.

RESULTS

The derivation and validation datasets included 203 (mean age 58.2 years; 101 men) and 239 patients (mean age 57.8 years; 80 men). The coefficients were distributed bimodally, with the first peak at T4 (coefficient, 0.78) and the second peak at L2-3 (coefficient 0.90). The top three correlations in the abdominal scan range were found for multi-slice waist averaging (0.92) and single-slice L3 and L2 (0.90, each), while those in the chest scan range were multi-slice T12-L1 averaging (0.89), single-slice L1 (0.89), and T12 (0.86). The model performance at the top three levels for estimating whole-body composition was similar in the derivation and validation datasets.

CONCLUSIONS

Single-slice L2-3 (abdominal CT range) and L1 (chest CT range) analysis best correlated with whole-body composition around 0.90 (coefficient). Multi-slice waist averaging provided a slightly higher correlation of 0.92.

Deep Learning for Automatic Bone Marrow Apparent Diffusion Coefficient Measurements From Whole-Body Magnetic Resonance Imaging in Patients With Multiple Myeloma: A Retrospective Multicenter Study

OBJECTIVES

Diffusion-weighted magnetic resonance imaging (MRI) is increasingly important in patients with multiple myeloma (MM). The objective of this study was to train and test an algorithm for automatic pelvic bone marrow analysis from whole-body apparent diffusion coefficient (ADC) maps in patients with MM, which automatically segments pelvic bones and subsequently extracts objective, representative ADC measurements from each bone.

MATERIALS AND METHODS

In this retrospective multicentric study, 180 MRIs from 54 patients were annotated (semi)manually and used to train an nnU-Net for automatic, individual segmentation of the right hip bone, the left hip bone, and the sacral bone. The quality of the automatic segmentation was evaluated on 15 manually segmented whole-body MRIs from 3 centers using the dice score. In 3 independent test sets from 3 centers, which comprised a total of 312 whole-body MRIs, agreement between automatically extracted mean ADC values from the nnU-Net segmentation and manual ADC measurements from 2 independent radiologists was evaluated. Bland-Altman plots were constructed, and absolute bias, relative bias to mean, limits of agreement, and coefficients of variation were calculated. In 56 patients with newly diagnosed MM who had undergone bone marrow biopsy, ADC measurements were correlated with biopsy results using Spearman correlation.

RESULTS

The ADC-nnU-Net achieved automatic segmentations with mean dice scores of 0.92, 0.93, and 0.85 for the right pelvis, the left pelvis, and the sacral bone, whereas the interrater experiment gave mean dice scores of 0.86, 0.86, and 0.77, respectively. The agreement between radiologists' manual ADC measurements and automatic ADC measurements was as follows: the bias between the first reader and the automatic approach was 49 × 10 -6 mm 2 /s, 7 × 10 -6 mm 2 /s, and -58 × 10 -6 mm 2 /s, and the bias between the second reader and the automatic approach was 12 × 10 -6 mm 2 /s, 2 × 10 -6 mm 2 /s, and -66 × 10 -6 mm 2 /s for the right pelvis, the left pelvis, and the sacral bone, respectively. The bias between reader 1 and reader 2 was 40 × 10 -6 mm 2 /s, 8 × 10 -6 mm 2 /s, and 7 × 10 -6 mm 2 /s, and the mean absolute difference between manual readers was 84 × 10 -6 mm 2 /s, 65 × 10 -6 mm 2 /s, and 75 × 10 -6 mm 2 /s. Automatically extracted ADC values significantly correlated with bone marrow plasma cell infiltration ( R = 0.36, P = 0.007).

CONCLUSIONS

In this study, a nnU-Net was trained that can automatically segment pelvic bone marrow from whole-body ADC maps in multicentric data sets with a quality comparable to manual segmentations. This approach allows automatic, objective bone marrow ADC measurements, which agree well with manual ADC measurements and can help to overcome interrater variability or nonrepresentative measurements. Automatically extracted ADC values significantly correlate with bone marrow plasma cell infiltration and might be of value for automatic staging, risk stratification, or therapy response assessment.

Deep Learning-Based Medical Images Segmentation of Musculoskeletal Anatomical Structures: A Survey of Bottlenecks and Strategies

By leveraging the recent development of artificial intelligence algorithms, several medical sectors have benefited from using automatic segmentation tools from bioimaging to segment anatomical structures. Segmentation of the musculoskeletal system is key for studying alterations in anatomical tissue and supporting medical interventions. The clinical use of such tools requires an understanding of the proper method for interpreting data and evaluating their performance. The current systematic review aims to present the common bottlenecks for musculoskeletal structures analysis (e.g., small sample size, data inhomogeneity) and the related strategies utilized by different authors. A search was performed using the PUBMED database with the following keywords: deep learning, musculoskeletal system, segmentation. A total of 140 articles published up until February 2022 were obtained and analyzed according to the PRISMA framework in terms of anatomical structures, bioimaging techniques, pre/post-processing operations, training/validation/testing subset creation, network architecture, loss functions, performance indicators and so on. Several common trends emerged from this survey; however, the different methods need to be compared and discussed based on each specific case study (anatomical region, medical imaging acquisition setting, study population, etc.). These findings can be used to guide clinicians (as end users) to better understand the potential benefits and limitations of these tools.