A study on whether deep learning models based on CT images for bone density classification and prediction can be used for opportunistic osteoporosis screening

The science and practice of imaging-based screening: What the radiologist needs to know

Imaging-based screening is an important public health focus and a fundamental part of Diagnostic Radiology. Hence, radiologists should be familiar with the concepts that drive imaging-based screening practice including goals, risks, biases and clinical trials. This review article discusses an array of imaging-based screening exams including the key epidemiology and evidence that drive screening guidelines for abdominal aortic aneurysm, breast cancer, carotid artery disease, colorectal cancer, coronary artery disease, lung cancer, osteoporosis, and thyroid cancer. We will provide an overview on societal interests in screening, screening-related inequities, and opportunities to address them. Emerging evidence for opportunistic screening and the role of AI in imaging-based screening will be explored. In-depth knowledge and formalized training in imaging-based screening strengthens radiologists as clinician scientists and has the potential to broaden our public health leadership opportunities. SUMMARY SENTENCE: An overview of key screening concepts, the evidence that drives today's imaging-based screening practices, and the need for radiologist leadership in screening policies and evidence development.

Effect of fully automatic classification model from different tube voltage images on bone density screening: A self-controlled study

PURPOSE

To develop two bone status prediction models combining deep learning and radiomics based on standard-dose chest computed tomography (SDCT) and low-dose chest computed tomography (LDCT), and to evaluate the effect of tube voltage on reproducibility of radiomics features and predictive efficacy of these models.

METHODS

A total of 1508 patients were enrolled in this retrospective study. LDCT was conducted using 80 kVp, tube current ranging from 100 to 475 mA. On the other hand, SDCT was performed using 120 kVp, tube current ranging from 100 to 520 mA. We developed an automatic thoracic vertebral cancellous bone (TVCB) segmentation model. Subsequently, 1184 features were extracted and two classifiers were developed based on LDCT and SDCT images. Based on the diagnostic results of quantitative computed tomography examination, the first-level classifier was initially developed to distinguish normal or abnormal BMD (including osteoporosis and osteopenia), while the second-level classifier was employed to identify osteoporosis or osteopenia. The Dice coefficient was used to evaluate the performance of the automated segmentation model. The Concordance Correlation Coefficients (CCC) of radiomics features were calculated between LDCT and SDCT, and the performance of these models was evaluated.

RESULTS

Our automated segmentation model achieved a Dice coefficient of 0.98 ± 0.01 and 0.97 ± 0.02 in LDCT and SDCT, respectively. Alterations in tube voltage decreased the reproducibility of the extracted radiomic features, with 85.05 % of the radiomic features exhibiting low reproducibility (CCC < 0.75). The area under the curve (AUC) using LDCT-based and SDCT-based models was 0.97 ± 0.01 and 0.94 ± 0.02, respectively. Nonetheless, cross-validation with independent test sets of different tube voltage scans suggests that variations in tube voltage can impair the diagnostic efficacy of the model. Consequently, radiomics models are not universally applicable to images of varying tube voltages. In clinical settings, ensuring consistency between the tube voltage of the image used for model development and that of the acquired patient image is critical.

CONCLUSIONS

Automatic bone status prediction models, utilizing either LDCT or SDCT images, enable accurate assessment of bone status. Tube voltage impacts reproducibility of features and predictive efficacy of models. It is necessary to account for tube voltage variation during the image acquisition.

Balancing Performance and Interpretability in Medical Image Analysis: Case study of Osteopenia

Multiple studies within the medical field have highlighted the remarkable effectiveness of using convolutional neural networks for predicting medical conditions, sometimes even surpassing that of medical professionals. Despite their great performance, convolutional neural networks operate as black boxes, potentially arriving at correct conclusions for incorrect reasons or areas of focus. Our work explores the possibility of mitigating this phenomenon by identifying and occluding confounding variables within images. Specifically, we focused on the prediction of osteopenia, a serious medical condition, using the publicly available GRAZPEDWRI-DX dataset. After detection of the confounding variables in the dataset, we generated masks that occlude regions of images associated with those variables. By doing so, models were forced to focus on different parts of the images for classification. Model evaluation using F1-score, precision, and recall showed that models trained on non-occluded images typically outperformed models trained on occluded images. However, a test where radiologists had to choose a model based on the focused regions extracted by the GRAD-CAM method showcased different outcomes. The radiologists' preference shifted towards models trained on the occluded images. These results suggest that while occluding confounding variables may degrade model performance, it enhances interpretability, providing more reliable insights into the reasoning behind predictions. The code to repeat our experiment is available on the following link: https://github.com/mikulicmateo/osteopenia .

Artificial intelligence-enhanced opportunistic screening of osteoporosis in CT scan: a scoping Review

PURPOSE

This scoping review aimed to assess the current research on artificial intelligence (AI)--enhanced opportunistic screening approaches for stratifying osteoporosis and osteopenia risk by evaluating vertebral trabecular bone structure in CT scans.

METHODS

PubMed, Scopus, and Web of Science databases were systematically searched for studies published between 2018 and December 2023. Inclusion criteria encompassed articles focusing on AI techniques for classifying osteoporosis/osteopenia or determining bone mineral density using CT scans of vertebral bodies. Data extraction included study characteristics, methodologies, and key findings.

RESULTS

Fourteen studies met the inclusion criteria. Three main approaches were identified: fully automated deep learning solutions, hybrid approaches combining deep learning and conventional machine learning, and non-automated solutions using manual segmentation followed by AI analysis. Studies demonstrated high accuracy in bone mineral density prediction (86-96%) and classification of normal versus osteoporotic subjects (AUC 0.927-0.984). However, significant heterogeneity was observed in methodologies, workflows, and ground truth selection.

CONCLUSIONS

The review highlights AI's promising potential in enhancing opportunistic screening for osteoporosis using CT scans. While the field is still in its early stages, with most solutions at the proof-of-concept phase, the evidence supports increased efforts to incorporate AI into radiologic workflows. Addressing knowledge gaps, such as standardizing benchmarks and increasing external validation, will be crucial for advancing the clinical application of these AI-enhanced screening methods. Integration of such technologies could lead to improved early detection of osteoporotic conditions at a low economic cost.

Femoral osteoporosis prediction model using autosegmentation and machine learning analysis with PyRadiomics on abdomen-pelvic computed tomography (CT)

Background

With the advancement of artificial intelligence technology and radiomics analysis, opportunistic prediction of osteoporosis with computed tomography (CT) is a new paradigm in osteoporosis screening. This study aimed to assess the diagnostic performance of osteoporosis prediction by the combination of autosegmentation of the proximal femur and machine learning analysis with a reference standard of dual-energy X-ray absorptiometry (DXA).

Methods

Abdomen-pelvic CT scans were retrospectively analyzed from 1,122 patients who received both DXA and abdomen-pelvic computed tomography (APCT) scan from January 2018 to December 2020. The study cohort consisted of a training cohort and a temporal validation cohort. The left proximal femur was automatically segmented, and a prediction model was built by machine-learning analysis using a random forest (RF) analysis and 854 PyRadiomics features. The technical success rate of autosegmentation, diagnostic test, area under the receiver operator characteristics curve (AUC), and precision recall curve (AUC-PR) analysis were used to analyze the training and validation cohorts.

Results

The osteoporosis prevalence of the training and validation cohorts was 24.5%, and 10.3%, respectively. The technical success rate of autosegmentation of the proximal femur was 99.7%. In the diagnostic test, the training and validation cohorts showed 78.4% vs. 63.3% sensitivity, 89.4% vs. 98.1% specificity. The prediction performance to identify osteoporosis within the groups used for training and validation cohort was high and the AUC and AUC-PR to forecast the occurrence of osteoporosis within the training and validation cohorts were 90.8% [95% confidence interval (CI), 88.4-93.2%] vs. 78.0% (95% CI, 76.0-79.9%) and 94.6% (95% CI, 89.3-99.8%) vs. 88.8% (95% CI, 86.2-91.5%), respectively.

Conclusions

The osteoporosis prediction model using autosegmentation of proximal femur and machine-learning analysis with PyRadiomics features on APCT showed excellent diagnostic feasibility and technical success.

Fully automated deep learning system for osteoporosis screening using chest computed tomography images

Background

Osteoporosis, a disease stemming from bone metabolism irregularities, affects approximately 200 million people worldwide. Timely detection of osteoporosis is pivotal in grappling with this public health challenge. Deep learning (DL), emerging as a promising methodology in the field of medical imaging, holds considerable potential for the assessment of bone mineral density (BMD). This study aimed to propose an automated DL framework for BMD assessment that integrates localization, segmentation, and ternary classification using various dominant convolutional neural networks (CNNs).

Methods

In this retrospective study, a cohort of 2,274 patients underwent chest computed tomography (CT) was enrolled from January 2022 to June 2023 for the development of the integrated DL system. The study unfolded in 2 phases. Initially, 1,025 patients were selected based on specific criteria to develop an automated segmentation model, utilizing 2 VB-Net networks. Subsequently, a distinct cohort of 902 patients was employed for the development and testing of classification models for BMD assessment. Then, 3 distinct DL network architectures, specifically DenseNet, ResNet-18, and ResNet-50, were applied to formulate the 3-classification BMD assessment model. The performance of both phases was evaluated using an independent test set consisting of 347 individuals. Segmentation performance was evaluated using the Dice similarity coefficient; classification performance was appraised using the receiver operating characteristic (ROC) curve. Furthermore, metrics such as the area under the curve (AUC), accuracy, and precision were meticulously calculated.

Results

In the first stage, the automatic segmentation model demonstrated excellent segmentation performance, with mean Dice surpassing 0.93 in the independent test set. In the second stage, both the DenseNet and ResNet-18 demonstrated excellent diagnostic performance in detecting bone status. For osteoporosis, and osteopenia, the AUCs were as follows: DenseNet achieved 0.94 [95% confidence interval (CI): 0.91-0.97], and 0.91 (95% CI: 0.87-0.94), respectively; ResNet-18 attained 0.96 (95% CI: 0.92-0.98), and 0.91 (95% CI: 0.87-0.94), respectively. However, the ResNet-50 model exhibited suboptimal diagnostic performance for osteopenia, with an AUC value of only 0.76 (95% CI: 0.69-0.80). Alterations in tube voltage had a more pronounced impact on the performance of the DenseNet. In the independent test set with tube voltage at 100 kVp images, the accuracy and precision of DenseNet decreased on average by approximately 14.29% and 18.82%, respectively, whereas the accuracy and precision of ResNet-18 decreased by about 8.33% and 7.14%, respectively.

Conclusions

The state-of-the-art DL framework model offers an effective and efficient approach for opportunistic osteoporosis screening using chest CT, without incurring additional costs or radiation exposure.

Traumatic rib fracture patterns associated with bone mineral density statuses derived from CT images

Background

The impact of decreased bone mineral density (BMD) on traumatic rib fractures remains unknown. We combined computed tomography (CT) and artificial intelligence (AI) to measure BMD and explore its impact on traumatic rib fractures and their patterns.

Methods

The retrospective cohort comprised patients who visited our hospital from 2017-2018; the prospective cohort (control group) was consecutively recruited from the same hospital from February-June 2023. All patients had blunt chest trauma and underwent CT. Volumetric BMD of L1 vertebra was measured by using an AI software. Analyses were done by using BMD categorized as osteoporosis (<80 mg/cm3), osteopenia (80-120 mg/cm3), or normal (>120 mg/cm3). Pearson's χ2, Fisher's exact, or Kruskal-Wallis tests and Bonferroni correction were used for comparisons. Negative binomial, and logistic regression analyses were used to assess the associations and impacts of BMD status. Sensitivity analyses were also performed.

Findings

The retrospective cohort included 2,076 eligible patients, of whom 954 (46%) had normal BMD, 806 (38.8%) had osteopenia, and 316 (15.2%) had osteoporosis. After sex- and age-adjustment, osteoporosis was significantly associated with higher rib fracture rates, and a higher likelihood of fractures in ribs 4-7. Furthermore, both the osteopenia and osteoporosis groups demonstrated a significantly higher number of fractured ribs and fracture sites on ribs, with a higher likelihood of fractures in ribs 1-3, as well as flail chest. The prospective cohort included 205 eligible patients, of whom 92 (44.9%) had normal BMD, 74 (36.1%) had osteopenia, and 39 (19.0%) had osteoporosis. The findings observed within this cohort were in concurrence with those in the retrospective cohort.

Interpretation

Traumatic rib fractures are associated with decreased BMD. CT-AI can help to identify individuals who have decreased BMD and a greater rib fracture rate, along with their fracture patterns.