Making Radiomics More Reproducible across Scanner and Imaging Protocol Variations: A Review of Harmonization Methods

Mastering CT-based radiomic research in lung cancer: a practical guide from study design to critical appraisal

Radiomics is a health technology that has the potential to extract clinically meaningful biomarkers from standard of care imaging. Despite a wealth of exploratory analysis performed on scans acquired from patients with lung cancer and existing guidelines describing some of the key steps, no radiomic-based biomarker has been widely accepted. This is primarily due to limitations with methodology, data analysis, and interpretation of the available studies. There is currently a lack of guidance relating to the entire radiomic workflow from study design to critical appraisal. This guide, written with early career lung cancer researchers, describes a more complete radiomic workflow. Lung cancer image analysis is the focus due to some of the unique challenges encountered such as patient movement from breathing. The guide will focus on CT imaging as these are the most common scans performed on patients with lung cancer. The aim of this article is to support the production of high-quality research that has the potential to positively impact outcome of patients with lung cancer.

Ensemble learning approach for prediction of early complications after radiotherapy for head and neck cancer using CT and MRI radiomic features

There are different side effects in radiotherapy of head and neck cancer (HNC) including xerostomia. The present study utilizes the addition of [Formula: see text] and [Formula: see text]-weighted Magnetic Resonance (MR) radiomic image features to typical Computed Tomography (CT)-based features and radiation dose-based characteristics and incorporates the evaluation and validation of individual and ensemble classifiers for prediction of early-onset xerostomia in radiotherapy of HNC. A total of 80 patients diagnosed with HNC were evaluated prospectively. The dataset was divided into two subsets: 70% for training and validation and 30% for testing. Stratified random sampling was used to ensure that the proportion of xerostomia cases was consistent across both subsets. This approach preserved the class balance and ensured a representative distribution of demographic, dosimetric, and radiomic features in both sets. MR and CT imaging, dosimetric, and demographic features of patients were used as model input data. Bilateral parotid radiomic features were extracted from CT, [Formula: see text], and [Formula: see text] weighted MR images. Pearson statistical tests were used for selection of features and Random Tree (RT), Neural Network (NN), Linear Support Vector Machine (LSVM) and Bayesian Network (BN) classifiers were evaluated. To prevent overfitting and data leakage, preprocessing was conducted. All features were normalized using the z-score technique, with the mean and standard deviation calculated from the training set and then applied to the test set. For the training dataset, the Synthetic Minority Oversampling Technique (SMOTE) was used to balance the minority class. The results suggest the extracted features from [Formula: see text] weighted images have superior prediction ability compared to [Formula: see text] weighted acquisitions. The RT and BN models based on [Formula: see text] weighted images show better performance than those obtained with [Formula: see text] weighted images. [Formula: see text] weighted image-based analysis shows area under the curve (AUC) values for The RT and BN models of 0.90 and 0.84, respectively, while corresponding values obtained from [Formula: see text] weighted images are 0.79 and 0.78 for RT and BN models respectively. Combined [Formula: see text] weighted image-based models RT-BN, RT-LSVM-BN and RT-NN-LSVM-BN also show good performance having AUC values 0.97, 0.92, and 0.90, respectively. These results show that radiomic features from MR images obtained before radiotherapy can be used in addition to other metrics as personalized and unique biomarkers for prediction of early-onset xerostomia. Ensemble classifiers are more efficient than individual classifiers in prediction of early xerostomia.

The influence of image selection and segmentation on the extraction of lung cancer imaging radiomics features using 3D-Slicer software

PURPOSE

Extracting image features can predict the prognosis and treatment effect of non-small cell lung cancer, which has been increasingly confirmed. However, the specific operation using 3D-Slicer still lacks standardization. For example, image segmentation is manually performed based on the lung window or automatically performed through the mediastinal window. The images used for feature extraction are either enhanced or plain scanned. It is questionable whether these influencing factors will affect the extraction results and which results will be affected. This article intends to preliminarily explore the above issues.

METHODS

This article downloaded images of 22 patients with lung cancer from The Cancer Imaging Archive (TCIA), including 11 cases of adenocarcinoma and 11 cases of squamous cell carcinoma. Perform tumor image segmentation on the lung window and mediastinal window of the plain scan image, and the lung window and mediastinal window of the enhanced image. Manual drawing is used on the lung window, and automatic drawing is used on the mediastinal window and make manual modifications. Extracting radiomics features using Python radiomics. Firstly, analyze the image features of the original sequence and perform the Shapiro test. If it follows a normal distribution, perform an analysis of variance. If it does not follow a normal distribution, perform the Friedman test. Compare the significantly different image features pairwise. Then, a preliminary analysis was conducted on the differences between squamous cell carcinoma and adenocarcinoma in each group.

RESULTS

A total of 88 sets of imaging features were extracted, with 107 features in each group. Among them, 33 features showed significant differences. Continuing with pairwise repeated testing, it was found that there were 2 significant differences between enhanced and plain lung windows. There were 12 significant differences between enhanced lung windows and plain mediastinal windows. There is one significant difference between plain scanning and enhancement mediastinal window. There are 14 significant differences between the plain lung window and the enhanced mediastinal window groups. There are 14 significant differences between the lung window and the mediastinal window in the plain scan. There are 13 significant differences between the enhanced lung window and the mediastinal window. According to pathological grouping testing, it was found that there 54 significant differences between squamous cell carcinoma and adenocarcinoma.

CONCLUSION

The enhancement of lung CT has a relatively small impact on extracting image features, while selecting lung or mediastinal windows for image segmentation has a significant impact on extracting image features. Therefore, choosing lung or mediastinal windows for feature extraction should be carefully considered, as the size of the image segmentation range has a significant impact on image features. The impact of lung squamous cell carcinoma and adenocarcinoma on imaging features is also significant, indicating a high possibility of distinguishing between squamous cell carcinoma and adenocarcinoma based on radiomics (Liu C, He Y, Luo J, The Influence of Image Selection and Segmentation on the Extraction of Lung Cancer Imaging Radiomics Features Using 3D-Slicer Software, 2024).

Radiomic Profiling of Orthotopic Mouse Models of Glioblastoma Reveals Histopathological Correlations Associated with Tumour Response to Ionising Radiation

BACKGROUND

Glioblastoma (GB) is a particularly malignant brain tumour which carries a poor prognosis and presents limited treatment options. MRI is standard practice for differential diagnosis at initial presentation of GB and can assist in both treatment planning and response assessment. MRI radiomics allows for discerning GB features of clinical importance that are not evident by visual analysis, augmenting the morphological and functional tumour characterisation beyond traditional imaging techniques. Given that radiotherapy is part of the standard of care for GB patients, establishing a platform for phenotyping radiation treatment responses using non-invasive methods is of high relevance.

METHODS

In this study, we modelled the responses to ionising radiation across four orthotopic mouse models of GB using diffusion and perfusion radiomics. We have identified the optimal set of radiomic features that reflect tumour cellularity, microvascularity, and blood flow changes brought about by radiation treatment in these murine orthotopic models of GB, and directly compared them with endpoint histopathological analysis.

RESULTS

We showed that the selected radiomic features can quantify textural information and pixel interrelationships of tumour response to radiation therapy, revealing subtle image patterns that may reflect intra-tumoural spatial heterogeneity. When compared to GB patients, similarities in selected radiomic features were noted between orthotopic murine tumours and non-enhancing central tumour areas in patients, along with several discrepancies in tumour cellularity and vascularization, denoted by distinct grey level intensities and nonuniformity metrics.

CONCLUSION

As the field evolves, radiomic profiling of GB may enhance the evaluation of targeted therapeutic strategies, accelerate the development of new therapies, and act as a potential virtual biopsy tool.

Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes

BACKGROUND

Medical imaging data frequently encounter image-generation heterogeneity and class imbalance properties, challenging strong generalized predictive performances with data-driven machine-learning methods. The purpose of this study was to investigate the impact of harmonization and oversampling methods on multi-center imbalanced datasets, with specific application to PET-based radiomics modeling for histologic subtype prediction in non-small cell lung cancer (NSCLC).

METHODS

The retrospective study included 245 patients with adenocarcinoma (ADC) and 78 patients with squamous cell carcinoma (SCC) from 4 centers. Utilizing 1502 radiomics features per patient, we trained, validated, and tested 4 machine-learning classifiers, to investigate the effect of no harmonization (NoH) or 4 feature harmonization methods, paired with no oversampling (NoO) or 5 oversampling methods on subtype prediction. Model performance was evaluated using the average area under the ROC curve (AUROC) and G-mean via 5 times 5-fold cross-validations. Statistical comparisons of the combined models against baseline (NoH + NoO) were performed for each fold of cross-validation using the DeLong test.

RESULTS

The number of cross-combinations with both AUROC and G-mean outperforming baseline in validation and testing was 15, 4, 2, and 7 (out of 29) for random forest (RF), linear discriminant analysis (LDA), logistic regression (LR), and support vector machine (SVM), respectively. ComBat harmonization combined with oversampling (SMOTE) via RF yielded better performance than baseline (AUROC and G-mean of validation: 0.725 vs. 0.608 and 0.625 vs. 0.398; testing: 0.637 vs. 0.567 and 0.506 vs. 0.287), though statistical significances were not observed.

CONCLUSIONS

Applying harmonization and oversampling methods in multi-center imbalanced datasets can improve NSCLC-subtype prediction, but the effect varies widely across classifiers. We have created open-source comparisons of harmonization and oversampling on different classifiers for comprehensive evaluations in different studies.

Integrating Clinical Variables, Radiomics, and Tumor-derived Cell-Free DNA for Enhanced Prediction of Resectable Esophageal Adenocarcinoma Outcomes

PURPOSE

The value of integrating clinical variables, radiomics, and tumor-derived cell-free DNA (cfDNA) for the prediction of survival and response to chemoradiation of patients with resectable esophageal adenocarcinoma is not yet known. Our aim was to investigate if radiomics and cfDNA metrics combined with clinical variables can improve personalized predictions.

METHODS AND MATERIALS

A cohort of 111 patients with resectable esophageal adenocarcinoma from 2 centers treated with neoadjuvant chemoradiation therapy was used for exploratory retrospective analyses. Models combining the clinical variables of the SOURCE survival model with radiomic features and cfDNA were built using elastic net regression and internally validated using 5-fold cross-validation. Model performance for overall survival (OS) and time to progression (TTP) were evaluated with the C-index and the area under the curve for pathologic complete response.

RESULTS

The best-performing baseline models for OS and TTP were based on the combination of SOURCE-cfDNA that reached a C-index of 0.55 and 0.59 compared with 0.44 to 0.45 with SOURCE alone. The addition of restaging positron emission tomography radiomics to SOURCE was the most promising addition for predicting OS (C-index: 0.65) and TTP (C-index: 0.60). Baseline risk stratification was achieved for OS and TTP by combining SOURCE with radiomics or cfDNA, log-rank P < .01. The best-performing combination model for the prediction of pathologic complete response reached an area under the curve of 0.61 compared with 0.47 with SOURCE variables alone.

CONCLUSIONS

The addition of radiomics and cfDNA can improve the performance of an established survival model. External validity needs to be further assessed in future studies together with the optimization of radiomic pipelines.

Clinical, Dosimetric and Radiomic Features Predictive of Lung Toxicity After (Chemo)Radiotherapy

BACKGROUND

Treatment of locally advanced non small cell lung cancer (LA-NSCLC) is based on (chemo)radiotherapy, which may cause acute lung toxicity: radiation pneumonitis (RP). Its frequency seems to increase by the use of adjuvant durvalumab therapy.

AIMS

To identify clinical, dosimetric, and radiomic factors associated with grade (G)≥2 RP and build a prediction model based on selected parameters.

PATIENTS AND METHODS

This is a retrospective multicenter cohort study including patients receiving radiation therapy between 2015 and 2019 for LA-NSCLC. Baseline computed tomography scanners were segmented to extract radiomic features from the "Lung - Tumor" volume. Variables associated with the risk of G≥2 RP in the descriptive analysis were then selected for explanatory analysis, followed by predictive analysis.

RESULTS

153 patients were included in 4 centers (51 with G≥2 RP). Factors associated with G≥2 RP included a high initial hemoglobin level, dosimetric factors (mean dose to healthy lungs, lung V20Gy and V13Gy), the addition of maintenance durvalumab, and 7 radiomic features (intensity distribution and texture). Other factors were associated with an increased risk of G≥2 RP in our explanatory model, such as older age, low Tiffeneau ratio, and a decreased initial platelet count. The best-performing predictive model was a random forest-based learning model using clinical, dosimetric, and radiomic variables, with an area under the ROC curve of 0.72 (95%CI [0.63; 0.80]) versus 0.64 for models using one type of data.

CONCLUSION

The addition of radiomic features to clinical and dosimetric ones improves prediction of the occurrence of G≥2 RP in patients receiving radiotherapy for lung cancer.

Can MRI radiomics distinguish residual adenocarcinoma from acellular mucin in treated rectal cancer?

BACKGROUND

We explored the use of MRI T2-weighted imaging (T2WI) radiomics to help distinguish acellular mucin from cellular mucin (residual tumor) to inform patient management.

METHODS

This retrospective, multi-institutional study included consecutive patients with rectal adenocarcinoma containing mucin on restaging MRI from March 2012-January 2020. Radiologists segmented 3-mm-thick T2WI. Data were split into training (n = 122), validation (n = 40), and test (n = 42) sets. Sensitivity, specificity, PPV, NPV, and AUC of the developed radiomic models were investigated on the test set. Histopathology was the reference standard. Two radiologists independently reviewed 42 patients using a unique five-point Likert scale for the probability of cellular mucin. DeLong-Delong or McNemar's test was used to compare the AUC and diagnostic performance, respectively, of the radiomics model to that of the human readers.

RESULTS

Of 204 patients (mean age, 56.6 ± 13.3 years; 129 men and 75 women), 39/204 (19 %) had acellular mucin whereas 165/204 (81 %) had cellular mucin. The radiomics model demonstrated a sensitivity of 90 % (27/30, 95 % CI: 74-97 %), specificity of 83 % (10/12, 95 % CI: 55-95 %), PPV of 93 % (27/29, 95 % CI: 78-98 %), NPV of 77 % (10/13, 95 % CI: 50-92 %), and AUC of 0.84 (95 % CI: 0.65-0.99), whereas the human readers showed inferior sensitivities of 47 % (14/30) (95 % CI: 28-66 %) and 50 % (15/30) (95 % CI: 31-69 %) and AUCs of 0.58 (95 % CI: 0.41-0.74) (DeLong's p = 0.06 [95 % CI: -0.55, 0.01]) and 0.73 (95 % CI: 0.57-0.88) (DeLong's p = 0.43 [95 % CI: -0.39, 0.16]).

CONCLUSION

The developed MRI radiomics model predicts cellular mucin post neoadjuvant chemoradiation with higher sensitivity than human readers.

The impact of the novel CovBat harmonization method on enhancing radiomics feature stability and machine learning model performance: A multi-center, multi-device study

PURPOSE

This study aims to assess whether the novel CovBat harmonization method can further reduce radiomics feature variability from different imaging devices in multi-center studies and improve machine learning model performance compared to the ComBat method.

MATERIALS

Non-contrast abdominal CT scans of 1,000 healthy subjects from three medical institutions (from four manufacturers and eight different models) were retrospectively included: Hospital A (n = 513), Hospital B (n = 338), and Hospital C (n = 149). 93 radiomics features were extracted from liver and spleen tissues using PyRadiomics. Performing a binary classification task of liver and spleen tissues on the pooled data from the three institutions: (1) Unharmonized, (2) ComBat, and (3) CovBat. Models were built separately for each radiomics feature classes (First-order, GLCM, GLRLM, GLSZM, NGTD, GLDM), as well as a combined model integrating all feature classes. The Kruskal-Wallis test and principal component analysis (PCA) were used to assess the variability of radiomics features among the groups. Multiple linear regression models were used to analyze the sources of variation. Accuracy, sensitivity, specificity, F1-score, and area under the curve (AUC) were used to evaluate model performance.

RESULTS

After ComBat and CovBat harmonization, the number of consistent features increased by 68.82 % and 73.12 %, respectively, and the feature variability due to hardware differences decreased from 12.32-25.38 % to 1.89-2.01 % with ComBat and 1.19-1.88 % with CovBat. The AUC of the machine learning models improved significantly: Combined (Unharmonized: 0.93, ComBat: 0.99, CovBat: 1.00), First-order (0.93, 0.98, 0.98), GLCM (0.81, 0.93, 0.98), GLRLM (0.78, 0.96, 0.98), NGTDM (0.75, 0.96, 0.98), GLSZM (0.78, 0.93, 0.97), and GLDM (0.83, 0.94, 0.97). DeLong's test showed that the results before and after harmonization were statistically significant (P < 0.05).

CONCLUSION

CovBat further reduced radiomics feature variability caused by different CT scanners and significantly improved the performance of machine learning models, although the degree of improvement varied across different feature categories.

Artificial intelligence for response prediction and personalisation in radiation oncology

Artificial intelligence (AI) systems may personalise radiotherapy by assessing complex and multifaceted patient data and predicting tumour and normal tissue responses to radiotherapy. Here we describe three distinct generations of AI systems, namely personalised radiotherapy based on pretreatment data, response-driven radiotherapy and dynamically optimised radiotherapy. Finally, we discuss the main challenges in clinical translation of AI systems for radiotherapy personalisation.

MRI in Oral Tongue Squamous Cell Carcinoma: A Radiomic Approach in the Local Recurrence Evaluation

(1) Background: Oral tongue squamous cell carcinoma (OTSCC) is a prevalent malignancy with high loco-regional recurrence. Advanced imaging biomarkers are critical for stratifying patients at a high risk of recurrence. This study aimed to develop MRI-based radiomic models to predict loco-regional recurrence in OTSCC patients undergoing surgery. (2) Methods: We retrospectively selected 92 patients with OTSCC who underwent MRI, followed by surgery and cervical lymphadenectomy. A total of 31 patients suffered from a loco-regional recurrence. Radiomic features were extracted from preoperative post-contrast high-resolution MRI and integrated with clinical and pathological data to develop predictive models, including radiomic-only and combined radiomic-clinical approaches, trained and validated with stratified data splitting. (3) Results: Textural features, such as those derived from the Gray-Level Size-Zone Matrix, Gray-Level Dependence Matrix, and Gray-Level Run-Length Matrix, showed significant associations with recurrence. The radiomic-only model achieved an accuracy of 0.79 (95% confidence interval: 0.69, 0.87) and 0.74 (95% CI: 0.54, 0.89) in the training and validation set, respectively. Combined radiomic and clinical models, incorporating features like the pathological depth of invasion and lymph node status, provided comparable diagnostic performances. (4) Conclusions: MRI-based radiomic models demonstrated the potential for predicting loco-regional recurrence, highlighting their increasingly important role in advancing precision oncology for OTSCC.

Myocardial perfusion SPECT radiomic features reproducibility assessment: Impact of image reconstruction and harmonization

BACKGROUND

Coronary artery disease (CAD) has one of the highest mortality rates in humans worldwide. Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) provides clinicians with myocardial metabolic information non-invasively. However, there are some limitations to interpreting SPECT images performed by physicians or automatic quantitative approaches. Radiomics analyzes images objectively by extracting quantitative features and can potentially reveal biological characteristics that the human eye cannot detect. However, the reproducibility and repeatability of some radiomic features can be highly susceptible to segmentation and imaging conditions.

PURPOSE

We aimed to assess the reproducibility of radiomic features extracted from uncorrected MPI-SPECT images reconstructed with 15 different settings before and after ComBat harmonization, along with evaluating the effectiveness of ComBat in realigning feature distributions.

MATERIALS AND METHODS

A total of 200 patients (50% normal and 50% abnormal) including rest and stress (without attenuation and scatter corrections) MPI-SPECT images were included. Images were reconstructed using 15 combinations of filter cut-off frequencies, filter orders, filter types, reconstruction algorithms, number of iterations and subsets resulting in 6000 images. Image segmentation was performed on the left ventricle in the first reconstruction for each patient and applied to 14 others. A total of 93 radiomic features were extracted from the segmented area, and ComBat was used to harmonize them. The intraclass correlation coefficient (ICC) and overall concordance correlation coefficient (OCCC) tests were performed before and after ComBat to examine the impact of each parameter on feature robustness and to assess harmonization efficiency. The ANOVA and the Kruskal-Wallis tests were performed to evaluate the effectiveness of ComBat in correcting feature distributions. In addition, the Student's t-test, Wilcoxon rank-sum, and signed-rank tests were implemented to assess the significance level of the impacts made by each parameter of different batches and patient groups (normal vs. abnormal) on radiomic features.

RESULTS

Before applying ComBat, the majority of features (ICC: 82, OCCC: 61) achieved high reproducibility (ICC/OCCC ≥ 0.900) under every batch except Reconstruction. The largest and smallest number of poor features (ICC/OCCC < 0.500) were obtained by IterationSubset and Order batches, respectively. The most reliable features were from the first-order (FO) and gray-level co-occurrence matrix (GLCM) families. Following harmonization, the minimum number of robust features increased (ICC: 84, OCCC: 78). Applying ComBat showed that Order and Reconstruction were the least and the most responsive batches, respectively. The most robust families, in a descending order, were found to be FO, neighborhood gray-tone difference matrix (NGTDM), GLCM, gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), and gray-level dependence matrix (GLDM) under Cut-off, Filter, and Order batches. The Wilcoxon rank-sum test showed that the number of robust features significantly differed under most batches in the Normal and Abnormal groups.

CONCLUSION

The majority of radiomic features show high levels of robustness across different OSEM reconstruction parameters in uncorrected MPI-SPECT. ComBat is effective in realigning feature distributions and enhancing radiomic features reproducibility.

Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?

The WHO grading of pancreatic neuroendocrine neoplasms (PanNENs) is essential in patient management and an independent prognostic factor for patient survival. Radiomics features from CE-CT images hold promise for the outcome and tumor grade prediction. However, variations in reconstruction parameters can impact the predictive value of radiomics. 127 patients with histopathologically confirmed PanNENs underwent CT scans with filtered back projection (B20f) and iterative (I26f) reconstruction kernels. 3190 radiomic features were extracted from tumors and pancreatic volumes. Wilcoxon paired tests assessed the impact of reconstruction kernels and ComBat harmonization efficiency. SVM models were employed to predict tumor grade using the entire set of radiomics features or only those identified as harmonizable. The models' performance was assessed on an independent dataset of 36 patients. Significant differences, after correction for multiple testing, were observed in 69% of features in the pancreatic volume and 51% in the tumor volume with B20f and I26f kernels. SVM models demonstrated accuracy ranging from 0.67 (95%CI: 0.50-0.81) to 0.83 (95%CI: 0.69-0.94) in distinguishing grade 1 cases from higher grades. Reconstruction kernels alter radiomics features and iterative kernel models trended towards higher performance. ComBat harmonization mitigates kernel impacts but addressing this effect is crucial in studies involving data from different kernels.

Machine Learning-Based Radiomics Analysis for Identifying KRAS Mutations in Non-Small-Cell Lung Cancer from CT Images: Challenges, Insights and Implications

Kirsten Rat Sarcoma viral oncogene homolog (KRAS) is a frequently occurring mutation in non-small-cell lung cancer (NSCLC) and influences cancer treatment and disease progression. In this study, a machine learning (ML) pipeline was applied to radiomic features extracted from public and internal CT images to identify KRAS mutations in NSCLC patients. Both datasets were analyzed using parametric (t test) and non-parametric statistical tests (Mann-Whitney U test) and dimensionality reduction techniques. Afterwards, the proposed ML pipeline was applied to both datasets using a five-fold cross-validation on the training set (70/30 train/test split) before being validated on the other dataset. The results show that the radiomic features are significantly different (Mann-Whitney U test; p < 0.05) between the two datasets, despite the use of identical feature extraction methods. Model transferability is therefore difficult to achieve, which became evident during external testing (F1 score = 0.41). Oversampling, undersampling, clustering and harmonization techniques were applied to balance and harmonize the datasets, but did not improve the classification of KRAS mutation presence. In general, due to only a single moderate result (highest test F1 score = 0.67), the accuracy of KRAS prediction is not sufficient for clinical application. In future work, the complexity of KRAS mutation might be addressed by taking submutations into consideration. Larger multicentric datasets with balanced tumor stages, including multi-scanner datasets, seem to be necessary for building robust predictive models.

Multiparametric MRI and artificial intelligence in predicting and monitoring treatment response in bladder cancer

Bladder cancer is the 10th most common and 13th most deadly cancer worldwide, with urothelial carcinomas being the most common type. Distinguishing between non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC) is essential due to significant differences in management and prognosis. MRI may play an important diagnostic role in this setting. The Vesical Imaging Reporting and Data System (VI-RADS), a multiparametric MRI (mpMRI)-based consensus reporting platform, allows for standardized preoperative muscle invasion assessment in BCa with proven diagnostic accuracy. However, post-treatment assessment using VI-RADS is challenging because of anatomical changes, especially in the interpretation of the muscle layer. MRI techniques that provide tumor tissue physiological information, including diffusion-weighted (DW)- and dynamic contrast-enhanced (DCE)-MRI, combined with derived quantitative imaging biomarkers (QIBs), may potentially overcome the limitations of BCa evaluation when predominantly focusing on anatomic changes at MRI, particularly in the therapy response setting. Delta-radiomics, which encompasses the assessment of changes (Δ) in image features extracted from mpMRI data, has the potential to monitor treatment response. In comparison to the current Response Evaluation Criteria in Solid Tumors (RECIST), QIBs and mpMRI-based radiomics, in combination with artificial intelligence (AI)-based image analysis, may potentially allow for earlier identification of therapy-induced tumor changes. This review provides an update on the potential of QIBs and mpMRI-based radiomics and discusses the future applications of AI in BCa management, particularly in assessing treatment response. CRITICAL RELEVANCE STATEMENT: Incorporating mpMRI-based quantitative imaging biomarkers, radiomics, and artificial intelligence into bladder cancer management has the potential to enhance treatment response assessment and prognosis prediction. KEY POINTS: Quantitative imaging biomarkers (QIBs) from mpMRI and radiomics can outperform RECIST for bladder cancer treatments. AI improves mpMRI segmentation and enhances radiomics feature extraction effectively. Predictive models integrate imaging biomarkers and clinical data using AI tools. Multicenter studies with strict criteria validate radiomics and QIBs clinically. Consistent mpMRI and AI applications need reliable validation in clinical practice.

Optimizing Glioblastoma, IDH-wildtype Treatment Outcomes : A Radiomics and Support Vector Machine-Based Approach to Overall Survival Estimation

OBJECTIVE

Glioblastoma multiforme (GBM), particularly the isocitrate dehydrogenase (IDH)-wildtype type, represents a significant clinical challenge due to its aggressive nature and poor prognosis. Despite advancements in medical imaging and its modalities, survival rates have not improved significantly, demanding innovative treatment planning and outcome prediction approaches.

METHODS

This study utilizes a support vector machine (SVM) classifier using radiomics features to predict the overall survival (OS) of GBM, IDH-wildtype patients to short (<12 months) and long (≥12 months) survivors. A dataset comprising multi-parametric magnetic resonance imaging scans from 574 patients was analyzed. Radiomic features were extracted from T1, T2, fluid-attenuated inversion recovery, and T1 with gadolinium (T1GD) sequences. Low variance features were removed, and recursive feature elimination was used to select the most informative features. The SVM model was trained using a k-fold cross-validation approach. Furthermore, clinical parameters such as age, gender, and MGMT (O6-methylguanine-DNA methyltransferase) promoter methylation status were integrated to enhance prediction accuracy.

RESULTS

The model showed reasonable results in terms of cross-validated area under the curve of 0.84 (95% confidence interval, 0.80-0.90) with (p<0.001) effectively categorizing patients into short and long survivors. Log-rank test (chi-square statistics) analysis for the developed model was 0.00029 along with the 1.20 Cohen's d effect size. Most importantly, clinical data integration further refined the survival estimates, providing a more fitted prediction that considers individual patient characteristics by Kaplan-Meier curve with p-value <0.0001.

CONCLUSION

The proposed method significantly enhances the predictive accuracy of OS outcomes in GBM, IDH-wildtype patients. By integrating detailed imaging features with key clinical indicators, this model offers a robust tool for personalized treatment planning, potentially improving OS.

[18F]FDG PET/CT Radiomics in Cervical Cancer: A Systematic Review

Background/Objectives: Cervical cancer is a significant global health concern, with high incidence and mortality rates, especially in less-developed regions. [18F]FDG PET/CT is now indicated at various stages of management, but its analysis is essentially based on SUVmax, a measure of [18F]FDG uptake. Radiomics, by extracting a multitude of parameters, promises to improve the diagnostic and prognostic performance of the examination. However, studies remain heterogeneous, both in terms of patient numbers and methods, so a synthesis is needed. Methods: This systematic review was conducted following PRISMA-P guidelines and registered in PROSPERO (CRD42024584123). Eligible studies on PET/CT radiomics in cervical cancer were identified through PubMed and Scopus and assessed for quality using the Radiomics Quality Score (RQS v2.0), with data extraction focusing on study design, population characteristics, radiomic methods, and model performances. Results: The review identified 22 studies on radiomics in cervical cancer, 19 of which focused specifically on locally advanced cervical cancer (LACC) and assessed various clinical outcomes, such as survival, relapse, treatment response, and lymph node involvement prediction. They reported significant associations between prognostic indicators and radiomic features, indicating the potential of radiomics to improve the predictive accuracy for patient outcomes in LACC; however, the overall quality of the studies was relatively moderate, with a median RQS of 12/36. Conclusions: While radiomic analysis in cervical cancer presents promising opportunities for survival prediction and personalized care, further well-designed studies are essential to provide stronger evidence for its clinical utility.

Methodological issues in radiomics: impact on accuracy of MRI for predicting response to neoadjuvant chemotherapy in breast cancer

AIM

To investigate whether methodological aspects may influence the performance of MRI-radiomic models to predict response to neoadjuvant treatment (NAT) in breast cancer (BC) patients.

MATERIALS AND METHODS

We conducted a systematic review until March 2023. A random-effects meta-analysis was performed to combine the area under the receiver operating characteristic curve (AUC) values. Publication bias was assessed using Egger's test and heterogeneity was estimated by I2. A meta-regression was conducted to investigate the impact of various factors, including scanner, features' number/transformation/type, pixel/voxel scaling, etc. RESULTS: Forty-two studies were included. The summary AUC was 0.77 (95% CI: 0.74-0.81). Substantial heterogeneity was observed (I2 = 81%) with no publication bias (p = 0.35). Radiomic model accuracy was influenced by the scanner vendor, with lower AUCs in studies using mixed scanner vendors (AUC; 95% CI: 0.70; 0.61-0.78) compared to studies including images obtained from the same scanner (AUC (95% CI): 0.83 (0.77-0.88), 0.74 (0.67-0.82), 0.83 (0.78-0.89) for three different vendors; vendors 1, 2, and 3, respectively; p-value = 0.03 for comparison with vendor 1). Feature type also seemed to have an impact on the AUC, with higher prediction accuracy observed for studies using 3D than 2D/2.5D images (AUC; 95% CI: 0.81; 0.78-0.85 and 0.73; 0.65-0.81, respectively, p-value = 0.03). Non-significant between-study heterogeneity was observed in the studies including 3D images (I2 = 33%) and Vendor 1 scanners (I2 = 40%).

CONCLUSION

MRI-radiomics has emerged as a potential method for predicting the response to NAT in BC patients, showing promising outcomes. Nevertheless, it is important to acknowledge the diversity among the methodological choices applied. Further investigations should prioritize achieving standardized protocols, and enhancing methodological rigor in MRI-radiomics.

KEY POINTS

Question Do methodological aspects influence the performance of MRI-radiomic models in predicting response to NAT in BC patients? Findings Radiomic model accuracy was influenced by the scanner vendor and feature type. Clinical relevance Methodological discrepancies affect the performance of MRI-radiomic models. Developing standardized protocols and enhancing methodological rigor in these studies should be prioritized.

Developmental trends and knowledge frameworks in the application of radiomics in prostate cancer: a bibliometric analysis from 2000 to 2024

BACKGROUND

This research utilized the bibliometrics method to analyze the published literature related to prostate cancer (PCa) imaging. Furthermore, current knowledge and research hotspots of radiomics in PCa diagnosis and treatment were comprehensively reviewed, as well as progress and emerging trends in field were explored.

METHODS

In this investigation, the relevant literature on radiomics, and PCa was retrieved from Web of Science Core Collection (WoSCC) databases from 2000 and 2024. Furthermore, a comprehensive bibliometric analysis was carried out using advanced tools like CiteSpace6.2, VOS viewer, and the 'bibliometrix' package of R software to visualize the annual distribution of publications across various aspects such as authors, countries, journals, institutions, and keywords.

RESULTS

This analysis included 593 from 58 countries including China and the United States. Chinese Academy of Sciences and Frontiers in Oncology were the institutions and journals that publish the most relevant articles, -while Radiology journal had the greatest number of co-cited publications. Furthermore, 3,621 authors published on this topic, of which Madabhushi Anant and Stoyanova Radka had the highest contributions. Moreover, Lambin, P. had the most co-citations. In addition, the diagnostic characteristics of radiomics in PCa imaging and treatment strategies are the current research focal points. The establishment of multi-functional imaging techniques and independent factor models warrants future investigation.

CONCLUSIONS

In summary, this analysis revealed that the research on PCa imaging is developing vigorously, focusing on the diagnostic methods and intervention measures of imaging in PCa diagnosis and treatment. In the future, there is an urgent need for improved collaboration and communication among countries and institutions.

Enhancing Radiomics Reproducibility: Deep Learning-Based Harmonization of Abdominal Computed Tomography (CT) Images

We assessed the feasibility of using deep learning-based image harmonization to improve the reproducibility of radiomics features in abdominal CT scans. In CT imaging, harmonization adjusts images from different institutions to ensure consistency despite variations in scanners and acquisition protocols. This process is essential because such differences can lead to variability in radiomics features, affecting reproducibility and accuracy. Harmonizing images minimizes these inconsistencies, supporting more reliable and clinically applicable results across diverse settings. A pre-trained harmonization algorithm was applied to 63 dual-energy abdominal CT images, which were reconstructed into four different types, and 10 regions of interest (ROIs) were analyzed. From the original 455 radiomics features per ROI, 387 were used after excluding redundant features. Reproducibility was measured using the intraclass correlation coefficient (ICC), with a threshold of ICC ≥ 0.85 indicating acceptable reproducibility. The region-based analysis revealed significant improvements in reproducibility post-harmonization, especially in vessel features, which increased from 14% to 69%. Other regions, including the spleen, kidney, muscle, and liver parenchyma, also saw notable improvements, although air reproducibility slightly decreased from 95% to 94%, impacting only a few features. In patient-based analysis, reproducible features increased from 18% to 65%, with an average of 179 additional reproducible features per patient after harmonization. These results demonstrate that deep learning-based harmonization can significantly enhance the reproducibility of radiomics features in abdominal CT, offering promising potential for advancing radiomics development and its clinical applications.

Advancing Hepatocellular Carcinoma Management Through Peritumoral Radiomics: Enhancing Diagnosis, Treatment, and Prognosis

Hepatocellular carcinoma (HCC) is the most common primary liver cancer and is associated with high mortality rates due to late detection and aggressive progression. Peritumoral radiomics, an emerging technique that quantitatively analyzes the tissue surrounding the tumor, has shown significant potential in enhancing the management of HCC. This paper examines the role of peritumoral radiomics in improving diagnostic accuracy, guiding personalized treatment strategies, and refining prognostic assessments. By offering unique insights into the tumor microenvironment, peritumoral radiomics enables more precise patient stratification and informs clinical decision-making. However, the integration of peritumoral radiomics into routine clinical practice faces several challenges. Addressing these challenges through continued research and innovation is crucial for the successful implementation of peritumoral radiomics in HCC management, ultimately leading to improved patient outcomes.

Robust machine learning challenge: An AIFM multicentric competition to spread knowledge, identify common pitfalls and recommend best practice

PURPOSE

A novel and unconventional approach to a machine learning challenge was designed to spread knowledge, identify robust methods and highlight potential pitfalls about machine learning within the Medical Physics community.

METHODS

A public dataset comprising 41 radiomic features and 535 patients was employed to assess the potential of radiomics in distinguishing between primary lung tumors and metastases. Each participant developed two classification models using: (i) all features (base model); (ii) only robust features (robust model). Both models were validated with cross-validation and on unseen data. The population stability index (PSI) was used as diagnostic metric for implementation issues. Performance was compared to reference. Base and robust models were compared in terms of performance and stability (coefficient of variation (CoV) of prediction probabilities).

RESULTS

PSI detected potential implementation errors in 70 % of models. The dataset exhibited strong imbalance. The average Gmean (i.e. an appropriate metric for imbalance) among all participants was 0.67 ± 0.01, significantly higher than reference Gmean of 0.50 ± 0.04. Robust models performances were slightly worse than base models (p < 0.05). Regarding stability, robust models exhibited lower median CoV on training set only.

CONCLUSION

AI4MP-Challenge models overperformed the reference, significantly improving the Gmean. Exclusion of less-robust features did not improve model robustness and it should be avoided when confounding effects are absent. Other methods, like harmonization or data augmentation, should be evaluated. This study demonstrated how the collaborative effort to foster knowledge on machine learning among medical physicists, through interactive sessions and exchange of information among participants, can result in improved models.

Radiomic feature robustness in CT scans affected by fiducial marker induced streak artifacts for patients with hepatocellular carcinoma

BACKGROUND

Stereotactic body radiation therapy for hepatocellular carcinoma necessitates the implantation of gold fiducial markers in the liver, resulting in artifacts on computed tomography (CT) scans, which affect radiomic feature values.

PURPOSE

This report aims to assess the effect of these artifacts on radiomic features and how removing CT slices affects radiomic features extracted from 3D regions of interest (ROI).

METHODS

First, the range variation in 38 tumor contours unaffected by artifacts was assessed after sequentially and randomly removing 25%, 50%, 75% of slices. Subsequently, the agreement of feature values before and after removing ROI slices containing artifacts from 186 patients' CT scans was assessed with Lin's concordance correlation coefficient (CCC) and a Wilcoxon signed-rank test.

RESULTS

In artifact-free tumor volumes, at least 71% of features remain robust with up to 50% of slices removed, while 56% remains robust with up to 75% of slices removed. When comparing contours before and after removing slices containing artifacts, around a third of features in the tumor contour and surrounding area remain robust (CCC > 0.9), compared to 44% in the healthy liver. Concerning the tumor, 13% (Gray Level Size Zone Matrix) to 61% (first order) of the features remain robust (CCC > 0.9). Over 90% of features differ significantly as assessed by Wilcoxon signed-rank test, however.

CONCLUSIONS

This study demonstrates that removing slices containing artifacts is a feasible solution for the CT fiducial problem in this patient population and provides insight into which features are affected.

GPT-Driven Radiology Report Generation with Fine-Tuned Llama 3

The integration of deep learning into radiology has the potential to enhance diagnostic processes, yet its acceptance in clinical practice remains limited due to various challenges. This study aimed to develop and evaluate a fine-tuned large language model (LLM), based on Llama 3-8B, to automate the generation of accurate and concise conclusions in magnetic resonance imaging (MRI) and computed tomography (CT) radiology reports, thereby assisting radiologists and improving reporting efficiency. A dataset comprising 15,000 radiology reports was collected from the University of Medicine and Pharmacy of Craiova's Imaging Center, covering a diverse range of MRI and CT examinations made by four experienced radiologists. The Llama 3-8B model was fine-tuned using transfer-learning techniques, incorporating parameter quantization to 4-bit precision and low-rank adaptation (LoRA) with a rank of 16 to optimize computational efficiency on consumer-grade GPUs. The model was trained over five epochs using an NVIDIA RTX 3090 GPU, with intermediary checkpoints saved for monitoring. Performance was evaluated quantitatively using Bidirectional Encoder Representations from Transformers Score (BERTScore), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Bilingual Evaluation Understudy (BLEU), and Metric for Evaluation of Translation with Explicit Ordering (METEOR) metrics on a held-out test set. Additionally, a qualitative assessment was conducted, involving 13 independent radiologists who participated in a Turing-like test and provided ratings for the AI-generated conclusions. The fine-tuned model demonstrated strong quantitative performance, achieving a BERTScore F1 of 0.8054, a ROUGE-1 F1 of 0.4998, a ROUGE-L F1 of 0.4628, and a METEOR score of 0.4282. In the human evaluation, the artificial intelligence (AI)-generated conclusions were preferred over human-written ones in approximately 21.8% of cases, indicating that the model's outputs were competitive with those of experienced radiologists. The average rating of the AI-generated conclusions was 3.65 out of 5, reflecting a generally favorable assessment. Notably, the model maintained its consistency across various types of reports and demonstrated the ability to generalize to unseen data. The fine-tuned Llama 3-8B model effectively generates accurate and coherent conclusions for MRI and CT radiology reports. By automating the conclusion-writing process, this approach can assist radiologists in reducing their workload and enhancing report consistency, potentially addressing some barriers to the adoption of deep learning in clinical practice. The positive evaluations from independent radiologists underscore the model's potential utility. While the model demonstrated strong performance, limitations such as dataset bias, limited sample diversity, a lack of clinical judgment, and the need for large computational resources require further refinement and real-world validation. Future work should explore the integration of such models into clinical workflows, address ethical and legal considerations, and extend this approach to generate complete radiology reports.

Recommendations for the creation of benchmark datasets for reproducible artificial intelligence in radiology

Various healthcare domains have witnessed successful preliminary implementation of artificial intelligence (AI) solutions, including radiology, though limited generalizability hinders their widespread adoption. Currently, most research groups and industry have limited access to the data needed for external validation studies. The creation and accessibility of benchmark datasets to validate such solutions represents a critical step towards generalizability, for which an array of aspects ranging from preprocessing to regulatory issues and biostatistical principles come into play. In this article, the authors provide recommendations for the creation of benchmark datasets in radiology, explain current limitations in this realm, and explore potential new approaches. CLINICAL RELEVANCE STATEMENT: Benchmark datasets, facilitating validation of AI software performance can contribute to the adoption of AI in clinical practice. KEY POINTS: Benchmark datasets are essential for the validation of AI software performance. Factors like image quality and representativeness of cases should be considered. Benchmark datasets can help adoption by increasing the trustworthiness and robustness of AI.

A Comprehensive Evaluation of Radiomic Features in Normal Brain Magnetic Resonance Imaging: Investigating Robustness and Region Variations

Background

Despite extensive research on various brain diseases, a few studies have focused on radiomic feature distribution in healthy brain images. The present study applied a novel radiomic framework to investigate the robustness and baseline values of radiomic features in normal brain magnetic resonance imaging (MRIs) regions.

Materials and Methods

Analyses were performed on T1 and T2 images including 276 normal brains and 14 healthy volunteers were scanned with three scanners using the same protocols. The images were divided into 1024 three-dimensional nonoverlap patches with the same pixel size. Seven patches located in the thalamus, putamen, hippocampus and brain stem were selected as volume of interest (VOI). Eighty-five radiomic features were generated. To investigate the variation of features across VOIs, the analysis of variance was performed and coefficient of variation (COV) and intraclass correlation coefficient (ICC) were explored to examine the features repeatability.

Results

Thalamus (right and left) and hippocampus (left) resulted in more stable features (COV ≤ 6%) in T1 and T2 images, respectively. The inter-scanner ICC analysis demonstrated the features of T2 sequences represented more repeatable results and the brain stem and thalamus (both T1 and T2) showed particularly high repeatability (higher ICC values). Robust results (ICC ≥ 0.9) were identified for energy and range features of the first order class and several textures features across different brain regions.

Conclusion

Our results indicated the baselines of the repeatable texture features in healthy brain structural MRI highlighting inter-scanner stability. According to the findings, MRI sequencing and VOI location impact feature robustness and should be considered in brain radiomic studies.

Recurrent Versus Metastatic Head and Neck Cancer: An Evolving Landscape and the Role of Immunotherapy

Squamous cell carcinoma of the head and neck (SCCHN) is among the ten most common cancers worldwide, with advanced SCCHN presenting with a 5-year survival of 34% in the case of nodal involvement and 8% in the case of metastatic disease. Disease-free survival at 2 years is 67% for stage II and 33% for stage III tumors, whereas 12-30% of patients undergo distant failures after curative treatment. Previous treatments often hinder the success of salvage surgery and/or reirradiation, while the standard of care for the majority of metastatic SCCHN remains palliative chemo- and immuno-therapy, with few patients eligible for locoregional treatments. The aim of this paper is to review the characteristics of recurrent SCCHN, based on different recurrence sites, and metastatic disease; we will also explore the possibilities not only of salvage surgery and reirradiation but also systemic therapy choices and locoregional treatment for metastatic SCCHN.

Impact of SUSAN Denoising and ComBat Harmonization on Machine Learning Model Performance for Malignant Brain Neoplasms

BACKGROUND AND PURPOSE

Feature variability in radiomics studies due to technical and magnet strength parameters is well-known and may be addressed through various preprocessing methods. However, very few studies have evaluated the downstream impact of variable preprocessing on model classification performance in a multiclass setting. We sought to evaluate the impact of Smallest Univalue Segment Assimilating Nucleus (SUSAN) denoising and Combining Batches harmonization on model classification performance.

MATERIALS AND METHODS

A total of 493 cases (410 internal and 83 external data sets) of glioblastoma, intracranial metastatic disease, and primary CNS lymphoma underwent semiautomated 3D-segmentation post-baseline image processing (BIP) consisting of resampling, realignment, coregistration, skull-stripping, and image normalization. Post-BIP, 2 sets were generated, one with and another without SUSAN denoising. Radiomics features were extracted from both data sets and batch-corrected to produce 4 data sets: (a) BIP, (b) BIP with SUSAN denoising, (c) BIP with Combining Batches, and (d) BIP with both SUSAN denoising and Combining Batches harmonization. Performance was then summarized for models using a combination of 6 feature-selection techniques and 6 machine learning models across 4 mask-sequence combinations with features derived from 1 to 3 (multiparametric) MRI sequences.

RESULTS

Most top-performing models on the external test set used BIP+SUSAN denoising-derived features. Overall, the use of SUSAN denoising and Combining Batches harmonization led to a slight but generally consistent improvement in model performance on the external test set.

CONCLUSIONS

The use of image-preprocessing steps such as SUSAN denoising and Combining Batches harmonization may be more useful in a multi-institutional setting to improve model generalizability. Models derived from only T1 contrast-enhanced images showed comparable performance to models derived from multiparametric MRI.

Deep learning for the harmonization of structural MRI scans: a survey

Medical imaging datasets for research are frequently collected from multiple imaging centers using different scanners, protocols, and settings. These variations affect data consistency and compatibility across different sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, and scanner calibration drift, as well as to ensure consistent data for medical image processing techniques. Given the critical importance and widespread relevance of this issue, a vast array of image harmonization methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal of this review paper is to examine the latest deep learning techniques employed for image harmonization by analyzing cutting-edge architectural approaches in the field of medical image harmonization, evaluating both their strengths and limitations. This paper begins by providing a comprehensive fundamental overview of image harmonization strategies, covering three critical aspects: established imaging datasets, commonly used evaluation metrics, and characteristics of different scanners. Subsequently, this paper analyzes recent structural MRI (Magnetic Resonance Imaging) harmonization techniques based on network architecture, network learning algorithm, network supervision strategy, and network output. The underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, as well as custom-designed network architectures. This paper investigates the effectiveness of Disentangled Representation Learning (DRL) as a pivotal learning algorithm in harmonization. Lastly, the review highlights the primary limitations in harmonization techniques, specifically the lack of comprehensive quantitative comparisons across different methods. The overall aim of this review is to serve as a guide for researchers and practitioners to select appropriate architectures based on their specific conditions and requirements. It also aims to foster discussions around ongoing challenges in the field and shed light on promising future research directions with the potential for significant advancements.

Impact of Preprocessing Parameters in Medical Imaging-Based Radiomic Studies: A Systematic Review

BACKGROUND

Lately, radiomic studies featuring the development of a signature to use in prediction models in diagnosis or prognosis outcomes have been increasingly published. While the results are shown to be promising, these studies still have many pitfalls and limitations. One of the main issues of these studies is that radiomic features depend on how the images are preprocessed before their computation. Since, in widely known and used software for radiomic features calculation, it is possible to set these preprocessing parameters before the calculation of the radiomic feature, there are ongoing studies assessing the stability and repeatability of radiomic features to find the most suitable preprocessing parameters for every used imaging modality.

MATERIALS AND METHODS

We performed a comprehensive literature search using four electronic databases: PubMed, Cochrane Library, Embase, and Scopus. Mesh terms and free text were modeled in search strategies for databases. The inclusion criteria were studies where preprocessing parameters' influence on feature values and model predictions was addressed. Records lacking information on image acquisition parameters were excluded, and any eligible studies with full-text versions were included in the review process, while conference proceedings and monographs were disregarded. We used the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) tool to investigate the risk of bias. We synthesized our data in a table divided by the imaging modalities subgroups.

RESULTS

After applying the inclusion and exclusion criteria, we selected 43 works. This review examines the impact of preprocessing parameters on the reproducibility and reliability of radiomic features extracted from multimodality imaging (CT, MRI, CBCT, and PET/CT). Standardized preprocessing is crucial for consistent radiomic feature extraction. Key preprocessing steps include voxel resampling, normalization, and discretization, which influence feature robustness and reproducibility. In total, 44% of the included works studied the effects of an isotropic voxel resampling, and most studies opted to employ a discretization strategy. From 2021, several studies started selecting the best set of preprocessing parameters based on models' best performance. As for comparison metrics, ICC was the most used in MRI studies in 58% of the screened works.

CONCLUSIONS

From our work, we highlighted the need to harmonize the use of preprocessing parameters and their values, especially in light of future studies of prospective studies, which are still lacking in the current literature.

Mitigating the impact of image processing variations on tumour [18F]-FDG-PET radiomic feature robustness

Radiomics analysis of [18F]-fluorodeoxyglucose ([18F]-FDG) PET images could be leveraged for personalised cancer medicine. However, the inherent sensitivity of radiomic features to intensity discretisation and voxel interpolation complicates its clinical translation. In this work, we evaluated the robustness of tumour [18F]-FDG-PET radiomic features to 174 different variations in intensity resolution or voxel size, and determined whether implementing parameter range conditions or dependency corrections could improve their robustness. Using 485 patient images spanning three cancer types: non-small cell lung cancer (NSCLC), melanoma, and lymphoma, we observed features were more sensitive to intensity discretisation than voxel interpolation, especially texture features. In most of our investigations, the majority of non-robust features could be made robust by applying parameter range conditions. Correctable features, which were generally fewer than conditionally robust, showed systematic dependence on bin configuration or voxel size that could be minimised by applying corrections based on simple mathematical equations. Melanoma images exhibited limited robustness and correctability relative to NSCLC and lymphoma. Our study provides an in-depth characterisation of the sensitivity of [18F]-FDG-PET features to image processing variations and reinforces the need for careful selection of imaging biomarkers prior to any clinical application.

The effect of harmonization on the variability of PET radiomic features extracted using various segmentation methods

PURPOSE

This study aimed to examine the robustness of positron emission tomography (PET) radiomic features extracted via different segmentation methods before and after ComBat harmonization in patients with non-small cell lung cancer (NSCLC).

METHODS

We included 120 patients (positive recurrence = 46 and negative recurrence = 74) referred for PET scanning as a routine part of their care. All patients had a biopsy-proven NSCLC. Nine segmentation methods were applied to each image, including manual delineation, K-means (KM), watershed, fuzzy-C-mean, region-growing, local active contour (LAC), and iterative thresholding (IT) with 40, 45, and 50% thresholds. Diverse image discretizations, both without a filter and with different wavelet decompositions, were applied to PET images. Overall, 6741 radiomic features were extracted from each image (749 radiomic features from each segmented area). Non-parametric empirical Bayes (NPEB) ComBat harmonization was used to harmonize the features. Linear Support Vector Classifier (LinearSVC) with L1 regularization For feature selection and Support Vector Machine classifier (SVM) with fivefold nested cross-validation was performed using StratifiedKFold with 'n_splits' set to 5 to predict recurrence in NSCLC patients and assess the impact of ComBat harmonization on the outcome.

RESULTS

From 749 extracted radiomic features, 206 (27%) and 389 (51%) features showed excellent reliability (ICC ≥ 0.90) against segmentation method variation before and after NPEB ComBat harmonization, respectively. Among all, 39 features demonstrated poor reliability, which declined to 10 after ComBat harmonization. The 64 fixed bin widths (without any filter) and wavelets (LLL)-based radiomic features set achieved the best performance in terms of robustness against diverse segmentation techniques before and after ComBat harmonization. The first-order and GLRLM and also first-order and NGTDM feature families showed the largest number of robust features before and after ComBat harmonization, respectively. In terms of predicting recurrence in NSCLC, our findings indicate that using ComBat harmonization can significantly enhance machine learning outcomes, particularly improving the accuracy of watershed segmentation, which initially had fewer reliable features than manual contouring. Following the application of ComBat harmonization, the majority of cases saw substantial increase in sensitivity and specificity.

CONCLUSION

Radiomic features are vulnerable to different segmentation methods. ComBat harmonization might be considered a solution to overcome the poor reliability of radiomic features.

Solving the Pervasive Problem of Protocol Non-Compliance in MRI using an Open-Source tool mrQA

Pooling data across diverse sources acquired by multisite consortia requires compliance with a predefined reference protocol i.e., ensuring different sites and scanners for a given project have used identical or compatible MR physics parameter values. Traditionally, this has been an arduous and manual process due to difficulties in working with the complicated DICOM standard and lack of resources allocated towards protocol compliance. Moreover, issues of protocol compliance is often overlooked for lack of realization that parameter values are routinely improvised/modified locally at various sites. The inconsistencies in acquisition protocols can reduce SNR, statistical power, and in the worst case, may invalidate the results altogether. An open-source tool, mrQA was developed to automatically assess protocol compliance on standard dataset formats such as DICOM and BIDS, and to study the patterns of non-compliance in over 20 open neuroimaging datasets, including the large ABCD study. The results demonstrate that the lack of compliance is rather pervasive. The frequent sources of non-compliance include but are not limited to deviations in Repetition Time, Echo Time, Flip Angle, and Phase Encoding Direction. It was also observed that GE and Philips scanners exhibited higher rates of non-compliance relative to the Siemens scanners in the ABCD dataset. Continuous monitoring for protocol compliance is strongly recommended before any pre/post-processing, ideally right after the acquisition, to avoid the silent propagation of severe/subtle issues. Although, this study focuses on neuroimaging datasets, the proposed tool mrQA can work with any DICOM-based datasets.

Multivariate testing and effect size measures for batch effect evaluation in radiomic features

While precision medicine applications of radiomics analysis are promising, differences in image acquisition can cause "batch effects" that reduce reproducibility and affect downstream predictive analyses. Harmonization methods such as ComBat have been developed to correct these effects, but evaluation methods for quantifying batch effects are inconsistent. In this study, we propose the use of the multivariate statistical test PERMANOVA and the Robust Effect Size Index (RESI) to better quantify and characterize batch effects in radiomics data. We evaluate these methods in both simulated and real radiomics features extracted from full-field digital mammography (FFDM) data. PERMANOVA demonstrated higher power than standard univariate statistical testing, and RESI was able to interpretably quantify the effect size of site at extremely large sample sizes. These methods show promise as more powerful and interpretable methods for the detection and quantification of batch effects in radiomics studies.

The impact of the combat method on radiomics feature compensation and analysis of scanners from different manufacturers

BACKGROUND

This study investigated whether the Combat compensation method can remove the variability of radiomic features extracted from different scanners, while also examining its impact on the subsequent predictive performance of machine learning models.

MATERIALS AND METHODS

135 CT images of Credence Cartridge Radiomic phantoms were collected and screened from three scanners manufactured by Siemens, Philips, and GE. 100 radiomic features were extracted and 20 radiomic features were screened according to the Lasso regression method. The radiomic features extracted from the rubber and resin-filled regions in the cartridges were labeled into different categories for evaluating the performance of the machine learning model. Radiomics features were divided into three groups based on the different scanner manufacturers. The radiomic features were randomly divided into training and test sets with a ratio of 8:2. Five machine learning models (lasso, logistic regression, random forest, support vector machine, neural network) were employed to evaluate the impact of Combat on radiomic features. The variability among radiomic features were assessed using analysis of variance (ANOVA) and principal component analysis (PCA). Accuracy, precision, recall, and area under the receiver curve (AUC) were used as evaluation metrics for model classification.

RESULTS

The principal component and ANOVA analysis results show that the variability of different scanner manufacturers in radiomic features was removed (P˃0.05). After harmonization with the Combat algorithm, the distributions of radiomic features were aligned in terms of location and scale. The performance of machine learning models for classification improved, with the Random Forest model showing the most significant enhancement. The AUC value increased from 0.88 to 0.92.

CONCLUSIONS

The Combat algorithm has reduced variability in radiomic features from different scanners. In the phantom CT dataset, it appears that the machine learning model's classification performance may have improved after Combat harmonization. However, further investigation and validation are required to fully comprehend Combat's impact on radiomic features in medical imaging.

Gastro-Esophageal Cancer: Can Radiomic Parameters from Baseline 18F-FDG-PET/CT Predict the Development of Distant Metastatic Disease?

We aimed to determine if clinical parameters and radiomics combined with sarcopenia status derived from baseline 18F-FDG-PET/CT could predict developing metastatic disease and overall survival (OS) in gastroesophageal cancer (GEC). Patients referred for primary staging who underwent 18F-FDG-PET/CT from 2008 to 2019 were evaluated retrospectively. Overall, 243 GEC patients (mean age = 64) were enrolled. Clinical, histopathology, and sarcopenia data were obtained, and primary tumor radiomics features were extracted. For classification (early-stage vs. advanced disease), the association of the studied parameters was evaluated. Various clinical and radiomics models were developed and assessed. Accuracy and area under the curve (AUC) were calculated. For OS prediction, univariable and multivariable Cox analyses were performed. The best model included PET/CT radiomics features, clinical data, and sarcopenia score (accuracy = 80%; AUC = 88%). For OS prediction, various clinical, CT, and PET features entered the multivariable analysis. Three clinical factors (advanced disease, age ≥ 70 and ECOG ≥ 2), along with one CT-derived and one PET-derived radiomics feature, retained their significance. Overall, 18F-FDG PET/CT radiomics seems to have a potential added value in identifying GEC patients with advanced disease and may enhance the performance of baseline clinical parameters. These features may also have a prognostic value for OS, improving the decision-making for GEC patients.

All you need is data preparation: A systematic review of image harmonization techniques in Multi-center/device studies for medical support systems

BACKGROUND AND OBJECTIVES

Artificial intelligence (AI) models trained on multi-centric and multi-device studies can provide more robust insights and research findings compared to single-center studies. However, variability in acquisition protocols and equipment can introduce inconsistencies that hamper the effective pooling of multi-source datasets. This systematic review evaluates strategies for image harmonization, which standardizes appearances to enable reliable AI analysis of multi-source medical imaging.

METHODS

A literature search using PRISMA guidelines was conducted to identify relevant papers published between 2013 and 2023 analyzing multi-centric and multi-device medical imaging studies that utilized image harmonization approaches.

RESULTS

Common image harmonization techniques included grayscale normalization (improving classification accuracy by up to 24.42 %), resampling (increasing the percentage of robust radiomics features from 59.5 % to 89.25 %), and color normalization (enhancing AUC by up to 0.25 in external test sets). Initially, mathematical and statistical methods dominated, but machine and deep learning adoption has risen recently. Color imaging modalities like digital pathology and dermatology have remained prominent application areas, though harmonization efforts have expanded to diverse fields including radiology, nuclear medicine, and ultrasound imaging. In all the modalities covered by this review, image harmonization improved AI performance, with increasing of up to 24.42 % in classification accuracy and 47 % in segmentation Dice scores.

CONCLUSIONS

Continued progress in image harmonization represents a promising strategy for advancing healthcare by enabling large-scale, reliable analysis of integrated multi-source datasets using AI. Standardizing imaging data across clinical settings can help realize personalized, evidence-based care supported by data-driven technologies while mitigating biases associated with specific populations or acquisition protocols.

Imaging Genomics and Multiomics: A Guide for Beginners Starting Radiomics-Based Research

Radiomics uses advanced mathematical analysis of pixel-level information from radiologic images to extract existing information in traditional imaging algorithms. It is intended to find imaging biomarkers related to the genomics of tumors or disease patterns that improve medical care by advanced detection of tumor response patterns in tumors and to assess prognosis. Radiomics expands the paradigm of medical imaging to help with diagnosis, management of diseases and prognostication, leveraging image features by extracting information that can be used as imaging biomarkers to predict prognosis and response to treatment. Radiogenomics is an emerging area in radiomics that investigates the association between imaging characteristics and gene expression profiles. There are an increasing number of research publications using different radiomics approaches without a clear consensus on which method works best. We aim to describe the workflow of radiomics along with a guide of what to expect when starting a radiomics-based research project.

Investigation of a potential upstream harmonization based on image appearance matching to improve radiomics features robustness: a phantom study

Objective. Radiomics is a promising valuable analysis tool consisting in extracting quantitative information from medical images. However, the extracted radiomics features are too sensitive to variations in used image acquisition and reconstruction parameters. This limited robustness hinders the generalizable validity of radiomics-assisted models. Our aim is to investigate a possible harmonization strategy based on matching image quality to improve feature robustness.Approach.We acquired CT scans of a phantom with two scanners across different dose levels and percentages of Iterative Reconstruction algorithms. The detectability index was used as a comprehensive task-based image quality metric. A statistical analysis based on the Intraclass Correlation Coefficient was performed to determine if matching image quality/appearance could enhance the robustness of radiomics features extracted from the phantom images. Additionally, an Artificial Neural Network was trained on these features to automatically classify the scanner used for image acquisition.Main results.We found that the ICC of the features across protocols providing a similar detectability index improves with respect to the ICC of the features across protocols providing a different detectability index. This improvement was particularly noticeable in features relevant for distinguishing between scanners.Significance.This preliminary study demonstrates that a harmonization based on image quality/appearance matching could improve radiomics features robustness and heterogeneous protocols can be used to obtain a similar image appearance in terms of the detectability index. Thus protocols with a lower dose level could be selected to reduce the amount of radiation dose delivered to the patient and simultaneously obtain a more robust quantitative analysis.

Pretreatment MRI-Based Radiomics for Prediction of Rectal Cancer Outcome: A Discovery and Validation Study

RATIONALE AND OBJECTIVES

Accurate prediction of local recurrence or distant metastasis is critical for developing individualized therapies for locally advanced rectal cancer (LARC) patients after standard therapy. This study aims to develop and validate a multiparameter MRI-based radiomics signature (RS) for prognostic prediction in LARC patients receiving neoadjuvant chemoradiotherapy (nCRT) and total mesorectal excision (TME) and to explore the ability of RS for personalized survival risk stratification.

MATERIALS AND METHODS

In this multi-center study, 454 patients who received nCRT and TME and completed 3 years of follow-up participated. RS was constructed for prognostic prediction based on features extracted from pretreatment multiparameter MRI in a training cohort (TC; n = 298), which was tested in an internal validation cohort (IVC; n = 75) and further validated in an independent external validation cohort (EVC; n = 81). Furthermore, the ability of RS for personalized survival risk stratification was explored using the Kaplan-Meier survival curves.

RESULTS

The RS model showed satisfactory accuracy for prognostic prediction with AUCs of 0.83, 0.81 and 0.82 in the TC, IVC and EVC, respectively. In addition, RS helped to refine risk stratification for LARC patients on the basis of significantly different 3-year disease-free survival rates, independent of their pathological stage, pre-surgery CEA, and even treatment modality.

CONCLUSIONS

The proposed RS can be used not only to predict local recurrence or distant metastasis but also to serve as an effective postoperative survival risk stratification tool for clinicians to facilitate decision-making for LARC patients receiving standard treatment.

The Effect of Image Resampling on the Performance of Radiomics-Based Artificial Intelligence in Multicenter Prostate MRI

BACKGROUND

Single center MRI radiomics models are sensitive to data heterogeneity, limiting the diagnostic capabilities of current prostate cancer (PCa) radiomics models.

PURPOSE

To study the impact of image resampling on the diagnostic performance of radiomics in a multicenter prostate MRI setting.

STUDY TYPE

Retrospective.

POPULATION

Nine hundred thirty patients (nine centers, two vendors) with 737 eligible PCa lesions, randomly split into training (70%, N = 500), validation (10%, N = 89), and a held-out test set (20%, N = 148).

FIELD STRENGTH/SEQUENCE

1.5T and 3T scanners/T2-weighted imaging (T2W), diffusion-weighted imaging (DWI), and apparent diffusion coefficient maps.

ASSESSMENT

A total of 48 normalized radiomics datasets were created using various resampling methods, including different target resolutions (T2W: 0.35, 0.5, and 0.8 mm; DWI: 1.37, 2, and 2.5 mm), dimensionalities (2D/3D) and interpolation techniques (nearest neighbor, linear, Bspline and Blackman windowed-sinc). Each of the datasets was used to train a radiomics model to detect clinically relevant PCa (International Society of Urological Pathology grade ≥ 2). Baseline models were constructed using 2D and 3D datasets without image resampling. The resampling configurations with highest validation performance were evaluated in the test dataset and compared to the baseline models.

STATISTICAL TESTS

Area under the curve (AUC), DeLong test. The significance level used was 0.05.

RESULTS

The best 2D resampling model (T2W: Bspline and 0.5 mm resolution, DWI: nearest neighbor and 2 mm resolution) significantly outperformed the 2D baseline (AUC: 0.77 vs. 0.64). The best 3D resampling model (T2W: linear and 0.8 mm resolution, DWI: nearest neighbor and 2.5 mm resolution) significantly outperformed the 3D baseline (AUC: 0.79 vs. 0.67).

DATA CONCLUSION

Image resampling has a significant effect on the performance of multicenter radiomics artificial intelligence in prostate MRI. The recommended 2D resampling configuration is isotropic resampling with T2W at 0.5 mm (Bspline interpolation) and DWI at 2 mm (nearest neighbor interpolation). For the 3D radiomics, this work recommends isotropic resampling with T2W at 0.8 mm (linear interpolation) and DWI at 2.5 mm (nearest neighbor interpolation).

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 2.

Investigating the role of imaging factors in the variability of CT-based texture analysis metrics

OBJECTIVE

This study assesses the robustness of first-order radiomic texture features namely interquartile range (IQR), coefficient of variation (CV) and standard deviation (SD) derived from computed tomography (CT) images by varying dose, reconstruction algorithms and slice thickness using scans of a uniform water phantom, a commercial anthropomorphic liver phantom, and a human liver in-vivo.

MATERIALS AND METHODS

Scans were acquired on a 16 cm detector GE Revolution Apex Edition CT scanner with variations across three different nominal slice thicknesses: 0.625, 1.25, and 2.5 mm, three different dose levels: CTDIvol of 13.86 mGy for the standard dose, 40% reduced dose and 60% reduced dose and two different reconstruction algorithms: a deep learning image reconstruction (DLIR-high) algorithm and a hybrid iterative reconstruction (IR) algorithm ASiR-V50% (AV50) were explored, varying one at a time. To assess the effect of non-linear modifications of images by AV50 and DLIR-high, images of the water phantom were also reconstructed using filtered back projection (FBP). Quantitative measures of IQR, CV and SD were extracted from twelve pre-selected, circular (1 cm diameter) regions of interest (ROIs) capturing different texture patterns across all scans.

RESULTS

Across all scans, imaging, and reconstruction settings, CV, IQR and SD were observed to increase with reduction in dose and slice thickness. An exception to this observation was found when using FBP reconstruction. Lower values of CV, IQR and SD were observed in DLIR-high reconstructions compared to AV50 and FBP. The Poisson statistics were more stringently noted in FBP than DLIR-high and AV50, due to the non-linear nature of the latter two algorithms.

CONCLUSION

Variation in image noise due to dose reduction algorithms, tube current, and slice thickness show a consistent trend across phantom and patient scans. Prospective evaluation across multiple centers, scanners and imaging protocols is needed for establishing quality assurance standards of radiomics.

Delta-radiomics features for predicting the major pathological response to neoadjuvant chemoimmunotherapy in non-small cell lung cancer

OBJECTIVES

To investigate if delta-radiomics features have the potential to predict the major pathological response (MPR) to neoadjuvant chemoimmunotherapy in non-small cell lung cancer (NSCLC) patients.

METHODS

Two hundred six stage IIA-IIIB NSCLC patients from three institutions (Database1 = 164; Database2 = 21; Database3 = 21) who received neoadjuvant chemoimmunotherapy and surgery were included. Patients in Database1 were randomly assigned to the training dataset and test dataset, with a ratio of 0.7:0.3. Patients in Database2 and Database3 were used as two independent external validation datasets. Contrast-enhanced CT scans were obtained at baseline and before surgery. The delta-radiomics features were defined as the relative net change of radiomics features between baseline and preoperative. The delta-radiomics model and pre-treatment radiomics model were established. The performance of Immune-Related Response Evaluation Criteria in Solid Tumors (iRECIST) for predicting MPR was also evaluated.

RESULTS

Half of the patients (106/206, 51.5%) showed MPR after neoadjuvant chemoimmunotherapy. For predicting MPR, the delta-radiomics model achieved a satisfying area under the curves (AUCs) values of 0.768, 0.732, 0.833, and 0.716 in the training, test, and two external validation databases, respectively, which showed a superior predictive performance than the pre-treatment radiomics model (0.644, 0.616, 0.475, and 0.608). Compared with iRECIST criteria (0.624, 0.572, 0.650, and 0.466), a mixed model that combines delta-radiomics features and iRECIST had higher AUC values for MPR prediction of 0.777, 0.761, 0.850, and 0.670 in four sets.

CONCLUSION

The delta-radiomics model demonstrated superior diagnostic performance compared to pre-treatment radiomics model and iRECIST criteria in predicting MPR preoperatively in neoadjuvant chemoimmunotherapy for stage II-III NSCLC.

CLINICAL RELEVANCE STATEMENT

Delta-radiomics features based on the relative net change of radiomics features between baseline and preoperative CT scans serve a vital support tool in accurately identifying responses to neoadjuvant chemoimmunotherapy, which can help physicians make more appropriate treatment decisions.

KEY POINTS

• The performances of pre-treatment radiomics model and iRECIST model in predicting major pathological response of neoadjuvant chemoimmunotherapy were unsatisfactory. • The delta-radiomics features based on relative net change of radiomics features between baseline and preoperative CT scans may be used as a noninvasive biomarker for predicting major pathological response of neoadjuvant chemoimmunotherapy. • Combining delta-radiomics features and iRECIST can further improve the predictive performance of responses to neoadjuvant chemoimmunotherapy.

Robustness of radiomic features in healthy abdominal parenchyma of patients with repeated examinations on dual-layer dual-energy CT

OBJECTIVES

Robustness of radiomic features in physiological tissue is an important prerequisite for quantitative analysis of tumor biology and response assessment. In contrast to previous studies which focused on different tumors with mostly short scan-re-scan intervals, this study aimed to evaluate the robustness of radiomic features in cancer-free patients and over a clinically encountered inter-scan interval.

MATERIALS AND METHODS

Patients without visible tumor burden who underwent at least two portal-venous phase dual energy CT examinations of the abdomen between May 2016 and January 2020 were included, while macroscopic tumor burden was excluded based upon follow-up imaging for all patients (≥3 months). Further, patients were excluded if no follow-up imaging was available, or if the CT protocol showed deviations between repeated examinations. Circular regions of interest were placed and proofread by two board-certified radiologists (4 years and 5 years experience) within the liver (segments 3 and 6), the psoas muscle (left and right), the pancreatic head, and the spleen to obtain radiomic features from normal-appearing organ parenchyma using PyRadiomics. Radiomic feature robustness was tested using the concordance correlation coefficient with a threshold of 0.75 considered indicative for deeming a feature robust.

RESULTS

In total, 160 patients with 480 repeated abdominal CT examinations (range: 2-4 per patient) were retrospectively included in this single-center, IRB-approved study. Considering all organs and feature categories, only 4.58 % (25/546) of all features were robust with the highest rate being found in the first order feature category (20.37 %, 22/108). Other feature categories (grey level co-occurrence matrix, grey level dependence matrix, grey level run length matrix, grey level size zone matrix, and neighborhood gray-tone difference matrix) yielded an overall low percentage of robust features (range: 0.00 %-1.19 %). A subgroup analysis revealed the reconstructed field of view and the X-ray tube current as determinants of feature robustness (significant differences in subgroups for all organs, p < 0.001) as well as the size of the region of interest (no significant difference for the pancreatic head with p = 0.135, significant difference with p < 0.001 for all other organs).

CONCLUSION

Radiomic feature robustness obtained from cancer-free subjects with repeated examinations using a consistent protocol and CT scanner was limited, with first order features yielding the highest proportion of robust features.

The Image Biomarker Standardization Initiative: Standardized Convolutional Filters for Reproducible Radiomics and Enhanced Clinical Insights

Filters are commonly used to enhance specific structures and patterns in images, such as vessels or peritumoral regions, to enable clinical insights beyond the visible image using radiomics. However, their lack of standardization restricts reproducibility and clinical translation of radiomics decision support tools. In this special report, teams of researchers who developed radiomics software participated in a three-phase study (September 2020 to December 2022) to establish a standardized set of filters. The first two phases focused on finding reference filtered images and reference feature values for commonly used convolutional filters: mean, Laplacian of Gaussian, Laws and Gabor kernels, separable and nonseparable wavelets (including decomposed forms), and Riesz transformations. In the first phase, 15 teams used digital phantoms to establish 33 reference filtered images of 36 filter configurations. In phase 2, 11 teams used a chest CT image to derive reference values for 323 of 396 features computed from filtered images using 22 filter and image processing configurations. Reference filtered images and feature values for Riesz transformations were not established. Reproducibility of standardized convolutional filters was validated on a public data set of multimodal imaging (CT, fluorodeoxyglucose PET, and T1-weighted MRI) in 51 patients with soft-tissue sarcoma. At validation, reproducibility of 486 features computed from filtered images using nine configurations × three imaging modalities was assessed using the lower bounds of 95% CIs of intraclass correlation coefficients. Out of 486 features, 458 were found to be reproducible across nine teams with lower bounds of 95% CIs of intraclass correlation coefficients greater than 0.75. In conclusion, eight filter types were standardized with reference filtered images and reference feature values for verifying and calibrating radiomics software packages. A web-based tool is available for compliance checking.

Strategies for Implementing Machine Learning Algorithms in the Clinical Practice of Radiology

Despite recent advancements in machine learning (ML) applications in health care, there have been few benefits and improvements to clinical medicine in the hospital setting. To facilitate clinical adaptation of methods in ML, this review proposes a standardized framework for the step-by-step implementation of artificial intelligence into the clinical practice of radiology that focuses on three key components: problem identification, stakeholder alignment, and pipeline integration. A review of the recent literature and empirical evidence in radiologic imaging applications justifies this approach and offers a discussion on structuring implementation efforts to help other hospital practices leverage ML to improve patient care. Clinical trial registration no. 04242667 © RSNA, 2024 Supplemental material is available for this article.

Radiomics Harmonization in Ultrasound Images for Cervical Cancer Lymph Node Metastasis Prediction Using Cycle-GAN

Background: Ultrasound (US) based radiomics is susceptible to variations in scanners, sonographers. Objective: To retrospectively investigate the feasibility of an adapted cycle generative adversarial networks (CycleGAN) in the style transfer to improve US based radiomics in the prediction of lymph node metastasis (LNM) with images from multiple scanners for patients with early cervical cancer (ECC). Methods: The CycleGAN was firstly trained to transfer paired US phantom images from one US device to another one; the model was then further trained and tested with clinical US images of ECC by transferring images from four US devices to one specific device; finally, the adapted model was tested with its effects on the radiomics feature harmonization and accuracy of LNM prediction in US based radiomics for ECC patients. Results: Phantom study demonstrated an increased radiomics harmonization using CycleGAN with an average Pearson correlation coefficient of 0.60 and 0.81 for radiomics features extracted from original and generated images in correlation with the target phantom images, respectively. Additionally, the image quality metric Peak Signal-to-Noise Ratio (PSNR) was increased from 11.18 for the original images to 15.45 for the generated image. Clinical US images of 169 ECC patients were enrolled for style transfer model training and validation. The area under curve (AUC) of LNM prediction radiomics models with features extracted from generated images of different style transfer models ranged from 0.73 to 0.85. The AUC was improved from 0.78 with features extracted from original images to 0.85 with style transferred images. Conclusions: The adapted CycleGAN network is able to increase the radiomics feature harmonization for images from different ultrasound equipment based on image domain and improve the LNM prediction accuracy for ECC.