A Chebyshev Confidence Guided Source-Free Domain Adaptation Framework for Medical Image Segmentation.

Source-free domain adaptation (SFDA) aims to adapt models trained on a labeled source domain to an unlabeled target domain without access to source data. In medical imaging scenarios, the practical significance of SFDA methods has been emphasized due to data heterogeneity and privacy concerns. Recent state-of-the-art SFDA methods primarily rely on self-training based on pseudo-labels (PLs). Unfortunately, the accuracy of PLs may deteriorate due to domain shift, thus limiting the effectiveness of the adaptation process. To address this issue, we propose a Chebyshev confidence guided SFDA framework to accurately assess the reliability of PLs and generate self-improving PLs for self-training. The Chebyshev confidence is estimated by calculating the probability lower bound of PL confidence, given the prediction and the corresponding uncertainty. Leveraging the Chebyshev confidence, we introduce two confidence-guided denoising methods: direct denoising and prototypical denoising. Additionally, we propose a novel teacher-student joint training scheme (TJTS) that incorporates a confidence weighting module to iteratively improve PLs' accuracy. The TJTS, in collaboration with the denoising methods, effectively prevents the propagation of noise and enhances the accuracy of PLs. Extensive experiments in diverse domain scenarios validate the effectiveness of our proposed framework and establish its superiority over state-of-the-art SFDA methods. Our paper contributes to the field of SFDA by providing a novel approach for precisely estimating the reliability of PLs and a framework for obtaining high-quality PLs, resulting in improved adaptation performance.

The Value of LI-RADS and Radiomic Features from MRI for Predicting Microvascular Invasion in Hepatocellular Carcinoma within 5 cm.

RATIONALE AND OBJECTIVES: To explore and compare the performance of LI-RADS® and radiomics from multiparametric MRI in predicting microvascular invasion (MVI) preoperatively in patients with solitary hepatocellular carcinoma (HCC)< 5 cm. METHODS: We enrolled 143 patients with pathologically proven HCC and randomly stratified them into training (n = 100) and internal validation (n = 43) cohorts. Besides, 53 patients were enrolled to constitute an independent test cohort. Clinical factors and imaging features, including LI-RADS and three other features (non-smooth margin, incomplete capsule, and two-trait predictor of venous invasion), were reviewed and analyzed. Radiomic features from four MRI sequences were extracted. The independent clinic-imaging (clinical) and radiomics model for MVI-prediction were constructed by logistic regression and AdaBoost respectively. And the clinic-radiomics combined model was further constructed by logistic regression. We assessed the model discrimination, calibration, and clinical usefulness by using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision-curve analysis respectively. RESULTS: Incomplete tumor capsule, corona enhancement, and radiomic features were related to MVI in solitary HCC＜5 cm. The clinical model achieved AUC of 0.694/0.661 (training/internal validation). The single-sequence-based radiomic model's AUCs were 0.753-0.843/0.698-0.767 (training/internal validation). The combination model exhibited superior diagnostic performance to the clinical model (AUC: 0.895/0.848 [training/ internal validation]) and yielded an AUC of 0.858 in an independent test cohort. CONCLUSION: Incomplete tumor capsule and corona enhancement on preoperative MRI were significantly related to MVI in solitary HCC＜5 cm. Multiple-sequence radiomic features potentially improve MVI-prediction-model performance, which could potentially help determining HCC's appropriate therapy.

Prediction of lymph node metastasis in stage T1-2 rectal cancers with MRI-based deep learning.

OBJECTIVES: This study aimed to investigate whether a deep learning (DL) model based on preoperative MR images of primary tumors can predict lymph node metastasis (LNM) in patients with stage T1-2 rectal cancer. METHODS: In this retrospective study, patients with stage T1-2 rectal cancer who underwent preoperative MRI between October 2013 and March 2021 were included and assigned to the training, validation, and test sets. Four two-dimensional and three-dimensional (3D) residual networks (ResNet18, ResNet50, ResNet101, and ResNet152) were trained and tested on T2-weighted images to identify patients with LNM. Three radiologists independently assessed LN status on MRI, and diagnostic outcomes were compared with the DL model. Predictive performance was assessed with AUC and compared using the Delong method. RESULTS: In total, 611 patients were evaluated (444 training, 81 validation, and 86 test). The AUCs of the eight DL models ranged from 0.80 (95% confidence interval [CI]: 0.75, 0.85) to 0.89 (95% CI: 0.85, 0.92) in the training set and from 0.77 (95% CI: 0.62, 0.92) to 0.89 (95% CI: 0.76, 1.00) in the validation set. The ResNet101 model based on 3D network architecture achieved the best performance in predicting LNM in the test set, with an AUC of 0.79 (95% CI: 0.70, 0.89) that was significantly greater than that of the pooled readers (AUC, 0.54 [95% CI: 0.48, 0.60]; p < 0.001). CONCLUSION: The DL model based on preoperative MR images of primary tumors outperformed radiologists in predicting LNM in patients with stage T1-2 rectal cancer. KEY POINTS: • Deep learning (DL) models with different network frameworks showed different diagnostic performance for predicting lymph node metastasis (LNM) in patients with stage T1-2 rectal cancer. • The ResNet101 model based on 3D network architecture achieved the best performance in predicting LNM in the test set. • The DL model based on preoperative MR images outperformed radiologists in predicting LNM in patients with stage T1-2 rectal cancer.

Performance of Radiomics Models Based on Coronary Computed Tomography Angiography in Predicting The Risk of Major Adverse Cardiovascular Events Within 3 Years: A Comparison Between the Pericoronary Adipose Tissue Model and the Epicardial Adipose Tissue Model.

RATIONALE AND OBJECTIVES: To compare the prediction performance of the epicardial adipose tissue (EAT) and pericoronary adipose tissue (PCAT) radiomics models based on coronary computed tomography angiography for major adverse cardiovascular events (MACE) within 3 years. MATERIALS AND METHODS: Our study included 288 patients (144 with MACE and 144 without MACE within 3 years) by matching age, gender, body mass index, and medication intake. Patients were randomly assigned either to the training (n = 201) or validation cohort (n = 87). A total of 184 radiomics features were extracted from EAT and PCAT images. Spearman's rank correlation coefficient and the gradient boosting decision tree algorithm were performed for feature selection. Five models were established based on PCAT or EAT radiomics features and clinical factors, including PCAT, EAT, clinical, PCAT-clinical, and EAT-clinical model (MPCAT, MEAT, Mclinical, MPCAT-clinical, and MEAT-clinical). Receiver operating characteristic curves, calibration curves, and the decision curve analysis were plotted to evaluate the model performance. RESULTS: The MPCAT achieved an area under the curve (AUC) of 0.703 in the validation cohort, which was better than MEAT with AUC of 0.538. The MPCAT-clinical showed better performance (AUC = 0.781) in predicting MACE than the Mclinical (AUC = 0.748) or MEAT-clinical (AUC = 0.745). CONCLUSION: Our results showed that the PCAT was better than the EAT in both single modality and combined models, and the MPCAT-clinical had the most significant clinical value in predicting the occurrence of MACE within 3 years.

Diagnostic value of whole-tumor apparent diffusion coefficient map radiomics analysis in predicting early recurrence of solitary hepatocellular carcinoma ≤ 5 cm.

PURPOSE: To evaluate the role of whole-tumor radiomics analysis of apparent diffusion coefficient (ADC) maps in predicting early recurrence (ER) of solitary hepatocellular carcinoma (HCC) ≤ 5 cm and compare the diagnostic efficiency of whole-tumor and single-slice ADC measurements. METHODS: One hundred and seventy patients with primary HCC were randomly divided into the training set (n = 119) and the test set (n = 51). The diagnostic efficiency was compared between the whole-tumor and single-slice ADC measurements. The clinical-radiological model was established by selected significant clinical characteristics and qualitative imaging features. The radiomics model was constructed using the least absolute shrinkage and selection operator (LASSO) logistic regression algorithm. The significant clinical-radiological risk factors and radiomics features were integrated to develop the combined model. Receiver operating characteristic (ROC) curves were used for evaluating the predictive performance. RESULTS: Cirrhosis, age, and albumin were significantly associated with ER in the clinical-radiological model selected by the random forest classifier. The diagnostic efficiency of the whole-tumor ADC measurements was slight higher than that of the single-slice (AUC = 0.602 and 0.586, respectively). The clinical-radiological model (AUC = 0.84 and 0.82 in the training and test sets, respectively) showed better diagnostic performance than the radiomics model (AUC = 0.70 and 0.69 in the training and test sets, respectively) in predicting ER. The combined model showed optimal predictive performance with the highest AUC values of 0.88 and 0.85 in the training and test sets, respectively. CONCLUSIONS: The whole-tumor ADC measurements performed better than the single-slice ADC measurements. The clinical-radiological model performed better than the radiomics model for predicting ER in patients with solitary HCC ≤ 5 cm.

Identification of patients with acute myocardial infarction based on coronary CT angiography: the value of pericoronary adipose tissue radiomics.

OBJECTIVE: To determine whether radiomics analysis of pericoronary adipose tissue (PCAT) captured by coronary computed tomography angiography (CCTA) could discriminate acute myocardial infarction (MI) from unstable angina (UA). METHODS: In a single-center retrospective case-control study, patients with acute MI (n = 105) were matched to patients with UA (n = 105) and all patients were randomly divided into training and validation cohorts with a ratio of 7:3. Fat attenuation index (FAI) and PCAT radiomics features selected by Max-Relevance and Min-Redundancy (mRMR) and least absolute shrinkage and selection operator (LASSO) around the proximal three major epicardial coronary vessels (LAD [left anterior descending artery], LCx [left circumflex artery], and RCA [right coronary artery]) were used to build logistic regression models. Finally, a FAI model, three radiomics models of PCAT (LAD, LCx, and RCA), and a combined model that used the scores of these independent models were constructed. The performance of the models was evaluated by identification, calibration, and clinical application. RESULTS: In training and validation cohorts, compared with the FAI model (AUC = 0.53, 0.50), the combined model achieved superior performance (AUC = 0.97, 0.95) while there was a significant difference of AUC between two models (p < 0.05). The calibration curves of the combined model demonstrated the smallest Brier score loss. Decision curve analysis suggested that the combined model provided higher clinical benefit than the FAI model. CONCLUSIONS: The CCTA-based radiomics phenotype of PCAT outperforms the FAI model in discriminating acute MI from UA. The combination of PCAT radiomics and FAI could further enhance the performance of acute MI identification. KEY POINTS: • Fat attenuation index based on CCTA can detect inflammation-induced changes in the ratio of lipid to aqueous phase in pericoronary adipose tissue. • Fat attenuation index cannot distinguish acute MI patients from UA patients, suggesting that the two groups have the same degree of ratio of lipid to aqueous phase in pericoronary adipose tissue. • Radiomics features of PCAT have the potential to distinguish acute MI patients from UA patients.