BACKGROUND & AIMS:Predicting retreatment response of viable tumours after transarterial chemoembolization (TACE) is critical for personalised treatment and prognosis assessment in hepatocellular carcinoma (HCC). We aimed to develop an MRI-based prediction model for viable tumour response. METHODS:This retrospective multicentre study included patients with HCC who presented with viable tumours 1 month after initial TACE between February 2015 and October 2022. In addition, data from a prospective clinical trial were reanalyzed as an external validation cohort. All patients underwent contrast-enhanced MRI at baseline (for initial HCC assessment), at 1 month (for evaluation of viable tumour characteristics) and at 6 months (for assessment of treatment response). The training set (n = 167) and test set (n = 59) were used to build lesion- and patient-level models predicting retreatment response at 6 months via logistic regression. Risk groups were stratified by the Youden index and analysed across retreatment subgroups. RESULTS:The Viable tumour Imaging Traits for Assessing Likelihood of response (VITAL) model incorporated four features: diffusion restriction (p = 0.050; 1 point), peritumoral hyperenhancement (p = 0.002; 1 point), heterogeneity (p < 0.001; 2 points) and mural nodule (p = 0.004; 2 points if absent). The model achieved AUCs of 0.821 (95% CI: 0.760-0.882) and 0.733 (95% CI: 0.602-0.865) in training and test cohorts. The patient-level VITAL-P Score was calculated as: 1 × (largest viable tumour size + viable tumour number) + 3 × VITAL Score. High-risk patients (VITAL-p ≥ 16) had significantly shorter overall survival (OS) in the overall cohort (p = 0.021) and shorter progression-free survival (PFS)and OS compared with low-risk patients in the locoregional therapy subgroup (p = 0.021 and 0.002, respectively). CONCLUSIONS:The prediction model based on imaging features of post-TACE viable HCCs demonstrates strong predictive power for tumour retreatment response at 6 months and correlates well with prognosis.

Background:The early diagnosis of abnormalities in transplanted kidney function is crucial for timely intervention in transplant patients. Non-invasive tests play a key role in this process. This study aimed to explore the value of blood oxygenation level-dependent (BOLD) and arterial spin labeling (ASL) techniques, based on magnetic resonance angiography (MRA) examination, in evaluating early renal allograft function. Methods:A total of 68 consecutive renal transplant recipients were prospectively recruited. Of them, 10 were excluded due to magnetic resonance imaging (MRI) contraindications, hydronephrosis, and renal artery stenosis. Finally, 58 patients were included. The recipients were separated into three groups based on their estimated glomerular filtration rate (eGFR): Group A, recipients with good renal allograft function (eGFR ≥60 mL/min/1.73 m2); Group B, recipients with mild-to-moderate impaired renal allograft function (30≤ eGFR <60 mL/min/1.73 m2); Group C, recipients with severe renal allograft function (eGFR <30 mL/min/1.73 m2). Some patients underwent biopsy. All patients underwent ASL, BOLD, and renal-MRA to assess the anastomotic status of the grafted renal artery and to analyze renal blood flow (RBF) and the apparent relaxation rate (R2*). Results:A total of 58 patients (Group A, 29 cases; Group B, 18 cases; and Group C, 11 cases) were included in this study. Groups B and C presented with significantly decreased RBF as compared with Group A (259.74±47.52 vs. 166.50±19.79 and 112.76±32.08 mL/100 g/min). R2* decreased in Group B (cortical/medullary: 10.503±1.136/11.609±1.665 sec-1) and Group C (cortical/medullary: 9.471±0.997/10.785±1.114 sec-1), compared with Group A (cortical/medullary: 10.933±0.996/12.689±1.348 sec-1). Correlation analysis revealed that cortical RBF, cortical R2*, and medullary R2* were positively correlated with eGFR (r=0.877, 0.536, and 0.359, respectively). The higher area under the curve (AUC) of BOLD and ASL for distinguishing Group A from Group B, Group B from Group C, and Group A from Group C were 0.973 [95% confidence interval (CI): 0.936-1.000; P<0.001], 0.914 (95% CI: 0.753-1.000; P<0.001), and 0.994 (95% CI: 0.977-1.000; P<0.001), respectively, exceeding the performance of BOLD alone. Conclusions:BOLD and ASL can evaluate the different functional transplanted kidneys' oxygenation status and perfusion level. ASL demonstrates superior diagnostic efficacy compared to BOLD. BOLD combined with ASL has high value in identifying different transplanted kidney functions in the early stage.

Macrotrabecular-massive (MTM+) hepatocellular carcinoma (HCC) is associated with poor prognosis and early recurrence. We developed and validated a nomogram integrating magnetic resonance imaging LI-RADS features with deep learning (DL) habitat radiomics for preoperative prediction of MTM + HCC and stratifying patients according to recurrence-free survival (RFS). In this retrospective multicenter study, 607 patients with early-stage HCC who underwent curative-intent surgical resection (mean age ± standard deviation, 59.6 ± 10.6 years; 474 males, 133 females) were divided into the training (n = 304), internal validation (n = 131), and external (n = 172) test sets. Liver Imaging Reporting and Data System (LI-RADS) features, LI-RADS categorization, and clinical features were analyzed. Finally, a nomogram integrating LI-RADS features, habitat radiomics, and DL features was developed using multivariate logistic regression. Multivariable Cox regression analyses were performed to identify independent prognostic factors. At multivariable analysis, habitat radiomics score, DL score, alpha-fetoprotein (AFP), alanine transferase (ALT), and fat mass were independent predictors of MTM + HCC. The DL habitat radiomics nomogram demonstrated powerful performance with areas under the receiver operating characteristic curve (AUCs) of 0.897 (95

Aim: To develop a deep learning radiomics (DLR) model based on Gd-EOB-DTPA- enhanced magnetic resonance imaging for preoperative prediction of dual-phenotype hepatocellular carcinoma (DPHCC) and patient prognosis. Methods: This study included 331 patients from two centers: 228 in the training cohort (center I, 93 DPHCC, 135 non-DPHCC) and 103 in the external validation cohort (center II, 47 DPHCC, 56 non-DPHCC). Conventional radiological features were analyzed to build a radiological model. In total, 1,316 radiomics features and 768 deep learning (DL) features were extracted from the volumes of interest in arterial phase (AP), portal venous phase (PP), and hepatobiliary phase images, respectively. Three machine learning classifiers were applied to develop radiomics, DL, and DLR models using single-phase and combined-phase (CP) data for DPHCC prediction. Recurrence-free survival (RFS) was assessed by the Kaplan-Meier method. Results: In the external validation cohort, rim AP hyperenhancement and non-smooth tumor margins were independent predictors of DPHCC (both P < 0.05), yielding a radiological model with an area under the curve (AUC) of 0.607 [95% confidence interval (CI), 0.511-0.704]. Among radiomics and DL models, the PP-DLR model achieved the highest discrimination (AUC: 0.805; 95%CI, 0.722-0.887), outperforming the radiological model, PP-Radiomics (AUC: 0.714; 95%CI, 0.615-0.813), CP-Radiomics (AUC: 0.639; 95%CI, 0.532-0.746), and CP-DLR models (AUC: 0.650; 95%CI, 0.543-0.758) (all P < 0.05). Patients classified as DPHCC by the PP-DLR model had significantly shorter median RFS than those classified as non-DPHCC (P < 0.05). Conclusion: The PP-DLR model, which integrates radiomics and DL features, may aid in the preoperative prediction of DPHCC. DPHCC may be associated with poorer postoperative outcomes.

Early and accurate diagnosis of focal liver lesions (FLLs) is crucial for guiding treatment strategies and improving patient outcomes. However, FLL diagnosis is often limited by the suboptimal sensitivity of ultrasound, as well as the high cost, gadolinium burden, and time-intensive acquisition and interpretation characteristic of dynamic contrast-enhanced MRI. To address these challenges, we introduce Plain MRI Recognition and Interpretable System for hepatic Malignancy (PRISM), a novel deep learning framework designed for automated, gadolinium-free FLL diagnosis. The framework was developed and validated using a large-scale, multicenter cohort of 12,823 patients from 9 institutions, incorporating multicenter retrospective data and single-center prospective data for comprehensive lesion segmentation and classification. PRISM employs a three-stage paradigm using noninvasive non-contrast MRI (NC-MRI), beginning with automated FLL detection and segmentation, followed by benign-malignant classification, and culminating in a precise four-subtype classification model for suspicious malignant findings. The binary and four-subtype classification models demonstrated exceptional performance, achieving mean accuracies of 0.981 and 0.859 across the five test cohorts, respectively. A multireader multicase study involving 13 radiologists of varying experience levels demonstrated that AI assistance realized significant gains in diagnostic accuracy for both benign-malignant (Protocol B vs. A, 4.4% increase) and four malignant subtype classification (Protocol E vs. C, 12.1% increase). These improvements were accompanied by 30.5% (10.9 s, 95% CI: 9.8–12.0) and 66.6% (61.8 s, 95% CI: 58.9–64.7) reductions in interpretation time, respectively. Furthermore, a single-center prospective study of 1,147 consecutive patients demonstrated PRISM’s utility in diagnostic triage, identifying 79.3% of the cohort as low-risk with a negative predictive value of 99.1%. By providing expert-level interpretable diagnosis on NC-MRI, PRISM offers a viable, non-invasive pathway to streamline the clinical diagnostic workflow for liver malignancies.

BACKGROUND AND OBJECTIVES:To develop two distinct models for predicting microvascular invasion (MVI) and vessels encapsulating tumor clusters (VETC) based on habitat imaging, and to integrate these models for prognosis assessment. METHODS:In this multicenter retrospective study, patients from two different institutions were enrolled and categorized for MVI (n=295) and VETC (n=276) prediction. Tumor and peritumoral regions on hepatobiliary phase images were segmented into subregions, from which all relevant features were extracted. The MVI and VETC predictive models were constructed by analyzing these features using various machine learning algorithms, and classifying patients into high-risk and low-risk groups. Cox regression analysis was utilized to identify risk factors for early recurrence. RESULT:The MVI and VETC prediction models demonstrated excellent performance in both the training and external validation cohorts (AUC: 0.961 and 0.838 for MVI; 0.931 and 0.820 for VETC). Based on model predictions, patients were classified into high-risk group (High-risk MVI/ High-risk VETC), medium-risk group (High-risk MVI/Low-risk VETC or Low-risk MVI/High-risk VETC), and low-risk group (Low-risk MVI/Low-risk VETC). Multivariable Cox regression analysis revealed that risk group, number of tumors, and gender were independent predictors of early recurrence. CONCLUSION:Models based on habitat imaging can be used for the preoperative, noninvasive prediction of MVI and VETC, offering valuable stratification and diagnostic insights for HCC patients.

Background:The optimal management strategy for end-stage renal disease is renal transplantation, graft function must be monitored regularly postoperatively. This cross-sectional study aimed to explore the value of combining functional magnetic resonance imaging (MRI) parameters with laboratory parameters in assessing chronic allograft dysfunction (CAD), and to compare whether a combined approach was superior to single-parameter indicators. Methods:A total of 86 subjects were enrolled in the study, of whom, 20 had stable renal function, and 66 had biopsy-confirmed CAD. Imaging was performed on a 1.5-T MRI system using T2-weighted imaging, arterial spin labeling (ASL), and diffusion tensor imaging (DTI). The serum creatinine, estimated glomerular filtration rate (eGFR), 24-hour urinary protein (24hUP), renal blood flow (RBF), and fractional anisotropy (FA) values of the subjects were measured. Correlation analysis was applied to assess MRI parameters' association with eGFR, while receiver operating characteristic (ROC) curves were used to evaluate the diagnostic efficacy of fMRI parameters and clinical parameters for CAD. Results:The subjects were categorized into CAD groups based on their eGFR levels. The control group had higher renal RBF [277.69±67.17 vs. 138.60 (99.54-193.51)] and FA values [cortex: 0.16 (0.14-0.16) vs. 0.13 (0.11-0.16); medulla: 0.32±0.06 vs. 0.24 (0.20-0.29)] than the CAD group (P<0.01). Cortical RBF decreased progressively across the CAD subgroups [group 1 (mild: 213.33±67.07) > group 2 (moderate: 151.14±53.21) > group 3 (severe: 92.89±35.62); all P<0.05]. Similarly, there was a gradual decrease in medullary FA across the CAD subgroups [group 1: 0.29±0.04; group 2: 0.24 (0.19-0.29); group 3: 0.20±0.06]. However, no statistically significant difference was found in medullary FA between groups 2 and 3 (P=0.102). The correlation analysis showed that cortical RBF and medullary FA were positively correlated with the eGFR in the CAD group (r=0.604, P<0.001; r=0.574, P<0.001). The combined RBF, medullary FA, 24hUP, and eGFR model (RBF-FA-24hUP-eGFR) had an area under the curve (AUC) of 0.95 [95% confidence interval (CI): 0.91-1.00], which was significantly better than the AUCs of the single indicators of 24hUP and medullary FA (AUC =0.78, 95% CI: 0.68-0.88; AUC =0.79, 95% CI: 0.69-0.89, P<0.05). Further, the combined RBF, medullary FA, and, 24hUP model (RBF-FA-24hUP) was significantly superior to single 24hUP in differentiating among the subgroups (all P<0.05). In the CAD subgroups, while the performance of RBF on its own was close to that of the RBF-FA-24hUP model, the AUC of the combined model showed an increasing trend compared with RBF. Notably, the RBF-FA-24hUP model (AUC =0.86, 95% CI: 0.76-0.97; P<0.001) also surpassed medullary FA alone (AUC =0.69, 95% CI: 0.54-0.85; P=0.023) in distinguishing between the subjects in group 2 and group 3 (P<0.05). Conclusions:In this study, two multiparametric MRI models (RBF-FA-24hUP-eGFR and RBF-FA-24hUP) were developed and shown to be superior to non-invasive CAD assessment tools. These models outperformed conventional single-parameter methods in diagnosis and moderate-to-severe subgroup stratification. To a certain extent, these models could prevent unnecessary puncture biopsies, and reduce the occurrence of complications such as bleeding and infection. RBF in particular and FA showed utility as non-invasive biomarkers for CAD and risk stratification.

ObjectivesTo develop a non-invasive model for the preoperative prediction of Cytokeratin 19 (CK19) expression in hepatocellular carcinoma (HCC) based on clinical, radiologic, habitat radiomics, and deep learning features using gadoxetic acid-enhanced MRI, and to assess its utility for RFS risk stratification.MethodsIn this retrospective study, 539 patients with HCC from two hospitals were divided into training (n = 266), internal (n = 114), and external (n = 159) test sets. Univariable and multivariable logistic regression analyses were conducted on clinical and radiologic features to develop a clinical-radiologic model. Habitat radiomics and deep learning (DL) features were extracted and selected to develop the Habitat and DL models, respectively. The DL-HR nomogram model incorporating clinical, radiologic, habitat radiomics, and deep learning features was developed and evaluated. The Kaplan-Meier survival analysis assessed recurrence-free survival (RFS) in the CK19-positive (CK19+) and CK19-negative (CK19-) patients.ResultsAFP level and arterial phase (AP) enhancement were identified as independent predictors of CK19 expression. The DL-HR nomogram model showed superior performance compared to the clinical-radiologic model in both internal and external test sets (all P < 0.05). The AUCs of the DL-HR nomogram and clinical-radiologic models were 0.794 [95% CI: 0.708-0.864] vs. 0.615 [95% CI: 0.520-0.705] for the internal test set and 0.744 [95% CI: 0.669-0.810] vs. 0.600 [95% CI: 0.520-0.677] for the external test set, respectively. RFS was significantly different between the DL-HR nomogram model-predicted CK19+ and CK19- HCC patients across all sets (all P < 0.05).ConclusionsThe DL-HR nomogram model integrating clinical, radiologic, habitat radiomics, and deep learning features effectively predicted the CK19 expression and served as an effective tool for RFS risk stratification in HCC.

Background:Although both magnetic resonance (MR) diffusion tensor imaging (DTI) and arterial spin labeling (ASL) have been demonstrated to be useful for the assessment of renal allograft fibrosis, their diagnostic value for renal allograft fibrosis is rarely compared. In this study, we collected a relatively large sample size to compare the value of DTI and ASL in the assessment of renal transplantation (RT) fibrosis. Methods:This study included 141 kidney transplant recipients who underwent DTI, ASL, and biopsy. The renal allograft fibrosis was divided into ci0, ci1, ci2, and ci3 fibrosis groups according to the biopsy results. The apparent diffusion coefficient (ADC), fractional anisotropy (FA), and renal blood flow (RBF) were calculated. One-way analysis of variance (ANOVA) was used to compare the differences of functional magnetic resonance imaging (MRI) parameters between different fibrosis subgroups. The area under the receiver operating characteristic curve (AUC) was calculated to evaluate diagnostic performance. Results:The medullary FA values in ci2 (0.27±0.04, P<0.001) and ci3 (0.21±0.03, P<0.001) groups were significantly lower than those in ci0 group (0.31±0.05). The medullary FA value in ci3 group (0.21±0.03) was significantly lower than that in ci1 group (0.30±0.07, P<0.001) and ci2 group (0.27±0.04, P<0.01). The AUC of DTI was found to be higher than that of ASL in accurately identifying renal allograft fibrosis, and the result was statistically significant in differentiating ci0-2 group and ci3 group (ci0 vs. ci1-3, 0.725 vs. 0.712, P>0.05; ci0-1 vs. ci2-3, 0.787 vs. 0.735, P>0.05; ci0-2 vs. ci3, 0.945 vs. 0.802, P<0.05). Conclusions:DTI has a higher diagnostic value than ASL in noninvasive identification of the degree of renal allograft fibrosis.

Microvascular invasion (MVI) is an important risk factor for early postoperative recurrence of hepatocellular carcinoma (HCC). Based on gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced magnetic resonance imaging (MRI) images, we developed a novel radiomics model. It combined bi-regional features and two machine learning algorithms. The aim of this study was to validate its potential value for preoperative prediction of MVI. This retrospective study included 304 HCC patients (training cohort, 216 patients; testing cohort, 88 patients) from three hospitals. Intratumoral and peritumoral volumes of interest were delineated in arterial phase, portal venous phase, and hepatobiliary phase images. Conventional radiomics (CR) and deep learning radiomics (DLR) features were extracted based on FeAture Explorer software and the 3D ResNet-18 extractor, respectively. Clinical variables were selected using univariate and multivariate analyses. Clinical, CR, DLR, CR-DLR, and clinical-radiomics (Clin-R) models were built using support vector machines. The predictive capacity of the models was assessed by the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. The bi-regional CR-DLR model showed more gains and gave better predictive performance than the single-regional models or single-machine learning models. Its AUC, accuracy, sensitivity, and specificity were 0.844, 76.9

OBJECTIVE:To construct and validate the optimal model for preoperative prediction of proliferative HCC based on habitat-derived radiomics features of Gd-EOB-DTPA-Enhanced MRI. METHODS:A total of 187 patients who underwent Gd-EOB-DTPA-enhanced MRI before curative partial hepatectomy were divided into training (n=130, 50 proliferative and 80 nonproliferative HCC) and validation cohort (n=57, 25 proliferative and 32 nonproliferative HCC). Habitat subregion generation was performed using the Gaussian Mixture Model (GMM) clustering method to cluster all pixels to identify similar subregions within the tumor. Radiomic features were extracted from each tumor subregion in the arterial phase (AP) and hepatobiliary phase (HBP). Independent sample t tests, Pearson correlation coefficient, and Least Absolute Shrinkage and Selection Operator (LASSO) algorithm were performed to select the optimal features of subregions. After feature integration and selection, machine-learning classification models using the sci-kit-learn library were constructed. Receiver Operating Characteristic (ROC) curves and the DeLong test were performed to compare the identified performance for predicting proliferative HCC among these models. RESULTS:The optimal number of clusters was determined to be 3 based on the Silhouette coefficient. 20, 12, and 23 features were retained from the AP, HBP, and the combined AP and HBP habitat (subregions 1, 2, 3) radiomics features. Three models were constructed with these selected features in AP, HBP, and the combined AP and HBP habitat radiomics features. The ROC analysis and DeLong test show that the Naive Bayes model of AP and HBP habitat radiomics (AP-HBP-Hab-Rad) archived the best performance. Finally, the combined model using the Light Gradient Boosting Machine (LightGBM) algorithm, incorporating the AP-HBP-Hab-Rad, age, and AFP (Alpha-Fetoprotein), was identified as the optimal model for predicting proliferative HCC. For the training and validation cohort, the accuracy, sensitivity, specificity, and AUC were 0.923, 0.880, 0.950, 0.966 (95% CI: 0.937-0.994) and 0.825, 0.680, 0.937, 0.877 (95% CI: 0.786-0.969), respectively. In its validation cohort of the combined model, the AUC value was statistically higher than the other models ( P <0.01). CONCLUSIONS:A combined model, including AP-HBP-Hab-Rad, serum AFP, and age using the LightGBM algorithm, can satisfactorily predict proliferative HCC preoperatively.

Background:Sintilimab plus a Bevacizumab biosimilar (IBI305) is an approved first-line regimen for unresectable hepatocellular carcinoma (uHCC) in China. However, data on its safety and efficacy in patients with impaired liver function remain limited. We assessed the clinical outcomes of this combination therapy in HCC patients with Child-Pugh class A (CP-A) and class B (CP-B) liver function. Methods:In this multicenter retrospective cohort study, 99 patients with advanced uHCC (73 CP-A; 26 CP-B) who received first-line Sin/Bev were included. Tumor response was assessed using modified RECIST criteria, and adverse events (AEs) were graded per CTCAE v5.0. Survival outcomes, including overall survival (OS), progression-free survival (PFS), and time to hepatic decompensation (TTD), were analyzed via Kaplan-Meier estimates and Cox proportional hazards models. Results:The objective response rates (ORR) of patients with CP-A and CP-B treated with Sin/Bev were 50.7% and 57.7%, respectively, and both could achieve good anti-tumor efficacy. CP-B had inferior survival: median OS (15 vs 22 months, p=0.044), PFS (8 vs 14 months, p=0.014), and TTD (7 vs 15 months, p<0.001). The CP-B cohort demonstrated comparable incidence rates of grade 3-4 AEs to the CP-A group (34.6% vs 34.2%). Hemorrhagic events and thrombocytopenia emerged as predominant grade 3-4 AEs in CP-B patients (15.4% for both). Conclusions:Sin/Bev demonstrated encouraging short-term anti-tumor activity in HCC of CP-A and CP-B, while survival outcomes were affected by differences in hepatic function. Although the regimen was generally well tolerated, patients with impaired liver reserve require vigilant monitoring and comprehensive supportive strategies to maximize therapeutic outcomes.

ObjectivesTo develop and validate radiomics and deep learning models based on contrast-enhanced MRI (CE-MRI) for differentiating dual-phenotype hepatocellular carcinoma (DPHCC) from HCC and intrahepatic cholangiocarcinoma (ICC).MethodsOur study consisted of 381 patients from four centers with 138 HCCs, 122 DPHCCs, and 121 ICCs (244 for training and 62 for internal tests, centers 1 and 2; 75 for external tests, centers 3 and 4). Radiomics, deep transfer learning (DTL), and fusion models based on CE-MRI were established for differential diagnosis, respectively, and their diagnostic performances were compared using the confusion matrix and area under the receiver operating characteristic (ROC) curve (AUC).ResultsThe radiomics model demonstrated competent diagnostic performance, with a macro-AUC exceeding 0.9, and both accuracy and F1-score above 0.75 in the internal and external validation sets. Notably, the vgg19-combined model outperformed the radiomics and other DTL models. The fusion model based on vgg19 further improved diagnostic performance, achieving a macro-AUC of 0.990 (95% CI: 0.965-1.000), an accuracy of 0.935, and an F1-score of 0.937 in the internal test set. In the external test set, it similarly performed well, with a macro-AUC of 0.988 (95% CI: 0.964-1.000), accuracy of 0.875, and an F1-score of 0.885.ConclusionsBoth the radiomics and the DTL models were able to differentiate DPHCC from HCC and ICC before surgery. The fusion models showed better diagnostic accuracy, which has important value in clinical application.Critical relevance statementMRI-based deep learning radiomics were able to differentiate DPHCC from HCC and ICC preoperatively, aiding clinicians in the identification and targeted treatment of these malignant hepatic tumors.Key PointsFusion models may yield an incremental value over radiomics models in differential diagnosis.Radiomics and deep learning effectively differentiate the three types of malignant hepatic tumors.The fusion models may enhance clinical decision-making for malignant hepatic tumors.

Purpose To assess the effectiveness of an explainable deep learning (DL) model, developed using multiparametric MRI (mpMRI) features, in improving diagnostic accuracy and efficiency of radiologists for classification of focal liver lesions (FLLs). Materials and Methods FLLs ≥ 1 cm in diameter at mpMRI were included in the study. nn-Unet and Liver Imaging Feature Transformer (LIFT) models were developed using retrospective data from one hospital (January 2018-August 2023). nnU-Net was used for lesion segmentation and LIFT for FLL classification. External testing was performed on data from three hospitals (January 2018-December 2023), with a prospective test set obtained from January 2024 to April 2024. Model performance was compared with radiologists and impact of model assistance on junior and senior radiologist performance was assessed. Evaluation metrics included the Dice similarity coefficient (DSC) and accuracy. Results A total of 2131 individuals with FLLs (mean age, 56 ± [SD] 12 years; 1476 female) were included in the training, internal test, external test, and prospective test sets. Average DSC values for liver and tumor segmentation across the three test sets were 0.98 and 0.96, respectively. Average accuracy for features and lesion classification across the three test sets were 93% and 97%, respectively. LIFT-assisted readings improved diagnostic accuracy (average 5.3% increase, P < .001), reduced reading time (average 34.5 seconds decrease, P < .001), and enhanced confidence (average 0.3-point increase, P < .001) of junior radiologists. Conclusion The proposed DL model accurately detected and classified FLLs, improving diagnostic accuracy and efficiency of junior radiologists. ©RSNA, 2025.

The study sought to develop and validate an MRI-based deep learning radiomics (DLR) nomogram for preoperative prediction of vessels encapsulating tumor clusters (VETC) and recurrence-free survival (RFS) in hepatocellular carcinoma (HCC). The dual-center study retrospectively enrolled 625 HCC patients who underwent preoperative Gd-EOB-DTPA-enhanced MRI, including training (n = 296), internal (n = 126), and external (n = 203) test sets. Clinical-radiologic characteristics were selected to develop a clinical-radiologic model. Habitat radiomics and deep learning (DL) features were extracted and selected to develop the habitat radiomics and DL models using the machine learning classifiers. The DLR nomogram model was ultimately constructed by integrating univariate-selected clinical-radiologic characteristics with habitat radiomics and DL scores. Both univariable and multivariable Cox regression analyses were performed to identify independent prognostic factors and develop a prognostic model for RFS. In the external test set, the DLR nomogram model yielded a higher area under the curve (AUC) than the clinical-radiologic model (0.752 vs 0.678; p = 0.004), while habitat radiomics (0.750) and DL models (0.748) showed comparable performance (both p > 0.05). The DLR nomogram consistently demonstrated the higher F1-scores across all three sets. The prognostic model incorporating AFP (hazard ratio (HR), 1.628 [95

To develop and validate radiomics and deep learning models based on Gd-EOB-DTPA enhanced MRI for differentiation between hepatocellular carcinoma (HCC) and focal nodular hyperplasia (FNH) showing iso- or hyperintensity in the hepatobiliary phase (HBP). 112 patients from three hospitals were collected totally. 84 patients from hospital a and b with 54 HCCs and 30 FNHs randomly divided into a training cohort (n = 59: 38 HCC; 21 FNH) and an internal validation cohort (n = 25: 16 HCC; 9 FNH). A total of 28 patients from hospital c (n = 28: 20 HCC; 8 FNH) acted as an external test cohort. 1781 radiomics features were extracted from tumor volumes of interest (VOIs) in the pre-contrast phase (Pre), arterial phase (AP), portal venous phase (PP) and HBP images. 512 deep learning features were extracted from VOIs in the AP, PP and HBP images. Pearson correlation coefficient (PCC) and analysis of variance (ANOVA) were used to select the useful features. Conventional, delta radiomics and deep learning models were established using machine learning algorithms (support vector machine [SVM] and logistic regression [LR]) and their discriminatory efficacy assessed and compared. The combined deep learning models demonstrated the highest diagnostic performance in both the internal validation and external test cohorts, with area under the curve (AUC) values of 0.965 (95

OBJECTIVE:To investigate the value of Gd-EOB-DTPA enhanced MRI radiomics and deep learning models with clinical-radiologic characteristics in predicting the pathological differentiation degree in hepatocellular carcinoma (HCC). METHODS:This study included 409 (training set: 304; validation set: 105) HCC patients who underwent preoperative Gd-EOB-DTPA-enhanced MRI from three hospitals. Clinical and radiological (CR) characteristics were selected using univariate and multivariate analyses. Radiomics and deep learning (DL) features were extracted based on FeAture Explorer software and the 3D ResNet-18 extractor, respectively. CR, radiomics, DL, radiomics combined DL (DLR) and CR-DLR models were built using machine learning algorithms. The predictive capacity of the models was assessed by the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. DCA curves and the calibration curves were used as model validation. RESULTS:Age, AFP, capsule, and peritumoral hypointensity in HBP were independent predictors of the differentiation degree in HCC. In the validation set, the CR-DLR model (AUC = 0.794) were higher than that of the radiomics (AUC = 0.771),DL (AUC = 0.669) CR (AUC = 0.606) and DLR (AUC = 0.776) models. And the differences were statistically significant between the CR-DLR and CR (P = 0.026) or DL models (P = 0.013). The calibration and DCA curves suggesting that the CR-DLR model has a high prediction accuracy and a good net benefit for predicting differentiation degree. CONCLUSION:Combining deep learning, radiomics models with clinical and conventional radiological features has certain value and can help doctors formulate more favorable treatment plans for patients.

Deep learning radiopathomics models based on MR images and pathologic images effectively helped predict the vessels encapsulating tumor clusters pattern in hepatocellular carcinoma and risk for early recurrence and progression-free survival.

To develop and validate a nomogram model based on Gd-EOB-DTPA enhanced MRI for differentiation between hepatocellular carcinoma (HCC) and focal nodular hyperplasia (FNH) showing iso- or hyperintensity in the hepatobiliary phase (HBP). A total of 75 patients with 49 HCCs and 26 FNHs randomly divided into a training cohort (n = 52: 34 HCC; 18 FNH) and an internal validation cohort (n = 23: 15 HCC; 8 FNH). A total of 37 patients (n = 37: 25 HCC; 12 FNH) acted as an external test cohort. The clinical and imaging characteristics between HCC and FNH groups in the training cohort were compared. The statistically significant parameters were included into the FAE software, and a multivariate logistic regression classifier was used to identify independent predictors and establish a nomogram model. Receiver operating characteristic (ROC) curves were used to evaluate the prediction ability of the model, while the calibration and decision curves were used for model validation. Subanalysis was used to compare qualitative and quantitative characteristics of patients with chronic hepatitis and cirrhosis between the HCC and FNH groups. In the training cohort, gender, age, enhancement rate in the arterial phase (AP), focal defects in uptake were significant predictors for HCC showing iso- or hyperintensity in the HBP. In the training cohort, area under the curve (AUC), sensitivity and specificity of the nomogram model were 0.989(95

To investigate the prognostic performance of radiomics analysis of lesion-specific pericoronary adipose tissue (PCAT) for major adverse cardiovascular events (MACE) with the guidance of CT derived fractional flow reserve (CT-FFR) in coronary artery disease (CAD). The study retrospectively analyzed 608 CAD patients who underwent coronary CT angiography. Lesion-specific PCAT was determined by the lowest CT-FFR value and 1691 radiomic features were extracted. MACE included cardiovascular death, nonfatal myocardial infarction, unplanned revascularization and hospitalization for unstable angina. Four models were generated, incorporating traditional risk factors (clinical model), radiomics score (Rad-score, radiomics model), traditional risk factors and Rad-score (clinical radiomics model) and all together (combined model). The model performances were evaluated and compared with Harrell concordance index (C-index), area under curve (AUC) of the receiver operator characteristic. Lesion-specific Rad-score was associated with MACE (adjusted HR = 1.330, p = 0.009). The combined model yielded the highest C-index of 0.718, which was higher than clinical model (C-index = 0.639), radiomics model (C-index = 0.653) and clinical radiomics model (C-index = 0.698) (all p < 0.05). The clinical radiomics model had significant higher C-index than clinical model (p = 0.030). There were no significant differences in C-index between clinical or clinical radiomics model and radiomics model (p values were 0.796 and 0.147 respectively). The AUC increased from 0.674 for clinical model to 0.721 for radiomics model, 0.759 for clinical radiomics model and 0.773 for combined model. Radiomics analysis of lesion-specific PCAT is useful in predicting MACE. Combination of lesion-specific Rad-score and CT-FFR shows incremental value over traditional risk factors.