IntroductionThis study aimed to construct a radiomics-based machine learning (ML) model for differentiation between non-clear cell and clear cell renal cell carcinomas (ccRCC) that is robust against institutional imaging protocols and scanners.Materials and methodsPreoperative unenhanced (UN), corticomedullary (CM), and excretory (EX) phase CT scans from 209 patients diagnosed with RCCs were retrospectively collected. After the three-dimensional segmentation, 107 radiomics features (RFs) were extracted from the tumor volumes in each contrast phase. For the ML analysis, the cases were randomly split into training and test sets with a 3:1 ratio. Highly correlated RFs were filtered out based on Pearson’s correlation coefficient (r > 0.95). Intraclass correlation coefficient analysis was used to select RFs with excellent reproducibility (ICC ≥ 0.90). The most predictive RFs were selected by the least absolute shrinkage and selection operator (LASSO). A support vector machine algorithm-based binary classifier (SVC) was constructed to predict tumor types and its performance was evaluated based-on receiver operating characteristic curve (ROC) analysis. The “Kidney Tumor Segmentation 2019” (KiTS19) publicly available dataset was used during external validation of the model. The performance of the SVC was also compared with an expert radiologist’s.ResultsThe training set consisted of 121 ccRCCs and 38 non-ccRCCs, while the independent internal test set contained 40 ccRCCs and 13 non-ccRCCs. For external validation, 50 ccRCCs and 23 non-ccRCCs were identified from the KiTS19 dataset with the available UN, CM, and EX phase CTs. After filtering out the highly correlated and poorly reproducible features, the LASSO algorithm selected 10 CM phase RFs that were then used for model construction. During external validation, the SVC achieved an area under the ROC curve (AUC) value, accuracy, sensitivity, and specificity of 0.83, 0.78, 0.80, and 0.74, respectively. UN and/or EX phase RFs did not further increase the model’s performance. Meanwhile, in the same comparison, the expert radiologist achieved similar performance with an AUC of 0.77, an accuracy of 0.79, a sensitivity of 0.84, and a specificity of 0.69.ConclusionRadiomics analysis of CM phase CT scans combined with ML can achieve comparable performance with an expert radiologist in differentiating ccRCCs from non-ccRCCs.

Background and Objectives: This study aims to evaluate artificial intelligence-calculated hepatorenal index (AI-HRI) as a diagnostic method for hepatic steatosis. Materials and Methods: We prospectively enrolled 102 patients with clinically suspected non-alcoholic fatty liver disease (NAFLD). All patients had a quantitative ultrasound (QUS), including AI-HRI, ultrasound attenuation coefficient (AC,) and ultrasound backscatter-distribution coefficient (SC) measurements. The ultrasonographic fatty liver indicator (US-FLI) score was also calculated. The magnetic resonance imaging fat fraction (MRI-PDFF) was the reference to classify patients into four grades of steatosis: none < 5%, mild 5–10%, moderate 10–20%, and severe ≥ 20%. We compared AI-HRI between steatosis grades and calculated Spearman’s correlation (rs) between the methods. We determined the agreement between AI-HRI by two examiners using the intraclass correlation coefficient (ICC) of 68 cases. We performed a receiver operating characteristics (ROC) analysis to estimate the area under the curve (AUC) for AI-HRI. Results: The mean AI-HRI was 2.27 (standard deviation, ±0.96) in the patient cohort. The AI-HRI was significantly different between groups without (1.480 ± 0.607, p < 0.003) and with mild steatosis (2.155 ± 0.776), as well as between mild and moderate steatosis (2.777 ± 0.923, p < 0.018). AI-HRI showed moderate correlation with AC (rs = 0.597), SC (rs = 0.473), US-FLI (rs = 0.5), and MRI-PDFF (rs = 0.528). The agreement in AI-HRI was good between the two examiners (ICC = 0.635, 95% confidence interval (CI) = 0.411–0.774, p < 0.001). The AI-HRI could detect mild steatosis (AUC = 0.758, 95% CI = 0.621–0.894) with fair and moderate/severe steatosis (AUC = 0.803, 95% CI = 0.721–0.885) with good accuracy. However, the performance of AI-HRI was not significantly different (p < 0.578) between the two diagnostic tasks. Conclusions: AI-HRI is an easy-to-use, reproducible, and accurate QUS method for diagnosing mild and moderate hepatic steatosis.

Liver tumors constitute a major part of the global disease burden, often making regular imaging follow-up necessary. Recently, deep learning (DL) has increasingly been applied in this research area. How these methods could facilitate report writing is still a question, which our study aims to address by assessing multiple DL methods using the Medical Open Network for Artificial Intelligence (MONAI) framework, which may provide clinicians with preliminary information about a given liver lesion. For this purpose, we collected 2274 three-dimensional images of lesions, which we cropped from gadoxetate disodium enhanced T1w, native T1w, and T2w magnetic resonance imaging (MRI) scans. After we performed training and validation using 202 and 65 lesions, we selected the best performing model to predict features of lesions from our in-house test dataset containing 112 lesions. The model (EfficientNetB0) predicted 10 features in the test set with an average area under the receiver operating characteristic curve (standard deviation), sensitivity, specificity, negative predictive value, positive predictive value of 0.84 (0.1), 0.78 (0.14), 0.86 (0.08), 0.89 (0.08) and 0.71 (0.17), respectively. These results suggest that AI methods may assist less experienced residents or radiologists in liver MRI reporting of focal liver lesions.