Table 2 Classification performance of the radiomics model for distinguishing low/intermediate- vs. high-grade tumors, showing relatively strong overall sensitivity and balanced predictive values across folds, averaged across all 30 models trained in the repeated, 3-fold cross validation. Metrics were calculated using a fixed probability cutoff of 0.5, providing a balanced threshold that enables a clear interpretation of the diagnostic performance. F1 score is shown as a weighted average to deal with any potential class imbalance. Balanced accuracy is defined as the average of the sensitivity obtained for each class, effectively representing the mean of the sensitivity and specificity in a binary classification.

From: MRI-based deep learning and radiomics pipeline for myxoid liposarcoma: a feasibility study in a rare sarcoma