Fig. 3

Overview of the deep learning modeling and evaluation flow. a Training and validation dataset (n = 243, slices = 3061, training slices:validation slices = 4:1). b Ground truth delineated by two radiologists. c Process of nnUNet Framework. d Prediction results using the trained DL model. e Evaluation metrics. f Test dataset (n = 41, slices = 408), TRG1 = 18, TRG2 = 19, and TRG3 = 4