Fig. 4.

SPCNet predictions on a sample patient in cohort C1‐Test shown from apex (top row) to base (bottom row). The per‐pixel AUCs were 0.92 for normal tissue, 0.90 for indolent cancer, and 0.93 for aggressive cancer. (Column a) T2w input; (Column b) ADC; (Column c) T2w image overlaid with grade annotations Indolent (Gleason pattern 3, Green) and Aggressive (Gleason patterns 4 or 5, Yellow); (Column d) T2w image overlaid with thresholded prediction from the SPCNet multiclass model cancer predictions (indolent — blue; aggressive — red). Cancer labels by the expert pathologist mapped from histopathology images onto MRI are outlined in black and white (Columns a–b).