Fig. 3: Core-based average probability of PTEN loss within internal cohort based on pathologist labels.

a validation set, N = 2048, and b testing set, N = 224; For training and evaluation purposes, cancer cores with partial and complete loss were grouped as “PTEN loss,” while both benign and cancer cores with intact staining were grouped as “PTEN intact.” Cores with AI-based probability > 26.5% were classified as “PTEN loss” (dashed gray line). Each individual cores is shown as a datapoint, except in the benign cores of validation set (panel a), due to high number of cores in these groups (n = 949 benign intact, n = 938 benign cancer). Any benign or cancer cores with intact PTEN staining classified as “PTEN loss” represent false positive by AI. Any cancer cores with partial or complete PTEN loss classified as “PTEN intact” represent false negative by AI.