An interpretable machine learning framework for automated mitosis detection in gastrointestinal stromal tumors

In grading gastrointestinal stromal tumors, the mitotic index is a critical indicator. The first automated machine learning framework for detecting and counting mitotic cells in GISTs was developed to improve the accuracy of the mitotic index. Addressing challenges from GIST spindle cell morphology, the SVM-RBF model, using a dual-scale approach, achieved strong performance (F1 up to 0.89). Slide-level counts correlated strongly with Ki-67 expression (r=0.6187), validating the method's favorable interpretability and clinical potential.