Deep Learning Predicts Patients Outcome And Mutations From Digitized Histology Slides In Gastrointestinal Stromal Tumor

Gastrointestinal stromal tumor (GIST) is the most common mesenchymal neoplasm of the gastrointestinal tract, characterized by a variable clinical behavior. GIST is caused by activating mutations in the tyrosine kinase coding genes KIT or platelet-derived growth factor receptor alpha (PDGFRA). Imatinib, a tyrosine kinase inhibitor, is a breakthrough in the treatment of advanced phases of the disease, although response to imatinib treatment is dependent on the type of mutation. The Armed Forces Institute of Pathology (AFIP)/Miettinen classification is used to evaluate relapse risk of GIST, but reproducibility between pathologists is not perfect. A retrospective study showed a crude proportion of agreement of 88.9% for the mitotic count with a discordance of 6.3% in the resulting Miettinen classification. Deep learning on virtual slides is a promising tool to improve the accuracy of the Miettinen classification.