Patient-Level Microsatellite Stability Assessment from Whole Slide Images by Combining Momentum Contrast Learning and Group Patch Embeddings

AbstractAssessing microsatellite stability status of a patient’s colorectal cancer is crucial in personalizing treatment regime. Recently, convolutional-neural-networks (CNN) combined with transfer-learning approaches were proposed to circumvent traditional laboratory testing for determining from hematoxylin and eosin stained biopsy whole slide images (WSI). However, the high resolution WSI practically prevent direct classification entire WSI. Current bypass by first classifying small patches extracted WSI, then aggregating patch-level logits deduce patient-level status. Such limit capacity capture important information which resides at data. We introduce an effective approach leverage momentum contrastive learning patch embeddings along training classifier on groups those embeddings. Our achieves up 7.4% better accuracy compared straightforward patient level aggregation higher (AUC, \(0.91 \pm 0.01\) vs. \(0.85 0.04\), p-value < 0.01). code can be found https://github.com/TechnionComputationalMRILab/colorectal_cancer_ai.KeywordsDigital pathologyColorectal cancerSelf-supervised learningMomentum contrast