Reunion Block for High Magnification Histopathology Microscopic Image Analysis

Abstract

The input image scale must be considered in the microsatellite instability recognition method through deep learning image analysis. Since pathological images can observe various features through high magnification, an image analysis method capable of analyzing high-resolution images is required. Although CNN has excellent image analysis capabilities, the size of input images is limited. If we want to analyze an area bigger than the input image size of the CNN, the area should be reduced or crop. In this paper, we propose a recombination block that extracts and combines features in patch units to handle microsatellite images made up of high-resolution images.