Glomerular Microscopic Image Segmentation Based on Convolutional Neural Network

Accurate glomerular microscopic medical image segmentation is important for renal pathology for obtaining reliable diseases diagnosis. In this study, we construct a pixel-level labeled glomerular microscopic medical image segmentation dataset and improve a classic algorithm, Mask R-CNN, for implementing automatic segmentation of glomerular microscopic medical images. The Improved Mask R-CNN algorithm consists of two parts: in the first part, in order to enhance the accuracy of model training, the anchors in region proposal network are scaled down. In the second part, we increase the number of deconvolution layers in the head mask branch to further improve the glomerular segmentation precision. The experimental results indicate that the algorithm we improve offers higher precision than the original Mask R-CNN algorithm and achieves state-of-the-art segmentation of the glomerular microscopic medical image dataset.