Human Kidney Tissue Image Segmentation by U-Net Models

Roman Statkevych, Sergii G. Stirenko, Yuri G. Gordienko

Abstract: Segmentation approaches based on deep neural networks are researched for the microscopical images of the human kidney tissues. Several existing methods, used for medical imaging analysis and based on neural networks, were examined. Among several U-Net architectures, which are widely used for image segmentation, some their variations demonstrated the quite high performance despite the 4 times lower model size. As a result, the reasonable precision was obtained by a rudimentary network architectures and limited train time augmentations. It will open the promising perspectives for their deployment of the Edge Computing devices with the limited computing resources.