Resumen de Sclerotic Glomerulus detection in kidney Whole Slide Images based on Convolutional Neural Networks
J. Gallego, O. Deniz, G. Bueno, Z. Swiderska Chadaj
Glomerular analysis is a key process in nephropathology to detect kidney disorders. Different diseases like diabetes, VIH, or hepatitis, among others, present symptoms in the kidney glomerular structure that can be studied to make a correct prognosis of the illness. Glomerular sclerosis is one of the most significant Glomerular pathologies. It appears when the Glomerulus blood filtering capability is highly compromised and medical actions are required. In this paper, we present a new deep learning-based approach that first, detects and segments the Glomeruli present in the scanned Whole Slide Images (WSI). In a second step, the detected Glomeruli are classified to detect the sclerotic ones. A dataset composed of 47 WSIs belonging to human kidney sections stained with PAS was used. We propose to apply two different Convolutional Neural Networks (CNNs) where first we use the U-Net model to achieve the pixel-wise segmentation of Glomerular regions and next, we use the segmented regions to perform a CNN classification between sclerotic, non-sclerotic Glomerulus and non-Glomerulus by using a fine-tuned AlexNet model. The results show the suitability of this CNN-based approach to detect the sclerotic and non-sclerotic Glomeruli in the WSIs.
