Fourier Transform Layer for Fast Foreground Segmentation in Samples' Images of Tissue Biopsies

Sjogren's Syndrome is a systemic disease, presenting itself in a spectrum of symptoms and signs throughout the body. Its diagnosis can include the examination of minor salivary gland biopsies. The image processing of digital images, specifically large-scale Whole Slide Images can aid in diagnosis, but unfortunately requires a lot of processing power and time for automatic analysis. This paper presents a new method for fast foreground (tissue) segmentation in fragments of Whole Slide Images containing biopsies of minor salivary glands taken from patients with diagnosed SjOgren's Syndrome. The method involves training a neural network with a novel type of layer, which calculates Fourier transform of data within the network. The neural network achieved accuracy of 89% on a test dataset.

Automated summary

Sjogren's Syndrome is a systemic disease that can be diagnosed through examination of minor salivary gland biopsies. This paper presents a new method using neural networks and Fourier transform for fast foreground segmentation in Whole Slide Images.