CNN Support to Diagnostics in Sjögren’s Syndrome

Abstract

Sjögren’s Syndrome is a systemic disease. Its diagnosis can be supported by histopathological examination of minor salivary glands. The value of the focus score equal or greater than one ensures the diagnosis of this disease. Human estimation is inherently subjective, which often leads to diverging results during the process of diagnosis. The paper proposes to use U-net CNN to find area of inflammation in WSI to support pathologist in the process of diagnosis by selecting foci for detailed inspection. Proposed neural network was trained based on tiles of size x pixels in magnification 400x from 13 digital slides stained with hematoxylin and eosin from patients with labial minor salivary gland biopsies. The ground truth was established by manual annotations done by a pathology resident. The accuracy and recall of proposed neural network model on testing dataset show the potential of machine learning for classification problem solving in this field.