Segmentation of Larynx Histopathology Images via Convolutional Neural Networks

Abstract

The larynx is a part of the throat, between the base of the tongue and the trachea. Larynx cancer is a disease in which malign cells form in the tissues of the larynx. It amounts to about 3% of newly diagnosed cancers and has a poor prognosis. The histopathologic analyses of larynx cancer have been performed manually by pathologists. It is time-consuming and correct diagnosis highly depends on pathologists’ experience. However, computerized methods are crucial for early detection and monitoring treatment progress in medicine, also free of human error. In this study, the Convolutional Neural Network-based U-Net model developed for the automatic segmentation of larynx histopathology images. The dataset comprises 55 laryngeal cancer cases. There was a total of 224 P63 stained images of different grades. Among them, 87 were Grade I, 73 Grade II, and 64 Grade III cases. According to the simulation results, the model can quickly and accurately differentiate cell structures on tissue and allows advanced image analysis operations. Moreover, it is suitable for use in a laboratory environment. It also helps pathologists in the decision-making process.