Nuclei Segmentation Using UNet on Breast Hematoxylin and Eosin Stained Histopathology Images

Pathology experts usually analyze digital versions of biopsy samples captured using digital microscope. Histopathological images contain adequate phenotypic information. Therefore, these images play an essential role in diagnosing and treating breast cancer. Pathologists perform a microscopic examination of tissue stained with Hematoxylin and Eosin stains. Nonetheless, the manual evaluation of histopathological images is a time-consuming job. With recent advancements in digital imaging, computer-aided analysis of histopathological slides has become essential. In order to perform image analysis using image processing, nuclei segmentation categorized as crucial initial stage. However, there are several challenges in segmenting nuclei images, including variations in color intensity, the presence of occluded objects, the wide distribution of cell clusters and the lack availability of appropriate annotated datasets makes it challenging to produce sufficient segmentation. This study present nuclei segmentation on histopathology images utilizing U-Net. From several tests conducted, the model shows promising result performance in cell segmentation with accuracy 92.70%, precision 87.10%, recall 84.07%, f1 score 85.24% and IoU 74.54%.