Automatic layer segmentation in H&E images of mice skin based on colour deconvolution and fuzzy C-mean clustering

Abstract

Abstract Skin is by far the largest organ in the mammalian body. It has a complex structure of multiple layers consisting of various distinguishable cells and other features. Dermatology specialists have long associated many diseases with changes in different skin layers. However, manually quantifying and analysing large volumes of images is an error-prone, time-consuming and costly endeavour in terms of the resources and staff training required. This paper presents an automated high-throughput solution for segmenting mice skin layers from images into epidermis, dermis, and adipose layers and further segmenting the epidermis into cornified and basal layers. Such segmentation can be considered the first step in automatically quantifying cutaneous features in different skin layers. The proposed method combines a colour deconvolution method with fuzzy C-mean (FCM) clustering to segment skin layers. A dataset of 7,000 mice skin images was used to evaluate the effectiveness of the proposed method. The images were hematoxylin and eosin (H&E) microscopic images taken at 20X magnification by the Mouse Genetics Project. A segmentation accuracy rate of 96% was achieved using the hybrid solution, and accuracy rates of 80% and 92% were achieved using colour deconvolution and FCM, respectively. The experimental results were then examined by domain experts who confirmed the viability of the hybrid solution.