Adaptive hybrid segmentation model based on watershed and enhanced u-net for endothelial human cornea cells

Abstract

Segmentation of the medical image plays a significant role when it comes to diagnosis using a computer-aided system. This study focused on the human corneal endothelium's health, one of the research areas that is particularly interested in human cornea health. Various pathological environments expedite the extermination of the endothelial cells, which abnormally decreases the cell density. Dead cells worsen the hexagonal design. In this study, medical feature extraction was obtained depending on the segmentation of the endothelial cell boundary. The task of segmentation of such objects is considered challenging due to the nature of the image captured during endothelium layer examination by ophthalmologists using confocal or specular microscopy. The resulting image suffers from various issues that affect the image's quality, such as noise, shadow, and blurry image. So, the study's primary goal was to propose and develop an automatic and robust model for the segmentation of endothelial cells of the human cornea obtained by in vivo microscopy and computation of the different clinical features of endothelial cells. A new scheme of image enhancement was proposed, such as The Contrast-Limited Adaptive Histogram Equalization (CLAHE) techniques to enhance contrast to achieve the goal of this study. After that, a new image denoising technique Enhanced Wavelet Transform Filter and Butterworth Bandpass for Segmentation (WTBBS) was employed. Subsequently, brightness level correction was applied by using the moving average filter and the CLAHE to reduce the effects of the non-uniform image lighting produced as a result of the previous step. The primary focus of this study was the segmentation stage. This stage involved precise detection of the endothelial contours. So, a new segmentation model was proposed, which is an Adaptive Hybrid Trainable Model for Segmenting Endothelial Cells (AHTMSEC). The AHTMSEC includes one crucial step: an Artificial Neural Network for Adaptive Segmenting (ANNAS) to identify the complexity of the image and the suitable algorithm. The output of this step was processed using either the Enhanced U-NET Approach for Endothelial Cell Segmentation (EU-NETAECS) or the Trainable Segmentation and Distance Transform (TDWS) to enhance the Watershed Transform for cell segmentation. In the segmentation stage, the shape of the cells was extracted, and the contours were highlighted. This stage was followed by clinical feature extraction and the used of the features for diagnosis. In this stage, several relevant clinical features such as Pleomorphism Mean Cell Perimeter (MCP), Mean Cell Density (MCD), Mean Cell Area (MCA), and Polymegathism were extracted. The role of these clinical features was crucial for the early detection of corneal pathologies and the evaluation of the health of the corneal endothelium layer. Every process was benchmarked against the best and upto- date segmentation and clinical features detection techniques found in the literature. The existing methods of image enhancement and segmentation have been enhanced considerably via original ideas. Significant contributions of the present study on medical feature extraction based on segmentation were enumerated and ranked from top to bottom according to the degree of importance. The accuracy of the adaptive segmentation model for images classification was 97.5 %. It can be observed that the values obtained using the manual and automated techniques did not exhibit statistically significant differences for any of the five clinical features. The manual and automated processes differences were below 2%, 2%, 1%, 1.5%, and 3.5% for MCD, MCA, Polymegathism, MCP, and Pleomorphism, respectively.