U-Net Deep Learning Model for the Retinal Vessel Segmentation of Latest Imaging Technology OCTA Images

Retinal vessel segmentation in Optical Coherence Tomography Angiography (OCTA) images constitutes a crucial task in ophthalmology and medical image analysis. It involves identifying and delineating blood vessels within the retinal layers captured by OCTA imaging. This segmentation task plays a pivotal role in various clinical applications, encompassing the diagnosis and monitoring of retinal diseases like diabetic retinopathy, macular degeneration, and glaucoma. Current research on retinal vessel segmentation primarily concentrates on enhancing the accuracy and efficiency of existing methods while also exploring novel approaches. The primary objective of this research is to evaluate the outcomes resulting from the application of the U-Net model on the ROSE (Retinal OCT-Angiography Segmentation) dataset. This involves a comparative analysis of its performance against alternative segmentation models such as FCN, DeepLab, and Gated-SCNN. Experimental results demonstrate that the U-Net model surpasses FCN, Gated-SCNN, and DeepLab, achieving an accuracy of 92.73%. Furthermore, this model exhibits notably high scores in Jaccard (0.2322), Dice (0.4645), and F1-score (0.3750), indicating substantial overlap between predicted and ground truth vessel regions. Moreover, the Area Under Curve (AUC) score of the U-Net model (0.68) surpasses that of other models, signifying its exceptional performance.