A Layered Pre-processing-Based Fair Effective Retinal Blood Vessels Segmentation

Effective AI-based clinical decision-making relies on accurate segmentation of medical images. Ensuring efficient and robust retinal blood vessels segmentation remains a challenge due to their fine, low-contrast nature. In response to this need, this paper introduces an efficient method to segment retinal blood vessels relying on a sequence of preprocessing steps. The proposed methodology combines the Locust Search algorithm, Contrast Limited Adaptive Histogram Equalization (CLAHE), and Standard Gamma correction, followed by a modified Fully Convolutional Network (FCN8) to enhance images and segment retinal blood vessels, respectively. In addition, as a benchmark in the field, the DRIVE dataset is employed to assess the presented model. The proposed model demonstrated its performance and highlights its effectiveness in segmenting retinal blood vessels. Specifically, the conceived model achieved a 0.7528 dice score, 0.8145 recall, and 0.8889 AUC score. Beyond what the existing literature has attained, our model achieves superior recall and dice scores.