MEDICAL INFORMATICS IV
EVALUATION OF IMAGE SEGMENTATION MODELS IN RETINAL VESSEL SEGMENTATION
Yayıncı:
İstanbul Üniversitesi Yayınları
Retinal vessel segmentation refers to the process of detecting and separating vessels within retinal images into specific regions. This process plays a crucial role in the diagnosis and monitoring of eye diseases such as diabetic retinopathy, hypertension, and macular degeneration. Ophthalmologists can analyze changes in the appearance of vascular structures to enable early disease diagnosis and track disease progression. The segmentation of retinal vessels holds significant importance in medical image processing, visual analysis studies, drug and treatment development processes, and clinical decision-making, contributing to the efficiency of healthcare services. Similar to other medical imaging techniques such as magnetic resonance imaging, computed tomography and X-rays, the segmentation of retinal vessels plays a critical role in diagnosing and treating diseases, including the detection of tumors, lesions, and other pathological structures through image segmentation methods. This study aims to investigate the segmentation of blood vessels from retinal images, which is the initial step in the disease diagnosis process. Towards this goal, existing articles in the literature have been thoroughly examined, and the methods used for retinal vessel segmentation have been systematically compiled. Throughout the process from the early works to the present day, various solution approaches to vessel segmentation have been evaluated based on different criteria. In this evaluation process, different segmentation models including Linknet, FPN (Feature Pyramid Network), UNET, and PSPNet (Pyramid Scene Parsing Network) were examined using the DRIVE, CHASEBASE and OCRIMA dataset. This study involves comparisons made by employing various foundational architectures (‘vgg16’, ‘efficientnetb0’, ‘resnet50’, etc.) for each model. The success of each model in retinal vessel segmentation and its performance using specific metrics have been comparatively evaluated. The obtained results have assisted in understanding which model is more effective under which conditions. In conclusion, this study aims to systematically evaluate the achievements of diverse image segmentation models for retinal vessel segmentation, providing guidance for future advancements in this field.
