A High-Accuracy Hybrid Method for Detecting Retinal Blood Vessel Changes Across Different Phases of Fluorescein Angiography in Diabetic Retinopathy Patients Publisher Pubmed



E Ramezanzadeh ESMAT ; N Shoeibi NASSER ; H Rabbani HOSSEIN ; Sm Hosseini Seyyedeh MARYAM ; Mra Astaneh Mohammad Reza ANSARI ; M Tavakoli MEYSAM ; H Tabesh HAMED ; H Zare HODA ; Mh Bahreynitoosi Mohammad HOSSEIN
Source: Biomedical Physics and Engineering Express Published:2025

Abstract

Objective. Fundus fluorescein angiography (FFA) remains the gold standard for retinal vascular imaging, especially for detecting leakage, neovascularization, and ischemia, despite advancements in non-invasive techniques like optical coherence tomography angiography (OCT-A) and color fundus photography (CFP). FFA’s unique role, particularly in late-phase imaging, is crucial for diagnosing and managing diabetic retinopathy (DR). This study introduces a novel dual-phase segmentation framework for FFA, enhancing the analysis of early and late-phase images. This study aims to overcome challenges such as noise and blurring in late-phase FFA by developing a method that enhances vascular map detection and monitors changes across two phases, enabling precise identification of lesions responsible for leakage. Validated through expert evaluations and quantitative metrics, this model enhances diagnostic accuracy for diabetic retinopathy and complements existing imaging technologies. Approach. A prospective randomized study of 280 images of 87 DR patients at various stages was included in this study. Our approach involved using four different image enhancement techniques including (1) histogram equalization (HE), (2) contrast limited adaptive HE (CLAHE), (3) recursive mean-square HE (RMSHE), and (4) the proposed method in this paper (CLAHE combined RMSHE). In addition, for robust noise reduction and edge sharpening in each enhancement method, combined median, match, and Hessian filters were used. Finally, four different thresholding methods, including, (i) C-means fuzzy thresholding, (ii) IsoData thresholding, (iii) modified active contour (MAC)+Otsu thresholding, and (iv) the proposed method in this paper (MAC+ IsoData) were used for vessel segmentation in FFA across different datasets. Main results. The most effective segmentation method, MAC+IsoData, was assessed using three metrics (DSC (early: 0.84 ± 0.05, late: 0.84 ± 0.03), Jaccard index (early:0.73 ± 0.06, late:0.74 ± 0.05), and Boundary F1 score (early: 0.98 ± 0.02, late: 0.97 ± 0.02)). The results were validated by three expert ophthalmologists. Significance. This work demonstrates that MAC+Otsu, following CLAHE enhancement, effectively delineates details necessary for the precise identification of lesions responsible for leakage across two different phases. Additionally, MAC+IsoData, with the proposed method enhancement for early phases and RMSHE enhancement for late phases, successfully reveals the vascular map. © 2025 Elsevier B.V., All rights reserved.