AFTM: Adaptive Fusion Transformation Map for Efficient Retinal Fundus Vessel Segmentation
Document Type
Article
Publication Date
1-1-2026
Abstract
Developing retinal vessel segmentation methods that generalize across diverse images is crucial for early disease detection, yet it remains challenging due to image variability, pathological changes, and limited expert annotations. While deep learning approaches achieve strong performance, they often require large labeled datasets, substantial computational resources, and frequent retraining, limiting their clinical scalability. To address these issues, we propose the Adaptive Fusion Transformation Map (AFTM) framework, a fully training-free, model-agnostic approach for adaptive retinal vessel segmentation. AFTM operates in three stages: 1) preprocessing for illumination correction and contrast enhancement; 2) fusion-based processing, which integrates complementary transformation maps to capture major and fine vessels; and 3) adaptive post-processing, employing filtering and morphological refinement to enhance vessel continuity and suppress noise. Ablation studies confirm the contribution of each stage to optimal performance. Extensive evaluation on five annotated datasets (DRIVE, CHASE_DB1, HRF, DRHAGIS, FIVES) demonstrates reliable segmentation across diverse and challenging imaging conditions, effectively capturing fine vessels. Qualitative assessment on unannotated datasets (FIRE, ODIR, DIARETDB1) confirms that AFTM reconstructs capillaries and branches across diverse imaging conditions. Comparative analyses show it outperforms traditional methods and rivals state-of-the-art learning-based approaches without training, demonstrating its effectiveness, robustness, and cross-dataset generalizability and suitability for large-scale clinical deployment.
Keywords
AFTM, cross-dataset generalization, medical image processing, Retinal vessel segmentation, transformation map fusion
Publication Title
IEEE Access
DOI
10.1109/ACCESS.2026.3670379
Recommended Citation
Alqahtani, Faisal Majed; Adwan, Somaya; Karman, Salmah Binti; and Ahmad, Mohd Yazed, "AFTM: Adaptive Fusion Transformation Map for Efficient Retinal Fundus Vessel Segmentation" (2026). Research Publications (2026 to 2030). 1601.
https://knova.um.edu.my/research_publications_2026_2030/1601
Volume
14
First Page
38927
Last Page
38950