Enhanced Diagnosis of Age-Related Macular Degeneration Using MedSAM on Optical Coherence Tomography Images

Advances in medical imaging technologies, especially optical coherence tomography (OCT), have greatly improved the diagnosis of retinal diseases. However, accurate segmentation of retinal structures in OCT images remains a challenging task. In this study, we refined the Medical Segment Anything (MedSAM 2) model using the AMD-SD dataset containing 3049 OCT scans annotated by ophthalmologists. The dataset includes annotations of five key anatomical structures (IRF, SRF, PED, SHRM, IS/OS), providing a robust training baseline. MedSAM 2, based on transformer architectures, was further trained and achieved a Dice coefficient of 0.893 and an intersection over union of 0.782, which outperformed baseline models. This highlights the effectiveness of fine-tuning to improve segmentation accuracy. We also developed a PyQt5 GUI for semi-automated OCT image annotation that supports semantic segmentation and mask export in a standardized format. This demonstrates the potential of MedSAM 2 as an accurate tool for medical image analysis in clinical settings. The results highlight the superiority of finely tuned MedSAM 2 for retinal image segmentation and its potential for application in medical imaging. Future work will focus on improving the algorithm and expanding its use in decision support systems in ophthalmology.