Enhanced Retinal-Choroidal Disorders Classification Model via Temporal Sequence Analysis of OCT Images Across Multiple Lines of Fovea

This study focuses on the classification of subretinal fluid (SRF) accumulation in conditions such as Central Serous Chorioretinopathy (CSC), Polypoidal Choroidal Vasculopathy (PCV), and Vogt-Koyanagi-Harada (VKH) disease. Optical coherence tomography (OCT) is a widely used imaging modality by ophthalmologists to diagnose these conditions; however, overlapping clinical features often make accurate differentiation challenging. This work aims to develop a robust classification model capable of making predictions from sequences of OCT scans, leveraging a dataset of 642 records that include over 23,000 retinal OCT images obtained from Srinagarind Hospital in Khon Kaen, Thailand. The model was trained and validated using stratified 10-fold cross-validation, achieving an overall classification accuracy of 96.57%, with per-class precision, recall, and F1 scores averaging 96.60%, 96.63%, and 96.60%, respectively. Additionally, gradient-weighted class activation mapping (Grad-CAM) was employed to highlight the anatomical regions relevant to the model's classification, which closely matched the diagnostic features identified by ophthalmologists. These findings indicate that the proposed approach holds substantial promise for advancing automated retinal-choroidal disorder analysis, thereby supporting clinical decision-making in ophthalmology.