BanglaOCT2025: A Population-Specific Fovea-Centric OCT Dataset with Self-Supervised Volumetric Restoration Using...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss.
• Existing Optical Coherence Tomography (OCT) datasets lack demographic diversity, particularly from South Asian populations, limiting AI generalizability.
• Raw OCT volumes contain noise and redundant spatial information, hindering analysis.

Purpose of the Study:

• Introduce BanglaOCT2025, the first clinically validated OCT dataset representing the Bengali population.
• Develop a novel preprocessing pipeline for fovea-centric volumetric OCT analysis.
• Improve AI-driven AMD classification accuracy through advanced denoising techniques.

Main Methods:

• Collected 1585 OCT volumes from Bangladesh National Institute of Ophthalmology and Hospital.
• Developed a constraint-based algorithm for automatic foveal center localization and fixed sub-volume extraction.
• Implemented a self-supervised Flip-Flop Swin Transformer (FFSwin) for volumetric speckle noise suppression.

Main Results:

• BanglaOCT2025 includes 857 expert-annotated AMD cases (DryAMD, WetAMD, NonAMD).
• Denoising preserved pathological biomarkers and showed no hallucination in clinical review.
• AMD classification accuracy improved from 69.08% to 99.88% with the proposed pipeline.

Conclusions:

• BanglaOCT2025 is the first OCT dataset representing the Bengali population for AMD research.
• Established a reproducible framework for fovea-centric volumetric preprocessing and restoration in OCT analysis.
• The developed methods show significant potential for enhancing diagnostic accuracy in AMD detection.