Can YOLO Detect Retinal Pathologies? A Step Towards Automated OCT Analysis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Ophthalmology

Background:

• Optical Coherence Tomography (OCT) provides non-invasive retinal visualization for disease identification.
• Increasing OCT data volume necessitates automated analysis methods.
• Manual analysis of complex OCT scans is becoming infeasible.

Purpose of the Study:

• To benchmark state-of-the-art object detection models for retinal pathology detection in OCT scans.
• To evaluate the performance of YOLO versions (v8-v12), YOLO-World, YOLOE, and RT-DETR.
• To identify optimal models for automated analysis of retinal OCT data.

Main Methods:

• Investigated object detection models including YOLOv8-v12, YOLO-World, YOLOE, and RT-DETR.
• Utilized two retinal OCT datasets: AROI (Age-related Macular Degeneration fluid detection) and OCT5k (diverse retinal pathologies).
• Assessed model accuracy and computational efficiency for pathological biomarker detection.

Main Results:

• YOLOv12 achieved a strong balance between detection accuracy and computational efficiency.
• YOLOE consistently outperformed other models across both datasets and most pathology classes.
• YOLOE showed particular strength in detecting smaller pathological areas.

Conclusions:

• This study provides a comprehensive benchmark for object detection models in retinal OCT analysis.
• YOLOE emerges as a highly capable model for identifying retinal pathologies from OCT scans.
• Findings offer a foundation for developing automated clinical analysis tools for retinal diseases.