| Jin H Kim

SD-OCT-based Epiretinal Membrane Diagnostic Assistant System

Introduction

This project presents a low-cost and efficient method for detecting and quantifying Epiretinal Membranes (ERM) using Spectral-Domain Optical Coherence Tomography (SD-OCT). By applying deep learning techniques—specifically, YOLO object detection—we generate en face "ERM Projection Images" from B-scan data, enabling intuitive visualization and accurate measurement of ERM areas. The method also introduces a novel approach to quantify the association between ERM and retinal thickness, enhancing clinical decision-making. Our approach aims to bridge the diagnostic performance gap between SD-OCT and Swept-Source OCT (SS-OCT) while maintaining accessibility and reducing diagnostic burden.

Overall pipeline architecture for ERM detection & quantification

YOLO Model Evaluation

We evaluated three YOLO-based models (v5, v8, v11) for ERM detection using SD-OCT B-scan images.
Each model was trained on two datasets (2,200 images for Full, 1,100 images for Half) and tested on 650 expert-labeled images.

Model	Size	Params (M)	Precision	Recall	mAP@50	mAP@50:95	Dataset Size
YOLOv5	S	7.02	0.752	0.703	0.722	0.423	Full
			0.694	0.642	0.664	0.376	Half
	M	20.87	0.783	0.734	0.752	0.444	Full
			0.723	0.685	0.701	0.396	Half
	L	46.14	0.813	0.762	0.784	0.463	Full
			0.745	0.704	0.726	0.414	Half
	X	86.22	0.836	0.784	0.802	0.485	Full
			0.763	0.725	0.743	0.437	Half
YOLOv8	S	11.14	0.781	0.736	0.764	0.447	Full
			0.723	0.676	0.701	0.393	Half
	M	25.86	0.813	0.762	0.791	0.466	Full
			0.748	0.705	0.724	0.412	Half
	L	43.63	0.844	0.792	0.823	0.482	Full
			0.774	0.731	0.754	0.436	Half
	X	68.15	0.867	0.814	0.842	0.504	Full
			0.793	0.752	0.772	0.454	Half
YOLOv11	S	9.43	0.804	0.752	0.783	0.468	Full
			0.746	0.692	0.714	0.417	Half
	M	20.05	0.846	0.794	0.821	0.493	Full
			0.774	0.736	0.757	0.443	Half
	L	25.31	0.873	0.823	0.854	0.524	Full
			0.807	0.773	0.793	0.476	Half
	X	56.87	0.902	0.857	0.882	0.556	Full
			0.836	0.803	0.826	0.507	Half

GitHub repository: github.com/jinkimh/SD-OCT-ERM-Quantification