Skip to main content Skip to main navigation

Publication

Enhancing retinal SELFF-OCT image quality: a deep-learning-based pipeline

Marc Steffen Seibel; Marc Rowedder; Julia Andresen; Richard Neffin; Tobias Neumann; Helge Sudkamp; Heinz Handels; Timo Kepp
In: Medical Imaging 2025: Image Processing. SPIE Medical Imaging (SPIE-2025), San Diego, United States, Vol. 13406, SPIE, 2025.

Abstract

Advances in the development of optical coherence tomographs will make it possible to monitor the progression of eye diseases at home. To this end, Self-Examination Low-Cost Full-Field Optical Coherence Tomography (SELFF-OCT) was recently developed. SELFF-OCT devices are easy to operate and allow patients to take images of the retina themselves without having a doctor present. However, images produced by these devices are of lower quality compared to traditional OCT devices. In this work, we propose a deep-learning assisted pipeline which enhances the quality of SELFF-OCT images. The pipeline consists of four steps: 1. Quality assessment, 2. Image denoising, 3. Registration and fusion of multiple OCT-scans, 4. Averaging of multiple neighboring B-scans. Our preprocessing pipeline enhances the image quality in terms of signal-to-noise ratio (SNR), artifacts and distortions, measured with the blind/referenceless image spatial quality (BRISQUE) evaluator, and detectability of retinal layers, measured with Fisher’s linear discriminant score. Starting from the recorded images, our method increases the SNR from 2.8 to 4.8, lowers the BRISQUE from 67 to 18 which indicates a reduced number of artifacts, and increases Fisher’s discriminant from 2.1 to 5.7 which indicates a better detectability of retinal layers. These results indicate that the proposed pipeline will be useful for improving the detection of biomarkers in future studies utilizing SELFF-OCT.