Abstract
Extensive research on retinal layer segmentation (RLS) using deep learning (DL) is mostly approaching a performance plateau, primarily due to reliance on structural information alone. To address the present situation, we conduct the first study on the impact of multi-spectral information (MSI) on RLS. Our experimental results show that incorporating MSI significantly improves segmentation accuracy for retinal layer optical coherence tomography (OCT) images. Furthermore, we investigate the primary factors influencing MSI, including the number of multi-spectral images, spectral bandwidth, and the different spectral combinations, to assess their impacts on the accuracy of RLS. Building upon this foundation, we have incorporated MSI into RLS methods, yielding exceptional performance in segmentation outcomes, and these findings have been validated in OCT images across both the near-infrared and visible-light spectral ranges. Fusing MSI provides a novel approach to improving RLS accuracy, further demonstrating the importance of open-source MSI information in OCT devices.