A Robust OCT Image Retinal Layer Segmentation Method

Optical coherence tomography (OCT) images of the posterior eye are valuable clinical information, which can be used to more accurately diagnose and monitor retinal diseases by detecting changes in retinal layer thickness. In order to quantify OCT images and observe the thickness of each layer and its related information, this paper proposes a deep learning-based OCT image retinal layer segmentation method the MD-UNet model, which can assist in segmenting the different layers of OCT images. The model uses multi-channel feature extraction to improve the U-Net network and enhance the model's robustness. At the same time, the MD-UNet model finely extracts the structural features of each layer and uses the mIoU coefficient as the judgment index for layer structure edge optimization during fusion, thereby improving the overall and local segmentation accuracy and boundary precision. Through ablation experiments, it was demonstrated that the mIoU of the multi-channel structure and improved U-Net structure were improved by 3.05% and 0.34%, respectively. Comparative experimental results showed that this method outperformed other methods in terms of Dice coefficient and boundary error coefficient, demonstrating the effectiveness of this method.