Swin-UNet with integrated attention mechanisms for precise retinal layer segmentation in optical coherence tomography images

Optical Coherence Tomography (OCT) is a non-invasive ophthalmic imaging technique that generates high-resolution cross-sectional images of the retina, with image analysis requiring the segmentation of retinal layers. Traditional methods often fall short in accuracy and efficiency with complex structures and high-resolution images. Recently, deep learning models, particularly those incorporating transformer architecture, have markedly enhanced segmentation performance. This paper introduces a new retinal layer segmentation approach by incorporating a channel attention (SE) module and a spatial attention (SimAM) module into the Swin-UNet architecture. The SE module adaptively recalibrates channel features to emphasize key features, while the SimAM module optimizes spatial feature transformations to enhance fine-grained segmentation accuracy. The experimental results showed that the model achieved a loss of 0.0557, accuracy of 0.9687, precision of 0.9693, recall of 0.9687, F1 score of 0.9689, and mIoU score of 0.9114, outperforming the baseline model. This study proposes a novel, efficient, and accurate method for OCT retinal layer segmentation and provides a reference for the optimization of other medical image segmentation tasks.