Retinal Fluid Segmentation from OCT B-Scan Using Swin-Unet

Macular edema (ME) is one of the main causes of visual impairment and even blindness. It is caused by various retinal diseases such as diabetic retinopathy (DR) and age-related macular degeneration (AMD), which lead to vascular damage and leakage, resulting in fluid accumulation in the macula. Optical coherence tomography (OCT) is widely uesd in clinical to observe ME due to its non-invasiveness and the ability to visualize cross-section of the retina. Quantitative analysis of the retinal fluid from OCT allows for tracking the progression of the condition and observing treatment responses. Therefore, segmenting the retinal fluid is crucial. This study employs swin-unet to address the challenges in retinal fluid segmentation, including low contrast of OCT, varying scales and quantities of retinal fluid. The proposed approach improves the accuracy of retinal fluid segmentation. On the UMN public dataset, the average intersection over union (IoU) is 0.706, and the average DICE coefficient is 0.814. These results demonstrate the effectiveness of the proposed method.