Automatic Segmentation of Deep Foveal Avascular Zone in Optical Coherence Tomography Angiography Using Deep Learning Method

Optical coherence tomography angiography (OCTA) is widely used for visualizing the retinal vascular network, allowing convenient quantification of the deep foveal avascular zone (dFAZ). Accurate segmentation of the dFAZ is essential for diagnosing many retinal-related diseases, but achieving precise segmentation results with clear boundaries poses challenges due to various factors. In this study, we propose a hierarchical and parallel dual-branch deep learning model with attention mechanisms for dFAZ ssegmentation. We tested and compared our method with other mainstream models using our dataset. The results obtained using our proposed model demonstrate clearer and smoother boundaries, with a Dice coefficient of 0.8573, an IoU of 0.8117, and a 95% Hausdorff distance of 19.7116, indicating accurate segmentation of the dFAZ and providing significant clinical utility.