RetiVas-MTNet: Multi-Task Learning Based on Retinal Vasculature

Eye diseases have become an important public health problem worldwide, and retinal vascular morphology and area changes are key diagnostic indicators for diabetic retinopathy, glaucoma, and other diseases. Traditional segmentation and detection methods suffer from insufficient accuracy and lack of quantitative analysis. In this study, the FR-Unet model is improved by introducing an efficient multi-scale convolutional attention decoding module (EMCAD), which improves the segmentation accuracy of microvessels through the dual-path attention mechanism of channel and space. With the innovative design of vessel area percentage quantification, we propose RetiVas-MTNet (Retina Vascular Multi-Task Network), a multi-task learning framework that integrates vascular segmentation and area computation, which provides an accurate and efficient solution for early automated diagnosis of ophthalmic diseases, and is of significant clinical translational value.