| Retinal fundus image analysis is a common task in medical imaging analysis.Retinal fundus blood vessel images are crucial for the diagnosis and treatment of ophthalmic diseases such as diabetic retinopathy and hypertensive retinopathy.Doctors need to manually analyze and diagnose these images,but due to the high complexity and large quantity of retinal fundus images,their workload is extremely heavy and diagnostic errors are prone to occur.Therefore,developing a fast and accurate automated retinal vascular image segmentation algorithm has become an important research direction in the field of medical image processing.Three datasets,DRIVE,STARE,CHASE_DB1 are chosen to be research objects.Various image segmentation algorithm experiments are constructed on these datasets.Firstly,a data augmentation scheme and a loss function improvement specific to this project are proposed.Based on transfer learning and attention mechanisms,PAU-Net is constructed as a baseline.Secondly,an improvement is made on this basis,proposing TMLP-Net,which is based on Transformers and multi-layer perceptrons.It uses a Transformer module that integrates convolutional layers to extract features from images at multiple scales and obtain multi-level feature information.The simple and efficient multi-layer perceptron decoder is used to perform fast and efficient pixel-level classification on the feature maps.Additionally,the loss function is further improved to enhance the algorithm’s performance.Finally,based on the requirements analysis and overall design,a deep learning-based intelligent ophthalmology medical system is developed to manage the entire process of ophthalmic patients,and the proposed TMLP-Net is utilized to assist clinical decision-making.TMLP-Net achieved excellent experimental results on multiple datasets and showed outstanding segmentation performance for retinal blood vessels.It provides new ideas for other researchers in the field of retinal image processing.The deep learning-based intelligent ophthalmic healthcare system based on TMLP-Net can assist doctors in the diagnosis of ophthalmic diseases and has certain practical value. |
