Objective:To develop an " AI-Doctor" collaborative intelligent model for precise segmentation of retinal non-perfusion areas and evaluate its effectiveness.Methods:Seventy-three retinal non-perfusion images were collected from diabetic retinopathy patients who visited Xuzhou Medical University Affiliated Xuzhou Municipal Hospital and underwent the ultra-widefield fluorescein angiography (UWFA) from December 2022 to January 2024.These images were divided into a training set of 38 images, a validation set of 10 images, and a test set of 25 images.A VGG-UNet model was created, which is an optimization of the combination of VGG-16 and U-Net.Large-scale and small-scale training datasets were created from the UWFA images, and the VGG-UNet was trained on each to obtain corresponding large-scale and small-scale networks.Initial segmentation of non-perfusion areas in UWFA images was conducted using the large-scale network.A physician interaction module was introduced to enhance local segmentation accuracy via the small-scale network, allowing for precise segmentation of non-perfusion areas in UWFA images.The efficacy of the " AI-Doctor" collaborative model was then compared with that of traditional physician annotation methods.This study adhered to the Declaration of Helsinki.The study protocol was approved by the Ethics Committee of Xuzhou Medical University Affiliated Xuzhou Municipal Hospital (No.xyy11[2023]069).Written informed consent was obtained from each subject.Results:The VGG-UNet model was generally able to accurately segment retinal non-perfusion areas.However, problems such as missegmentation, omission, and imprecision were observed at the edge of the eyeball.After the introduction of the physician interaction module, the average segmentation accuracy was improved to 90.36%, showing a significant improvement over conventional methods.Based on the VGG-UNet, a collaborative intelligent segmentation model of " AI-Doctor" was constructed, which can accurately segment images of the non-perfusion area of retinal blood vessels.The validation results showed that the average time of " AI-Doctor" collaborative annotation was about 3.0 minutes, which was significantly shorter than the 29.6 minutes of the traditional annotation method, and the efficiency was improved by about 10 times, and the segmentation accuracy reached 90.36%.Conclusions:An intelligent segmentation model with " AI-Doctor" collaboration is successfully established to achieve efficient and accurate segmentation of the non-perfused area of retinal blood vessels.