Improved U-Net Retinal Segmentation forDisease Diagnosis

Retina segmentation is an essential preprocessing step for computer-aided detection and recognition algorithms. As one of thedeep learning models, U-Net offers the potential to perform classification and segmentation in one step. Evaluation of the UNetmodel for retinal image segmentation was performed on ۱۲۰۰ images from the MESSIDOR dataset. We suggest thatclassification and segmentation tasks be performed in one step. In this case, network training is done based on all the images,and the images that contain disease are entered into the segmentation stage and the segmentation map is extracted.Therefore, we used a separate classifier to filter images so that only images containing disease were segmented with U-Net.Disease-free images remain as output and do not enter the segmentation stage. Our approach increased the Dice coefficient to۰.۸۱۲۱ and the average accuracy to ۷۲.۱۲. However, our results show that improvements are needed before U-Net can achievea precision comparable to a stand-alone classifier

کلیدواژه ها:

U-Net ، Segmentation ، Classification ، Retina ، Disease Recognition.