{"title":"MPLNet: Multi-task supervised progressive learning network for diabetic retinopathy grading","authors":"","doi":"10.1016/j.compeleceng.2024.109746","DOIUrl":null,"url":null,"abstract":"<div><div>Diabetic Retinopathy (DR) is a retinal disease resulting from diabetes. In severe cases, it can lead to irreversible damage to the retina or even blindness. Employing deep learning models to assist in DR diagnosis and classification can alleviate the burden of screening. However, challenges such as the tendency of models to overlook subtle lesions (e.g., microaneurysms) in retinal images and the imbalance in DR data distribution hinder accurate grading. To address these issues, this paper proposes a multi-task supervised progressive learning network (MPLNet) consisting of a Lesion-aware feature extraction Module (LFM) and a Category feature extraction Module (CFM). The network utilizes two progressive tasks – DR identification and DR grading – to guide the LFM and CFM in extracting comprehensive lesion information and then learning discriminative features for each category, thereby enhancing the performance of DR grading. Additionally, to improve the feature extraction capabilities of the two modules, this paper introduces the Detail Attention Module (DAM) and the Category Attention Module (CAM). DAM enhances the detection ability of tiny abnormal areas in the retinal images from both channel and spatial dimensions. The CAM thoroughly explores the critical features of each category from multiple dimensions, thereby reducing the impact of data imbalance. The proposed method achieved kappa scores of 87.0%, 88.2%, and 93.0% on the DDR, Messidor-2, and APTOS datasets, respectively. Experimental results demonstrate that MPLNet outperforms other DR grading methods. T-SNE and Grad-CAM visualization techniques verify the interpretability of the model.</div></div>","PeriodicalId":50630,"journal":{"name":"Computers & Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":4.0000,"publicationDate":"2024-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers & Electrical Engineering","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0045790624006736","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE","Score":null,"Total":0}
引用次数: 0
Abstract
Diabetic Retinopathy (DR) is a retinal disease resulting from diabetes. In severe cases, it can lead to irreversible damage to the retina or even blindness. Employing deep learning models to assist in DR diagnosis and classification can alleviate the burden of screening. However, challenges such as the tendency of models to overlook subtle lesions (e.g., microaneurysms) in retinal images and the imbalance in DR data distribution hinder accurate grading. To address these issues, this paper proposes a multi-task supervised progressive learning network (MPLNet) consisting of a Lesion-aware feature extraction Module (LFM) and a Category feature extraction Module (CFM). The network utilizes two progressive tasks – DR identification and DR grading – to guide the LFM and CFM in extracting comprehensive lesion information and then learning discriminative features for each category, thereby enhancing the performance of DR grading. Additionally, to improve the feature extraction capabilities of the two modules, this paper introduces the Detail Attention Module (DAM) and the Category Attention Module (CAM). DAM enhances the detection ability of tiny abnormal areas in the retinal images from both channel and spatial dimensions. The CAM thoroughly explores the critical features of each category from multiple dimensions, thereby reducing the impact of data imbalance. The proposed method achieved kappa scores of 87.0%, 88.2%, and 93.0% on the DDR, Messidor-2, and APTOS datasets, respectively. Experimental results demonstrate that MPLNet outperforms other DR grading methods. T-SNE and Grad-CAM visualization techniques verify the interpretability of the model.
期刊介绍:
The impact of computers has nowhere been more revolutionary than in electrical engineering. The design, analysis, and operation of electrical and electronic systems are now dominated by computers, a transformation that has been motivated by the natural ease of interface between computers and electrical systems, and the promise of spectacular improvements in speed and efficiency.
Published since 1973, Computers & Electrical Engineering provides rapid publication of topical research into the integration of computer technology and computational techniques with electrical and electronic systems. The journal publishes papers featuring novel implementations of computers and computational techniques in areas like signal and image processing, high-performance computing, parallel processing, and communications. Special attention will be paid to papers describing innovative architectures, algorithms, and software tools.