An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images
Diabetic retinopathy (DR) is a leading cause of visual blindness. However if DR can be diagnosed and treated early, 90% of DR causing blindness can be prevented significantly. Microaneurysms (MAs) and exudates (EXs), as signs of DR, can be used for early DR diagnosis. However, MAs and EXs segmentati...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2020-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9214883/ |
| _version_ | 1818727492157964288 |
|---|---|
| author | Caixia Kou Wei Li Zekuan Yu Luzhan Yuan |
| author_facet | Caixia Kou Wei Li Zekuan Yu Luzhan Yuan |
| author_sort | Caixia Kou |
| collection | DOAJ |
| description | Diabetic retinopathy (DR) is a leading cause of visual blindness. However if DR can be diagnosed and treated early, 90% of DR causing blindness can be prevented significantly. Microaneurysms (MAs) and exudates (EXs), as signs of DR, can be used for early DR diagnosis. However, MAs and EXs segmentation is a challenging task due to the low contrast of the lesions, the interference of noises, and the imbalance between the lesion areas and the background. In this paper, an enhanced residual U-Net (ERU-Net) for MAs and EXs segmentation is proposed. ERU-Net obtains three U-paths, which are composed by three upsampling paths together with one downsampling path. With such three U-paths structure, ERU-Net can enhance the corresponding features fusion and capture more details of fundus images. Also, a residual block is constructed in ERU-Net to extract more representative features. In the experiments, we evaluate the performance of ERU-Net for MAs and EXs segmentation on three public datasets, E-Ophtha, IDRiD, and DDR. The ERU-Net obtains the AUC values of 0.9956, 0.9962, 0.9801, 0.9866, 0.9679, 0.9609 for MAs and EXs segmentation on these three datasets, respectively, which are greater than that of the original U-Net. Compared with some traditional methods, convolutional neural networks and other recent U-Nets, ERU-Net also performs competitively. Besides, we have applied ERU-Net to segment optic disc (OD) on the DRISHTI-GS1 dataset, achieving the highest Jaccard index of 0.994 compared with the existing methods. The numerical results indicate that ERU-Net is a promising network for medical image segmentation. |
| first_indexed | 2024-12-17T22:14:57Z |
| format | Article |
| id | doaj.art-22ca58b840b043b795902cf10d45a3f7 |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-17T22:14:57Z |
| publishDate | 2020-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-22ca58b840b043b795902cf10d45a3f72022-12-21T21:30:38ZengIEEEIEEE Access2169-35362020-01-01818551418552510.1109/ACCESS.2020.30291179214883An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus ImagesCaixia Kou0https://orcid.org/0000-0001-8945-8613Wei Li1Zekuan Yu2https://orcid.org/0000-0003-3655-872XLuzhan Yuan3School of Sciences, Beijing University of Posts and Telecommunications, Beijing, ChinaSchool of Sciences, Beijing University of Posts and Telecommunications, Beijing, ChinaAcademy for Engineering and Technology, Fudan University, Shanghai, ChinaSchool of Sciences, Beijing University of Posts and Telecommunications, Beijing, ChinaDiabetic retinopathy (DR) is a leading cause of visual blindness. However if DR can be diagnosed and treated early, 90% of DR causing blindness can be prevented significantly. Microaneurysms (MAs) and exudates (EXs), as signs of DR, can be used for early DR diagnosis. However, MAs and EXs segmentation is a challenging task due to the low contrast of the lesions, the interference of noises, and the imbalance between the lesion areas and the background. In this paper, an enhanced residual U-Net (ERU-Net) for MAs and EXs segmentation is proposed. ERU-Net obtains three U-paths, which are composed by three upsampling paths together with one downsampling path. With such three U-paths structure, ERU-Net can enhance the corresponding features fusion and capture more details of fundus images. Also, a residual block is constructed in ERU-Net to extract more representative features. In the experiments, we evaluate the performance of ERU-Net for MAs and EXs segmentation on three public datasets, E-Ophtha, IDRiD, and DDR. The ERU-Net obtains the AUC values of 0.9956, 0.9962, 0.9801, 0.9866, 0.9679, 0.9609 for MAs and EXs segmentation on these three datasets, respectively, which are greater than that of the original U-Net. Compared with some traditional methods, convolutional neural networks and other recent U-Nets, ERU-Net also performs competitively. Besides, we have applied ERU-Net to segment optic disc (OD) on the DRISHTI-GS1 dataset, achieving the highest Jaccard index of 0.994 compared with the existing methods. The numerical results indicate that ERU-Net is a promising network for medical image segmentation.https://ieeexplore.ieee.org/document/9214883/U-Netmicroaneurysmsexudatesmedical image segmentation |
| spellingShingle | Caixia Kou Wei Li Zekuan Yu Luzhan Yuan An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images IEEE Access U-Net microaneurysms exudates medical image segmentation |
| title | An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images |
| title_full | An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images |
| title_fullStr | An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images |
| title_full_unstemmed | An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images |
| title_short | An Enhanced Residual U-Net for Microaneurysms and Exudates Segmentation in Fundus Images |
| title_sort | enhanced residual u net for microaneurysms and exudates segmentation in fundus images |
| topic | U-Net microaneurysms exudates medical image segmentation |
| url | https://ieeexplore.ieee.org/document/9214883/ |
| work_keys_str_mv | AT caixiakou anenhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT weili anenhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT zekuanyu anenhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT luzhanyuan anenhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT caixiakou enhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT weili enhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT zekuanyu enhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages AT luzhanyuan enhancedresidualunetformicroaneurysmsandexudatessegmentationinfundusimages |