TY - GEN
T1 - Hemorrhage Segmentation in Retinal Images Using Modified FCN-8
AU - Nurul Qomariah, Dinial Utami
AU - Tjandrasa, Handayani
AU - Alam, Basuki Rachmatul
N1 - Publisher Copyright:
© 2021 IEEE.
PY - 2021
Y1 - 2021
N2 - Chronic diabetes mellitus affecting the retina can lead to diabetic retinopathy. Advanced diabetic retinopathy can cause vision loss in patients. Therefore, automatic retinal diagnosis can detect diabetic retinopathy for further prevention. Several signs are effective in detecting diabetic retinopathy automatically, one of which is hemorrhage. Hemorrhages are significant lesions for the diagnosis of diabetic retinopathy. Hemorrhage detection is still a challenge because of the intensity of the color that resembles blood vessels. In this study, we propose a modification of the FCN-8 architecture by adding a modified identity mapping (convolution and batch normalization) to enrich the features of the hemorrhage segmentation process. In addition, contrast limited adaptive histogram equalization (CLAHE) and the extended version of CLAHE are applied to enhance the input images for our proposed method. We use the IDRiD dataset in the evaluation process. Using CLAHE enhancement, our proposed method achieved specificity, sensitivity, and accuracy with values of 98.81%,55.45%, and 98.25, respectively. The sensitivity value is higher than the sensitivity value of other methods.
AB - Chronic diabetes mellitus affecting the retina can lead to diabetic retinopathy. Advanced diabetic retinopathy can cause vision loss in patients. Therefore, automatic retinal diagnosis can detect diabetic retinopathy for further prevention. Several signs are effective in detecting diabetic retinopathy automatically, one of which is hemorrhage. Hemorrhages are significant lesions for the diagnosis of diabetic retinopathy. Hemorrhage detection is still a challenge because of the intensity of the color that resembles blood vessels. In this study, we propose a modification of the FCN-8 architecture by adding a modified identity mapping (convolution and batch normalization) to enrich the features of the hemorrhage segmentation process. In addition, contrast limited adaptive histogram equalization (CLAHE) and the extended version of CLAHE are applied to enhance the input images for our proposed method. We use the IDRiD dataset in the evaluation process. Using CLAHE enhancement, our proposed method achieved specificity, sensitivity, and accuracy with values of 98.81%,55.45%, and 98.25, respectively. The sensitivity value is higher than the sensitivity value of other methods.
KW - FCN
KW - diabetic retinopathy
KW - hemorrhages
KW - identity mapping
KW - residual network
UR - http://www.scopus.com/inward/record.url?scp=85124393597&partnerID=8YFLogxK
U2 - 10.1109/ICVEE54186.2021.9649686
DO - 10.1109/ICVEE54186.2021.9649686
M3 - Conference contribution
AN - SCOPUS:85124393597
T3 - Proceedings - 4th International Conference on Vocational Education and Electrical Engineering: Strengthening Engagement with Communities through Artificial Intelligence Application in Education, Electrical Engineering and Information Technology, ICVEE 2021
BT - Proceedings - 4th International Conference on Vocational Education and Electrical Engineering
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th International Conference on Vocational Education and Electrical Engineering, ICVEE 2021
Y2 - 2 October 2021 through 3 October 2021
ER -
