Publication: GlaucoVIZ: Assisting System for Early Glaucoma Detection Using Mask R-CNN
Issued Date
2020-06-01
Resource Type
Other identifier(s)
2-s2.0-85091886011
Rights
Mahidol University
Rights Holder(s)
SCOPUS
Bibliographic Citation
17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, ECTI-CON 2020. (2020), 364-367
Suggested Citation
Srisupa Palakvangsa-Na-Ayudhya, Thanan Sapthamrong, Kritsada Sunthornwutthikrai, Darin Sakiyalak GlaucoVIZ: Assisting System for Early Glaucoma Detection Using Mask R-CNN. 17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, ECTI-CON 2020. (2020), 364-367. doi:10.1109/ECTI-CON49241.2020.9158128 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/59945
Research Projects
Organizational Units
Authors
Journal Issue
Thesis
Title
GlaucoVIZ: Assisting System for Early Glaucoma Detection Using Mask R-CNN
Other Contributor(s)
Abstract
© 2020 IEEE. Glaucoma is a chronic eye disease which is the second leading cause of blindness all over the world and a great number of people have been experiencing it. Glaucoma is a group of diseases that cause the damage to the optic nerves which leads to irreversible partial or full vision loss. Currently, its causes cannot be identified; hence, early detection of glaucoma is critical and highly beneficial in preventing further deterioration. One of the most valuable methods of structural diagnosis in the glaucoma screening is the measurement of the Cup-to-Disc Ratio (CDR). To segment optic cup and optic disc, and the determination of the CDR currently require specialists and this process is very time consuming. Therefore, automated screening solution will be very useful in this situation. Due to this, GlaucoVIZ is proposed as an automated system to assist in the preliminary glaucoma screening process to help general ophthalmologists by calculating the CDR with the use of the deep leaning technique, Mask Regional - Convolutional Neural Network (Mask R-CNN). With the use of four public datasets available, we conduct two experiments to understand the characteristics of the datasets and train our model to be more generalize by combining all public datasets together. With the use of the interpolated mean Average Precision (mAP) as our evaluation criteria, our model trained from the combined datasets can achieve mAP of 0.78 by using the original fundus images and the augmented images in the total of about 3,390 images.