| 대한안과학회 학술대회 발표 연제 초록 |
| 신경 F-007 |
| 안저사진을 이용한 시신경 뒤틀림의 인공지능 딥러닝을 통한 진단 |
성균관대학교 의과대학 삼성서울병원 안과학교실 삼성서울병원 스마트헬스케어 센터, 의공학 연구 센터 |
| 박경아(1), 이다영(2), 오세열(1), 조백환(2), 이가인(1), 노훈(1), 정준교(1), 강민채(1) |
목적 : To develop a machine learning model that detect tilted optic disc using fundus photography. 방법 : This study used 198 fundus photographs of normal and patients with myopic tilted disc, collected from Samsung Medical Center between April 2016 and December 2018. We developed an automated computer-aided detection system for optic disc tilt on fundus photographs via a convolutional neural network algorithm. The performance of this algorithm was compared with a manual detection. 결과 : The fully automated deep learning system achieved a sensitivity of 98 % and a specificity of 97 % for detecting optic disc tilt in colour fundus images. The mean AUC area was 0.98. The overall validation accuracy of the deep learning system with 50 folds augmentation was 96%. 결론 : We developed an automated deep learning system detecting optic disc tilt in this study. It demonstrated excellent agreement with human examiner and achieved promising results for future developing programs to adjust the effect of optic disc tilt on ophthalmic measurements. |