| 대한안과학회 학술대회 발표 연제 초록 |
| 녹 F-026 |
| 딥러닝 분류기를 이용한 스펙트럼영역 빛간섭단층촬영에서 녹내장과 압박성 시신경병증의 감별 |
1. 서울대학교 의과대학 안과학교실 2. 서울대학교병원 안과 3. 한림대학교춘천성심병원 안과 |
| 이진호(1,2), 김진수(3), 이행진(1,2), 김성준(1,2), 김영국(1,2), 박기호(1,2), 정진욱(1,2) |
목적 : To assess the performance of a deep learning classifier for differentiation of glaucomatous optic neuropathy (GON) from compressive optic neuropathy (CON) based on ganglion cell-inner plexiform layer (GCIPL) and retinal nerve fiber layer (RNFL) spectral-domain optical coherence tomography (SD-OCT). 방법 : Eighty-four (84) SD-OCT image sets from 84 eyes of 84 GON patients along with 81 SD-OCT image sets of 54 eyes of 54 CON patients were recruited. The bottleneck features extracted from the GCIPL thickness map, GCIPL deviation map, RNFL thickness map, and RNFL deviation map were used as predictors for the deep learning classifier. The area under the receiver operating characteristic curve (AUC) with the deep learning classifier was compared with those for conventional diagnostic parameters including temporal raphe sign, SD-OCT thickness profile, and standard automated perimetry (SAP). 결과 : The deep learning system achieved an AUC of 0.986 (95% CI, 0.973-0.997) with a sensitivity of 94.9% and a specificity of 92.6% in a 5-fold cross-validation testing, which was significantly larger than the AUCs with the other parameters: 0.836 (95% CI, 0.736-0.936) with temporal raphe sign, 0.864 (95% CI, 0.770-0.959) with superonasal GCIPL, and 0.788 (0.665-0.911) with superior GCIPL thicknesses (P = 0.002, 0.003, and <0.001, respectively). 결론 : The deep learning classifier can outperform the conventional diagnostic parameters for discrimination of GON and CON on SD-OCT. |