| SP F-023 |
| Application of automated computer-based disease severity assessment for monitoring disease progression for ROP |
| Department of Ophthalmology, The Catholic University of Korea, Seoul St. Mary's Hospital (1)
Department of Ophthalmology, Samsung Medical Center, Sungkyunkwan University (2)
Department of Ophthalmology, Casey Eye Institute, Oregon Health & Science University (3)
Department of Ophthalmology, University of Illinois at Chicago (4) |
| Shin Hae Park (1), Sang Jin Kim (2), J. Peter Campbell(3), Michael F Chiang(3), R.V.P Chan (4) |
|
목적 : To characterize the alterations in ROP assessments by incorporating fully automated disease severity assessment of three sequential retinal images using deep learning.
방법 : A previously validated computer-based deep learning algorithm was used for scoring total 45 exam images from 15 eyes. The severity scores in different time points were noted as T1, T2 and T3 (present). The degrees of retinal vascular changes over time in each case were described as the differences in severity scores, such as T3-T2 and T3-T1. Survey responses from 7 graders were scored as follows; neither (1), pre-plus (2), plus (3) in terms of diagnosis and 2 weeks (1), 1 week (2), less than 1 week (3), treat (4) in terms of management. The differences of survey responses in ROP assessment were calculated in each case by subtracting the total scores when viewing single images from those when viewing serial images. Correlation analyses were performed by calculating Pearson’s correlation coefficients between the differences of survey responses and the degrees of changes in severity scores.
결과 : Increasing severity scores over relatively recent time based on the timing of the evaluation (T3-T2) was associated with a diagnostic trend toward more severe disease (r=0.458) and toward more aggressive management (0.362). Instead, the changes in severity scores over the total period (T3-T1) had poor correlation with ROP diagnosis (r=0.065) and management (r=-0.024). The answer changes in diagnosis and management toward less severe disease were noted even in eyes with progressive vascular changes based on severity scores.
결론 : An automated algorithm can be used as a monitoring tool to monitor the progression of plus disease in infants undergoing screening for ROP. Relatively recent changes in retinal vasculatures did affect ROP assessments. The tempo or degree of retinal vascular changes as well as whether there is vascular progression or not might be the influential factors in ROP assessment.
|
| |