Jooyoung Chang1, Ahryung Ko2, Sang Min Park3, Seulggie Choi1, Kyuwoong Kim1, Sung Min Kim1, Jae Moon Yun2, Uk Kang4, Il Hyung Shin5, Joo Young Shin6, Taehoon Ko7, Jinho Lee8, Baek-Lok Oh8, Ki Ho Park8. 1. Department of Biomedical Sciences, Seoul National University Graduate School, Seoul, South Korea. 2. Department of Family Medicine, Seoul National University Hospital, Seoul, South Korea. 3. Department of Biomedical Sciences, Seoul National University Graduate School, Seoul, South Korea; Department of Family Medicine, Seoul National University Hospital, Seoul, South Korea. Electronic address: smpark.snuh@gmail.com. 4. InTheSmart Co., Ltd., Seoul, South Korea; Seoul National University Hospital Biomedical Research Institute, Seoul, South Korea. 5. InTheSmart Co., Ltd., Seoul, South Korea. 6. Department of Ophthalmology, Seoul Metropolitan Government Seoul National University Boramae Medical Center, Seoul, Korea. 7. Office of Hospital Information, Seoul National University Hospital, Seoul, South Korea. 8. Department of Ophthalmology, Seoul National University Hospital, Seoul, South Korea.
Abstract
PURPOSE: The prediction of atherosclerosis using retinal fundus images and deep learning has not been shown possible. The purpose of this study is to develop a deep learning model which predicts atherosclerosis using retinal fundus images and to verify its clinical implications by conducting a retrospective cohort analysis. DESIGN: Retrospective cohort study. METHODS: The database at Health Promotion Center of Seoul National University Hospital (HPC-SNUH) was used. The deep learning model was trained on 15,408 images to predict carotid artery atherosclerosis, which we named the deep learning-funduscopic atherosclerosis score (DL-FAS). We constructed a retrospective cohort of participants aged 30-80 years who had completed elective health check-ups at HPC-SNUH. Using DL-FAS the as the main exposure, we followed participants for the primary outcome of death due to CVD until Dec. 31st, 2017. RESULTS: For predicting carotid artery atherosclerosis among testing-set subjects, the model achieved an AUROC, AUPRC, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of 0.713, 0.569, 0.583, 0.891, 0.404, 0.465, and 0.865 respectively. The cohort comprised of 32,227 participants, 78 CVD deaths, and 7.6-year median follow-up. Those with DL-FAS greater than 0.66 had an increased risk of CVD deaths compared to DL-FAS<0.33 (HR, 95%CI; 8.83, 3.16-24.7). Risk association was significant among intermediate and high Framingham risk score (FRS) subgroups. The DL-FAS improved the concordance by 0.0266 (95% CI, 0.0043-0.0489) over the FRS-only model. Relative integrated discrimination index (IDI) was 20.45% and net reclassification index (NRI) was 29.5%. CONCLUSIONS: We developed a deep learning model which can predict atherosclerosis from retinal fundus images. The resulting DL-FAS was an independent predictor of CVD deaths when adjusted for FRS and added predictive value over FRS.
PURPOSE: The prediction of atherosclerosis using retinal fundus images and deep learning has not been shown possible. The purpose of this study is to develop a deep learning model which predicts atherosclerosis using retinal fundus images and to verify its clinical implications by conducting a retrospective cohort analysis. DESIGN: Retrospective cohort study. METHODS: The database at Health Promotion Center of Seoul National University Hospital (HPC-SNUH) was used. The deep learning model was trained on 15,408 images to predict carotid artery atherosclerosis, which we named the deep learning-funduscopic atherosclerosis score (DL-FAS). We constructed a retrospective cohort of participants aged 30-80 years who had completed elective health check-ups at HPC-SNUH. Using DL-FAS the as the main exposure, we followed participants for the primary outcome of death due to CVD until Dec. 31st, 2017. RESULTS: For predicting carotid artery atherosclerosis among testing-set subjects, the model achieved an AUROC, AUPRC, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of 0.713, 0.569, 0.583, 0.891, 0.404, 0.465, and 0.865 respectively. The cohort comprised of 32,227 participants, 78 CVD deaths, and 7.6-year median follow-up. Those with DL-FAS greater than 0.66 had an increased risk of CVD deaths compared to DL-FAS<0.33 (HR, 95%CI; 8.83, 3.16-24.7). Risk association was significant among intermediate and high Framingham risk score (FRS) subgroups. The DL-FAS improved the concordance by 0.0266 (95% CI, 0.0043-0.0489) over the FRS-only model. Relative integrated discrimination index (IDI) was 20.45% and net reclassification index (NRI) was 29.5%. CONCLUSIONS: We developed a deep learning model which can predict atherosclerosis from retinal fundus images. The resulting DL-FAS was an independent predictor of CVD deaths when adjusted for FRS and added predictive value over FRS.
Authors: Carol Y Cheung; Valérie Biousse; Pearse A Keane; Ernesto L Schiffrin; Tien Y Wong Journal: Nat Rev Dis Primers Date: 2022-03-10 Impact factor: 52.329
Authors: Simon Mueller; Maximilian W M Wintergerst; Peyman Falahat; Frank G Holz; Christian Schaefer; Nadjib Schahab; Robert P Finger; Thomas Schultz Journal: Sci Rep Date: 2022-01-26 Impact factor: 4.379
Authors: Nergis C Khan; Chandrashan Perera; Eliot R Dow; Karen M Chen; Vinit B Mahajan; Prithvi Mruthyunjaya; Diana V Do; Theodore Leng; David Myung Journal: Diagnostics (Basel) Date: 2022-07-14