Predicting glaucoma prior to its onset using deep learning

Abstract

<jats:title>Abstract</jats:title><jats:sec><jats:title>Purpose</jats:title><jats:p>To assess the accuracy of deep learning models to predict glaucoma development from fundus photographs several years prior to disease onset.</jats:p></jats:sec><jats:sec><jats:title>Design</jats:title><jats:p>A deep learning model for prediction of glaucomatous optic neuropathy or visual field abnormality from color fundus photographs.</jats:p></jats:sec><jats:sec><jats:title>Participants</jats:title><jats:p>We retrospectively included 66,721 fundus photographs from 3,272 eyes of 1,636 subjects to develop deep leaning models.</jats:p></jats:sec><jats:sec><jats:title>Method</jats:title><jats:p>Fundus photographs and visual fields were carefully examined by two independent readers from the optic disc and visual field reading centers of the ocular hypertension treatment study (OHTS). When an abnormality was detected by the readers, subject was recalled for re-testing to confirm the abnormality and further confirmation by an endpoint committee. Using OHTS data, deep learning models were trained and tested using 85% of the fundus photographs and further validated (re-tested) on the remaining (held-out) 15% of the fundus photographs.</jats:p></jats:sec><jats:sec><jats:title>Main Outcome Measures</jats:title><jats:p>Accuracy and area under the receiver-operating characteristic curve (AUC).</jats:p></jats:sec><jats:sec><jats:title>Results</jats:title><jats:p>The AUC of the deep learning model in predicting glaucoma development 4-7 years prior to disease onset was 0.77 (95% confidence interval 0.75, 0.79). The accuracy of the model in predicting glaucoma development about 1-3 years prior to disease onset was 0.88 (0.86, 0.91). The accuracy of the model in detecting glaucoma after onset was 0.95 (0.94, 0.96).</jats:p></jats:sec><jats:sec><jats:title>Conclusions</jats:title><jats:p>Deep learning models can predict glaucoma development prior to disease onset with reasonable accuracy. Eyes with visual field abnormality but not glaucomatous optic neuropathy had a higher tendency to be missed by deep learning algorithms.</jats:p></jats:sec>

Predicting the risk of developing diabetic retinopathy using deep learning

“…For instance, models using CFPs have been shown capable of detecting anemia, 33 predicting cardiovascular risk factors, 34 and determining DME and glaucoma-related damage that otherwise needed optical coherence tomography (OCT) or additional tests for a diagnosis, 17,35,36 predict the progression of DR, 37 progression of age-related macular degeneration, [38][39][40] and development of glaucoma. 41 There are also several examples of algorithms for stratifying risk of DR. Estil et al used an algorithm to reduce the frequency of screening visits based on individual risk factors for DR progression. 42 Cunha-Vaz et al, 43 found that in eyes with mild non-proliferative DR, increased microaneurysm turnover rate and central macular thickness were correlated with an increased risk of progression to DME.…”