Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning

Abstract

Traditionally, medical discoveries are made by observing associations and then designing experiments to test these hypotheses. However, observing and quantifying associations in images can be difficult because of the wide variety of features, patterns, colors, values, shapes in real data. In this paper, we use deep learning, a machine learning technique that learns its own features, to discover new knowledge from retinal fundus images. Using models trained on data from 284,335 patients, and validated on two independent datasets of 12,026 and 999 patients, we predict cardiovascular risk factors not previously thought to be present or quantifiable in retinal images, such as such as age (within 3.26 years), gender (0.97 AUC), smoking status (0.71 AUC), HbA1c (within 1.39%), systolic blood pressure (within 11.23mmHg) as well as major adverse cardiac events (0.70 AUC). We further show that our models used distinct aspects of the anatomy to generate each prediction, such as the optic disc or blood vessels, opening avenues of further research.

Current state and future prospects of artificial intelligence in ophthalmology: a review

“…63 Another looked at contact lenses for non-invasively detecting Staphylococcus aureus. 72 Ultimately AI is being researched for its benefits across a wide variety of applications in ophthalmology and beyond. 71 Interestingly, some studies are using the ophthalmic datasets for utilization across other aspects of health.…”

“…71 Interestingly, some studies are using the ophthalmic datasets for utilization across other aspects of health. 72 Ultimately AI is being researched for its benefits across a wide variety of applications in ophthalmology and beyond.…”

Development of an artificial intelligence system to classify pathology and clinical features on retinal fundus images

“…The system also allows researchers to access the TensorFlow backend to deepen their understanding of the AI architecture and training process. Our resulting proof‐of‐concept system used a modular architecture allowing for additional modules to be added (eg, for detecting haemoglobin A1c values, or systolic blood pressure, which have previously been developed by Poplin et al) and existing modules to be updated with further training. An algorithm could be conceivably trained to more rigorously classify disease—such as using the standardized diabetic retinopathy grading system used in New Zealand…”

Screening und Management retinaler Erkrankungen mittels digitaler Medizin

“…Hierfür sind allerdings weit größere Datensätze erforderlich. Auf Basis von fast 300.000 Farbfotos entwickelten Forscher der Firma Google ein AI-Modell, mit dem unter anderem das Alter, Geschlecht, Raucherstatus, Blutdruck und das kardiovaskuläre Risiko mit erstaunlich hoher Präzision bestimmt werden konnten [19].…”