Traditionally, medical discoveries are made by observing associations, making hypotheses from them and then designing and running experiments to test the hypotheses. However, with medical images, observing and quantifying associations can often be difficult because of the wide variety of features, patterns, colours, values and shapes that are present in real data. Here, we show that deep learning can extract new knowledge from retinal fundus images. Using deep-learning models trained on data from 284,335 patients and validated on two independent datasets of 12,026 and 999 patients, we predicted cardiovascular risk factors not previously thought to be present or quantifiable in retinal images, such as age (mean absolute error within 3.26 years), gender (area under the receiver operating characteristic curve (AUC) = 0.97), smoking status (AUC = 0.71), systolic blood pressure (mean absolute error within 11.23 mmHg) and major adverse cardiac events (AUC = 0.70). We also show that the trained deep-learning models used anatomical features, such as the optic disc or blood vessels, to generate each prediction.
Automated Detection of Ocular & Systemic Disease Via Retinal Fundus Imaging
Retinal imaging in the form of fundus photography and OCT are well-established diagnostic tools for eye diseases such as diabetic retinopathy, glaucoma and age-related macular degeneration. It has also been suggested as a prognostic tool for the severity of systemic disease such as diabetes, stroke, and dementia. Our work centers around using machine learning and computer vision to (1) automate the detection of eye diseases which are currently diagnosed via retinal imaging and (2) identify novel features in retinal imaging that may be predictors or early signs of eye disease as well as systemic disease.
If successful, this work will help improve the detection of eye diseases and potentially other systemic diseases. Automated detection also has the potential of increasing efficiency and reducing costs. Using labeled fundus and OCT images as the main inputs, we will train computer algorithms to automatically predict image labels using machine learning and computer vision. We are requesting data from all patients that have had the retinal imaging performed. Per the UK Biobank look-up tool, this consists of 67,711 patients that make up the collection of 68,151 paired colour retinal photographs and optical coherence tomography (OCT) scans.
|Lead investigator:||Mr Philip Nelson|
|Lead institution:||Google LLC|