Machine Learning for Glaucoma Prediction
Supervised machine learning allows scientists to use data to make predictions about the future, including predicting the risk that an individual may develop a disease. This project aimed to produce a glaucoma risk prediction algorithm to aid early diagnosis and management of primary open-angle glaucoma.
Glaucoma has both genetic and environmental risk factors. In this project, we obtained genetic and clinico-demographic data from the UK Biobank, a longitudinal study of approximately 500,000 people recruited across the UK between 2006 and 2010. The goal was to use machine learning algorithms to produce a model that could predict whether an individual would develop primary open-angle glaucoma.
Several machine learning models were trained and tuned on different sets of features from the UK Biobank
The best performing model for prediction of primary open angle glaucoma was a ‘gradient boosting model’, which gave high levels of accuracy. Some of the predictive factors are well reocgnised such as increasing age, higher baseline intra-ocular pressure (the fluid pressure inside the eye), and greater baseline vertical cup-to-disc ratio (a measure comparing the vertical diameter of the cup part of the optic nerve with the total diameter of the optic nerve). Other measures we identified, that are less well recognised, included corneal measures, metabolic factors, waist circumference, and local pollution levels.
“Glaucoma often remains undetected until vision loss becomes severe; knowing which individuals are at high risk of developing glaucoma will help us know which individuals may need to be screened to detect glaucoma early.”
This research project was carried out by medical student Christopher Mayo as part of a Vision Research Foundation summer studentship in 2021-2022, as well as a 2022 elective dissertation project, supervised by Professor Helen Danesh-Meyer and Dr William Schierding.
Christopher Mayo is a final year medical student at the University of Auckland. His interests are . data science, computer science and machine learning, artificial intelligence, and their applications in medicine. “In not-too-distant future, machine learning will revolutionise clinical decision-making, improving diagnosis and management of patients. I have hands-on machine learning and data analysis experience in Python, having completed a number of previous machine learning and data analysis projects. I hope to combine my passion for AI and my passion for medicine in order to improve the workflows of clinicians and the lives of patients. As part of this, I have a strong passion for research, and I am hoping to eventually pursue a PhD in this area.” Mayo said to Vision Research Foundation.
