Pioneering AI to predict outcomes in macular degeneration 

Artificial intelligence (AI) has recently been recognised as one of the main game-changers in medicine for the years ahead. Recent advances have transformed the technology industry and researchers have been increasingly finding exciting ways to apply AI to medicine, and more particularly in eye health. 

Dr Konstantinos Balaskas, director, Moorfields Image Reading Centre, received a Springboard award to pilot a machine learning tool that could help clinicians better predict how patients with age-related macular degeneration (AMD) would respond to different treatment options, and which treatment works best for them.

Macular degeneration is a common eye condition where cells in the middle of the retina (the light-sensitive layer at the back of your eye) become damaged. This leads to the loss of central vision, making it difficult to see fine details clearly. The most common form of macular disease is age-related macular degeneration (AMD), which generally affects people over the age of 50. AMD comes in two forms: dry and wet AMD.

Although there are treatments to slow down the loss of vision caused by wet AMD, numerous factors influence how well patients respond to these treatments. Clinical information about the patient, such as their age and the stage of their condition, and each person’s genetic make-up all play a role. However, we don’t yet have the capacity to bring all this complex information together to predict which treatment option an individual patient will respond best to.

AI represents a breakthrough in computing, enabling systems to recognise patterns in large quantities of data, and learn and make informed decisions traditionally associated with human intelligence.

Being the largest eye hospital in the UK, Moorfields Eye Hospital collects vast amounts of clinical and scientific data on a daily basis, and herein lies a clear opportunity to gain new scientific insights by analysing the trends in this information. These datasets, however, are typically considered to be​‘raw’ data, which can be difficult to assess using traditional statistical techniques due to the sheer volume collected.

New advances in AI, in particular machine learning, can help make sense of all this data. Machine learning uses artificial neural networks as a computational model to discover complex structures in large, high-dimensional datasets, such as the ones collected at the hospital. 

To harness the potential of AI in this setting, Dr Balaskas’s team had to firstly ensure that all the data being used was in a standardised, digital format that is suitable for machine learning analysis. 

The team collected clinical, genetic and imaging information from over 600 Moorfields patients with wet AMD. Through the imaging data gathered, a novel machine learning segmentation tool was developed and applied to this dataset using various image analysis platforms. 

The study’s end goal was to establish a machine learning model that uses this high-dimensional data to predict the course of wet AMD in each patient, by revealing which factors play a key role in how they respond to treatment.

This helped to show that developing a large dataset of clinical, imaging and genetic data in a form that is suitable for machine learning can lead to exciting new insights into the behaviour of wet AMD.

The team leading this research project is currently focused on creating an extensive pipeline for artificial intelligence. This will expand its application beyond wet AMD and help diagnose various commonly occurring eye conditions. The ultimate objective is to enable doctors to provide personalised treatments that cater to the needs of each patient.