An innovative, artificial intelligence program that could be used to help clinicians diagnose juvenile arthritis is showing promise.
Currently there is no single test to diagnose juvenile arthritis and, in some cases, it can take up to 10 months for families to find an answer.
But thanks to your donations to The Hospital Research Foundation Group – Arthritis, Dr Zhibin Liao and his team at the Australian Institute for Machine Learning have been developing a model that could help make getting a diagnosis easier.
Dr Liao said the model is trained to identify visual patterns from a sample of 220,000 paediatric x-rays and has been able to identify arthritis with an accuracy over 90%.
“The team has worked extensively to process 220,000 paediatric X-rays and curate a clean dataset suitable for AI development,” he said.
“Using the remaining data from the 220,000 X-rays, the team trained an AI foundation model with a robust understanding of X-ray images. This method enables the AI model to link imaging data with clinical information from radiology reports, which is a cutting-edge AI technique proven to enhance performance, particularly when data is limited.
“With this foundation model, our AI achieved 84% AUC and 93% accuracy in classifying arthritis directly from raw paediatric X-rays.”
The technology works by providing prediction scores and highlighting the location of potential arthritis symptoms to help clinicians make a diagnosis earlier.
While AI is a highly intelligent tool, some models have what’s called a ‘black box’, which means that it can arrive at a conclusion without showing how it got there.
Dr Liao said the team has implemented a visualization algorithm which will allow clinicians to review the model’s work.
“The algorithm highlights areas of the X-ray the AI focuses on when predicting arthritis, which provides insights into the model’s decision-making process and helps clinicians assess whether the AI is targeting the correct joints,” he said.
The technology is set to be tested in a clinical setting to assess its diagnostic capabilities early this year, and we look forward to keeping you updated on how it progresses!
