We’re creating a faster, less invasive way to diagnose endometriosis through machine learning

Endometriosis is a debilitating disease with serious impacts on a person’s quality of life far beyond the extreme pain it causes. It can affect them financially, cause disruption to their work, social lives and relationships.

By the age of 44, one in nine Australian women (and those assigned female at birth) are diagnosed with endometriosis. In 2016/17 it hospitalised 34,000 patients.

Endometriosis occurs where tissue similar to the lining of the uterus, grows outside the uterus often causing intense pain and for some fertility problems. Diagnosis is often delayed, with an average of 6.4 years between onset of symptoms and diagnosis.

The only reliable way of currently diagnosing endometriosis is to perform keyhole surgery to see the endometriosis lesions inside the abdomen, ideally then verified by microscopic examination of the tissue.

This method is considered the gold standard for the diagnosis of endometriosis, but surgery can be problematic, difficult to access, and add to delays. Non-surgical diagnosis can be particularly tricky, especially when doctors aren’t specifically trained to identify endometriosis in ultrasound or MRI.

Researchers from the Robinson Research Institute and the Australian Institute for Machine Learning (AIML) are collaborating to harness artificial intelligence to facilitate less invasive and quicker diagnosis of endometriosis.

Professor Gustavo Carneiro of the Australian Institute for Machine Learning is supervising the design and implementation of a program that can read specialist scans and recognise certain imaging markers seen in endometriosis. It will help doctors provide surgery-free diagnosis with initial tests showing the software is capable of diagnostic accuracy approaching that of a specialist doctor.

Co-lead researcher Professor Louise Hull of the Robinson Research Institute says the IMAGENDO project will provide a cost-effective, accessible, and accurate method to non-invasively diagnose endometriosis.

“We’re using machine learning to combine the diagnostic capabilities of pelvic ultrasound scans and magnetic resonance imaging (MRI) to identify endometriosis lesions,” says Professor Hull.

Co-lead researcher Dr Jodie Avery explains machine learning is an application of artificial intelligence that provides systems the ability to automatically learn and improve from experience.

“Machine learning is an iterative process – as you give more and more training samples, the accuracy of the system improves,” Dr Avery said.

What’s next?

The IMAGENDO project is ongoing and will evolve as more data becomes available.

Professor Carneiro says machine learning algorithms like this could hasten identification of endometriosis when a specialist isn’t available, fast-tracking delivery of surgical, medical and fertility care.

“We hope that our approach will soon mean patients from all over Australia will have access to high quality, non-invasive screening for endometriosis,” says Professor Carneiro.