0
0
Researchers at the University of Adelaide have introduced a pioneering approach to endometriosis classification, integrating machine learning models with human expertise. This breakthrough system, named Human-Artificial Intelligence Collaborative Multi-modal Multi-rater Learning (HAICOMM), was developed by the IMAGENDO team from the University’s Robinson Research Institute and the Australian Institute for Machine Learning (AIML). Their findings have been published in Physics in Medicine & Biology.
Addressing Diagnostic Challenges
HAICOMM has been designed to eliminate three key challenges in endometriosis diagnosis by incorporating artificial intelligence (AI) and human perspectives at multiple stages:
- Multi-rater learning: The system refines and combines multiple inconsistent or ‘noisy’ labels for each training sample, resulting in clearer and more reliable labeling.
- Multi-modal learning: By leveraging T1- and T2-weighted MRI images during training and testing, the system enhances its diagnostic accuracy and understanding.
- Human-AI collaboration: HAICOMM integrates predictions from clinicians with AI-generated outputs, ensuring more accurate and reliable classifications than either could achieve independently.
A major diagnostic challenge in endometriosis is detecting Pouch of Douglas obliteration, a condition where the space between the uterus and rectum is blocked. Even experienced clinicians often struggle to identify this condition accurately in MRI images, with manual classification accuracy ranging from only 61.4% to 71.9%.
The Urgent Need for AI-Enhanced Diagnosis
Endometriosis is a condition where tissue similar to the uterine lining grows outside the womb, affecting approximately 14% of individuals assigned female at birth. The condition is often difficult to diagnose, with patients waiting an average of 6.4 years for a formal diagnosis, typically confirmed through imaging or invasive laparoscopic surgery.
“The long waiting period significantly reduces patients’ quality of life,” says Dr. Yuan Zhang, an IMAGENDO researcher at the Robinson Research Institute. “Furthermore, the reliance on invasive diagnostic methods increases healthcare costs and burdens both patients and medical systems.”
Expanding AI Capabilities
With HAICOMM, the IMAGENDO team aims to enhance diagnostic accuracy and reliability. The next step involves integrating HAICOMM into the IMAGENDO-patented algorithm, which will utilize both MRI and transvaginal ultrasound images to further improve detection rates.
“Beyond Pouch of Douglas obliteration, we plan to expand HAICOMM’s capabilities to detect bowel nodules, endometriomas, and uterosacral ligament endometriosis,” says Dr. Zhang. “The upcoming phase will focus on testing the system on diverse datasets and validating its effectiveness in clinical settings.”
Future Prospects
The team believes that HAICOMM’s combination of AI precision and human expertise represents a significant step toward non-invasive, accurate, and faster diagnosis of endometriosis. This development has the potential to alleviate diagnostic delays, reduce patient suffering, and optimize healthcare resources.
Disclaimer: The research findings are still in the early stages, and further clinical validation is required before HAICOMM can be widely implemented in healthcare settings. Patients should consult medical professionals for diagnosis and treatment.
