Human-AI collaborative multi-modal multi-rater learning for endometriosis diagnosis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Women's Health

Background:

• Endometriosis affects 10% of individuals assigned female at birth and is difficult to diagnose.
• Current diagnosis relies on laparoscopy or MRI, with MRI being faster but less accurate.
• Accurate classification of pouch of Douglas (POD) obliteration on MRI is challenging for clinicians and AI models.

Purpose of the Study:

• To introduce a novel Human-AI Collaborative Multi-modal Multi-rater Learning (HAICOMM) methodology.
• To address the challenge of accurate endometriosis diagnosis using MRI, specifically POD obliteration.
• To develop a system that combines clinician and AI predictions for improved diagnostic accuracy.

Main Methods:

• The HAICOMM methodology integrates multi-rater learning to refine labels from noisy data.
• It employs multi-modal learning, utilizing both T1 and T2 MRI images.
• Human-AI collaboration is central, combining expert and AI predictions.

Main Results:

• The HAICOMM model demonstrated superior performance on a multi-rater T1/T2 MRI endometriosis dataset.
• It significantly outperformed ensembles of clinicians, noisy-label learning, and multi-rater learning models.
• The methodology validated its effectiveness in classifying POD obliteration.

Conclusions:

• HAICOMM offers a significant advancement in diagnosing endometriosis from MRI.
• The approach has the potential to enhance diagnostic accuracy for POD obliteration.
• Improved diagnosis can lead to better management of this widespread condition.