Advancing endometriosis detection in daily practice: a deep learning-enhanced multi-sequence MRI analytical model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Women's Health

Background:

• Endometriosis affects 5-10% of reproductive-age women, posing diagnostic challenges.
• Deep learning (DL) offers potential for improving medical diagnoses.
• Accurate imaging is crucial for endometriosis detection and management.

Purpose of the Study:

• To evaluate deep learning (DL) tools for enhancing multi-sequence MRI accuracy in endometriosis detection.
• To compare DL model performance against experienced radiologists.

Main Methods:

• A cohort of 395 endometriosis patients and 356 controls underwent multi-sequence MRI (T1W FS, T2W).
• A 3D-DenseNet-121 DL model was trained and tested using a patient-level split (12.5% test set).
• Seven radiologists and one fellow reviewed images with and without AI assistance.

Main Results:

• The DL model achieved high performance (F1 Score 0.881, AUROCC 0.911).
• AI assistance improved radiologist sensitivity from 84.48% to 87.93% and inter-reader agreement (Fleiss' kappa from 0.5718 to 0.6839).
• Multi-sequence MRI (T2W, T1W FS pre/post-contrast) yielded the most accurate predictions.

Conclusions:

• The study presents the first DL model for multi-sequence MRI-based endometriosis detection in a large cohort.
• DL model performance is comparable to trained human readers.
• AI tools can significantly aid radiologists in endometriosis diagnosis.