A machine learning pipeline for predicting bone marrow oedema along the sacroiliac joints on magnetic resonance imaging

Researchers developed and validated a fully automated machine learning algorithm to predict bone marrow edema (BMO) on a quadrant-level in sacroiliac joint MRI scans. The algorithm utilized a computer vision workflow to locate the joints, segment regions of interest, and predict the presence of BMO based on T1/T2-weighted MRI slices. The algorithm achieved high precision in joint detection and segmentation. In cross-validation, the inflammation classifier performed well with an area under the curve (AUC) of 94.5% and balanced accuracy (B-ACC) of 80.5%. On a patient-level, the model achieved a B-ACC of 81.6% in the cross-validation dataset. This automated pipeline shows potential for assisting in diagnosing and monitoring inflammatory conditions.