Unsupervised Anomaly Detection of Diseases in the Female Pelvis for Real-Time MR Imaging
New AI flags hard-to-detect pelvic diseases in women during live MRI scans.
Deep Dive
Researchers developed an AI that learns from healthy pelvic MRI scans to spot anomalies in real-time, without needing labeled disease data. Trained on 294 healthy scans and synthetic data, it achieved 83% sensitivity and 69% specificity in detecting conditions like fibroids and endometriosis. It processes images at 93 frames per second, establishing a benchmark for fast, generalizable disease detection to aid earlier diagnosis where anatomical variability complicates analysis.
Why It Matters
This could enable faster, more accurate diagnosis of common but often missed pelvic conditions in women.