Researchers introduced BioTTA, an AI method that adjusts a pretrained model to each new, unlabeled fetal brain MRI scan. In a retrospective study, they tested it on 394 reconstructed scans from pregnancies at 20–39 weeks, using 11 measurements annotated by radiologists.
The authors report that BioTTA outperformed several comparison methods across datasets with differences in MRI sequence, center and country, and matched models fine-tuned with labeled data. Its reported diagnostic accuracy was 82.1%–95.0%, comparable to a senior radiologist in the study; the paper does not establish how it would perform in routine clinical use.
Why it matters
For radiology teams working with fetal MRI from different scanners or hospitals, BioTTA could help standardize measurements without target-site labels, but its clinical performance outside this retrospective study remains unknown.
