Semi-supervised Learning of Fetal Anatomy from Ultrasound
This work addresses the problem of applying semi-supervised learning to non-ideal medical imaging data for clinicians, but it is incremental as it highlights limitations rather than breakthroughs.
The study investigated whether semi-supervised learning methods, effective on standard benchmarks, translate to fetal 2D ultrasound anatomy classification, finding that including a challenging background class can be harmful and that benefits are limited to distinct classes, sometimes harming similar ones.
Semi-supervised learning methods have achieved excellent performance on standard benchmark datasets using very few labelled images. Anatomy classification in fetal 2D ultrasound is an ideal problem setting to test whether these results translate to non-ideal data. Our results indicate that inclusion of a challenging background class can be detrimental and that semi-supervised learning mostly benefits classes that are already distinct, sometimes at the expense of more similar classes.