Fine-tuning a Foundation Model for Fetal Orientation Prediction From Blind-Sweep Ultrasound Videos

Chun Kit Wong, Mary Le Ngo, Jakob Ambsdorf, Anders Nymark Christensen, Mads Nielsen, Martin Grønnebæk Tolsgaard, Aasa Feragen

Keywords: fetal ultrasound, navigation, foundation model

Registration Requirement: Yes

Abstract: Previous work on determining fetal orientation from blind ultrasound sweeps has relied on anatomical heuristics, specifically fetal head detection and segmentation. However, clinical validation reveals that this approach is not robust, frequently failing when the head is malpositioned. To address this, we propose a temporally-aware framework that leverages a foundation model to encode video frames into a 768-dimensional embedding space, followed by an ensemble of 1D-convolutional classification heads to predict fetal orientation directly.

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 74