PRISM-CTG: Self-Supervised Foundation Model for Cardiotocography Analysis
Researchers propose PRISM-CTG, a self-supervised foundation model for automated CTG analysis. It uses a multi-view self-supervised framework with three pretext objectives: masked signal reconstruction, clinical variable prediction, and feature classification. The model leverages large-scale unlabelled recordings to learn transferable representations, overcoming limitations of supervised models on curated datasets.
Key facts
- PRISM-CTG stands for Physiology-aware Representation Learning via Integrated Self-supervision and Metadata for CTG.
- It is a clinically grounded self-supervised foundation model for cardiotocography analysis.
- The model uses a multi-view self-supervised framework with three complementary pretext objectives.
- The three objectives are: random-projected guided masked signal reconstruction, clinical variable prediction, and feature classification.
- Each objective has a dedicated task-specific token for specialized representation learning.
- The model is pretrained on large-scale unlabelled clinical recordings.
- It aims to learn transferable domain-level representations.
- Supervised deep learning models for CTG are limited by narrowly curated labelled datasets and small patient cohorts.
Entities
—