Today is World Heart Day. Don’t miss a beat.

This interdisciplinary work Noble Group and Nuffield Dept of Women's & Reproductive Health led by Prof. Alison Noble and Prof. Aris Papageorghiou and our international partners, aims make fetal heart screening more consistent and accessible - so every tiny heartbeat gets the attention it deserves.

Why this matters
• CHD is the most common birth defect. Early detection during routine pregnancy scans can change care plans and outcomes. 👶
• World Heart Day (29 September) unites the world around prevention, early screening, and access to care. 2025 marks 25 years of this global campaign with the theme “Don’t miss a beat.” (world-heart-federation.org)

What we’re building in CAIFE
• A quick, standardized Sweep: during the routine scan, the sonographer performs a short sweep across the standard fetal heart views. ⏱️
• Instant triage: our AI flags “looks normal” or “needs expert review,” helping flag scans that may require fetal cardiologist review faster. It supports clinicians and does not replace them. 🧑‍⚕️🧠
• Learn “what healthy looks like”: trained primarily on normal fetal heart videos, so the system can spot unusual patterns without extensive abnormal labels.
• Privacy-first collaboration: each hospital keeps data on site. We merge models, not data, so the tool benefits from many sites while protecting patient privacy and generalizing across scanners and populations. 🔐🌍

Presence at MICCAI 2025
• We presented this work at the MICCAI 2025 main conference. Thank you to everyone who engaged with the talk and shared thoughtful feedback. 🙌

Follow-up work: DISCOVR (accepted at NeurIPS 2025 🎉)
• A self-supervised video foundation model for echocardiography that spans fetal, pediatric, and adult populations. One backbone for multiple tasks: zero-shot anomaly detection, classification, and cardiac structure segmentation. It also supports EF prediction with light fine-tuning. ❤️

Learn more
• Zero-shot CHD in fetal ultrasound (STUD + DivMerge): https://arxiv.org/pdf/2503.07799
• DISCOVR (echo video representations): https://arxiv.org/pdf/2506.11777