Research record · Not a deployment caseClinical-AI research · Privacy-aware data

Edwards Lifesciences research

Clinical-AI research under privacy constraints.

Partner / contextEdwards Lifesciences
Binding constraint

Make ultrasound usable for research without removing the diagnostic evidence the research depended on.

Binding constraintMake ultrasound usable for research without removing the diagnostic evidence the research depended on.
01Source study
02Preserve diagnostic cone
03Mask identifying exterior
04Research-ready output
05Reproducible benchmark

Research across privacy-aware cardiac-ultrasound preparation, weak-supervision experiments, and an independent EchoNet-Dynamic reimplementation.

3 methodsInstance · semantic · classical
4 stagesWeak-supervision curriculum
IndependentBenchmark reimplementation
ResearchNo deployment or clinical claim

Cardiac-ultrasound frames can carry identifying material outside the diagnostic cone. A blunt crop could protect identity while erasing color-Doppler evidence the research needed.

Preparation therefore treated de-identification as evidence preservation, not generic image cleaning.

Nixense contributed privacy-aware preparation, experimental implementation, and independent benchmark work in a collaborative research setting. Collaboration is not presented as deployment or sole authorship.

Instance segmentation, semantic segmentation, and classical projection were compared against whether the exterior was masked while Doppler information inside the cone survived.

The experiment progressed through increasingly complex synthetic bags and inspected positive/negative instance balance before acceptance.

The independent EchoNet-Dynamic path followed segmentation, video regression, cardiac-cycle extraction, and beat-level aggregation. It established a reproducible research reference — not clinical validation or superiority over Stanford.

Evidence boundary

No Edwards product deployment, clinical-care effect, regulatory approval, validated clinical performance, or sole Nixense authorship is claimed.

Record close
The work established a privacy-aware data path, weak-supervision experiments, and a reproducible benchmark reference. It did not establish clinical performance or production deployment.

Working on a problem like this?

Tell us what the system needs to do, what exists today, and where you need help.

Discuss a system
Principal review[email protected]Pakistan · Working internationally