Echo-POSED guides echocardiography with 8.2° accuracy, no expert labels
AI guides ultrasound probes in real-time using only 2D slices from 3D scans.
Echo-POSED is a self-supervised framework for real-time transthoracic echocardiography (TTE) guidance that recommends probe adjustments directly from 2D ultrasound images—without expert-labelled views or tracked probe trajectories. It trains on 2D views sliced from routinely acquired 3D volumes and enforces equivariance to probe motions while remaining invariant to cardiac phase. Across a held-out split and public external datasets (including vendor shift), it maintains geometric consistency under virtual perturbations and enables both intra- and inter-patient guidance simulations, achieving a combined mean angular error of 8.2 degrees in intra-patient simulations with cardiac motion.
- Trains on 2D slices from routine 3D echocardiography volumes—no expert labels or tracked probe trajectories needed.
- Achieves mean angular error of 8.2° in intra-patient guidance simulations under cardiac motion.
- Maintains geometric consistency across different vendors (vendor shift) and works for both intra- and inter-patient guidance.
Why It Matters
Echo-POSED could democratise cardiac ultrasound by removing the need for expert labelling, reducing training time and operator dependence.