P2Echo segments multi-view echocardiography from text prompts with one model
Original title: Prompt2Echo (P2Echo): Text-Prompted Multi-View Echocardiography Segmentation with One Model
A routing module switches decoder pathways by view and structure, matching specialists at lower compute.
Researchers present P2Echo, a single model that segments diverse cardiac structures across standard echocardiographic views using structured view-aware text prompts. Dynamic Image-Text Alignment uses differentiable routing and token differential operators to modulate decoder features without architectural changes. On a large internal dataset and three external datasets, it matches or exceeds view-specific specialists and universal models at a fraction of their cost. Code is available on GitHub.
- DyITA conditions segmentation on the requested view and target structure.
- Held-out evaluation covers three external datasets plus a large internal set.
- Authors report specialist-level accuracy at a fraction of the compute.
- Code is published at github.com/moeinheidari7829/P2Echo.
From the original article
1 min read at huggingface.co
Automated segmentation in multi-view echocardiography is essential for comprehensive cardiac assessment, yet remains challenging due to substantial anatomical and appearance variability across standard acoustic windows.
By Moein Heidari, Alvin Kimbowa, Mahdi Tabatabaei · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.