--- title: P2Echo · Text-Prompted Echo Segmentation emoji: 🫀 colorFrom: purple colorTo: green sdk: gradio sdk_version: 6.32.0 app_file: app.py short_description: Text-prompted multi-view echocardiography segmentation python_version: "3.12" startup_duration_timeout: 30m --- # P2Echo · Text-Prompted Echocardiography Segmentation **P2Echo** segments cardiac structures (LV, MYO, LA, RV, RA) across four echocardiography views (2CH, 3CH, 4CH, PSAX) with a single 37.5M-parameter model controlled by structured text prompts. Pick the view, select the structures to segment, and the demo builds the paper's view-aware prompt (e.g. *"Apical 4-Chamber (4CH): … Segment the LV: The LV appears as a bullet-shaped chamber forming the apex."*) and runs one forward pass. - **Paper**: [Prompt2Echo (P2Echo): Text-Prompted Multi-View Echocardiography Segmentation with One Model](https://arxiv.org/abs/2611.13425) - **Code**: [github.com/moeinheidari7829/P2Echo](https://github.com/moeinheidari7829/P2Echo) - **Weights**: [moein99/P2Echo](https://huggingface.co/moein99/P2Echo) ## How it works 1. The requested structures are turned into the paper's view-aware text prompts. 2. A frozen **Qwen3-Embedding-0.6B** text encoder embeds each prompt. 3. The **P2Echo** segmentation network (PVTv2-B2 encoder + DyITA prompt decoder) processes the frame with all structure prompts in one forward pass. 4. Each structure's logit map is thresholded and colour-blended over the input. ## Example images The bundled example frames are from the Yale University echocardiography atlas ([four chamber](https://commons.wikimedia.org/wiki/File:Apical_4_chamber_view.png), [two chamber](https://commons.wikimedia.org/wiki/File:Apical2Chamber.png), [short axis](https://commons.wikimedia.org/wiki/File:LeftVentricleShortAxis.gif)), via Wikimedia Commons, © Patrick J. Lynch and C. Carl Jaffe, Yale Center for Advanced Instructional Media, licensed under [CC BY 2.5](https://creativecommons.org/licenses/by/2.5/). ## Disclaimer P2Echo is a research prototype. It has **not** been validated for clinical use and must not be used for diagnosis or any clinical decision-making.