Prompt2Echo (P2Echo): Text-Prompted Multi-View Echocardiography Segmentation with One Model
The authors report matching or beating specialist segmentation models at a fraction of their computational cost.
P2Echo uses structured text prompts to specify the echocardiographic view and cardiac structure to segment. Its Dynamic Image-Text Alignment module adjusts decoder features to the request, letting one model switch between views without architectural changes. Evaluation covered an internal dataset used during training and three held-out external datasets. The abstract reports performance matching or exceeding both view-specific specialists and leading universal models, but gives no numerical results or compute savings.
HF Daily Papers' note
P2Echo uses structured text prompts to specify the echocardiographic view and cardiac structure to segment. Its Dynamic Image-Text Alignment module adjusts decoder features to the request, letting one model switch between views without architectural changes. Evaluation covered an internal dataset used during training and three held-out external datasets. The abstract reports performance matching or exceeding both view-specific specialists and leading universal models, but gives no numerical results or compute savings.
HF Daily Papers' note
score 4