Core Concepts
P-Mamba combines Perona-Malik Diffusion with Mamba for efficient pediatric echocardiographic left ventricular segmentation.
Abstract
1. Introduction
Importance of accurate cardiac function assessment in pediatric cardiology.
Challenges in echocardiography due to noise interference and segmentation efficiency.
AI advancements in pediatric cardiology.
2. P-Mamba
Combines vision mamba layers and DWT-based PMD blocks for efficient segmentation.
Achieves superior accuracy and efficiency compared to existing models.
3. Methodology
Vision Mamba encoder branch and DWT-based PMD encoder branch.
Decoders for segmentation masks generation supervised by loss functions.
4. DWT-based PMD Block
Utilizes Perona-Malik Diffusion for noise suppression while preserving boundary details.
Anisotropic diffusion equation used for noise reduction and edge preservation.
5. Vision Mamba Block
Improves computing and memory efficiency with global dependency modeling.
Transformation of input patches into vectors with position embedding.
6. Loss Function
Cross-entropy loss functions used exclusively for primary and auxiliary losses.
7. Experiment Results
Dataset from Lucile Packard Children's Hospital Stanford used for evaluation.
Model performance evaluated on precision, recall, and Dice coefficients.
8. Conclusion
P-Mamba offers higher accuracy and efficiency in pediatric cardiac imaging compared to existing models.