Echo-SMADS: A hierarchical planning model for predicting ejection fraction using echocardiography.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Echocardiography Analysis
• Deep Learning for Cardiovascular Assessment

Background:

• Current deep learning models for ejection fraction (EF) prediction often use end-to-end regression, lacking interpretability and clinical workflow alignment.
• Limited transparency in AI-driven EF assessment hinders clinical adoption and trust.
• Need for a system that mirrors physician diagnostic processes for improved reliability.

Purpose of the Study:

• To design a clinically aligned, modular system (Echo-SMADS) for ejection fraction prediction.
• To improve interpretability, stability, and real-world applicability of AI-based EF assessment.
• To emulate the diagnostic workflow of physicians for enhanced transparency.

Main Methods:

• Proposed Echo-SMADS system utilizes hierarchical planning from artificial intelligence.
• Decomposed EF prediction into three clinically relevant subtasks: structure identification, phase selection, and volume estimation.
• Implemented as decoupled, independently optimized functional modules with interpretable intermediate outputs for transparent reasoning.

Main Results:

• Echo-SMADS achieved a mean absolute error of 5.48 ± 0.17 and a root mean square error of 7.64 ± 0.20 on the EchoNet-Dynamic dataset.
• Demonstrated improved performance stability compared to traditional end-to-end models.
• Provided meaningful intermediate outputs, enhancing interpretability and trustworthiness of EF predictions.

Conclusions:

• Echo-SMADS integrates modular, clinically aligned components reflecting real-world diagnostic workflows.
• The system combines interpretability, physical grounding, and performance stability for reliable EF prediction.
• Presents a promising approach for future clinical application in cardiovascular assessment.