Automated Myocardial Wall Motion Classification using Handcrafted Features vs a Deep CNN-based mapping.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Ultrasound image analysis is challenging due to sonographer expertise, machine variability, and image artifacts.
• Manual segmentation of cardiac walls shows significant inter-observer variability.
• Accurate wall motion analysis is crucial for diagnosing cardiac conditions.

Purpose of the Study:

• To develop a fully automated image analysis framework for wall motion abnormality classification in 2D+T ultrasound images.
• To compare traditional machine learning with deep learning approaches for this task.
• To evaluate the impact of Feature Asymmetry preprocessing on deep learning performance.

Main Methods:

• Exploration of Random Forests with handcrafted features.
• Implementation of a spatio-temporal Convolutional Neural Network (CNN) for hierarchical feature aggregation.
• Integration of Feature Asymmetry (FA) for local phase information retrieval in video preprocessing.

Main Results:

• The deep learning CNN approach demonstrated superior performance compared to traditional methods.
• Preprocessing ultrasound videos with Feature Asymmetry significantly enhanced the CNN's ability to identify relevant endocardial features.
• The automated framework successfully classified wall motion abnormalities.

Conclusions:

• A fully automated framework for cardiac ultrasound analysis and wall motion abnormality classification is feasible.
• Deep learning, particularly with Feature Asymmetry preprocessing, offers a robust solution to overcome ultrasound imaging limitations.
• This approach has the potential to reduce diagnostic variability and improve cardiac assessment.