Classification of aortic stenosis using conventional machine learning and deep learning methods based on...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Aortic stenosis (AS) diagnosis often relies on invasive or complex imaging techniques.
• Non-invasive methods for AS assessment are crucial for early detection and patient monitoring.
• Cardio-mechanical signals offer a novel data source for analyzing cardiac conditions.

Purpose of the Study:

• To develop and evaluate a machine learning framework for classifying aortic stenosis (AS) using non-invasive cardio-mechanical signals.
• To compare the performance of various machine learning algorithms and deep learning approaches for AS classification.
• To assess the efficacy of feature reduction techniques in enhancing classification accuracy.

Main Methods:

• Collected cardio-mechanical signals from 21 AS patients and 13 non-AS subjects using wearable inertial sensors.
• Applied Elastic Net for feature selection on data processed by continuous wavelet transform (CWT), achieving 95.47% feature reduction.
• Compared machine learning algorithms (Decision Tree, Random Forest, MLP, XGBoost) and developed 2D-CNN models (custom and MobileNet transfer learning).

Main Results:

• Random Forest achieved the highest accuracy (0.96) among traditional ML algorithms.
• XGBoost and a simple neural network showed strong performance with accuracies of 0.95 and 0.91, respectively.
• The 2D-CNN models demonstrated competitive results, with MobileNet transfer learning reaching 0.91 accuracy and a custom CNN achieving 0.89.

Conclusions:

• The proposed feature selection and classification framework effectively identifies aortic stenosis using non-invasive sensor data.
• Machine learning and deep learning models, particularly Random Forest and XGBoost, show significant potential for accurate AS classification.
• This study highlights the promise of leveraging deep learning tools for non-invasive diagnosis and management of aortic stenosis.