Phonocardiogram transfer learning-based CatBoost model for diastolic dysfunction identification using multiple...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Left ventricular diastolic dysfunction detection is crucial for cardiac screening.
• Non-invasive methods are highly desirable for evaluating cardiac function.

Purpose of the Study:

• To develop and validate a phonocardiogram (PCG) transfer learning-based CatBoost model for non-invasive detection of diastolic dysfunction.
• To compare the performance of the proposed model against other machine learning classifiers.

Main Methods:

• Utilized Short-Time Fourier Transform (STFT), MFCCs, S-transform, and gammatonegram for PCG signal representation.
• Employed pre-trained Convolutional Neural Networks (CNNs) (VGG16, Xception, ResNet50, InceptionResNetv2) for feature extraction via transfer learning.
• Fused extracted deep features and applied Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) before classification with CatBoost.

Main Results:

• The CatBoost model with LDA-based feature fusion achieved superior performance: AUC of 0.911, accuracy of 0.882, sensitivity of 0.821, specificity of 0.927, and F1-score of 0.892.
• Deep features from gammatonegram extracted by ResNet50 showed the highest contribution to classification.
• The proposed model outperformed other tested classifiers like multilayer perceptron, SVM, and random forest.

Conclusions:

• The developed PCG transfer learning-based CatBoost model offers a promising non-invasive approach for diastolic dysfunction detection.
• This method can significantly contribute to the non-invasive evaluation of diastolic cardiac function.