A deep multimodal early fusion architecture for automated phonocardiogram classification achieved a quantitative accuracy of 99.99% for heart valve disease data and 92.78% for CinC data.
Does a deep multimodal early fusion architecture improve the automated detection of heart valve diseases from PCG signals?
A novel deep multimodal fusion architecture for PCG signals achieves high accuracy and interpretability in the automated detection of heart valve diseases.
The early and accurate detection of heart valve disease is crucial in cardiovascular diagnosis. With the introduction of phonocardiogram (PCG), automated diagnosis of heart diseases is employed with advanced deep learning techniques. In this paper, we propose a novel framework for automated PCG classification using deep multimodal early fusion architecture. The proposed method integrates two heterogeneous modalities of PCG such as 1D temporal and 2D time frequency representations using deep multimodal fusion architecture. Two public PCG databases PhysioNet/Computing in Cardiology Challenge (CinC) 2016 database (binary data) and Yaseen Heart Valve Disease (HVD) database (multiclass data) are used for the experimental analysis. With extensive ablation studies and comprehensive evaluation with state-of-the-art methods, the proposed method achieved a high quantitative accuracy of 99.99% for HVD data and 92.78% for CinC data demonstrating the effectiveness of proposed approach. To assure clinically aligned interpretability of the proposed method, we implemented explainable artificial intelligence (XAI) techniques like Local Interpretable Model-Agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (Grad-CAM) and Shapley Additive exPlanations (SHAP). To measure the consistency and significance of the proposed model several statistical analysis are implemented, which showcased high calibrated results proving the reliability and generalizability of the proposed model across real time clinical scenarios. In summary, in this work we introduce a strong and interpretable diagnostic pipeline that efficiently performs heart valve disease classification with trustful AI assistance.
K.P et al. (Thu,) conducted a other in Heart valve disease. Deep multimodal early fusion architecture vs. State-of-the-art methods was evaluated on Quantitative accuracy. A deep multimodal early fusion architecture for automated phonocardiogram classification achieved a quantitative accuracy of 99.99% for heart valve disease data and 92.78% for CinC data.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: