A Robust Multilevel DWT Densely Network for Cardiovascular Disease Classification

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Medicine

Background:

• Cardiovascular disease is a leading global cause of mortality, necessitating precise and timely diagnosis.
• Existing methods for heartbeat classification often struggle with class imbalance and require extensive data preprocessing.
• The intra-patient paradigm, while common, may not fully represent real-world diagnostic challenges.

Purpose of the Study:

• To develop a robust classification system for accurate detection of normal heartbeats and cardiovascular diseases including coronary artery disease (CAD), myocardial infarction (MI), and congestive heart failure (CHF).
• To overcome limitations of existing methods, specifically addressing class imbalance and reliance on preprocessing.
• To evaluate the system's performance under both intra-patient and inter-patient diagnostic scenarios.

Main Methods:

• A novel multilevel discrete wavelet transform densely network (MDD-Net) was developed for feature extraction and classification.
• An adaptive sample frequency segmentation algorithm (ASFS) was employed to segment raw ECG signals into uniform segments.
• Fusion features were extracted using MDD-Net to enhance classification performance, minimizing reliance on traditional preprocessing steps.

Main Results:

• The MDD-Net achieved high diagnostic accuracy, with average metrics including 99.74% accuracy, 99.09% positive predictive value, 98.67% sensitivity, and 99.83% specificity in the intra-patient paradigm.
• Under the more challenging inter-patient paradigm, the model demonstrated strong performance with 96.92% accuracy, 92.17% positive predictive value, 89.18% sensitivity, and 97.77% specificity.
• Experimental results confirmed the model's robustness against noise and class imbalance issues inherent in ECG data.

Conclusions:

• The proposed MDD-Net offers a robust and accurate system for classifying normal heartbeats and detecting various cardiovascular diseases from ECG signals.
• The method effectively handles class imbalance and noise, outperforming traditional approaches and demonstrating significant potential for clinical application.
• The system's strong performance in both intra-patient and inter-patient settings highlights its adaptability and reliability for real-world cardiovascular diagnostics.