Lightweight Multireceptive Field CNN for 12-Lead ECG Signal Classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiogram (ECG) interpretation is crucial for diagnosing heart conditions but faces challenges due to cardiologist shortages and interpretation complexity.
• Existing computer-aided ECG analysis methods, particularly deep learning models like 1D-CNN, can be data-intensive and computationally demanding.
• The need for efficient and accurate automated ECG interpretation systems is critical for widespread clinical application.

Purpose of the Study:

• To develop and evaluate a novel deep learning architecture for improved ECG signal classification.
• To address the limitations of traditional deep learning models in ECG analysis by incorporating multi-receptive field concepts.
• To enhance the accuracy and efficiency of computer-aided diagnosis for cardiovascular diseases (CVDs) using ECG data.

Main Methods:

• Designed a custom multireceptive field Convolutional Neural Network (MRF-CNN) architecture tailored for 1D ECG time-series data.
• Utilized the PTB-XL dataset, a comprehensive ECG database, for training and evaluating the proposed MRF-CNN model.
• Compared the performance of the MRF-CNN against established methods for classifying ECG signals across different diagnostic granularities (superclasses, subclasses, and all diagnostic classes).

Main Results:

• The MRF-CNN architecture demonstrated improved performance in ECG classification tasks.
• Achieved an F1 score of 0.72 and AUC of 0.93 for classifying 5 ECG superclasses.
• Obtained an F1 score of 0.46 and AUC of 0.92 for 20 subclasses, and an F1 score of 0.31 and AUC of 0.92 for all diagnostic classes on the PTB-XL dataset.

Conclusions:

• The proposed MRF-CNN model effectively captures semantic context within ECG signals, leading to enhanced classification performance.
• This approach offers a promising solution for automated ECG interpretation, potentially alleviating the burden on healthcare professionals.
• The MRF-CNN architecture represents a significant advancement in applying deep learning for accurate and reliable cardiovascular disease detection from ECGs.