Automated ECG classification based on 1D deep learning network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• The 12-lead electrocardiogram (ECG) is a vital, non-invasive tool for diagnosing cardiac conditions.
• Manual ECG interpretation is time-consuming, requires specialized expertise, and can be strenuous.
• Deep learning (DL) shows promise for automating medical image analysis, including ECG interpretation.

Purpose of the Study:

• To develop and evaluate an automated system for classifying normal and abnormal 12-lead ECG signals.
• To implement a novel multi-channel, multi-scale deep neural network (DNN) for end-to-end ECG signal classification.
• To assess the performance of the proposed DNN model in identifying cardiac abnormalities from ECG data.

Main Methods:

• A multi-channel, multi-scale deep neural network (DNN) architecture was designed for end-to-end ECG classification.
• Convolutional layers were utilized for feature extraction, complemented by Long Short-Term Memory (LSTM) and attention mechanisms.
• The model was trained and validated using a 12-lead ECG dataset from Kaohsiung Medical University Hospital (KMUH).

Main Results:

• The proposed DNN system demonstrated high recognition rates in classifying ECG signals.
• The integration of LSTM and attention mechanisms enhanced the model's performance in ECG analysis.
• The end-to-end approach eliminated the need for manual feature extraction, simplifying the classification process.

Conclusions:

• The developed automated system effectively classifies normal and abnormal ECG signals using a sophisticated DNN.
• This approach offers a promising, efficient, and accurate method for cardiac condition screening and diagnosis.
• The study highlights the potential of deep learning in revolutionizing ECG interpretation and improving patient care.