[Paper Review] A Wearable ECG Monitor for Deep Learning Based Real-Time Cardiovascular Disease Detection

Peng Wang, Zihuai Lin+5 more|arXiv (Cornell University)|Jan 25, 2022

ECG Monitoring and Analysis26 citations

TL;DR

The paper presents a wearable single-lead ECG patch (IREALCARE) and a semi-supervised Confident-ResNet model to achieve 90.2% average accuracy in real-time ECG classification on noisier IoT-obtained data.

ABSTRACT

Cardiovascular disease has become one of the most significant threats endangering human life and health. Recently, Electrocardiogram (ECG) monitoring has been transformed into remote cardiac monitoring by Holter surveillance. However, the widely used Holter can bring a great deal of discomfort and inconvenience to the individuals who carry them. We developed a new wireless ECG patch in this work and applied a deep learning framework based on the Convolutional Neural Network (CNN) and Long Short-term Memory (LSTM) models. However, we find that the models using the existing techniques are not able to differentiate two main heartbeat types (Supraventricular premature beat and Atrial fibrillation) in our newly obtained dataset, resulting in low accuracy of 58.0 %. We proposed a semi-supervised method to process the badly labelled data samples with using the confidence-level-based training. The experiment results conclude that the proposed method can approach an average accuracy of 90.2 %, i.e., 5.4 % higher than the accuracy of conventional ECG classification methods.

Motivation & Objective

• Develop a portable wireless real-time single-lead ECG patch (IREALCARE) for long-term monitoring.
• Address noisy, imperfectly labeled IoT ECG data with a semi-supervised, confidence-level training scheme.
• Improve ECG rhythm classification accuracy over conventional CNN/LSTM approaches using a deep residual network.
• Enable cloud-based analysis and clinician access for remote cardiovascular disease detection.

Proposed method

• Design of a small, 13 g single-lead ECG patch with 250 Hz sampling and 24-bit ADC.
• Data preprocessing including data normalization, sliding-window data augmentation, and wavelet-based denoising (DWT).
• Adoption of a ResNet-based backbone with six residual blocks for ECG classification.
• Introduction of a confidence-level based training to select clean data from imperfect labels.
• Comparison against standard CNN/LSTM/AlexNet/VGG16/WBCNN classifiers to demonstrate performance gains.
• Training uses 5 rhythm classes mapped to AAMI EC57 with 80/20 train/test split and 227,680 training beats.

Experimental results

Research questions

• RQ1 Can a wearable single-lead ECG patch reliably collect long-term ECG data suitable for deep learning classification in real-world conditions?
• RQ2 Does a confidence-level based semi-supervised training improve rhythm classification accuracy on noisy, imperfectly labeled IoT ECG data?
• RQ3 How does a ResNet-based model with confidence-based training compare to existing deep learning models for ECG rhythm classification?
• RQ4 What is the achievable real-time classification accuracy and robustness of the proposed system in an IoT/cloud ecosystem?

Key findings

• The proposed Confident-ResNet achieved 90.2% average accuracy on the test set, outperforming CNN, LSTM, AlexNet, VGG16, and other baselines by 3–10 percentage points.
• Confidence-level based training with an 80% threshold yielded the best validation accuracy (89.9%), demonstrating effective handling of imperfect labels.
• The system used 62 patients, 300 Hz sampling, and 1-second segments mapped to 5 rhythm classes (N, V, S, A, Q) per AAMI EC57 with a large training set of 227,680 beats.
• Denoising via discrete wavelet transform and data augmentation via sliding windows helped address noisy IoT patch data.
• Compared to conventional ECG classification methods, the Confident-ResNet method showed improved accuracy on the implemented dataset.