[Paper Review] Deep Recurrent Neural Networks for ECG Signal Denoising

Karol Antczak|arXiv (Cornell University)|Jul 30, 2018

ECG Monitoring and Analysis31 references81 citations

TL;DR

The paper presents a deep recurrent denoising neural network for ECG signals that uses synthetic data for pretraining and fine-tunes on real data to outperform reference methods on heavily-noised signals.

ABSTRACT

Electrocardiographic signal is a subject to multiple noises, caused by various factors. It is therefore a standard practice to denoise such signal before further analysis. With advances of new branch of machine learning, called deep learning, new methods are available that promises state-of-the-art performance for this task. We present a novel approach to denoise electrocardiographic signals with deep recurrent denoising neural networks. We utilize a transfer learning technique by pretraining the network using synthetic data, generated by a dynamic ECG model, and fine-tuning it with a real data. We also investigate the impact of the synthetic training data on the network performance on real signals. The proposed method was tested on a real dataset with varying amount of noise. The results indicate that four-layer deep recurrent neural network can outperform reference methods for heavily noised signal. Moreover, networks pretrained with synthetic data seem to have better results than network trained with real data only. We show that it is possible to create state-of-the art denoising neural network that, pretrained on artificial data, can perform exceptionally well on real ECG signals after proper fine-tuning.

Motivation & Objective

• Motivate the need for effective denoising of ECG signals degraded by various noises.
• Propose a four-layer deep recurrent denoising neural network architecture.
• Investigate transfer learning by pretraining on synthetic ECG data and fine-tuning on real data.
• Evaluate the approach on real noisy ECG data and compare with reference methods.

Proposed method

• Utilize a four-layer deep recurrent neural network for denoising ECG signals.
• Pretrain the network on synthetic data generated by a dynamic ECG model.
• Fine-tune the pretrained network on real ECG data.
• Assess the impact of synthetic pretraining on performance for real signals.
• Compare denoising performance against reference methods, especially for heavily noised signals.

Experimental results

Research questions

• RQ1 Can a deep recurrent denoising network outperform reference methods on heavily-noised ECG signals?
• RQ2 Does pretraining on synthetic ECG data improve denoising performance on real ECG signals after fine-tuning?
• RQ3 What is the effect of synthetic data quality and transfer learning on real-world ECG denoising outcomes?

Key findings

• A four-layer deep recurrent network can outperform reference methods for heavily noised ECG signals.
• Pretraining with synthetic data tends to yield better results than training on real data alone.
• The approach demonstrates it is possible to create a state-of-the-art denoising network pretrained on artificial data and fine-tuned for real signals.