Exploring Best Practices for ECG Signal Processing in Machine Learning

Amir Salimi,Sunil Vasu Kalmady,Abram Hindle,Osmar Zaiane,Padma Kaul

2023-11-03

Abstract:In this work we search for best practices in pre-processing of Electrocardiogram (ECG) signals in order to train better classifiers for the diagnosis of heart conditions. State of the art machine learning algorithms have achieved remarkable results in classification of some heart conditions using ECG data, yet there appears to be no consensus on pre-processing best practices. Is this lack of consensus due to different conditions and architectures requiring different processing steps for optimal performance? Is it possible that state of the art deep-learning models have rendered pre-processing unnecessary? In this work we apply down-sampling, normalization, and filtering functions to 3 different multi-label ECG datasets and measure their effects on 3 different high-performing time-series classifiers. We find that sampling rates as low as 50Hz can yield comparable results to the commonly used 500Hz. This is significant as smaller sampling rates will result in smaller datasets and models, which require less time and resources to train. Additionally, despite their common usage, we found min-max normalization to be slightly detrimental overall, and band-passing to make no measurable difference. We found the blind approach to pre-processing of ECGs for multi-label classification to be ineffective, with the exception of sample rate reduction which reliably reduces computational resources, but does not increase accuracy.

Signal Processing,Machine Learning

What problem does this paper attempt to address?

### The Problem Addressed by the Paper

This paper aims to explore the optimal preprocessing methods for electrocardiogram (ECG) signals in machine learning to improve the performance of classifiers used for diagnosing heart diseases. Although existing deep learning algorithms have achieved significant results in certain heart disease classifications, there is a lack of consensus on preprocessing methods. Specifically, the paper attempts to answer the following questions:

1. **Do different conditions and architectures require different preprocessing steps to achieve optimal performance?**
2. **Have state-of-the-art deep learning models made preprocessing unnecessary?**

To answer these questions, the authors applied preprocessing methods such as downsampling, normalization, and filtering to three different multi-label ECG datasets and measured the impact of these methods on three high-performance time series classifiers.

### Main Findings

1. **Downsampling**:
- A sampling rate as low as 50Hz can produce results comparable to or even better than the commonly used 500Hz sampling rate. This has significant implications for reducing training time and hardware requirements.

2. **Normalization**:
- Although normalization is widely used in other fields, this paper found that min-max normalization is slightly detrimental overall.

3. **Band-pass Filtering**:
- Band-pass filtering has no significant impact on model performance.

### Conclusion

The study emphasizes that the optimal method for ECG preprocessing is not universal but depends on specific labels and model architectures. Therefore, researchers should consider the specific task requirements when choosing preprocessing methods. Additionally, lower sampling rates can significantly reduce training costs without sacrificing performance.