TLDR: This research paper evaluates real-time preprocessing methods for AI-based Electrocardiogram (ECG) signal analysis, focusing on their applicability in edge computing environments. It compares various filtering techniques, QRS detection algorithms, and signal quality assessment methods based on criteria like energy efficiency, processing capability, and real-time performance. The study identifies adaptive filters, an extended Pan-Tompkins method, ICA+LDA, a signal quality-aware technique (SQAw), and an artefact detection layer as particularly promising for developing energy-efficient, low-latency, and secure AI solutions for long-term ECG monitoring at the edge, as part of projects like FACE.
Cardiovascular diseases remain a leading cause of death globally, making continuous heart monitoring crucial for promoting healthy lifestyles. Electrocardiograms (ECGs) are vital tools for assessing heart health, especially for long-term monitoring. However, manually analyzing the vast amount of data from long-term ECGs is incredibly labor-intensive and prone to human error, with even experienced cardiologists achieving only up to 75% accuracy in some interpretations.
To address these challenges, data-driven Artificial Intelligence (AI) models have emerged, capable of matching or even surpassing human cardiologists in classification performance. Traditionally, these AI models are trained and tested on powerful cloud servers. While effective, this approach presents issues related to data privacy, response times, and processing costs.
The Rise of Edge Computing in ECG Analysis
This is where edge computing offers a promising solution. By processing data closer to its source – on devices at the ‘edge’ of the network – it can significantly reduce latency, enhance data security, and lower power consumption. The FACE project, for instance, aims to develop an innovative machine learning solution for long-term ECG analysis that combines the strengths of both edge and cloud computing. A key aspect of this project, as detailed in the research paper “Evaluation of Real-Time Preprocessing Methods in AI-Based ECG Signal Analysis”, is the selection and evaluation of suitable real-time preprocessing methods for ECG signals in an edge environment, focusing on energy efficiency, processing capability, and real-time performance.
Key Preprocessing Steps Explored
The paper systematically reviews various state-of-the-art preprocessing techniques, categorizing them into filtering, QRS detection, and signal quality assessment.
Filtering
ECG signals are often noisy, requiring filtration to remove artifacts. Standard bandpass filters, like the Butterworth filter, are energy-efficient and fast, making them suitable for real-time applications. However, their fixed frequency limits can be problematic for the naturally time-varying nature of ECG signals. Adaptive filters, such as the Adaptive Notch Filter (ANF), offer a superior solution by effectively tracking frequency variations and eliminating power line interference without requiring high filter orders. Another method, the MaMeMi filter, is effective at eliminating baseline wander with minimal resources, though it can be slower compared to other techniques.
QRS Detection
Detecting the QRS complex, which represents ventricular depolarization, is fundamental for ECG analysis. The Pan-Tompkins algorithm is widely used for this. An optimized version, which replaces the multiplier in the squaring step with a more power-efficient squarer, significantly reduces energy consumption while maintaining high processing speed. While the original Pan-Tompkins is fast, its accuracy can drop significantly with low-quality ECG signals. The Hamilton-Tompkins algorithm, an enhancement, uses patient-specific thresholds but can be slower. The paper also highlights the Independent Component Analysis (ICA) combined with Linear Discriminant Analysis (LDA) as a method with low computational complexity and reduced information loss, though it can be vulnerable to motion artifacts and Gaussian amplitude distributions.
Signal Quality Assessment (SQA)
ECG signals are susceptible to various interferences like baseline wander, motion artifacts, and power line interference, which can obscure diagnostic information. The paper examines several SQA techniques. Fourier and Wavelet Transformations, while offering different signal views or aiding compression, are often computationally intensive and less real-time capable. A Signal Quality-Aware (SQAw) technique, based on Discrete Fourier Transform, offers a rapid solution for quality analysis, though it can have high battery consumption. An artefact detection layer, utilizing a rule-based algorithm, proves to be power-efficient and effective in filtering out normal and noisy heartbeats before transmission, thereby saving energy and time. Signal Quality Indices (SQIs), particularly frequency-based and statistical-based SQIs, are noted for their speed and simplicity, with QRS detection-based SQIs like tSQI, corSQI, and eSQI also performing well and being easy to implement.
Also Read:
- Advancing Cardiac Diagnosis: 12-Lead ECG Reconstruction from Three Leads
- Advanced AI Models for Anemia Detection: A Review
Conclusion and Future Directions
The research concludes that adaptive filters, such as the ANF, are highly effective for real-time tasks, especially when dealing with variable ECG signals. The extended Pan-Tompkins method and the ICA+LDA approach are identified as promising, real-time-capable solutions for QRS complex extraction with low power consumption. For signal quality assessment, SQAw and the artefact detection layer are highlighted as the most promising. While SQAw is faster than QRS detection-based SQIs, the artefact detection layer offers superior power efficiency and overall performance, despite being slightly slower than some SQIs and SQAw.
Future work within the FACE project will involve evaluating these promising techniques using a common reference dataset and standardized hardware, such as the NVIDIA Jetson AGX Orin, to enable a comprehensive and objective comparison. This ongoing research is funded by the German Federal Ministry of Economic Affairs and Climate Action (BMWK) and the European Union “NextGenerationEU”, in collaboration with several medical and academic partners.
