Wavelet-Based ECG Delineation on a Wearable Embedded Sensor Platform

Abstract

Abstract-The analysis of the electrocardiogram (ECG) is widely used for diagnosing many cardiac diseases. Since most of the clinically useful information in the ECG is found in characteristic wave peaks and boundaries, a significant amount of research effort has been devoted to the development of accurate and robust algorithms for automatic detection of the major ECG characteristic waves (i.e., the QRS complex, P and T waves), socalled ECG wave delineation.One of the most salient ECG wave delineation algorithms is based on the wavelet transform (WT). This work is dedicated to the sensible optimization and porting of this WT-based ECG wave delineator to an actual wearable embedded sensor platform with limited processing and storage resources. The porting was successful and the implementation was extensively validated using a standard manually annotated database. Interestingly, our results show that, despite the limitations of the embedded sensor platform, careful optimization allows to achieve comparable or even better delineation results than the original offline algorithm.

Development and Evaluation of Multilead Wavelet-Based ECG Delineation Algorithms for Embedded Wireless Sensor Nodes

“…More specifically, this work revisits the originally offline WTbased ECG delineation algorithm, first introduced in [5] and further developed and extensively evaluated in [7], and which we previously nonstraightforwardly optimized and ported for real-time single-lead ECG delineation [16] on the Shimmer embedded sensor node. It then extends our online WT-based single-lead delineation algorithm [16] for the multilead scenario, including the prerequisite baseline removal filtering and the sensible algorithmic modifications and parameter optimization to achieve real-time execution on the Shimmer notwithstanding its limited processing and storage resources.…”

“…It then extends our online WT-based single-lead delineation algorithm [16] for the multilead scenario, including the prerequisite baseline removal filtering and the sensible algorithmic modifications and parameter optimization to achieve real-time execution on the Shimmer notwithstanding its limited processing and storage resources. As a matter of fact, an embedded wireless sensor node typically has only a few kilobytes of memory [12], [14], [15] (e.g., 10 kB for Shimmer), is equipped with an ultra-low-power microcontroller running at a maximum clock speed between 8 and 16 MHz, and does not include hardware support for division and floating-point operation.…”

Ultralow power processor employing block instruction for ECG applications

“…We also rewrote the memory structure. The open source time domain ECG analysis software designed by PhysioNet [3] is used as typical time-domain algorithms and DWT software proposed by Nicolas Boichat [4] is used as typical frequency-domain algorithms to evaluate the effectiveness of the proposed design The block scheduler assembled the traditional atomic instruction into block instruction in four passes. After rough blocking pass, the average size of raw block is 10.94.…”

“…It can be generally classified into two categories: the time-domain manipulation and the frequency-domain analysis [2]. To analyze the program characteristics and power consumption of QRS detection algorithms, a time-domain manipulation [3] and a frequency-domain analysis [4] were tested as two typical algorithms on a low power RISC processor CK802 [5]. More detailed information regarding the experimental method employed can be found in Section 5.…”

Inexact‐aware architecture design for ultra‐low power bio‐signal analysis

“…Traditionally, such an estimation is performed by applying fast-Fourier transform (FFT). Instead, and similarly to [24], in this paper we use a wavelet-based FFT (WFFT), which reduces substantially (up to 28% w.r.t the state-of-the-art) the complexity of the fast-Lomb method [25], while also introducing sparsity in the data being processed [26].…”