Lossless electrocardiogram signal compression: A review of existing methods
Introduction
Electrocardiography (ECG) is a non-invasive measure of heart activity, structure and function. For example, the atrial and ventricular depolarization and repolarization of the heart are represented on the ECG as a series of waves, namely the P wave, followed by the QRS complex, and, lastly, the T wave [1], as depicted by Fig. 1 . The time intervals between these various peaks and differences in amplitudes are known to contain important clinical information [2]. As such, long-term ECG monitoring remains the go-to method of detecting, diagnosing, and monitoring various heart diseases [3] [4] [5].
Since its inception by Holter [7] in the early sixties, long-term ECG monitoring has been used to monitor patients in critical conditions in order to gather information about pathological events that may occur only sporadically over long periods of time. With hardware advances and lowered storage costs, so-called Holter monitors have shrunk in size and can now weigh just a fraction of a pound. Typical monitors record data from up to three ECG leads during 24-48 hours on a single battery charge. The data is stored on the device and has to be downloaded by the physician at the hospital for analysis and offline diagnosis.
Recently, however, there has been a push to use ambulatory 12-lead Holter monitors, particularly to monitor patients after myocardial infarction, as it has shown to provide significantly higher detection rates of e.g., ischaemia [8]. Moreover, longer recordings, in the order of a week or more, have been suggested for more accurate detection of e.g., atrial fibrillation and flutter [9]. A four-fold increase in the number of ECG leads and a 2-to-4-fold increase in recording duration, however, poses a serious threat to the portability of existing devices, as storage and battery life become an issue. Typically, lower resolution ECG is recorded, thus resulting in reduced performance relative to a conventional 12-lead digital ECG monitor [10].
Moreover, the last few years has seen an emergence of portable, unobtrusive ECG monitoring devices capable of streaming live data to the physician [11], thus bypassing the need to wait several days for clinical analysis of the recorded patient data. Smart garments, smart watches, and chest straps have been developed to stream real-time ECG data. Such transmission capabilities impose additional constraints on battery life and the need for reliable data compression algorithms has now become a reality [12].
Data compression can significantly reduce storage and transmission requirements, thus positively impacting battery life, device size and portability. Compression algorithms reduce the data size by removing redundancies present in the data and have been widely used in audio, video, and speech applications. Compression algorithms can be classified as lossy or lossless. As the name suggests, lossy techniques are capable of achieving high compression ratios as a true replica of the original signal is not needed after decompression. This has been very popular in audio-video applications, as perfect reconstruction is not needed for human perception. Lossless compression, in turn, results in lower compression ratios, but allows for perfect signal reconstruction. While consumer applications can rely on lossy ECG compression, clinical applications require lossless compression for several reasons, including:
- 1.
Medical regulatory boards worldwide advocate for lossless compression [13], [14], [15].
- 2.
With advances in the medical fields, the definition of diagnostically relevant information is ever expanding. Any signal patterns which might have been discarded previously as random heart activity might eventually be shown to contain medically relevant information [16].
- 3.
Lossy algorithms may remove segments of the data that may not be relevant to a trained clinician's eyes. With the emergence of new big data and machine learning tools, however, algorithms may find diagnostically relevant information in these removed segments.
- 4.
Reconstruction error may be confused as a diagnostically relevant feature, such as QRS complex amplitudes, ST segment slopes, T and P wave amplitudes, amongst others. While existing methods have explored a combination of lossy and lossless compression, the latter being used on relevant regions of the signal [17], as well as more relevant distortion measures [18], the abovementioned issues may still hold.
For ECGs, compression algorithms typically fall under three categories, namely: (1) direct compression methods, which operate on the time domain signals, (2) transform based methods, which operated in some transformed domain, such as frequency or time-frequency, and (3) parameter extraction methods, which directly extract relevant information from the ECG signal, such as heart rate, heart rate variability and RR intervals, and then discards the ECG [19], [20]. Lossless compression algorithms typically fall within the first two categories. In this paper, we review the literature to characterize existing ECG lossless compression methods and gauge their effectiveness for clinically relevant applications.
In the past, two important reviews on ECG compression have been published. In [19], a systematic review of both lossy and lossless ECG compression methods was presented. Since then, however, several new and more efficient compression methods have emerged, along with new publicly-available databases. More recently, a review paper was published on wavelet-based ECG compression methods [20]. As most wavelet based methods are lossy in nature, the review bypassed lossless methods. Given the importance of lossless compression methods in clinically-relevant applications and the burgeoning of monitoring devices, an up-to-date review of existing lossless ECG compression methods is lacking. The present review aims to fill this gap.
The remainder of this paper is organized as follows: Section 2 first introduces the typical ECG compression pipeline. Section 3 then describes the methodology used to perform the literature review. Section 4 presents and discusses our findings and proposes recommendations for future lossless ECG compression studies. Finally, conclusions are drawn in Section 5.
Section snippets
ECG compression pipeline
When analyzing ECG compression papers, four main components are typically discussed, as depicted by Fig. 2: the databases being used, type of pre-processing performed, the compression algorithm per se, and the performance evaluation figures of merit. In the sections to follow, these components are described in more detail.
Methodology used for the literature review
English peer-reviewed journal articles published between 1990 and 2017 were chosen as the target of this review. Three major science and engineering bibliographic databases were queried: Web of Science (seven databases), Engineering Village (Compendex and Inspec), and SCOPUS. In addition, reference sections from the selected articles were scanned to find similar journal articles that might have not been found in the database queries. The last search was run on 9 September 2017. Some conference
Results and discussion
The database queries provided 58 hits that matched the search results. The reference section for these articles helped identify five additional articles. Thirteen articles were excluded based on their title and abstract, and nineteen were removed after a full text assessment for not meeting the inclusion criteria. Therefore, for the final analysis, 31 articles were chosen. The results have been provided for the six different data items to allow a direct comparison between the articles.
Conclusions
In this review, 31 articles were surveyed in the field of ECG lossless compression. Important aspects, such as databases used, pre-processing algorithms, compression domains and encoding paradigms, online versus offline implementations, performance evaluation metrics, and prior art comparisons were explored. To the best of the authors’ knowledge, this is the first review to focus solely on lossless ECG compression methods.
It was found that most studies (22 out of 31) made use of the MIT-BIH
Conflict of interest
None to declare.
Acknowledgements and declarations
The authors wish to acknowledge funding from the Natural Sciences and Engineering Research Council of Canada and to discussions with Hexoskin (Montreal, Canada) on lossless ECG compression for wearable applications.
References (67)
- et al.
Using contexts and r-r interval estimation in lossless ecg compression
Comput. Methods Programs Biomed.
(2002) - et al.
Ecg signal compression using ascii character encoding and transmission via sms
Biomed. Signal Process. Control
(2013) - et al.
Wavelet-based electrocardiogram signal compression methods and their performances: A prospective review
Biomed. Signal Process. Control
(2014) Lossless ecg encoding
Comput. Methods Programs Biomed.
(1997)- et al.
Compact digital storage of ecgs
Comput. Progr. Biomed.
(1979) - et al.
A lossless ecg data compression technique using ascii character encoding
Comput. Electr. Eng.
(2011) - et al.
Making sense of the ECG: a hands-on guide
(2014) - et al.
Ectopic beats in approximate entropy and sample entropy-based HRV assessment
Int. J. Syst. Sci.
(2012) - et al.
Recommendations for standardization of leads and of specifications for instruments in electrocardiography and vectorcardiography
Circulation
(1967) - et al.
Research in electrocardiography and magnetocardiography
Proc. IEEE
(1972 March)