Abstract
Background: An increasing number of wearable patch based electrocardiographic devices are being developed. These devices record non-standardized short-distance bipolar (SDB) ECG-leads. The SDB ECG-leads that are recorded by these devices can produce ECG morphologies that are different when compared to standard ECG-leads. A comprehensive performance assessment of these SDB devices requires a sufficiently large sample of device-specific SDB ECG-leads. Gathering such a large sample of device-specific data is cumbersome. Linear ECG-lead transformations, that estimate SDB ECG-leads from the standard 12-lead ECG, were recently proposed as an alternative to the collection of large device-specific SDB ECG databases. More precisely, it has been proposed to utilize estimated SDB ECG-leads for the performance assessment of SDB devices. SDB devices are typically developed for rhythm monitoring applications that utilize upon QRS or P-QRS data. It is therefore desirable that the used linear ECG-lead transformations accurately estimate these ECG intervals. Whether the utilization of different parts of the P-QRS-T complex, for the development of the linear ECG-lead transformations, has an influence on the estimation performance of these transformations has not yet been reported. We have investigated whether a specific part of the P-QRS-T complex should be used when developing linear ECG transformations for the performance assessment of SDB devices.
Method: We extracted standard 12-lead ECGs and two SDB ECG-leads from body surface potential maps (BSPM) of n=726 subjects (left ventricular hypertrophy, n=232; old myocardial infarction, n=265; normal subjects, n=229). Our ECG dataset was randomly divided into one training (DTrain, n=545) and one testing dataset (DTest, n=181). DTrain was used to generate four linear ECG-lead transformation matrices for each of the two SDB ECG-leads. Transformation matrices were developed using P-wave, QRS-complex P-QRS and QRS-T data.
Results: Table 1 details the RMSE differences between recorded and derived SDB ECG-leads.
Conclusions: Our findings suggest that it is possible to optimize linear ECG-lead transformations for the P-wave. However, this comes at the cost of a substantially reduced QRS estimation performance. We therefore recommend developing the linear ECG transformations for the performance assessment of SDB devices using QRS, P-QRS or QRS-T data.
Method: We extracted standard 12-lead ECGs and two SDB ECG-leads from body surface potential maps (BSPM) of n=726 subjects (left ventricular hypertrophy, n=232; old myocardial infarction, n=265; normal subjects, n=229). Our ECG dataset was randomly divided into one training (DTrain, n=545) and one testing dataset (DTest, n=181). DTrain was used to generate four linear ECG-lead transformation matrices for each of the two SDB ECG-leads. Transformation matrices were developed using P-wave, QRS-complex P-QRS and QRS-T data.
Results: Table 1 details the RMSE differences between recorded and derived SDB ECG-leads.
Conclusions: Our findings suggest that it is possible to optimize linear ECG-lead transformations for the P-wave. However, this comes at the cost of a substantially reduced QRS estimation performance. We therefore recommend developing the linear ECG transformations for the performance assessment of SDB devices using QRS, P-QRS or QRS-T data.
Original language | English |
---|---|
Pages (from-to) | S101 |
Journal | Journal of Electrocardiology |
Volume | 57 |
DOIs | |
Publication status | Published (in print/issue) - 6 Dec 2019 |
Event | International Society for Computerised Electrocardiology - Atlantic Beach, Jacksonville, United States Duration: 10 Apr 2019 → 14 Dec 2019 |
Keywords
- ECG
- Data analytics
- Signal processing
- lead transformations