Annals of Noninvasive Electrocardiology (Sep 2022)

A universal, high‐performance ECG signal processing engine to reduce clinical burden

  • Austin Gibbs,
  • Matthew Fitzpatrick,
  • Mark Lilburn,
  • Holly Easlea,
  • Jonathan Francey,
  • Rebecca Funston,
  • Jordan Diven,
  • Stacey Murray,
  • Oliver G. J. Mitchell,
  • Adrian Condon,
  • Andrew R. J. Mitchell,
  • Benjamin Sanchez,
  • David Steinhaus

DOI
https://doi.org/10.1111/anec.12993
Journal volume & issue
Vol. 27, no. 5
pp. n/a – n/a

Abstract

Read online

Abstract Background Electrocardiogram (ECG) signal conditioning is a vital step in the ECG signal processing chain that ensures effective noise removal and accurate feature extraction. Objective This study evaluates the performance of the FDA 510 (k) cleared HeartKey Signal Conditioning and QRS peak detection algorithms on a range of annotated public and proprietary ECG databases (HeartKey is a UK Registered Trademark of B‐Secur Ltd). Methods Seven hundred fifty‐one raw ECG files from a broad range of use cases were individually passed through the HeartKey signal processing engine. The algorithms include several advanced filtering steps to enable significant noise removal and accurate identification of the QRS complex. QRS detection statistics were generated against the annotated ECG files. Results HeartKey displayed robust performance across 14 ECG databases (seven public, seven proprietary), covering a range of healthy and unhealthy patient data, wet and dry electrode types, various lead configurations, hardware sources, and stationary/ambulatory recordings from clinical and non‐clinical settings. Over the NSR, MIT‐BIH, AHA, and MIT‐AF public databases, average QRS Se and PPV values of 98.90% and 99.08% were achieved. Adaptable performance (Se 93.26%, PPV 90.53%) was similarly observed on the challenging NST database. Crucially, HeartKey's performance effectively translated to the dry electrode space, with an average QRS Se of 99.22% and PPV of 99.00% observed over eight dry electrode databases representing various use cases, including two challenging motion‐based collection protocols. Conclusion HeartKey demonstrated robust signal conditioning and QRS detection performance across the broad range of tested ECG signals. It should be emphasized that in no way have the algorithms been altered or trained to optimize performance on a given database, meaning that HeartKey is potentially a universal solution capable of maintaining a high level of performance across a broad range of clinical and everyday use cases.

Keywords