Performance Challenges in Current Multi-lead QRS Detection Systems

Abstract Modern QRS detectors in patient monitoring systems must provide accurate results even when interfering noise is present. This task is complicated by the heterogeneous and non-uniform nature of the patient population. Certain beat morphologies demand a significant increase in the complexity of beat detection algorithms. At the same time, it is important to take a system-wide approach to the beat detection problem and include artifact rejection and baseline correction challenges. Here we emphasize the importance of such an approach to be able to successfully handle non-trivial input data. We also explain how this approach can improve the QRS detection performance in commercial patient monitoring systems. The performance benchmarks are presented by being tested on proprietary and public ECG databases. Introduction This paper outlines some performance challenges faced by QRS detection systems in commercial patient monitors. The QRS detection system is a vital part of the patient monitoring system. It is tasked with detecting ventricular contractions from surface electrocardiograms (ECG). Such information is used by advanced functions of a monitor, i.e. QRS classification and arrhythmia detection. Thus, together with ECG pre-processing, the QRS detection system confronts strict performance requirements. Failure to correctly detect QRS complexes propagates further into the analysis and undermines performance of the system as a whole. Modern patient monitoring systems must operate under stringent performance requirements and client demands. On one hand patient monitors are regulated by standards established by AAMI [1], these standards outline minimum expected performance from the monitors. On the other hand improving computational capacity of embedded systems raises expectations of users in terms of additional features and capabilities, especially in high-end patient monitors. The patient population is also constantly changing. Introduction of various cardiac assist devices, pacemakers and implantable defibrillators into the population affects the expected surface ECG in patient monitors. For example, ventricular assist devices change both the morphology and the amplitude of QRS complexes as seen on a surface ECG. To ensure reliable performance, manufacturers of patient monitors must be aware of these new clinical methods for treatment of heart related diseases. In this paper, we discuss performance challenges faced by current multi-lead QRS detection systems. We particularly focus on QRS double-counting and false asystoles. Both challenges make significant contribution to overall satisfaction of a customer with a patient monitor. However, the challenges entail solving contradictory problems and demand additional algorithmic complexity in QRS detection systems. This paper addresses these issues in detail. QRS double-counting The authors use the term of QRS double-counting to define a QRS detection scenario when non-QRS elements of the surface ECG get detected as QRS complexes. One example is tall, abnormally strong P or T-waves that are counted as QRS complexes. Established methodology of QRS detection widely adopted by the patient monitoring industry The task of distinguishing tall T-waves from ventricular contractions by differences in their morphology is, however, non-trivial in a real-worl