Matched Filtering for Heart Rate Estimation on Compressive Sensing ECG Measurements

Objective: Compressive sensing (CS) has recently been applied as a low-complexity compression framework for long-term monitoring of electrocardiogram (ECG) signals using wireless body sensor networks. Longterm recording of ECG signals can be useful for diagnostic purposes and to monitor the evolution of several widespread diseases. In particular, beat-to-beat intervals provide important clinical information, and these can be derived from the ECG signal by computing the distance between QRS complexes (R-peaks). Numerous methods for R-peak detection are available for uncompressed ECG. However, in the case of compressed sensed data, signal reconstruction can be performed with relatively complex optimization algorithms, which may require significant energy consumption. This paper addresses the problem of heart rate estimation from CS ECG recordings, avoiding the reconstruction of the entire signal. Methods: We consider a framework, where the ECG signals are represented under the form of CS linear measurements. The QRS locations are estimated in the compressed domain by computing the correlation of the compressed ECG and a known QRS template. Results: Experiments on actual ECG signals show that our novel solution is competitive with methods applied to the reconstructed signals. Conclusion: Avoiding the reconstruction procedure, the proposed method proves to be very convenient for real-time low-power applications.