Impact of Motion Artifact on Detection of Atrial Fibrillation in Compressively Sensed ECG using a Deterministic Matrix

Abstract

Early detection of Atrial Fibrillation (AFib) is warranted to reduce the chances of patients developing complications. Compressive sensing (CS) of electrocardiogram (ECG) will facilitate long term monitoring with detection of AFib in the compressed domain eliminating the need for the expensive operation of reconstructing the ECG. This paper presented an AFib detector in the compressed domain and studied the effect of noise on it. ECG records from the Long-Term Atrial Fibrillation Database were contaminated with motion artifact from the MIT-BIH Noise Stress Database and compressed to 50%, 75%, and 95% levels. A 100 tree random forest was used to detect AFib in the uncompressed and compressed ECG at different noise levels. The random forest was evaluated using 5-fold cross validation and patient hold-out method. The random forest achieved a maximum of 81.87% F1 score at the 3 dB Signal to Noise Ratio (SNR) and 75% compression level in cross validation. Changing the SNR to -10 dB reduced the F1 score by 3.25%. The random forest achieved a maximum of 61.03% at 3 dB SNR and on uncompressed ECG in the hold-out test. Changing the SNR to -10 dB reduced the F1 score by 6.55%. The results show that it is possible to detect AFib in the compressed domain with noise impacting the performance.