MS

thesisAccurate QRS detection is essential in online computerized rhythm monitoring systems. A major cause of error in QRS detection schemes arises from artifacts superimposed on the input signal. To a lesser extent identification of P or T waves as QRS complexes can represent another source of error. In an effort to reduce the incidence of false and missed alarms generated by the rhythm monitoring system currently used in the LDS Hospital Coronary Care Unit, a project was undertaken to improve the accuracy and reliability of the QRS detection algorithm, specifically in contaminated single lead electrocardiographic data. The algorithm uses a dual scan of the sample data combined with a peak detection scheme to locate a reference point on a QRS candidate. The candidate is then checked for evidence of baseline shift or an excessively low signal-to-noise ratio. If neither of these criteria is met, the candidate is assumed to be QRS and a fiducial point is located on the complex. To assess the sensitivity and specificity of the QRS detection algorithm, an off-line evaluation was performed on forty-one patient records collected in the Coronary Care Unit. Arrhythmias included in the evaluation were fast ventricular and atrial rhythms and heart block. Over 90 percent of the data base was contaminated with excessive muscle artifacts. Of a total of 7,205 beats used in the evaluation, and positive predictive accuracy were .9641 and .9573, respectively. Of the error, 92.16 percent of the false positives and 84.17 percent of the false negatives were due to excessive noise spike superimposition on the data. None of the false positive error (.0071) was due to P or T wave misidentification of a QRS complex. These results indicate that a signal-in-noise approach to automated QRS detection is effective in identifying QRS complexes in the contaminated single lead electrocardiogram with minimal error