Premature ventricular contraction detection and estimation of daily burden by an insertable cardiac monitor

Figure 1. A: A 2-stage PVC algorithm efficiently uses ICM processor resources via a PVC candidate trigger stage followed by a morphology-based classifier on the reduced set of beats. Different beat classifier configurations are used depending on PVC sequence type. B: Example of ICM-stored ECG signal segment overlayed with PVC algorithm results. V denotes ICM algorithm PVC detections, arrows are interim algorithm-triggered PVC candidates. ECG = electrocardiogram; ICM = insertable cardiac monitor; PVC = premature ventricular contraction.

Figure 2. PVC burden performance of ICM episodes played back through LUX-Dx II+ ICM device versus adjudicated (true) PVC burden. Both linear fit and identity reference lines (dashed) are shown. ICM = insertable cardiac monitor; PVC = premature ventricular contraction.

Figure 3. A: Performance of ICM PVC algorithm (via validated model) on the Holter 24-hour Holter dataset. The unity line (dashed) is also shown for comparison. The arrow identifies an example of a patient with underestimation of PVC burden caused by PVCs with narrow QRS projections that are mistaken for PACs (see Figure 4 for additional analysis). B: Bland-Altman plot comparison of ICM and Holter PVC burden estimates. ICM = insertable cardiac monitor; PVC = premature ventricular contraction.

Figure 4. Example of underestimation of PVC burden (for the patient denoted by the arrow in Figure 3) caused by narrow projection of PVCs onto ICM vector. The blue arrows above indicate true PVCs clearly apparent in Lead I from Holter monitoring. The bottom trace shows the ICM vector simulated by V2-V3 and filtered by the ICM device for PVC detection. The ICM detected the 2nd beat of a couplet and the 4th bigeminal PVC (V symbols), but others were missed because of high similarity to NSR beats. ICM = insertable cardiac monitor; NSR = normal sinus rhythm; PVC = premature ventricular contraction.