Projects/06
Automated ARI Extraction
Wavelet-guided activation and recovery detection from intracardiac electrograms
We developed an automated Python pipeline to estimate activation-recovery intervals (ARI) from intracardiac electrograms, using surface ECG timing to guide wavelet-based detection. Evaluated in nine chronically infarcted swine, the method supported comparisons of local repolarization between sinus rhythm and pacing, and between healthy and scar border-zone tissue.
Measuring local repolarization
The activation-recovery interval (ARI) measures the time between local electrical activation and recovery in a unipolar electrogram. It is used as a surrogate for action potential duration. After myocardial infarction, differences in repolarization across healthy tissue and the scar border zone can help characterize the substrate for ventricular arrhythmias.
From electrogram to ARI
- Establish timing references. We filtered the intracardiac electrograms and surface ECG with a 100 Hz low-pass FIR filter. Wavelet-based delineation of surface lead II located R-peaks and P- and T-wave boundaries to guide the searches in each electrogram.
- Detect local activation and recovery. A multilevel Symlet-4 discrete wavelet transform located activation events. Within the resulting search windows, the steepest negative electrogram slope defined activation time, and the steepest positive slope during repolarization defined recovery time, following Wyatt's method. Recovery windows were adapted for sinus rhythm and pacing.
- Compare the intervals. ARI was calculated as recovery time minus activation time and averaged per channel. We compared distributions by rhythm and tissue type using Mann-Whitney U tests. Bipolar amplitudes above 1.5 mV identified healthy tissue; values from 0.5 to 1.5 mV identified border-zone tissue.
Evaluation and findings
We evaluated the pipeline on nine chronically infarcted swine, with approximately 2,900 intracardiac electrograms per animal recorded at 1 kHz using CARTO3 and a Pentaray catheter. The recordings covered sinus rhythm and right ventricular pacing.
ARI distributions differed between rhythms in every case (p < 0.0001). Healthy and border-zone tissue differed significantly in five of the nine cases, while their pooled distributions were statistically similar. These results show why both rhythm and individual variation matter when interpreting repolarization maps.
The pipeline processed thousands of electrograms per case. Formal detection accuracy still requires expert-annotated ground truth. The method also requires a simultaneous surface ECG, and its filtering and wavelet settings were tuned for 1 kHz recordings.