High specificity IEGM beat detection by combining morphological and temporal classification for a cardiac neuromodulation system

Elevated heart rate is known to be an independent risk factor for a higher overall mortality, especially for patients suffering from coronary artery disease, e.g. from heart failure. Since pharmacological approaches can not exclusively address heart rate, we investigated a cardiac neuromodulation technique lowering elevated heart rate by means of electrical neurostimulation. The idea is to exclusively modulate the parasympathetic tone in the sinoatrial node area to decrease heart rate. However, electrical stimulation of the heart may pose a specific risk as one temporally misplaced stimulation can cause atrial and especially ventricular fibrillation. Accordingly, we aimed to trigger on the intracardiac electrogram in the upper right atrium and present two algorithms satisfying the requirements of highly specific, secure real-time detection within one heart beat: Decision tree and neural network. Both algorithms were combined with a heart rate prediction estimating upcoming action potentials to maximize beat recognition against artifacts. The combined algorithms were validated on human intracardiac electrograms from electrophysiological examinations with promising results (specificity: 100%, sensitivity_tree: 70.2%, sensitivity_net: 87.3%) for secure neurostimulation.