On-chip, non-invasive detection of all atrial arrhythmias for immediate and accurate characterization in adult rats

Abstract

Abstract Background To monitor and record the occurrence and duration of cardiac arrhythmias in freely moving animals, ECG telemetry devices are often used. However, traditional telemetry devices require complex equipment unsuitable for on-site application in most animal housing facilities. Additionally, current atrial arrhythmias detection algorithms are mainly based on ECG R-R interval variability for atrial fibrillation (AF) detection. Such algorithms do not allow the detection of stable R-R interval atrial arrhythmias such as flutter. Thus there is currently a lack of non-invasive methods allowing automated detection of all atrial arrhythmias subtypes, thereby precluding in-depth research into general and subtype-specific mechanisms and treatment. Purpose Here we aim to create a non-invasive system allowing fast and accurate on-site detection and characterization of all atrial arrhythmias in rats. Methods Our atrial arrhythmia detector system was made of an on-skin ECG sensor, a low-power microprocessor unit, a large data storage unit and a battery. All components were assembled within a lightweight rat jacket. The filtered ECG signal was sent to the microprocessor allowing algorithm-based, real-time detection of atrial arrhythmias. The device allowed storage of ECG traces as well as arrhythmia characteristics (including type of arrhythmia and its duration). Atrial arrhythmias were induced by transesophageal atrial burst pacing in Wistar rats. These arrhythmia recordings were used to develop an AI detection method. Next, real-time detection algorithms based on R-R variability and AI were both integrated on-chip and compared using the same ECG data. Results ECG traces measured with our new system were similar to classical ECG measurements. Stored ECG traces provided the raw data for the development of an AI-based detection algorithm. The AI training was done using ECG data recorded from five different rats and reached 99.5% self-testing accuracy to detect all sub-types of atrial arrhythmias based on pre-categorized data. Using uncategorized data (n=25 from 5 rats) different from the training dataset, the AI algorithm could detect >80% of all types of atrial arrhythmias within 3s after initiation, while the rest could be detected within 10s. Though R-R interval based algorithm showed a high detection rate of 88% of AF within 10s, this method did not allow the detection of flutter or AF presenting regular R-R intervals within 1 minute after initiation. Conclusion Our atrial arrhythmia detection and analysis system provides a novel method for non-invasive investigation into atrial arrhythmias in rats and could easily be scaled for other species. It allows on-site fast and accurate detection and characterization of atrial arrhythmia subtypes without external equipment, thereby opening new possibilities for mechanistic studies and therapeutic testing, including closed-loop applications aiming for arrhythmia termination.