Heart Arrhythmia Classification

Personalized health monitoring is becoming ubiquitous with the development of AI models, spanning clinical-grade remote patient monitoring to commercial-grade health and fitness applications. Most leading consumer products offer similar electrocardiograms (ECG) for common types of heart arrhythmia. Ambiq’s HeartKit is a reference AI model that demonstrates analyzing 1-lead ECG data to enable a variety of heart applications, such as detecting heart arrhythmias and capturing heart rate variability metrics. Furthermore, by analyzing individual beats, the model can identify irregular beats, such as premature and ectopic beats originating in the atrium or ventricles.

By leveraging a modern multi-head network architecture coupled with Ambiq's low-power SoC, the model is designed to be efficient, explainable, and extensible. The current architecture consists of an ECG segmentation model followed by three upstream heads: HRV head, arrhythmia head, and beat head. The ECG segmentation model serves as the backbone and is used to annotate every sample as either P-wave, QRS, T-wave, or none. The arrhythmia head is used to detect the presence of Atrial Fibrillation (AFIB) or Atrial Flutter (AFL). The HRV head is used to calculate heart rate, rhythm (e.g., bradycardia), and heart rate variability from the R peaks. Lastly, the beat head is used to classify individual beats as either normal, premature/ectopic atrial contraction (PAC), or premature/ectopic ventricular contraction (PVC).

While the pre-trained model is ready to use on Ambiq platforms, it also includes software to train, convert, and deploy customized models where needed. The HeartKit has been released under the permissive BSD-3 license for ease of deployment and development. Available now as a Technical Preview at [https://github.com/AmbiqAI/heartkit] to download and start developing AI today. GitHub Link: https://github.com/AmbiqAI/heartkit NeuralSPOT Link: https://github.com/AmbiqAI/neuralSPOT