DeepRhythmAI

DeepRhythmAI is cloud-based software that uses artificial intelligence algorithms to analyze cardiac arrhythmias from single- or two-lead ECG data in adult patients. It is designed for integration by healthcare solution providers into web and mobile applications, allowing qualified healthcare professionals to review and confirm automated analysis results from Holter monitors, event recorders, and other ECG devices. The results are advisory only and not intended as the sole means of diagnosis.

The primary technological change is modification of the arrhythmia detection algorithm to process both two-lead and single-lead ECG recordings, specifically including single-lead patch recorders placed on the upper mid/left chest. The device maintains the same core architecture: a web API interface and automated proprietary deep-learning algorithm (CNN and transformer models) with no primary graphical user interface, requiring integration with external visualization software.

Performance testing was conducted according to ANSI/AAMI EC57:2012 and ANSI/AAMI/IEC 60601-2-47:2012/(R)2016. Software development and validation followed ANSI/AAMI/IEC 62304 and the FDA's January 2002 guidance on General Principles of Software Validation. Testing included unit, integration, and system-level verification, cybersecurity assessment, and validation against a proprietary database (MDG validation db) containing recordings from the intended patient population.

The subject device is substantially equivalent because it maintains identical intended use, patient population (adults), classification, and core functional capabilities (QRS detection, heart rate determination, arrhythmia interpretation, atrial fibrillation detection) as the predicate. Although the algorithm was modified to expand ECG processing capability from two-lead only to both two-lead and single-lead formats, performance validation demonstrated that this technological difference does not raise different safety or effectiveness questions, as the same international standards were applied and no deficiencies were identified in verification and validation testing.