Abstract: Methods, systems, and apparatuses are described for automated external defibrillation for determining shock/no-shock decisions based on ECG readings taken when a patient is undergoing cardiopulmonary resuscitation (CPR). A device may filter an ECG signal of a patient to reduce artifacts from the signal caused by CPR being performed on the patient. The device may use the filtered ECG signal to determine whether to deliver a therapeutic shock to a patient even if the patient is undergoing CPR.

Abstract: The invention provides, in some aspects, an automated external defibrillator (AED) that includes defibrillation circuitry capable of delivering a therapeutic shock to a patient. The AED also includes an input that receives a patient ECG signal representing real-time electrical activity of a heart of the patient and that may include artifacts resulting from cardiopulmonary resuscitation of the patient. The AED generates a real-time alert from a patient data image that is based on the patient ECG signal to effect application of the therapeutic shock to the patient. The AED generates that alert based at least in part on a shock recommendation of an artificial intelligence (AI) engine that analyzes the patient data image using a machine learning (ML) model trained with training data images. The patient data image comprises pixels whose values are a function of magnitudes of components of representations of respective time-slices of the patient ECG signal in a frequency domain.