Artificial Intelligence

The future of electrocardiography lies within AI-assisted analysis

Traditional Algorithms

  • Work like a decision tree
  • Based on a limited number of pre-defined rules
  • Hard to detect P-waves due to low signal-to-noise ratio, possible overlapping with T-waves, variability with specific morphologies or diseases
  • Lack specificity, creating a high rate of false alarms, which are a burden to manage

Cardiologs AI

  • Based on deep learning technology that interprets the whole ECG¹
  • Uses a similar thought process as the brain, handling multiple abnormalities and complex patterns at once
  • Our delineation Deep Neural Network (DNN) segments different electrical waves (P, QRS, T). It was built upon 20M+ recordings²
  • Improved specificity, dividing analysis times by two³
Read more about our Deep Neural Network

Our algorithm detects more than 20 types of events

Including the main arrhythmias:
  • Pause
  • Atrioventricular Block (second-degree Mobitz types I and II, third-degree, advanced high-grade AV block)
  • Atrial Fibrillation or Atrial Flutter
  • Ventricular Tachycardia
  • Premature Supraventricular Complexes (PSVCs), PSVC Couplets
  • Premature Ventricular Complexes (PVCs)
  • Ventricular Couplets, Ventricular Bigeminy, Ventricular Trigeminy
  • Sinus rhythm, Bradycardia, Tachycardia
Performance of our algorithms was tested and validated following requirements of the standard IEC 60601-2-25: 2011(Part 2-25). It was also tested on MIT-BIH, AHA and NST databases, according to the requirements of the ANSI AAMI EC57 - 2012 and IEC 60601-2-47:2012 standards.

Validation of our AI algorithm

10 times reduction in AFib False Positives in Holter

Using the MIT-BIH public database, our Cardiologs AI algorithm was compared to the results of a reference RR interval-based method⁴. Our AI algorithm was shown to improve the specificity in Afib detection from 82.8% to 96.9% with comparable sensitivity.

Read the publication

Reduction of False Positives by up to 70% in ICM

In a clinical study including 425 patients with ICMs, our algorithm was applied as a filter to all episodes detected by the ICM as Afib. It resulted in a reduction of false positive episodes by almost 70%, thereby considerably reducing the clinical burden of managing the data generated by those devices⁵.

Read the publication
Dr. Suneet Mittal, BA, MD
Director of the Electrophysiology Laboratory and Associate Chief of Cardiology at the Valley Hospital in Ridgewood, NJ.

“The high false positive rate of AFib detected by Implantable Loop Recorders (ILRs) has created a significant clinical burden. Since ILRs transmit data daily, these false positives are one of the Achilles heels of remote cardiac monitoring. This study validates that Cardiologs’ advanced AI can filter 2/3 of false positive AF episodes, which should improve clinical efficiency”