Medtronic Leverages MATLAB to Develop Deep Learning Model for Atrial Fibrillation Detection in Implantable Cardiac Monitors

Medtronic has developed an implantable cardiac monitor that enables continuous ECG monitoring to detect conditions such as atrial fibrillation (AF). Detecting AF is clinically important because it can help prevent stroke and heart failure. However, the intermittent and asymptomatic nature of AF makes accurate long-term monitoring extremely challenging. While existing device algorithms achieve high sensitivity, they also generate a large number of false-positive results, leading to doctors spending hundreds of hours annually on manual review.

To address this challenge, the team quickly built and benchmarked a prototype of an AI-based post-processing solution using MATLAB®. They combined ECG features, such as Lorenz plots, P-wave morphology, atrial rate estimation, and RR interval variability, into a single transformed image, then trained a simple convolutional neural network using Deep Learning Toolbox™ and Parallel Computing Toolbox™. In just three weeks, they developed a model that reduced false atrial fibrillation detections by over 90% while maintaining minimal loss in true positive detection rates, thereby improving the efficiency of clinical workflows.

"MATLAB makes it very easy for us to complete this task, learn the entire process, and be able to place models and evaluate them in a highly efficient manner. Because you have both the Parallel Computing Toolbox and the Deep Learning Toolbox, and you can combine them together."

—— Shantanu Sarkar, Medtronic

Key Results

• Medtronic used MATLAB and Deep Learning Toolbox to create and validate a high-performance AI model in just three weeks, with a validation accuracy rate of 97%.
• Medtronic utilized MATLAB for signal processing and feature engineering to develop a customized feature ensemble based on ECG atrial fibrillation characteristics.
• Medtronic Reduced False Positive Detection Rate of Atrial Fibrillation by More Than 90%, Minimizing Physicians' Review Time.

Follow the MATLAB Official Account ↓↓↓