[Paper Review] A Novel real-time arrhythmia detection model using YOLOv8

Guang Jun Nicholas Ang, Aditya Goil+7 more|arXiv (Cornell University)|May 26, 2023

ECG Monitoring and AnalysisMedicine3 citations

TL;DR

This paper proposes a real-time arrhythmia detection system using a loss-modified YOLOv8 model to classify single-lead ECG signals as arrhythmia or normal beats. By treating ECG segments as object detection tasks, the model achieves 99.5% test accuracy and 0.992 mAP@50 with inference in just 0.002 seconds on a V100 GPU, demonstrating high-accuracy, low-latency deployment potential for home-based cardiac monitoring.

ABSTRACT

In a landscape characterized by heightened connectivity and mobility, coupled with a surge in cardiovascular ailments, the imperative to curtail healthcare expenses through remote monitoring of cardiovascular health has become more pronounced. The accurate detection and classification of cardiac arrhythmias are pivotal for diagnosing individuals with heart irregularities. This study underscores the feasibility of employing electrocardiograms (ECG) measurements in the home environment for real-time arrhythmia detection. Presenting a fresh application for arrhythmia detection, this paper leverages the cutting-edge You-Only-Look-Once (YOLO)v8 algorithm to categorize single-lead ECG signals. We introduce a novel loss-modified YOLOv8 model, fine-tuned on the MIT-BIH arrhythmia dataset, enabling real-time continuous monitoring. The obtained results substantiate the efficacy of our approach, with the model attaining an average accuracy of 99.5% and 0.992 mAP@50, and a rapid detection time of 0.002 seconds on an NVIDIA Tesla V100. Our investigation exemplifies the potential of real-time arrhythmia detection, enabling users to visually interpret the model output within the comfort of their homes. Furthermore, this study lays the groundwork for an extension into a real-time explainable AI (XAI) model capable of deployment in the healthcare sector, thereby significantly advancing the realm of healthcare solutions.

Motivation & Objective

• To enable real-time, self-administered arrhythmia detection using single-lead ECG signals in home environments.
• To address the limitations of traditional ECG analysis, such as high computational cost and manual interpretation fatigue.
• To adapt object detection techniques from computer vision to ECG signal classification for improved speed and accuracy.
• To develop a deployable, lightweight model suitable for edge devices and future integration into wearable health systems.

Proposed method

• Transformed single-lead ECG signals into 2D image-like representations using time-frequency mapping for input into YOLOv8.
• Fine-tuned YOLOv8n on the MIT-BIH arrhythmia dataset with a novel loss function combining dynamic inverse class frequency and Wise IoU to address class imbalance.
• Utilized a lightweight YOLOv8n architecture to ensure real-time inference performance on consumer-grade hardware.
• Applied data augmentation and normalization techniques to improve robustness and generalization across diverse ECG waveforms.
• Validated the model using 10-fold cross-validation and tested on unseen data to ensure reliability and consistency.
• Generated visual outputs with bounding boxes and class labels to provide interpretable feedback for end-users.

Experimental results

Research questions

• RQ1 Can YOLOv8 be effectively adapted for real-time arrhythmia detection in single-lead ECG signals using object detection principles?
• RQ2 How does the proposed loss-modified YOLOv8 model perform in terms of accuracy and inference speed compared to existing deep learning models for ECG classification?
• RQ3 To what extent does the use of dynamic inverse class frequency and Wise IoU improve detection performance on minority arrhythmia classes?
• RQ4 Can the model achieve real-time inference with minimal latency suitable for deployment on portable or edge devices?
• RQ5 What are the challenges and limitations in deploying such a model in real-world, noisy home environments?

Key findings

• The model achieved a test accuracy of 99.5% and a mean average precision of 0.992 mAP@50 on the MIT-BIH dataset, demonstrating state-of-the-art performance for object detection-based ECG classification.
• Inference time was measured at 0.002 seconds per ECG segment on an NVIDIA Tesla V100 GPU, confirming real-time capability.
• The use of dynamic inverse class frequency and Wise IoU loss significantly improved detection performance on minority arrhythmia classes, reducing class imbalance bias.
• The model maintained consistent performance across 10-fold cross-validation, indicating robustness and generalization across different data splits.
• The visual output with bounding boxes and confidence scores enables intuitive interpretation by non-expert users, supporting patient self-monitoring.
• Despite high performance, the model’s decision-making process remains a black box, highlighting the need for future explainable AI (XAI) integration in clinical deployment.