Meta-Learning for ECG Abnormality Detection During Driving Using a Smart Steering Wheel Device

Cardiac arrhythmias are abnormal heart rhythms caused by irregular impulse generation or conduction. Electrocardiogram (ECG) is a non-invasive tool for detecting these abnormalities. However, current models face challenges in generalizing to unseen ECG samples. This paper introduces a meta-learning approach designed to address these challenges, particularly in non-quiescent ECGs, which are often overlooked in previous researches. Using a smart steering wheel device, we collect massive non-quiescent ECG data from drivers which can record drivers' resembled Lead-I ECGs continuously during valid driving sessions. Meta learning algorithms are utilized to transfer meta-knowledge to detect abnormalities during driving. After noise reduction and waveform evaluation, qualified ECGs, including ECG records from public dataset, are input into Convolutional Neural Networks (CNNs) of meta learning framework. Experimental results show an accuracy of 94% in binary classification, outperforming traditional methods. The proposed framework provides a reliable tool for real-time ECG abnormality detection during driving.