Artificial intelligence-based approach for atrial fibrillation detection using normalised and short-duration time-frequency ECG

Atrial fibrillation (Afib) is a heart arrhythmia that is linked to a number of other cardiac-related issues. The incidence of Afib increases with age, causing high risks of stroke. Accurate and reliable detection of Afib remains a challenge and is valuable for clinical diagnosis. This work presents a novel approach for the detection of Afib using both 1-D electrocardiogram signal and its time-frequency representation as an image (2D). The signal was pre-processed utilising a 2-stage median filter and least-square filter followed by normalisation before applying it to artificial intelligence–based models for classification. Bi-directional long short-term memory network was trained and tuned to attain high accuracy. Our proposed method shows favourable performances applying ECG segment as short as 4 s. And it has achieved an accuracy of 98.85% in the 2-D time-frequency representation. Better classification accuracy and use of short-duration ECG signal compared to existing state-of-art methods make this method suitable for an automated, reliable, and timely detection of Afib.