Automatic Atrial Fibrillation Detection using Artificial Neural Network

Atrial fibrillation is one of the serious heart diseases in which the heartbeats are
irregular. Patients with this disease usually have shortness of breath, dizziness, and
tiredness. Atrial fibrillation is considered a serious disease because the heart may
develop clots of blood that might travel to the brain and cause a stroke, which may
lead to death.
Since it is heart disease, atrial fibrillation can be detected by observing the
electrocardiograph (ECG) of the patient. The ECG is usually characterized by its
peaks, intervals, and segments, and using the ECG, many features have been extracted
to detect atrial fibrillation. Atrial fibrillation can be identified by observing the heart
rate variability and the atrial activity from the ECG.
In this work, I used an open-source feature extraction program that uses a modified
version of the Pan-Tompkins algorithm and several other open-source algorithms to
detect QRS complex and R peaks. Using the extracted features, I built an artificial
neural network for automatic atrial fibrillation detection. Moreover, I measured the
effect of using feature selection algorithms in enhancing the classification results.
Also, I was able to overcome the challenge of the unbalanced dataset by using the
weighted neural network and under-sampling the dataset to be almost balanced.
Based on the achieved results, and using feature selection tools, the performance of
the classification could be improved. The obtained performance results are compared
with other powerful tools. In this work and using an artificial neural network, I got an
overall F1 score of 76.5% on the test dataset which is in the range of results achieved
by other researchers. The complexity of the proposed system is light and fast which
makes it a good choice to be used in portable devices and real-time applications.