ECG-based Arrhythmia Classification & Clinical Suggestions: An Incremental Approach of Hyperparameter Tuning

Cardiovascular diseases (CVD) are the principal cause of death globally. Electrocardiography (ECG) is a widely adopted tool to quantify heart activities to detect any heart abnormalities. Arrhythmia is one of these CVDs that heavily relies on continuous ECG recordings in order to detect and predict irregularities in the heart rhythms. Various Deep Learning (DL) approaches has been heavily used to classify and predict different heart rhythms. However, most of the proposed works do not consider the various hyperparameter optimization and tuning to get the full potential of the DL model and achieve higher accuracy. Besides, very few works implemented the full monitoring cycle and close the loop to propose some clinical and non-clinical recommendations. Therefore, in this paper, we adopt the Convolutional Neural Network (CNN) model and we apply various parameter optimization to capture various properties of the data, the training, and the model. We also close the monitoring loop and suggest tailored recommendations for each category of arrhythmia that go beyond simple to more deeper diagnosis using the Global Registry of Acute Coronary Events (GRACE), and the European Guidelines on CVDs prevention in clinical practice (ESC/EAS 2016). We conducted a set of experiments to evaluate our model and the set of hyperparameter optimization we have experienced and the results we have obtained showed significant improvement in the prediction accuracy after a couple of optimization iterations.