A NOVEL TRANSFORMER-BASED APPROACH FOR CARDIOVASCULAR DISEASE DETECTION

Nimra Noor ¹ Muhammad Bilal ¹ Saadullah Farooq Abbasi ^2* Omid Pournik ² Theodoros N Arvanitis ²

• ¹ Rare Sense Inc, Covina, United States
• ² University of Birmingham, Birmingham, United Kingdom

According to the world health organization (WHO), cardiovascular diseases (CVDs) accounts for approximately an estimated 17.9 million deaths annually. CVDs are referred to disorders of heart and blood vessels like Arrhythmia (ARR), Atrial Fibrillation (AFF), Congestive heart failure (CHF) and Normal Sinus Rhythm (NSR) etc. Early prediction of these diseases can significantly reduce the number of annual deaths. This study proposes a novel, efficient and low cost transformer-based algorithm for CVD classification. Initially, 56 features were extracted from electrocardiography (ECG) recordings using 1200 cardiac ailment records, with each of the four diseases represented by 300 records. Then, random forest has been used to select the 13 most prominent features. Finally, a novel transformer based algorithm has been developed to classify 4 classes of cardiovascular diseases. The proposed study achieved an accuracy, precision, recall, f1-score of 0.9979, 0.9959, 0.9958 and 0.9959, respectively. The proposed study outperformed all the existing state-of-the-art algorithms for CVDs classification.