Abstract:
Cardiac diseases are among the leading causes of deaths all around the world. About 23.5% of people die every year as a consequence of heart diseases mainly cardio vascular diseases (CVDs) such as ischemic strokes, heart attacks and cardiac arrhythmias. Cardiac arrhythmia refers to a condition of abnormal heart rhythm or electric activity of the cardiac conduction system resulting into an irregular heartbeat barely discernible.
Signal processing is a significant tool used in biomedical engineering and diagnostics. It has a dominant contribution in feature extraction and pattern analysis for efficient diagnosis of various diseases by using fMRIs, EEGs and EMGs recorded from the subjects. Numerous cardiac conditions can be analyzed using one of the oldest and the simplest tools called ECG. Amongst its various types, some cardiac arrhythmias can be fatal and may lead to Sudden Cardiac Death Syndrome (SCD). Using calipers for ECG analysis is a tedious and unreliable process which may result into misdiagnosis and hence mistreatment of the arrhythmia. The dilemma is that despite the availability of latest equipment and various treatments of arrhythmias in our country, the discussed situation leads to an increased death ration.
Despite the availability of enriched literature, mature methods were only tested on only a few categories in the past, not paying any attention to many important arrhythmias such as Pre-Excitation, Paced rhythms, Bigeminy, Idioventricular rhythm and others. Considering how unexplored various abnormal rhythms have been in previous researches, this thesis explains in detail the development of a self- diagnostic and automated cardiac arrhythmia recognition system using hybrid features extracted in time, frequency and wavelet domains. Signal acquisition, processing and correct event classification for 15 types of arrhythmias is carried out. The system is capable of automatic diagnosis with a very high degree of certainty which stratifies the risks of life-threatening arrhythmic abnormalities by recognizing trends in ECGs. This will assist the physicians in correct medical diagnosis by taking ECG as an input and identifying the status of the heart rhythm.
The algorithm is tested on the benchmark MIT-BIH Cardiac Arrhythmia Database. An overall accuracy of 98.12%, sensitivity of 97.79% and specificity of 92.96% justify the performance of the suggested technique over existing methods.
