Comparison of Time and Frequency Methods for Cardiac Anomalies Recognition

The main aim of this paper is a contribution to the design of an Intelligent Diagnosis System for Cardiac anomalies detection in acquired ECG (Electrocardiogram) signals. To attain this goal, the authors have developed two approaches to extract the most relevant and significant parameters that can best classify ECGs into two classes namely, Normal and Abnormal from real Electrocardiogram signals. The first approach is a time method which consists in modeling ECGs using RBF (Radial Basis Function) neural network algorithm, whereas, the second approach is a frequency based method which relies on the DWT (Discrete Wavelet Transform) that projects the signal into a time-scale (time-frequency) plane. In both approaches, the resulting data are submitted to the same SVM (Support Machine Vector) Classifier. The results obtained have shown a good adjustment of relevant parameters for each approach and have revealed the most efficient combined processing-classification algorithm that has achieved classification rates up to 100 % allowing at a time a reduced number of parameters. This fact had considerably simplified the hardware for a real time implementation. Finally, the authors compared both approaches and have identified some implementation constraints, such as the sampling frequency for the frequency based approach and preprocessing and cutting for the time approach.