ANN based Multilevel Classification Technique with Optimum Measurement Period for Accurate Diagnosis using Biomedical Signals
Paul Thomas1, R.S. Moni2

1Paul Thomas, Department of Electronics and Communication Engineering, Mar Baselios College of Engineering and Technology, Nalanchira, Trivandrum (Kerala), India.
2Dr. Moni R.S., Department of Electronics and Communication Engineering, Marian Engineering College, Menamkulam, Trivandrum (Kerala), India.

Manuscript received on 10 December 2016 | Revised Manuscript received on 18 December 2016 | Manuscript Published on 30 December 2016 | PP: 63-70 | Volume-6 Issue-2, December 2016 | Retrieval Number: B4790126216/16©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (

Abstract: Biomedical signals are representations of the mechanical and electrical activities within the human body. These signals contain a lot of information on the state of health of a person and their analysis have a significant role in the diagnosis of various health disorders and medical abnormalities, such as activation levels and the biomechanics of the muscles and other human organs. Of the many Biomedical signals, focus of this work is on Electro-cardiogram (ECG) and Electro-myogram (EMG). ECG provides information on the rhythm and functioning of the heart. EMG is the recording of human muscular activity. ECG signals used in this work are taken from the standard MIT-BIH, and CU data bases of PhysioNet database and EMG signals are taken from the EMGLab and PhysioNet database. Automated analysis of Biomedical signals can largely assist the physicians in their diagnostic process. The extracted spectral and temporal features represent the diverse characteristics of a Biomedical signal. In this work, more emphasis is given to spectral features since a lot of critical information on the health of a person are hidden in the spectral content of the signal. A subset from a larger set of available features is experimentally selected for optimum performance. The feature vector has a size of 11 for ECG signal analysis and a size of 9 for EMG signal analysis. Accuracy of detecting a health disorder depends on the quality of the features extracted from a Biomedical signal. A few techniques are proposed to achieve improved quality for the features. Also a method is developed to arrive at the optimum length of the Biomedical signal to be used for analysis. Accordingly, the length of the ECG signal used in this work is 10 s and the length of the EMG signal is 11 s. It is observed that the variance of the features is minimum when the signal for analysis is taken from the mid portion of the whole Biomedical signal. To make the value of a feature close to its true value, each feature value is taken as the average of the values of the feature extracted from 20 consecutive signal segments. A technique is also proposed to reduce the effect of wild points in the computation of spectral parameters. It is observed that classification accuracy also depends on the sampling rate of the Biomedical signal. The sampling rate of ECG signal in this work is 128 Hz and that of EMG signal is 750 Hz. Classifying a Biomedical signal is the process of attaching the signal to a disease state or healthy state. The work proposes a Multi level classification approach for Biomedical signals. Each classifier is a cascade of two ANN classifiers, the first ANN has a linear transfer function and the second ANN has a sigmoid transfer function. First level classification is to the broad categories of the disorders. In the second level, these disorders are drilled down to more specific categories. This concept can be extended further to achieve finer classification of Biomedical signals. In this work the classification is demonstrated to two levels for ECG signals and one level for EMG signals. The performance of the proposed method is evaluated using the standard parameters of specificity, sensitivity and classification accuracy (CA). The performance is found to be better than the reported figures in the case of both ECG and EMG signals.
Keywords: ECG, EMG, FFT, DWT, Pattern Recognition ANN, Feature Extraction, Multilevel Classification, Wavelet, Physionet Database, CA, Atrial Arrhythmias, Ventricular Arrhythmias, NSR, MI, MUAP, Myopathy, ALS.

Scope of the Article: Pattern Recognition