COMPUTER AIDED ANALYSIS AND INTERPRETATION OF EMG SIGNAL FOR DISEASE DIAGNOSTICS

Abstract

he Electromyograph (EMG) is useful to know the state of a patient under medical diagnosis and treatment. As the number of neuromuscular patients is increasing, it is not possible to take care of all the neuromuscular patients by carrying out manual investigations under all the conditions. After realizing this problem about three decades back, a good amount ofwork has been carried out on computer aided analysis and interpretation ofEMG signal. In building an Expert System, one of the major problem encountered is to extract the knowledge from human experts. That is why, identification and encoding of the knowledge are two most complex and arduous tasks encountered in the construction of an Expert System. To handle uncertainties in symptoms descriptions and data, fuzzy logic and fuzzy neural networks have come upas a helpful tools in Expert System based diagnosis. The EMG data acquisition and preprocessing, detection of Motor Unit Action Potentials (MUAPs), decompositions of EMG, feature extraction of all MUAPs and their usage in disease classification and diagnostics are the important stages in computer aided EMG analysis and interpretation. After carrying out detailed survey, it was found that there is a gap between what is ideally required and what has been achieved so far in the area of computer aided EMG analysis and interpretation. Thus after finding the gap, the problem was formulated to carryout the work on the automated analysis and interpretation of the EMG signal. The objective in the work had been to develop knowledge-based system for computer aided analysis and interpretation ofEMG signal for disease diagnostics. For the present work, myopathy, motoneuron disorders and Hansen disease have been considered for detailed study and development of knowledge based system for their identification, classification and diagnosis. Separate knowledge base systems have been developed for myopathy and motor neuron disorder category of disease and for hansen disease category as these are two different classes of diseases and do not overlap each other. Aknowledge base has been developed for each of these classes. EMG signal recorded by the needle electrode is used for the analysis and identification of myopathy and motor neuron disorder type of diseases and EMG signal recorded by surface electrode is used for the analysis and identification of hansen disease. The first stage of the work deals with the decomposition ofEMG signals using Wavelet Transform (WT). In this study, four decomposition algorithms, which have been modified, implemented, analyzed and evaluated for their performance for separating the MUAPs from the EMG Signal. The performance has been evaluated keeping in view the parameters like accuracy, speed, reliability and efficiency of algorithms for extracting clean MUAPs even from those EMG signals, which are recorded for limited duration and contained noise and interference. Both synthetic and real time EMG signals have been used for testing. The classification success rate is 98.9% with both statistical pattern recognition and cross-correlation approaches, 99.2% with Kohonen neural network, and 99.8% with wavelet transform. Although the success rate is higher of Kohonen network, still the wavelet transform method has been recommended as there is no need of correction for baseline drift or high frequency noise. The WT method allows fast extraction of the localized frequency components, provides good time-resolution, and is capable of tracking rapid changes in MUAPs. The superimposed signal, which could not be separated by this technique, has been decomposed by using cross-correlation and Euclidean distance. All the algorithms have been successfully implemented and tested for decomposition of EMG signals recorded from subjects having normal (NOR) state of muscle and having motor neuron disorder (MND) and myopathy (MYO) disease. In the second stage of work, all the clinically important parameters namely duration, spike duration, amplitude, area, spike area, phases and turns of the MUAPs in time domain and moments oforder zero, one and two; median frequency, maximum frequency, quality factor and bandwidth in frequency domain have been identified and extracted. Generally, three types of MUAP morphologies (class I, II & III), which are normally present with the slight variations in different category of muscle disorders, have been taken into consideration. Each disease has MUAPs with some variations, which can be categorized in any one of the three classes. Therefore, we have taken the all MUAPs having same morphology. Clinical important parameters in time and frequency domain for each class were extracted and for each parameter, lower, mean and upper value have been determined. In this way, features were obtained in fuzzified form. Total 58 MUAPs sets are computed from 25 subjects. 17 MUAPs sets were rejected on the basis of distortion, noise and repeatability. The crisp values of these parameters have also been obtained by taking single MUAP from each class in each subject (i.e. one MUAP from each 58 MUAPs sets) randomly. In the next phase of work, a fuzzy neural network based knowledge base system for the diagnostics of neuromuscular disorders has been developed. The key feature of the knowledge based system is the processing of the information using fuzzy AND and fuzzy OR gates. The successive reduction approach has been used for reducing the number of rules for interpretation. in Out of the two available methods to obtain the possibility measure, look-up table method has been used in this work. The neural network has been trained by optimizing the local weights between input and hidden layer and global weights between hidden layer and output layer. The optimized local weights have been used to obtain the contribution ofindividual symptoms, whereas, the global weights have been used to finalize disease diagnosis. The knowledge based system has been successfully trained and tested using synthetic and real MUAPs for normal (NOR), motor neuron disorder (MND) and myopathy (MYO). On the basis of exhaustive test results, it can be claimed that the developed knowledge based system can be successfully used in real hospital environment for the neuromuscular disease diagnostics. In this knowledge based system, sometimes the lower, middle and upper values of the parameters are overlapping, therefore, it becomes very difficult to finalize category ofdisease; also the number of rules becomes higher, therefore, knowledge based system takes more time to take final decision, thereby, training ofthe fuzzy neural network is very complex manner and limits the performance ofthe network. This knowledge based system is only applicable for crisp measurements i.e. when data are available in crisp form. Due to these limitations of this knowledge based system, the work has been carried out on a second knowledge based system which works under both the conditions i.e. when data are available in crisp or fuzzy form. The second knowledge based system uses time and frequency-domain parameters of the Motor Unit Action Potential (MUAP) and the symptoms of a patient. The time- and frequencydomain parameters and the symptoms are fuzzified to bring-in the procedure of the diagnostics exactly in line with the process being adopted by the human experts. The time- and frequencydomain parameters are handled by distance measure criteria while the symptoms are handled by Error Back Propagation-Neural Network (EBP-NN). The final diagnostics is carried out by taking an overall view of the diagnostics made individually from time and frequency domain parameters and the symptoms. The system has been tested using MUAP data ofNormal (NOR) muscles as well as ofmuscles with Motor Neuron Disorder (MND) and Myopathy (MYO). Once myopathies had been classified, then EBP-NN is used for further sub-classification of myopathies on the basis of only symptoms. The test results are highly encouraging, and after carrying out exhaustive testing in the clinical environment, the developed system can be reliably used for diagnostics in the hospitals/ healthcare centres. Before developing these knowledge based system, the knowledge base was created by carrying out in depth study of the available literature and also by interaction with the medical IV experts in hospitals. The link has been established to have such interaction at Medical College Bhopal, India. The data base for recorded real time EMG signal for 7normal (NOR) subjects and for 8subjects suffering from Motor Neuron disorder (MND) and for 8subjects suffering from Myopathy (MYO) was obtained from the Department of Computer Science, University of Cyprus. The EMG signal was acquired from the biceps brachii muscle using aconcentric needle electrode. The signal was sampled at 20kHz for aduration of 5sec. The MUAPs were recorded at different points in the muscle using monopolar and concentric needle electrodes in normal subjects and subjects with MYO and MND disorders. The same knowledge base and database have been used in both the knowledge based systems. Clinical electromyography (EMG) also provides useful information for the diagnosis of Hansen Disease (HD) or leprosy. AComputer Based knowledge based system has been developed for the diagnosis of HD using the features of EMG signal recorded from the disease site and the symptoms obtained from the observation and interaction with the patient. In this study, the wavelet transform has been applied for feature extraction. The knowledge based system uses the rms value and frequency domain parameters of the EMG signal and the symptoms under both the healthy and leprosy conditions of the subjects for the disease diagnostics using an artificial neural-network. The test results are promising and show that the system can be successfully used for the identification of leprosy condition right from its initiation stage and a large number of subjects can be saved from falling in the deadly trap of this disease. The recorded data available in the Department ofElectrical Engineering, IIT Roorkee, India, has been used for the development of the knowledge based system. The data was recorded by DISA 1500 EMG SYSTEM (a microprocessor based device) at the EMG laboratory of the Schieffelin Leprosy Research and Training Centre Karigiri, India. Three knowledge based systems for the identification and detection of myopathy, motor neuron and hansen diseases were trained with the known cases. First two systems have been developed for automatic identification of different muscle disorders and third system has been developed for Hansen disease diagnostics. The validation of these systems was done by interacting with the physicians on the basis ofEMG signal and patient history.