سیستم هوشمند ترکیبی برای رده بندی آریتمی های قلبی با سیستم فازی مرکب از قاعده همسایه K نزدیکترین فازی و شبکه های عصبی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|5580||2012||9 صفحه PDF||سفارش دهید||4200 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 39, Issue 3, 15 February 2012, Pages 2947–2955
In this paper we describe a hybrid intelligent system for classification of cardiac arrhythmias. The hybrid approach was tested with the ECG records of the MIT-BIH Arrhythmia Database. The samples considered for classification contained arrhythmias of the following types: LBBB, RBBB, PVC and Fusion Paced and Normal, as well as the normal heartbeats. The signals of the arrhythmias were segmented and transformed for improving the classification results. Three methods of classification were used: Fuzzy K-Nearest Neighbors, Multi Layer Perceptron with Gradient Descent and momentum Backpropagation, and Multi Layer Perceptron with Scaled Conjugate Gradient Backpropagation. Finally, a Mamdani type fuzzy inference system was used to combine the outputs of the individual classifiers, and a very high classification rate of 98% was achieved.
An electrocardiogram or ECG represents the electrical activity of the heart, as a waveform graph. An ECG signal contains important information that can help medical diagnosis, reflecting cardiac activity of a patient, if it is normal or failing heart that has certain pathologies. The ECG is the standard tool used in diagnosing heart disease (Health, 2009). The physicians get those signals easily and noninvasively by adding electrodes to the patient’s body. The Holter device is frequently used for ECG recording. Physicians apply the Holter device to the patient when ECG monitoring is required to find the existence of abnormal heartbeats in a one day ECG. A person can register about 100,000 heartbeats in one day (Health, 2009). The ECG shows each heartbeat as a series of electrical waves. The contractions that pump blood are represented by the P wave, the QRS complex and T wave. The P wave represents activity in the upper chambers of the heart. The QRS complex and T wave represents activity in the lower chambers (Health, 2009) (see Fig. 1).By arrhythmia we mean any alteration in the activity of the heart rhythm in amplitude, duration or shape of the rhythm. The MIT-BIH Arrhythmia Database is a set of 48 ECG records with 30 min duration each, and each record corresponds to a patient. In this database there are different types of arrhythmias such as: L-Left Bundle Branch Block (LBBB), R-Right Bundle Branch Block (RBBB), A-Atrial Premature Beat, a-aberrated Premature Atrial Beat Premature junctional Nodal J-Beat, Fusion of Ventricular and Normal Beat, I-Ventricular Flutter Wave, J-junctional Nodal Escape Beat, E-Ventricular Escape Beat Supra-ventricular Premature Beat S-, f-Fusion of an Paced Beat Normal and normal heartbeats (MIT-BIH Arrhythmia Database. PhysioBank, 2000). Many solutions have been proposed to develop automated recognition and classification of ECG. Some processing methods have been applied to the ECG signal: Statistical and Syntatic, MultiLayer Perceptron (MLP), Self-Organizing Maps (SOM), Learning Vector Quantization (LVQ), Linear Discriminant System, Fuzzy or Neuro-Fuzzy Systems, Support Vector Machines (SVM), Bayesian approach, Experts Systems, Markov Models, Hybrid system use a combination of different solutions to improve performance (Acharya et al., 2004, Alzate and Giraldo, 2006, Anuradha and Veera Reddy, 2008, Barbosa et al., 2001, Belgacem et al., 2003, Cepek, 2007, Ceylan et al., 2009, Clifford et al., 2006, de Chazal and Reilly, 1998, Engin, 2004, Khadra et al., 1997, Maglaveras et al., 1998, Nabney et al., 2001, O’Dwyer et al., 2000, Ozbay et al., 2006, Patra et al., 2009, Sun and Chan, 2000, Tsipouras and Fotiadis, 2003 and Werbos, 1994). In this paper, we describe a hybrid intelligent system for classifying cardiac arrhythmia classification using three methods: Fuzzy KNN, MLP Gradient Descent with momentum Backpropagation and MLP Scaled Conjugate Gradient Backpropagation, and finally combine these outputs with a Mamdani fuzzy inference system that improves performance by achieving a very high classification rate.
نتیجه گیری انگلیسی
Based on the performed experiments we noticed that in the three classifiers used we achieved good results individually. But by combining their outputs using the Mamdani fuzzy inference system with the appropriate membership functions and rules we achieved a very high classification rate of 98%. We applied the hypothesis testing to compare the error rate of the three classification methods and we found that not exist enough statistical evidence to reject the nulls hypothesis in all made comparisons, so the three methods are good, but they capture (like an expert) different knowledge about classification. Therefore, the aggregation of the three classifiers with a fuzzy system provides a better overall classifier.