Diagnosis of long QT syndrome via support vector machines classification

Halil Bisgin,Ahmet Ugur,Orhan U Kilinc,Volkan Tuzcu,Xiaowei Xu

doi:10.4236/jbise.2011.44036

Abstract

Congenital Long QT Syndrome (LQTS) is a genetic disease and associated with significant arrhythmias and sudden cardiac death. We introduce a noninva-sive procedure in which Discrete Wavelet Trans-form (DWT) is used to extract features from elec-trocardiogram (ECG) time-series data first, then the extracted features data is classified as either abnormal or unaffected using Support Vector Machines (SVM). A total of 26 genetically identified patients with LQTS and 19 healthy controls were studied. Due to the limited number of samples, model selection was done by training 44 instances and testing it on remaining one in each run. The proposed method shows reasonably high average accuracy in LQTS diagnosis when combined with best parameter selection process in the classifica-tion stage. An accuracy of 80%is achieved when Sigmoid kernel is used in v-SVM with parameters v = 0.58 and r = 0.5. The corresponding SVM model showed a classification rate of 21/26 for LQTS pa-tients and 15/19 for controls. Since the diagnosis of LQTS can be challenging, the proposed method is promising and can be a potential tool in the correct diagnosis. The method may be improved further if larger data sets can be obtained and used.

Highlights

We introduce a noninvasive procedure in which Discrete Wavelet Transform (DWT) is used to extract features from electrocardiogram (ECG) time-series data first, the extracted features data were classified as either abnormal or unaffected using Support Vector Machines (SVM)
The outcome of every set of experiments was evaluated on a leave-one-out cross-validation (LOOCV) technique and we obtained series of accuracies where SVM models can successfully work
This study demonstrates that a signal processing technique, DWT and a data mining method, SVM can contribute in the diagnosis of Long QT Syndrome (LQTS)

Summary

Introduction

QT interval consists of depolarization and repolarization time periods. Zation, QT interval mostly gives information about repolarization. Various subtypes of LQTS can show significant differences in QT intervals. The diagnosis of LQTS can be tricky since patients do not always exhibit significantly prolonged QT intervals. Healthy individuals can have slightly prolonged QT intervals. Besides the assessment of an expert, which may not always be timely, the need of reliable automated analysis is inevitable. Such noninvasive techniques will make the process faster, and reduce the cost

Objectives

Methods

Results

Discussion

Conclusion