Real-time monitoring of ST change for telemedicine

Abstract

Modern medical breakthroughs and general improvements in environmental and social conditions have raised the global life expectancy. As the world’s population is aging, the incidence and prevalence of chronic diseases increases. Dramatic increase in the numbers of chronically ill patients is profoundly affects the healthcare system. Care at home provides benefits not only to patients but also the community and the health care providers. A telemedicine system utilizing today’s information and mobile communication technologies plays a crucial role in providing care at home. Currently, diverse telemedicine projects are progress in the most countries. A telemedicine project is supported by The Scientific and Technological Research Council of Turkey (TÜBİTAK) under Grant 114E452in Turkey. This project aims end to end remote monitoring of patients with chronic diseases such as heart failure, diabetes, asthma, and high blood pressure. A clinical decision support system integrated to telemedicine improves prognosis and quality of life in patients. The mainstay of a decision support system is early detection of important clinical signs and prompts medical intervention. Cardiovascular diseases are the leading cause of death globally. People with cardiovascular disease need early detection. An effective decision support system is needed to detect ECG arrhythmia before a serious heart failure occurs. One of the aims of the project is to develop decision support system which will detect whether a beat is normal or arrhythmia. The ECG signals in MIT-BIH arrhythmia database and Long Term ST Database are used for training and testing the algorithm. A total of 103026 beat samples attributing to fifteen ECG beat types are selected for experiments in MIT-BIH arrhythmia database. 103026 RR intervals with ST segment change were selected from the Long Term ST Database. ST segment changes detection is just based on the signal between two consecutive R peaks. The features are obtained from Wigner-Ville transform of this signal. The classification algorithms provided by the MATLAB Classification Learner Toolbox were tested. The Cubic SVM achieved best results with accuracy of 98.03%, sensitivity of 98.04%, specificity of 98 % and positive predictive value of 98%.