A robust high accuracy cardiovascular disease detection system based on ECG energy concentration time-frequency analysis supported by threshold and intelligent classifier

Globally, cardiovascular diseases (CVDs) are the primary cause of deaths. According to the most recent statistics of the World Health Organization (WHO), CVDs mortality rates are expected to range between 246 deaths for 100,000 population in 2015 to 264 for 100,000 population in 2030. Reportedly, nearly half of them do not indicated any prior symptoms or experienced any pain of heart attack. Moreover, about 25% of CVDs patients were unable to get timely medical aid at the critical time especially those who experienced heart problems at the late stages and who live in remote places. High accuracy out-of-hospital detection of CVDs is, therefore, vital to prevent complications of the heart that may lead to sudden death or disability. Electrocardiogram (ECG) represents cardiac condition as electrical signal waveforms. However, the interpretation of these waveforms is still very challenging because the signals are mainly composite of eight different signals from various heart components namely atriums, ventricles, sinus node, AV-node, and common bundles. The non- stationary and multi-frequency nature of ECG signal waveforms makes the use of Time-Frequency Distributions (TFDs) for analysis, inevitable. The main aim of this study is to develop a high accuracy scheme for CVDs detection, including ischemia and arrhythmia, for multi-lead and long intervals ECG signal waveforms. The scheme is based on non-linear TFD analysis supported by threshold technique and intelligent machine learning classifier namely Support Vector Machine (SVM). In addition to the new TFD scheme, the use of multi-leads instead of single lead, and 1-minute interval instead of beats or frames for classification, contributes to the improvement of detection performance. In addition to the venerable MIT database, a 7-lead low power ECG device is also designed and implemented. It is used for raw ECG data acquisition to further evaluate the proposed scheme for the ECG data outside the MIT ECG database and enable the real-time CVDs detection capability. The ECG data collected from this device have also been evaluated for both normal and abnormal cases. The proposed scheme is examined and evaluated with various normal and abnormal ECG cases that cover CVDs namely arrhythmia and ischemia. The datasets used in this study comes mainly from MIT ECG database where it is used for the classifier training and performance evaluation as well. The proposed scheme contributes to a very high overall accuracy, sensitivity and specificity of more than 99% for CVDs detection. The results for arrhythmia detection are 99.39% accuracy, 99.38% sensitivity, and 99.44% specificity. The results for ischemia detection are 99.10% accuracy, 99.09% sensitivity, and 99.13% specificity. These results indicate that the proposed scheme is suitable for CVDs detection and can be an excellent platform for automated CVDs detection systems providing on-demand or continuous monitoring for long time duration at high accuracy