Enhanced Ischemia Classifier for ECG Signals

Abstract

Myocardial ischemia (MI) is the most common type of coronary heart disease (CAD) and cause of the heart attacks. The ischemia is the result of the accumulation of cholesterol and plaque building narrows the lumen of the arteries of the heart, hence heart becomes starved of oxygen and the vital nutrients, it needs to pump properly. It usually leads to angina (chest pain), myocardial ischemia or sometimes Infraction. The non-invasive method, Electrocardiogram (ECG), is a universally adopted method for diagnosis of ischemia. The ECG is a procedure of recording of the electrical activity of the myocardial fibers of the heart over a period of time. The ECG has a characteristic morphology comprises of P wave, QRS complex and T wave. The ST segment deviations are produced by the flow of injury currents, generated by the voltage gradients between the ischemic and non-ischemic myocardium during the plateau and resting phases of the ventricular action potentials. This is manifested as elevated or depressed ST segments in the ECG. Elevated ST segments appear typically in transmural ischemia, while depressed ST segment appears subendocardial ischemia. During recording of ECG, artifacts are incorporated on the ECG.

Commonly noticed artifacts include baseline wandering, power line interference and muscle tremors. Artifacts are extremely common, and knowledge of them and elimination is primarily required to prevent mis-diagnosis of ischemia. There are various ischemia detection methods available in the literature exhibiting certain disadvantages, like classifier trainings and the involvement of the complex calculations for decision. Hence, a simple, reliable and rugged method is required to address these disadvantages. Recent developments in signal processing methods, multi-resolution analysis, wavelet transform is found more suitable for ECG analysis, preprocessing and delineation technique. Using wavelet transform toolbox in MATLAB programming, we preprocessed the ECG records and then delineates the ECG characteristic points. The proposed method shows better performance than the existing methods. Simulated results show that the optimal value of the wavelet function for eliminating of artificially added artifacts as, db7 at 8th decomposition level for baseline wander with percentage root mean square difference (PRD) of 0.1636%, coif2 at a 5th decomposition level for power line interference with PRD of 0.3980% and db7 at 8th decomposition level for muscle tremor with PRD of 0.8959% in selected records of European ST-T database. Moreover, it also addresses the issues of conventional methods. Similarly, for delineation process, the proposed method result shows 99.72% sensitivity (SE), 99.78% positive predictivity (+P) for 10 records. Then detected ST segment has been classified as normal or ischemic based on proposed isoelectric energy function (IEEF), an alternate method to detect the elevation or depression of ST segment with respect to isoelectric reference. The proposed method is a mathematical function that detects these deviations in a rugged manner. The motivation behind developing this function is to go in for a simpler but reliable threshold based classification for detecting myocardial ischemia. The isoelectric energy threshold is used to classify ischemic beats from normal beats. The method is validated for 43,876 ST segments for one lead of annotated European ST-T database (EDB) records. The results show 98.12% average sensitivity, 98.16% average specificity, 98.36% average positive predictivity and 100% sensitivity for ischemic episodes detection process has been achieved for these records. These results are significantly better than those of existing methods cited in the literature. In addition, we have also proposed a simple classifier to classify the detected ischemia episode as transmural or subendocardial ischemia. Filtering and delineation has been performed in such a well-organized manner, which makes possible to attain significantly better results than the existing methods. The proposed method has following advantages. The method could provide an interpretation of the results. This is of great importance for a medical decision support for the patient in the critical care unit (CCU) without knowing past references. The developed method filters out the spurious beats in a record. Another advantage includes direct analysis based on isoelectric energy without involvement of any complicated algorithm.