A Concept Learning-Based Patient-Adaptable Abnormal ECG Beat Detector for Long-Term Monitoring of Heart Patients

Abstract

n this chapter, a new concept learning-based approach is presented for abnormal ECG beat detection to facilitate long-term monitoring of heart patients. The novelty in our approach is the use of complementary concept—“normal” for the learning task. The concept “normal” can be learned by a v-support vector classifier (v-SVC) using only normal ECG beats from aspecific patient to relieve the doctors from annotating the training data beat by beat to train a classifier. The learned model can then be used to detect abnormal beats in the long-term ECG recording of the same patient. We have compared with other methods, including multilayer feedforward neural networks, binary support vector machines, and so forth. Experimental results on MIT/BIH arrhythmia ECG database demonstrate that such a patient-adaptable concept learning model outperforms these classifiers even though they are trained using tens of thousands of ECG beats from a large group of patients.