Manifold learning for premature ventricular contraction detection

Prompt diagnosis of abnormally shaped wave forms in ECG signal is an important component in the early diagnosis of cardiac arrhythmias, improving the quality of life of patients. Meanwhile, detection models for Premature Ventricular Contractions (PVC) are widely investigated, a less studied problem is data analysis and visualization. In this paper, we propose an approach for PVC detection and data visualization by exploiting the intrinsic geometry of the high-dimensional data using manifold learning and Support Vector Machines (SVM). ISOMAP forms a neighborhood-preserving projection which allows to uncover the low-dimensional manifold and is used here as a pre-processing step. Then by incorporating training labels the method is capable of recognizing PVC patterns with comparable accuracy of kernel learning machines.