LAUSR

Health monitoring devices are integral parts of smart health in the era of smart connected communities. In recent years, remote heart monitoring systems are developed to harness advanced machine learning methods to identify heart disorders by processing electrocardiogram (ECG) signals. However, current technologies suffer from two important drawbacks: i) the lack of prediction capacity to predict heart abnormalities ahead of time, and ii) failure in capturing inter-patient variability. In this paper, we propose a novel two-step predictive framework for ECG signal processing, where a global classifier recognizes severe abnormalities (red alarms) by comparing the signal against a universal reference model. The seemingly normal signal samples undergo a subsequent deviation analysis and yellow alarms are called by identifying mild and yet informative signal morphology distortions comparing to the learned patient-specific baseline that can be indicative of upcoming heart conditions. To facilitate an accurate deviation analysis, a controlled nonlinear transformation with optimized parameters is proposed to increase the symmetry of signals for different abnormality classes in the feature space. The proposed method achieves a classification accuracy of 96.6% and provides a unique feature of predictive analysis by generating precaution warning messages about the elevated risk of heart abnormalities to take preventive actions according to physician orders. In particular, the chance of observing a severe problem (in terms of a red alarm) is raised by about 5% to 10% after observing a yellow alarm of the same type. The main goal of this technology is providing quality healthcare for elderly and high-risk heart-patients, however, the developed methodology is general and applicable to other biomedical signals such as EEG, Pleth, and PPG.