State-Based General Gamma CUSUM for Modeling Heart Rate Variability Using Electrocardiography Signals

Traditional approaches based on short-term heart rate variability for cardiovascular disease diagnosis fail to capture the long-term dynamic information and individual effect from electrocardiography signals among subjects when examining the physiological condition. These shortages may lead to incorrect disease detection and weaken diagnosis performance. To address these problems, this paper proposes a new disease detection approach by considering the long-term dynamics and meanwhile the individual effect existing among subjects. Specifically, a multistate general Gamma cumulative sum (GGCUSUM) scheme is developed for signal state detection. Further, a backward elimination algorithm based on the exponential likelihood ratio test (ELRT) is proposed to reduce the risk of incorrect detection of change points. A general disease severity index is then designed based on our approach to satisfy the clinical requirement for disease diagnosis. A real clinical case from one of cardiovascular diseases, is given to validate the proposed approach, of which the result demonstrates the effectiveness with a satisfactory detection performance.