Signal Modeling and Detection in Nephrologic and Cardiac Applications

Abstract: This doctoral thesis is comprised of five parts in the field of biomedical signal processing with focus on methods for use in hemodialysis as well as in cardiac applications. The problem of predicting hypotension is the main concern in the parts were data from patients undergoing hemodialysis are used. In Part I, a newly developed method for heart rate variability (HRV) analysis in the presence of ectopic beats, based on the recently published heart timing (HT) signal, is presented. The derived HRV method deals efficiently with the ectopic beats and is shown to have better performance than other existing methods as well as to be computationally very efficient. In Part II, the method from Part I is used to analyze the problem of hypotension on a database acquired from patients during hemodialysis. In addition, Part II also investigates other aspects from the ECG signal, namely, heart rate turbulence (HRT) and ectopic beat count (EBC). A method for early detection of hypotension, involving HRV and EBC analysis, is introduced, found to detect the cases of acute dialysis induced hypotension. It is suggested that the LF/HF ratio of the HRV spectrum and HRT are useful quantities for classifying patients as being either resistant or prone to hypotension. The integral pulse frequency modulation (IPFM) model is extended to account for the presence of ectopic beats and HRT in Part III. Based on this model, a new test statistic to detect and characterize HRT is presented. Three simulations were performed for the purpose of studying the influence of signal-to-noise ratio (SNR), QRS jitter, and ECG sampling rate on detector performance. The results show that the test statistic performs better in all simulations than do the commonly used parameters turbulence onset (TO) and turbulence slope (TS). In Part IV, the detector structure presented in Part III is further developed. A new detector, obtained from introducing a priori information to the detector structure, is presented. The a priori information consists of the average HRT shape and magnitude reflected in a weight vector. The results showed that the performance of the new detector outperformed that of the previously presented test statistic and TS on both simulations and real ECG data. Part V introduces a new method for prediction of intradialytic hypotension based on pulse oximetry. The method is based on a measure denoted relative magnitude of capillary pulse (RMCP), which reflects capillary vasoconstriction and cardiac output. The proposed method is able to predict all the cases in this study with acute intradialytic hypotension without producing any false alarms. In general, the prediction occurs early in time, allowing clinical staff to take actions to prevent the onset of hypotension or to alleviate symptoms.