Abstract
In this thesis, a detection and quantification system of congestive heart failure (CHF) based on heart rate variability was proposed. Since the majority of studies focused on 24-hour electrocardiogram (ECG) data analysis for risk assessment, we offered a faster manner to achieved this goal. For the long-term analysis, we accessed 4-hour data; for the short-term analysis, we captured 5-minute data. All the ECG data was acquired from the 24-hour recording of the clinical trial in National Taiwan University Hospital. The proposed detection and quantification system was composed of four portions: data processing, feature extraction, feature selection, and classification. First, we retrieved 4-hour and 5-minute daytime RR interval time series for the subjects and checked the position and beat type of every R peak. To extract more physiological traits from limited amount of ECG data, not only the conventional linear methods but also the non-linear measurements, such as the multiscale entropy, detrended fluctuation analysis, and multi dynamic trend analysis, were applied in our research. To further dig out the relation between the features and congestive heart failure, we employed the statistical method to verify whether there existed a significant difference in these features between CHF and Control, also mild CHF and severe CHF. After excluding the features that were not effective enough, we chose the sequential forward selection (SFS) to search for the best feature subset for the short- and long-term analysis, respectively. Afterwards, the support vector machine (SVM) was applied to search for the best support hyperplanes in order to organize the subjects into three categories: no risk, mild risk, and severe risk. The comparisons of short- and long-term analysis are also discussed in this thesis. The recognition accuracy of risk assessment is up to 91.67% for long-term analysis, and for short-term analysis achieved 96.67%.