| Among all types of noise occur in the Holter system, motion artifact is the most common type and particularly hard to handle. Currently, motion artifact has already become one of the major factors affecting the interpretation accuracy and diagnosis efficiency of the Dynamic Electrocardiogram (DCG). To handle the motion artifacts, previous computational efforts have largely relied on motion artifacts removal, but significantly paid seldom attention to motion artifacts identification up to now. After analyzing the technical limitations of motion artifact removal and the deficiencies of the existing identification techniques, this thesis has done in-depth researches on the key technologies and algorithms for automatic identification of motion artifact in the Holter system.The main contents and innovations of this thesis mainly include:1. An acceleration signal acquistion circuit was designed and integrated into the Holter recorders, and a variety of exercise tests were designed for different healthy subjects in order to investigate the application of acceleration signal to motion artifact identification. After processing the collcetced triaxial acceleration signals by binomial fitting, high-pass filter and RMS summation to obtain the motion reference signal (REF), a spectral mapping method was applied to REF to help clinicians quickly locate the examiner’s movement periods.2. A novel method for automatic identification of motion artifact segments was proposed. This thesis first analyzes the QRS complex shape variation law of the beats located in the motion artifact segments, and then calculates the consequent beats shape mutation curve (cSMC) based on the results of beat template clustering. In the cSMC, fuzzy-logic criterion and local integration were employed to find out the start and end position of the motion artifact segments. Using the optimal threshold analysis in the MIT-BIH Noise Stress Test Database, the accuracy of the proposed method is: sensitivity=97.95%. NFR=0.91%and VFR=3.33%. Moreover, the total cost time of our novel method applied to24hours recordings is less than4seconds using the Intel2.5G/2GB PC machine.3. A combined Higher-order statistics method for motion artifacts identification was proposed. By analyzing the kurtosis variation characteristics of the four common clinical morphological types of heart beats contaminated by the motion artifacts with different signal-to-noise ratio levels, a multi-parameter hierarchical strategy combined kurtosis and beat features was porposed to classify motion artifacts and clean beats. Combine the strategy with the previous motion artifact segment identification method, the NFR and VFR have been effectively reduced while the identification sensitivity keeps alomost unchanged:sensitivity=96.66%, NFR=0.19%and VFR=1.74%.4. Using the MIT-BIH Noise Stress Test Database and the clinical data gathering from the ECG department of the hospital, the identification performance of the whole method was evaluated, and the accuracy was characterized as a function of signal-to-noise ratio comply with the ANSI/AAMI EC38:1998recommended practice. Compared with the existing identification methods, the results show that the proposed method achieves higher accuracy, and is highly efficient that can be conveniently used in clinical applications.The proposed method for automatic identification of motion artifacts has been integrated into the clinical Holter system and has been used for many DCG diagnosis applications in need of discrimination between clean and motion artifact contaminated segments, such as atrial fibrillation detection, classification of cardiac arrhythmias and so on. |
PDF Full Text Request