RESPIRATION EXTRACTION USING PRINCIPAL COMPONENT ANALYSIS

Breathing Rate (BR), measured in cycles per minute (cpm) is an important physiological parameter which provides valuable diagnostic and prognostic information which is benefited in predicting lower respiratory tract infections, indication of severity of pneumonia and associated with mortality in intensive care unit (ICU) patients and is also useful in treatment of many common disease such as cardiac disorder, asthma, pulmonary diseases and sleep apnea. Current routine practices for obtaining BR measurement outside critical care involves manual counting chest movement using spirometers and respiratory effort belt these practices is time consuming, inaccurate with motion artefacts. For this reason it is necessary to use accurate automated method to measure BR in ambulatory patients. Many algorithms have been made in estimating BR from ECG signals but have not yet widely adopted into clinical practices. ECG Derived Respiration (EDR) is an indirect method and attractive procedure to extract the respiration from ECG using principal component analysis (PCA), which is the aim of this thesis. In this thesis, 33 subjects clinical data were used with the mean length of the signal were 118.18±4.15 mins, 1-minute windowing procedure was used for all the signals. The correlation coefficient (ρ) between the extracted EDR and the direct respiration signals is computed. In relation to the correlation coefficient, the window is classified as incorrectly estimated (|ρ|<0.5) and correctly estimated (|ρ|>0.5). The number of incorrectly estimated windows is 29±14 [24±12%] for each subject, associated with a distribution of correlation coefficients equal to 0.37±0.03. On the contrary, the number of correctly estimated windows is 89±15 [76±12%] for each subject associated with a distribution of correlation coefficients equal to 0.74±0.02. In the conclusion, Our PCA-based procedure is a promising method for EDR extraction.