English abstract
Abstract
One of the most promising methods to detect and/or predict chronic heart dis eases is the long-term remote monitoring of the electrocardiogram (ECG) signal using wireless body area network (WBAN) technology. WBAN allows mobility and flexibility for both the patients and medical staff, along with improving the quality of healthcare services. It is also expected to reduce the hospital stays, pressure on emergency rooms, burden on the hospital's staff as well as decrease number of deaths due to delays in clinical interventions. Ischaemic heart disease or coronary heart disease (CHD) is the leading cause of mortality and disability around the world. There are many research works on developing wearable telecardiology systems with automatic detection algorithms to detect the cardiac disease at earlier stages. Those systems involve data trans mission and features extraction from the ECG signal. Data transmission requires compression of the raw ECG data to reduce the band-width, power consumption and storage memory, while detection of some cardiac diseases requires a fast and robust feature detection algorithm. This thesis seeks to contribute to the field of wearable healthcare technology by proposing two methods which are: a new ECG compression technique and an R-peak detection method. The proposed compression technique is designed for time-critical applications, which require small size of payload packets to guarantee fast transmission. The method employs discrete wavelet transform (DWT) for its excellent compression capability, bit-field preserving (BFP) method to reduce the error in the reconstructed data and running-length encoding (RLE) scheme to reduce the redundancy in the data. The simulation results revealed that the new "DWT-BFP-RLE compression scheme" has better performance than similar existing methods in terms of compression ratio (CR), percentage root mean square error (PRD) and quality score (QS). Moreover, to solve the issue of small payload packets, the DWT-BFP-RLE compression scheme was modified to dynamically control the size of the compressed segments that fill the payload, whilst the quality of the diagnostic information is preserved. The simulation
results showed that the modified compression scheme was able to increase the amount of packet reduction (PR) to 85.39%. The second part of this research introduces an automated detection technique to detect the R-peaks of the ECG signal. By adopting ECG segmentation, filters and adaptive thresholding, the new technique is designed to be fast and with high detection accuracy. It attempts to strike a balance between sensitivity (Se), positive predictivity (+P), and processing time. The simulation results showed that the proposed algorithm achieved a Se of 99.63% and a +P of 99.50%. The processing time used to detect the R-peaks in 30 min long records was close to the best obtained in other studies. The aim of the third part of this research was to evaluate the effect of compression on the quality of the recovered signal. This was done by evaluating the classifica tion results before and after compression. The classification was conducted using short-term heart rate variability (HRV), which is generated from the R-peaks. The short-term HRVs were able to differentiate the cardiac arrhythmias from the normal ECG rhythms using a simple binary classifier. The best classification threshold was obtained using LF/HF ratio. It achieved a specificity of 91.67%, sensitivity of 100% and accuracy of 94.87% in case of the original uncompressed ECG segments, while it achieved a specificity of 87.50%, sensitivity of 93.33% and accuracy of 89.74% from the recovered compressed segments. The results showed that the proposed compression scheme does not significantly affect the overall classification results.
