Abstract
Ischemic heart disease is the single most frequent cause of death in the world today. Electrocardiogram (ECG) exercise testing is a common tool for diagnosing this disease. With the ECG exercise test one can compare the recording of the electrical activity in the heart during exercise and rest. A shift between the rest and exercise recording in the segment of the heartbeat called the ST segment is used both for diagnosing ischemia, and as input to cardiac computation methods.
Body surface potential mappings (BSPM) are ECG recordings at a greater number of locations on the torso, and provides better detection and localisation properties than the traditional 12-lead ECG. In BSPM and ECG, noise and drift from various sources are recorded in addition to the signal propagating from the heart. A model for this is: BSPM=signal+noise+drift. Before accurate measurements of the ST segments in a BSPM can be made, the noise and drift in the recording must be reduced while keeping the signal unchanged. In this thesis, an automatic algorithm for post processing raw BSPM data recordings was made.
The following methods have been developed, implemented and tested as part of the algorithm: First, noise reduction methods using frequency based filtering techniques was implemented and tested. Second, an algorithm for detecting the BSPM signal peaks was developed. This method was used to locate the interesting parts of each heartbeat. Third, methods for removing the baseline drift is discussed. Four methods were selected, implemented and evaluated against each other. A method using cubic spline interpolation as an approximation to the drift was deemed best and used in the automatic algorithm. Even after this initial processing, there may be noisy or corrupted signal parts present in a BSPM. Hence a framework for removing such parts of the BSPM was developed as the fourth step of the algorithm. In the fifth step, a robust method for computing the ST segment shifts at each electrode location from a processed BSPM was made. Finally, a tool for visualising these shifts was created.
The algorithm developed in this thesis was applied to BSPM recordings of real patients. Before processing, it was not possible to compute neither reliable nor correct ST shifts from these recordings. After the automatic algorithm was applied to these recordings, all the resulting BSPMs were physically realistic, and showed signs of being close to the true values. The computed ST shifts from the processed data showed promising results for diagnosing ischemia in the limited set of BSPMs available. In addition, a comparison between traditional ECG and BSPM was made.