ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Žiga Pirnar (2002) . Prešeren awards for students.

Full text not available from this repository.


The latest developments in the filed of computer ECG pattern analysis have made automated detection of myocardial ischemia, which is the main cause of heart stroke, and in turn preventive treatment of related heart disease possible. Automated software for assesing the extent of myocardial ischemiausing lon-term ECG recordings comprises a preprocessing stage, construction of uniform sequences of feature vectors (i.e. resampled time-series of diagnostics and morphology feature vectors), detection of significant ST change episodes and exclusion of non-ischemic intervals due to changes in heart rate, conduction changes, slow drift of ST segment level or sudden shifts of the mean heart electrical axis. The latter is the cause of most problems connected to differentiation from real ischemic episodes. Development of efficient algorithms for detecting these axis shifts was not possible due to absence of long-term ECG recordings that include a sufficient number of non-ischemic events to make their characterization possible, and enable objective performance evaluation of developed algorithms. Inaccurate exclusion of non-ischemic intervals leads to poor performance of automated ischemia detectors and therefore false detection of heart disease thus limiting their practical application. In this study we describe the process of characterization of diagnostic and morphology feature vector patterns around non-ischemic events using the new international Long-term ST Database (otherwiseknown as LTST DB), a collection of eighty six 24-hour ECG recordings including a large number of significant ischemic and non-ischemic ST change episodes. In the introduction, we present the main characteristics of ambulatory ECGs, indicators of ischemic disease in ECGs, a detailed description of axis shifts, structure of the LTST database and describe the objectives of our work. We also describe the discrete Karhunen-Loeve transform and derive time-series of important feature vectors in both the Karhunen-Loeve space, and in time domain. We have developed an interactive time-series visualization tool for the Microsoft Windows operating system with a user interface that follows the principles of direct manipulation and is relatively easy to use. It makes use of time-series cache and several resampling tehniques to enable faster viewing and it also allows the user to customize the user interface, display annotations in WFDB files, visualize ECG signals and average heart beats, manipulate axis shift annotations and export the current workspace to image files. Details of its use and design are described in the second chapter of this document. We later expanded LTST DB recordswith irrelugar and noisy beat density functions, projections of the mean heart electrical vectors to lead axis, trajectories of morphology feature vectors and a derivate of the Mahalanobis distance function in the Karhunen-Loeve space. We have described thses feature vectors in the third chapter. The LTST Db has not been used previously to characterize changes in described feature vectors due to axis shifts and this fact led us to to the developmentof software for extracting and averaging intervals of selected time-series centered on thses events. We concluded that axis shifts are accompanied by short intervals of rapid significant in noisy beat density and heart rate, and by synchronous step changes in most original and derived ST and QRS feature vector time-series. We have also characterized changes due to true ischemic episodes and noticed a slow increase in heart rate just seconds pripor to onset of ischemic episodes, increased heart rate throughout the episode, increased density of irregular heart beats and triamgle shaped changes in the ST level function. The following findings are presented in the fourth chapter of this document. In the last chapter we present the possibilties of future use of our work for development of efficient axis-shift detection algorithms, describe the process of development of such algorithms and conclude our work by presenting pointers for the future work.

Item Type: Thesis (Prešeren awards for students)
Number of Pages: 86
Language of Content: Slovenian
Mentor / Comentors:
Name and SurnameIDFunction
prof. dr. Franc Jager235Mentor
Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=51012&select=(ID=3413076)
Institution: University of Ljubljana
Department: Faculty of Computer and Information Science
Item ID: 3765
Date Deposited: 02 Feb 2017 16:28
Last Modified: 13 Feb 2017 10:03
URI: http://eprints.fri.uni-lj.si/id/eprint/3765

Actions (login required)

View Item