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

Miha Amon (2003) . Prešeren awards for students.

Full text not available from this repository.


For development and performance evaluation of algorithms and analyzers for automtatic detection of transient ischaemic ST segment episodes of electrocardiogram (ECG), high quality reference ECG signal records with transient ischaemic episodes and other non-ischaemic ST segment events manually annotated by expert cardiologists are needed. Special protocols are defined for annotating and automatic detection of transient ST segment episodes. Criteria of these protocols form the basis for deciding whether a particular event is a transient ST segment episode. In the field of analysis of transient ST segment level changes, two standardized databases of such records are in use at present: the European Society of Cardiology ST-T Database (ESC DB) and the Long-Tem ST Database (LTST DB). The expert annotations of both databases are quite comparable, however certain significant differences do appear. These differences can be categorized according to the following possible causes: differences in quality of source signals originating from the process of sampling during database creation, differences in annotation protocols, and consequently differences in actual annotations of the events. If these differences are actually present, we can conclude that results of evaluations of ECG analyzers using different databases will not be quite the same. This fact could lead to serious consequences. Main objectives of this work are therefore to categorize all possible kinds of differences between both databases, to evaluate these differences, and to find the answer to the key question on how comparable can the resuts of evaluations of analyzers obtained on these two databases be. For this purpose, it was possible to exploit a helpful circumstance that ten records of both databases come from the same source signals. However, direct comparison of these records is not possible since there are deviations in sampling frequency among them. For this reason, we developed an effective method for temporal synchronization of signals. Moreover, records of the LTST DB are much longer from those of the ESC DB. It was therefore necessary to search within LTST DB records for predse segments that match the records of the ESC DB. As a result of these procedures on the LTST DB, a new database of records ready for comparison with those of the ESC DB was derived. We named it the C database (C DB) and it is now available for further studies on comparability between the databases. For comparison of the databases and their expert annotations, we developed an interactive graphic user interface using Kylix/LINUX graphic environment. The interface enables use of a wide set of analytic functions for analysis and comparison of ECG signals which we developed in this study, eg. synchronization of signals, temporal navigation, different displays and views of signals and time series of ST segment level features, together with display of expert annotations and episodes for two records simultaneously. New metric of reference annotations matching were developed for accurate comparison of the expert annotations of the records from both databases and for statistical evaluation of their possible discrepancies. These matching metrics were implemented in the graphic user interface developed as well. Results of the analyses showed that the expert annotations of both databases are comparable, but moderate discrepancies can be expected during assessing performance of the analyzers.

Item Type: Thesis (Prešeren awards for students)
Number of Pages: 94
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=4033620)
Institution: University of Ljubljana
Department: Faculty of Computer and Information Science
Item ID: 3723
Date Deposited: 05 Jan 2017 16:56
Last Modified: 13 Feb 2017 09:41
URI: http://eprints.fri.uni-lj.si/id/eprint/3723

Actions (login required)

View Item