The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9718

Title:Feature-Detection in Brain-Injury Monitoring Data
Date:Sep 1997
Abstract:Head injury patients are routinely monitored electronically, and it is important to react quickly to 'insults' - that is, parameters (such as intracranial oxygen pressure) moving too far away from normal for more than a brief period. Such insults are known to be animportant factor in affecting the mortality and morbidity of head injury patients. for years, many doctors and researchers try to develop some prevention strategy for these insults. Lots of clinical data have been collecte from past cases and analysis on these data are being done in different approaches. The potential of solving complicated problems with evolutionary means has been proved on many areas. This study tries to use Genetic Programming as a tool to search for a possible pattern of feature(s) in the database of head injury patients to predict the secondary insult grade that may happen. The approach is compared to another machine learning system, C4.5. The performance of the two systems could then be compared. Unfortunately, the solution found is not yet too encouraging. The accuracy of the prediction tree could only reach up to a bit more than 50 percent of the test cases. A similar result is obtained from C4.5. Nvertheless, the tests included in this study are only a small part of the possible tests that could be used to tackle the problem. In the future, the same problem could possibly be solved in many other options that have not yet been tried out in this study. This report could be used as a reference for the future work in the same subject.

[Search These Pages] [DAI Home Page] [Comment]