Related Experiment Videos
Application of autonomous neural network systems to medical pattern classification tasks
C P Lim1, R F Harrison, R L Kennedy
1School of Industrial Technology, Universiti Sains Malaysia, Penang, Malaysia.
Artificial Intelligence in Medicine
|December 31, 1997
Summary
This study applies multiple neural network systems for medical pattern classification. The system shows promise as a clinical diagnostic tool, improving upon single classifiers in patient prognosis and survival prediction.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Medical Informatics
Background:
- Earlier work developed a hybrid neural network for on-line learning and probability estimation, achieving optimal Bayes classification rates.
- Multiple classifier systems are proposed to enhance single classifier performance in pattern classification.
Purpose of the Study:
- To apply autonomously learning multiple neural network systems to medical pattern classification tasks.
- To assess the system's applicability in two medical domains: coronary care unit patient prognosis and trauma patient survival prediction.
Main Methods:
- Implementation of three decision combination algorithms to create a multiple neural network classifier system.
- Assessment using patient records from two distinct medical domains.
- Comparison of results with logistic regression models.
Main Results:
- The multiple neural network system was evaluated for its effectiveness in medical pattern classification.
- Performance was assessed in predicting coronary care unit patient prognosis and trauma patient survival.
- Results were benchmarked against traditional logistic regression models.
Conclusions:
- The developed multiple neural network system demonstrates potential as a valuable clinical diagnostic tool.
- The study highlights the advantages of multiple classifier systems over single classifiers in medical applications.
- Further discussion on the implications for clinical practice is provided.