Related Experiment Video
Updated: Aug 16, 2026
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Mining for diagnostic information in body surface potential maps: a comparison of feature selection techniques
Dewar D Finlay1, Chris D Nugent, Paul J McCullagh
1School of Computing and Mathematics, Faculty of Engineering, University of Ulster, Shore Road, Belfast, UK. d.finlay@ulster.ac.uk
Biomedical Engineering Online
|September 6, 2005
Summary
Data mining techniques can reduce the number of electrocardiographic recording sites needed for accurate detection of Myocardial Infarction (MI). Feature selection methods identified optimal lead subsets for Body Surface Potential Map (BSPM) analysis.
Area of Science:
- Cardiology
- Biomedical Engineering
- Data Science
Background:
- Body Surface Potential Mapping (BSPM) offers superior cardiac diagnostics over traditional electrocardiography.
- Clinical adoption of BSPM is hindered by the perceived complexity of its acquisition process.
- Research aims to find a balance between information richness and the number of recording sites.
Purpose of the Study:
- To apply data mining techniques for feature selection in BSPM analysis.
- To evaluate the diagnostic capability of reduced electrocardiographic lead subsets for Myocardial Infarction (MI) detection.
- To compare filter and wrapper approaches for identifying optimal recording sites.
Main Methods:
- Employed Single Variable Classifier (SVC) filter and Sequential Forward Selection (SFS) wrapper methods for feature selection.
- Evaluated subsets of 3, 6, 9, 12, 24, and 32 leads from 192-lead BSPMs.
- Assessed classification accuracy for Myocardial Infarction (MI) presence or absence in 116 subjects.
Main Results:
- The SFS wrapper approach with a 5-nearest neighbor classifier identified 24 leads for 82.8% accurate MI classification.
- The SVC filter method achieved comparable performance with 79.3% classification accuracy.
- Investigated classifier specificity of selected features and non-uniqueness of lead subsets.
Conclusions:
- Both filter and wrapper methods effectively guide the selection of recording sites for MI detection.
- Selected sites optimize disease detection but may not be ideal for body surface potential distribution estimation.
- Data mining offers a viable strategy to simplify BSPM acquisition for clinical use.
