Related Experiment Videos
A neural network design for event-related potential diagnosis
F Y Wu1, J D Slater, L S Honig
1Department of Electrical & Computer Engineering, University of Miami, Coral Gables, FL 33124.
Computers in Biology and Medicine
|May 1, 1993
Summary
Artificial neural networks (ANNs) improve the classification accuracy of multiple sclerosis (MS) using electroencephalogram (EEG) evoked potentials. ANNs offer enhanced diagnostic potential compared to traditional P300 latency analysis for detecting neuropsychological impairment.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Medical Diagnostics
Background:
- Electroencephalogram (EEG) abnormalities are common in neuropsychological impairment, but normal EEG patterns vary widely.
- Cerebral evoked potentials, specifically P300 auditory evoked potentials, show promise for diagnosing dementing illnesses.
- Previous studies using P300 latency and waveform criteria achieved limited classification accuracy for multiple sclerosis (MS) detection.
Purpose of the Study:
- To evaluate the efficacy of artificial neural networks (ANNs) in classifying MS patients versus control subjects using EEG evoked potential data.
- To compare the diagnostic performance of ANNs against traditional nearest neighbor (NN) classifiers and P300 latency criteria.
- To analyze the classification strategy identified by ANNs through weight pattern analysis.
Main Methods:
- EEG data from MS patients and control subjects were analyzed using P300 auditory evoked potentials.
- Classification was performed using a nearest neighbor (NN) classifier and an artificial neural network (ANN).
- ANN classification strategies were further investigated using weight pattern analysis.
Main Results:
- The ANN classifier achieved a classification accuracy of 75% for MS patients and 87% for control subjects.
- This represents an improvement in average prediction accuracy compared to NN classifiers (65% for MS, 91% for controls) and P300 statistical analysis.
- Weight pattern analysis provided insights into the ANN's decision-making process, distinct from P300 latency criteria.
Conclusions:
- Artificial neural networks demonstrate superior performance in classifying MS patients based on EEG evoked potentials compared to conventional methods.
- ANNs offer a more accurate and potentially more insightful approach to diagnosing neuropsychological impairments indicated by EEG.
- Further research into ANN-based analysis of EEG data may enhance early and accurate diagnosis of neurological conditions.