Neuropsychological test using machine learning for cognitive impairment screening.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Neuropsychological tests (NPTs) are crucial for assessing cognitive function but require specialized interpretation.
• Interpreting NPTs can be time-consuming, necessitating efficient diagnostic tools.

Purpose of the Study:

• To develop and evaluate machine learning models for detecting normal cognition (NC), cognitive impairment (CI), and dementia.
• To assess the performance of Support Vector Machine (SVM) algorithms in classifying cognitive states using NPT data.

Main Methods:

• Utilized a dataset of 14,927 subjects from the Seoul Neuropsychological Screening Battery (SNSB), including 44 NPTs, age, education, and diagnosis.
• Trained supervised machine learning models using the SVM classifier 30 times to predict cognitive states (NC vs. CI, NC vs. dementia, NC vs. CI vs. dementia).
• Evaluated model performance using accuracy, sensitivity, and specificity, with confusion matrices generated from the testing dataset.

Main Results:

• The NC vs. dementia model achieved the highest mean accuracy (97.74% ± 5.78%), sensitivity (97.99% ± 5.78%), and specificity (96.08% ± 4.33%).
• The model for distinguishing NC, CI, and dementia achieved a mean accuracy of 83.85% ± 4.33%.
• SVM demonstrated suitability for imbalanced datasets, outperforming other algorithms in prediction accuracy.

Conclusions:

• SVM-based machine learning models trained on NPT data can effectively assist neuropsychologists in classifying cognitive function.
• The developed models show significant potential for improving the efficiency and accuracy of diagnosing cognitive impairment and dementia.
• The study highlights the utility of AI in augmenting clinical decision-making for neurological disorders.