A diagnosis model of dementia via machine learning

Area of Science:

• Gerontology and Neurology
• Artificial Intelligence in Healthcare

Background:

• Aging populations present significant challenges, increasing the prevalence and societal impact of dementia.
• Current dementia detection methods rely on complex, time-consuming tests and lengthy questionnaires.
• There is a critical need for efficient and cost-effective dementia screening tools.

Purpose of the Study:

• To develop a machine learning-based diagnostic model for dementia using questionnaire data.
• To employ feature selection techniques to identify and reduce the number of essential questions in dementia assessment.
• To decrease the medical and time costs associated with dementia diagnosis.

Main Methods:

• Utilized Clinical Dementia Rating (CDR) data for model development and feature selection.
• Applied various machine learning algorithms for diagnostic modeling.
• Employed feature selection methods, including Principal Component Analysis (PCA), to identify key diagnostic indicators.
• Used a confusion matrix to evaluate model performance.

Main Results:

• The bagging method demonstrated superior performance in the diagnostic model, achieving an accuracy rate of 80% compared to the true diagnosis rate.
• Principal Component Analysis (PCA) proved to be the most effective feature selection method among those tested.
• The study successfully identified key questions, enabling a reduction in questionnaire length.

Conclusions:

• Machine learning, particularly the bagging method, offers a promising approach for accurate and efficient dementia diagnosis.
• Feature selection techniques like PCA can significantly streamline dementia questionnaires, reducing patient burden and healthcare costs.
• This approach can aid clinicians in faster and more cost-effective dementia diagnosis, especially in aging populations.