Deep learning based low-cost high-accuracy diagnostic framework for dementia using comprehensive neuropsychological...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Gerontology

Background:

• Conventional neuropsychological tests for dementia lack optimal diagnostic accuracy despite their data richness.
• There is a need for cost-effective and highly accurate dementia diagnosis using neuropsychological batteries.

Purpose of the Study:

• To propose a novel, cost-effective, and precise two-stage framework for dementia diagnosis.
• To leverage deep learning for enhanced diagnostic performance in dementia screening and assessment.

Main Methods:

• A two-stage classification procedure using Mini Mental Status Examination (MMSE) for screening and the KLOSCAD Neuropsychological Assessment Battery for diagnosis.
• Implementation of deep neural networks (DNNs) for classification, incorporating redundant variable evaluation and k-nearest-neighbor imputation for missing data.

Main Results:

• The proposed DNNs achieved superior accuracy compared to other classifiers in the two-stage dementia diagnosis.
• Removal of 49 redundant variables improved diagnostic performance, indicating potential for assessment simplification.
• The framework yielded an 8.06% increase in diagnostic accuracy over MMSE alone and a 64.13% cost reduction compared to KLOSCAD-N alone.

Conclusions:

• The developed framework offers improved robustness, precision, and cost-effectiveness for dementia early detection.
• This approach is applicable to general dementia early detection programs, enhancing diagnostic capabilities.