Enhancing Dementia Classification for Diverse Demographic Groups: Using Vision Transformer-Based Continuous Scoring...

Area of Science:

• Gerontology
• Artificial Intelligence in Medicine
• Neuroscience

Background:

• Alzheimer's disease and related dementias (ADRD) significantly affect older adults' well-being.
• The clock-drawing test (CDT) is a common dementia screening tool, but manual coding is laborious and error-prone.
• Current CDT scoring methods are typically ordinal, limiting precision in large-scale studies.

Purpose of the Study:

• To develop a continuous clock-drawing test (CDT) score using a deep learning neural network (DLNN).
• To evaluate the DLNN-generated continuous CDT score's effectiveness in classifying dementia in older adults.
• To compare the performance of continuous CDT scores against traditional ordinal scores.

Main Methods:

• Deep learning models were trained on CDT images from the National Health and Aging Trends Study (NHATS).
• Both ordinal and continuous CDT scores were generated.
• Area Under the Receiver Operating Characteristic Curve (AUC) was computed using a modified NHATS dementia classification algorithm as a benchmark.

Main Results:

• Continuous CDT scores yielded more granular thresholds for dementia classification compared to ordinal scores.
• Demographic-specific thresholds were identified, with lower thresholds noted for Black individuals, those with less education, and individuals aged 90+.
• Continuous scores offered a more balanced sensitivity and specificity profile than ordinal scores.

Conclusions:

• DLNN-generated continuous CDT scores show promise for enhancing dementia classification accuracy.
• The identification of demographic-specific thresholds suggests a more inclusive and adaptive approach to dementia screening.
• This methodology could inform improved guidelines for utilizing the CDT in clinical practice and research settings.