A Task Selection Method for Alzheimer's Disease Patient Prediction to Reduce Patients’ Burden

Open Access
Article
Conference Proceedings
Authors: Soma Furukawa, Masaomi Kimura
Abstract

Alzheimer’s disease (AD) is one of the leading causes of dementia, and its diagnosis can impose a burden on patients due to examinations requiring specialized equipment and sustained effort. Handwriting-based assessment has attracted attention as a simplified diagnostic approach; however, such assessments may still require older adults suspected of having AD to perform multiple tasks. This study proposes a task selection method to reduce the number of handwriting tasks while maintaining classification performance. For each task-feature combination, measurements from 165 participants were compressed into a one-dimensional latent representation using an Autoencoder (AE). The resulting representations were organized into a task-by-feature matrix, and Principal Component Analysis (PCA) was applied to characterize the tasks. Tasks considered to contribute little to discrimination were identified based on their principal component scores and removed. Classification performance was evaluated using Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron (MLP). The results showed that accuracy decreased for SVM and MLP as the number of tasks was reduced, whereas RF maintained relatively stable accuracy even with fewer tasks. These results suggest that the proposed method can identify tasks that contribute little to discrimination while reducing the number of tasks required for AD prediction. However, the reduction in examination burden was inferred from the reduction in the number of tasks and was not directly evaluated in terms of task difficulty. Future work should therefore develop and validate objective indices of task difficulty to further evaluate the effectiveness of the proposed task reduction method.

Keywords: Alzheimer's Disease, Handwriting Tasks, Task Selection, Deep Learning

DOI: 10.54941/ahfe1008439

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Utilization of Multi-Stakeholder Visualization in Medical Management Strategy Development: A Case Study of the BSP Cardiovascular Detection ProjectToward Robust Patient-Level Parkinson’s Disease Classification: Evaluating Stability Across Participant Compositions Using Wearable-Sensor and Questionnaire Information →
View all articles in Human Factors in Design, Engineering, and Computing →