Handwriting-Based Alzheimer’s Disease Detection Using VGG-19

Alzheimer’s disease is a progressive neurological disorder that impairs cognitive function and daily activities. Conventional diagnostic methods are often invasive, costly, and reliant on specialized clinical facilities, limiting their suitability for large-scale early screening. This study proposes a handwriting-based early detection approach using a Convolutional Neural Network built on the VGG-19 architecture. Experiments were conducted on the DARWIN-I dataset comprising 174 participants (89 Alzheimer’s patients and 85 healthy controls) across two handwriting tasks (TASK_04 and TASK_05), considering both raw and cleansed image subsets. Preprocessing strategies included standard VGG-19 preprocessing as well as enhanced techniques combining Gaussian Blur, normalization, CLAHE, and Otsu thresholding. Hyperparameter optimization was performed using Hyperband, followed by 5-fold stratified cross-validation to ensure stable performance. The best configuration was retrained and evaluated on a hold-out test set. The highest experimental accuracies reached 87% for TASK_04 and 90% for TASK_05. For deployment, the final model was selected based on stability and robustness, achieving 81% and 90% accuracy for TASK_04 and TASK_05, respectively. The selected model was deployed in a cloud-based environment and integrated into a mobile application for real-time prediction. A usability evaluation involving 21 participants showed positive responses in usefulness, ease of use, ease of learning, and satisfaction. These findings demonstrate that the proposed approach is effective and suitable for supporting practical early-stage Alzheimer’s disease screening.