Artificial Intelligence for Alzheimer’s disease screening via handwriting images acquired by smartphones

Alzheimer’s disease is a common neurodegenerative disorder that progressively impairs cognitive and motor functions. Early diagnosis can slow the progression and improve the quality of life. However, many diagnostic tools are invasive or costly, making them unsuitable for large-scale screening. Although Artificial Intelligence applied to medical imaging can aid diagnosis, it often relies on specialised hardware, controlled environments, and expert operators, limiting scalability. This study presents a smartphone-acquired handwriting screening system evaluated on data acquired in everyday settings. We design an end-to-end validation of a deployable, task-specific pipeline for unconstrained smartphone image captures. This pipeline integrates region-of-interest extraction that is robust to uneven lighting and paper transparency, convolutional feature extraction, multiple per-task classifiers, and late-fusion ensemble strategies to improve robustness across handwriting tasks. The engineering application is a practical screening tool for Alzheimer’s disease that enables remote, low-cost data collection via standard smartphones, without specialised equipment or professional supervision. Evaluation on a dataset of 174 participants (89 patients and 85 healthy controls) acquired with a 25-task protocol shows that smartphone-acquired images yield task-wise accuracies of up to 76%, comparable to those from high-resolution scans, while ensemble methods achieve accuracy exceeding 84%, with precision of 90%. These results indicate a scalable and practical route to real-world screening beyond clinical environments.