Diagnosing Alzheimer’s disease from on-line handwriting: A novel dataset and performance benchmarking

Neurodegenerative diseases are caused by the progressive degeneration of nerve cells that affect motor skills and cognitive abilities with increasing severity. Unfortunately, there is no cure for this type of disease and their impact can only be slowed down with specific pharmacological and rehabilitative therapies. Early diagnosis, therefore, remains the primary means to delay brain damage and improve the quality of life of people affected. Neurodegenerative diseases also affect movement fine control. Consequently, the analysis of handwriting dynamics can represent an effective tool to support an early diagnosis of these diseases. While many methods have been proposed in the literature based on the use of a wide range of handwriting tasks, researchers have not yet defined a universally accepted standard experimental protocol to collect data. Furthermore, although some databases containing handwriting data have been produced, only a few of them were designed specifically for research on neurodegenerative diseases, and, in most cases, they involve a small number of participants performing a few tasks. Here, we introduce the DARWIN (Diagnosis AlzheimeR WIth haNdwriting) dataset to overcome these drawbacks, which contains handwriting samples from people affected by Alzheimer’s and a control group. The dataset includes data from 174 participants, acquired during the execution of handwriting tasks, performed according to a protocol specifically designed for the early detection of Alzheimer’s. We report the results of the experiments performed to evaluate the effectiveness of the proposed tasks and features in capturing the distinctive aspects of handwriting that support the diagnosis of Alzheimer’s disease. • We introduce the DARWIN dataset (Diagnosis AlzheimeR WIth haNdwriting). • The dataset contains handwriting data from people affected by Alzheimer’s. • The dataset is the largest publicly available in terms of number of participants and tasks. • We investigated the effectiveness of the proposed tasks and the features extracted.