A new algorithm for kinematic analysis of handwriting data; towards a reliable handwriting-based tool for early detection of alzheimer's disease

•An efficient measure of handwriting data is presented for early detection of AD.•Singular value decomposition and sparse coding methods are successfully developed.•The effect of the proposed method is evaluated on the variety of time profiles.•The feasibilities of single task compared with dual-task conditions are explored. Early detection of Alzheimer's disease (AD) has attracted the attention of scientific and clinical community because of its application in control, early care, and treatment. The development of a cost-effective but reliable method is a challenge in this field. To address this challenge, in this study an effort was made to represent an efficient algorithm based on the analysis of handwriting data. Detection of premonitory symptoms using the handwriting data could be more difficult due to individual differences, effects of different sources of variability and noises. For this purpose, a noise-robustness paradigm was adopted that was independent of small variations. It was based on the singular value decomposition technique and sparse non-negative least-square classifier. To find out the best results, the effects of single and dual-task conditions as well as several handwriting time series such as horizontal, vertical and absolute velocity, acceleration, pressure, and trajectory curvature were studied. The discriminant capability of the proposed method was studied in 13 subjects with mild cognitive impairment, 15 with AD, and 15 healthy participants. They performed four writing tasks under single and dual task conditions. The new feature extracted from vertical acceleration yielded high average accuracy rates of 100% in classification between healthy controls and subjects with MCI. The average accuracy rate of 93.5% was also obtained in discriminating between healthy controls and AD patients. More investigations confirmed that using the proposed features under dual-task condition could enhance the detection rate. Achieving high performance using a relatively simple and cost-effective method demonstrated that it could potentially be used in clinical devices.