A novel computer vision approach to kinematic analysis of handwriting with implications for assessing neurodegenerative diseases. (31st December 2021)

• Record Type: Journal Article
• Authors: Nachum, Ron; Jackson, Kyle; Duric, Zoran; Gerber, Naomi Lynn
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-31
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.049601
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Handwriting kinematic analysis with digitizing tablets has been established as a biomarker for use in discriminative analysis between patients with Parkinson's disease (PD), Alzheimer's disease (AD), and healthy controls (HC). However, vision‐based systems offer a rapid diagnostic assessment and screening solution for neurodegenerative diseases (ND) in low‐income areas and resource‐poor health systems, while also providing further data about movements to increase assessment accuracy. Method: 146 handwriting videos were captured from a single subject to demonstrate feasibility of extracting kinematic information from videos. Measured tasks included Archimedean spiral drawing, and tracing of l's and e's. To enable statistical comparison, video data was captured by a smartphone camera and simultaneously quantified by a digitizing tablet on which the writing template was overlaid. The developed novel system for fine motor movement analysis using computer vision leveraged recurrent region of interest feature matching to quantify kinematic information from video frames. To assess discriminative value of kinematic data collected with computer vision, the PaHaW dataset, consisting of 38 HCs and 37 PD patients, was down‐sampled to 60 Hz and filtered to simulate vision‐based data. This preprocessed data was then used to derive features describing movement fluidity, which were then tested for statistical significance with t‐tests to select final features for classification. An ensemble classifier consisting of a neural network, support vector machine, and random forest was trained on the final feature set using 10‐fold cross validation to prevent over‐fitting and provide patient diagnostic assessments. Result: Using the collected dataset of synchronized videos and digitizing tablet data, positional accuracy within 0.5 mm (+/‐ 0.088 mm 95% confidence interval) was achieved in fine motor movement quantification (n=146). Quantified speed and acceleration had mean absolute errors of less 1.1% (+/‐ 0.169 mm/s, +/‐ 0.177 mm/s^2 95% confidence intervals respectively) (n=146). Classification of PD patients was demonstrated with 74% accuracy (n=75), pending further data collection to confirm classification structure's literature‐supported use for AD and mild cognitive impairment diagnostic assessment. Conclusion: Computer vision‐based kinematic analysis of handwriting with standard cameras offers an accurate and accessible solution for neurodegenerative disease diagnostic assessment, with potential to improve diagnostic accuracy through additional features and movement data.