[Paper Review] The Dynamics of Handwriting Improves the Automated Diagnosis of Dysgraphia

Konard Zolna, Thibault Asselborn+5 more|arXiv (Cornell University)|Jun 12, 2019

Writing and Handwriting Education48 references12 citations

TL;DR

This study proposes a tablet-based, RNN-driven digital diagnostic system that analyzes the dynamic features of handwriting—such as stroke order, speed, and pressure—to automatically detect dysgraphia in children. The model achieves 90% accuracy in identifying dysgraphic children, significantly outperforming static image-based methods, and offers a fast, objective, and scalable alternative to traditional BHK testing.

ABSTRACT

Handwriting disorder (termed dysgraphia) is a far from a singular problem as nearly 8.6% of the population in France is considered dysgraphic. Moreover, research highlights the fundamental importance to detect and remediate these handwriting difficulties as soon as possible as they may affect a child's entire life, undermining performance and self-confidence in a wide variety of school activities. At the moment, the detection of handwriting difficulties is performed through a standard test called BHK. This detection, performed by therapists, is laborious because of its high cost and subjectivity. We present a digital approach to identify and characterize handwriting difficulties via a Recurrent Neural Network model (RNN). The child under investigation is asked to write on a graphics tablet all the letters of the alphabet as well as the ten digits. Once complete, the RNN delivers a diagnosis in a few milliseconds and demonstrates remarkable efficiency as it correctly identifies more than 90% of children diagnosed as dysgraphic using the BHK test. The main advantage of our tablet-based system is that it captures the dynamic features of writing -- something a human expert, such as a teacher, is unable to do. We show that incorporating the dynamic information available by the use of tablet is highly beneficial to our digital test to discriminate between typically-developing and dysgraphic children.

Motivation & Objective

• To develop an objective, automated method for diagnosing dysgraphia that reduces reliance on subjective human evaluation.
• To leverage the dynamic kinematic features of handwriting—captured via graphics tablets—to improve diagnostic accuracy beyond static visual assessment.
• To create a scalable, low-cost digital solution that enables early detection of handwriting difficulties in children.
• To compare the performance of dynamic (RNN) versus static (CNN) models in distinguishing typically developing children from those with dysgraphia.
• To explore the potential of using such a system in educational applications, such as personalized handwriting tutoring via robot-assisted learning.

Proposed method

• Children were asked to write all letters of the alphabet and digits 0–9 on a Wacom graphics tablet, which recorded x, y, time, pressure, and pen inclination.
• The system used a Recurrent Neural Network (RNN) to process sequential handwriting data, capturing temporal dynamics such as stroke velocity and trajectory.
• A Convolutional Neural Network (CNN) was used as a baseline, analyzing only the final static image of the written trace.
• The RNN model was trained to classify each child as either typically developing (TD) or dysgraphic based on the dynamic features of their handwriting.
• The system evaluated performance using standard metrics, including sensitivity and specificity, comparing RNN and CNN outcomes.
• Future integration of pressure and inclination data was considered, though current models used only x, y, and time coordinates.

Experimental results

Research questions

• RQ1 Can a dynamic, data-driven approach using handwriting kinematics improve the automated diagnosis of dysgraphia compared to static image-based methods?
• RQ2 How does the inclusion of temporal dynamics (e.g., speed, stroke order) enhance the ability to distinguish between typically developing and dysgraphic children?
• RQ3 To what extent does an RNN-based model outperform a CNN-based model in classifying dysgraphia using the same handwriting data?
• RQ4 Can a digital, tablet-based system reduce the time and subjectivity associated with traditional BHK testing for dysgraphia diagnosis?
• RQ5 What are the implications of using such a system in educational contexts, such as personalized learning or robot-assisted tutoring?

Key findings

• The RNN model achieved 90% accuracy in correctly identifying dysgraphic children, significantly outperforming the CNN model, which labeled fewer than 30% of dysgraphic cases correctly.
• The dynamic features of handwriting—captured through time-resolved x, y, and time coordinates—were essential for discrimination, as static image analysis alone was insufficient.
• The tablet-based system delivered a diagnosis in a few milliseconds, offering a fast, scalable alternative to the 10-minute manual scoring of BHK tests.
• The study demonstrated that human experts miss critical dynamic cues in handwriting, such as irregular stroke timing and unstable pressure, which the RNN model successfully captured.
• The system's performance suggests that digital diagnostics can reduce diagnostic delays and subjectivity, enabling earlier intervention.
• Future work should explore integrating pen pressure and inclination data, and assess whether results generalize across different tablet types, including low-cost tactile devices.