Deep Learning and Procrustes Analysis for Early Dysgraphia Risk Detection with a Tablet Application.

Area of Science:

• Neurodevelopmental disorders
• Developmental psychology
• Computational neuroscience

Background:

• Classical dysgraphia diagnosis relies on evaluating mastered handwriting, leading to delayed detection.
• Current diagnostic methods are language-dependent and not always accessible.
• Delayed diagnosis of dysgraphia can negatively impact academic and daily life.

Purpose of the Study:

• To develop an early screening tool for dysgraphia.
• To identify children at risk before handwriting is fully developed.
• To prevent negative consequences associated with delayed dysgraphia diagnosis.

Main Methods:

• Utilized the Play-Draw-Write iPad application to collect handwriting-related data from children.
• Employed dimensionality reduction techniques (autoencoder, Time2Vec) for signal processing.
• Developed a deep learning meta-model (ensemble techniques, Quasi-SVM) with feature extraction (Procrustes Analysis).

Main Results:

• The dysgraphia classifier achieved 84.62% accuracy.
• The classifier demonstrated 100% precision in identifying at-risk children.
• The method enables screening more than two years earlier than current techniques.

Conclusions:

• The developed deep learning model offers a promising solution for early dysgraphia screening.
• Anticipating dysgraphia identification can mitigate adverse effects on children's development.
• This approach provides a more accessible and timely method for identifying handwriting-related neurodevelopmental disorders.