Handwriting Evaluation Using Deep Learning with SensoGrip.

Area of Science:

• Pediatric neurology
• Educational psychology
• Machine learning in healthcare

Background:

• Handwriting learning disabilities, such as dysgraphia, significantly hinder children's academic performance and well-being.
• Early identification of dysgraphia is crucial for timely and effective intervention.
• Previous research primarily used classical machine learning with manual feature extraction and binary classification for dysgraphia detection.

Purpose of the Study:

• To investigate the fine-grained assessment of handwriting capabilities using deep learning.
• To predict the SEMS score (0-12) for a more nuanced evaluation of dysgraphia.
• To leverage advanced technology for improved detection of handwriting difficulties.

Main Methods:

• Employed deep learning models for handwriting analysis.
• Utilized automatic feature extraction and selection, eliminating manual processes.
• Implemented a smart pen (SensoGrip) equipped with sensors to capture dynamic writing data.
• Focused on predicting a continuous SEMS score instead of binary classification.

Main Results:

• Achieved a root-mean-square error (RMSE) of less than 1 in predicting the SEMS score.
• Demonstrated the effectiveness of deep learning in fine-grading handwriting capabilities.
• Showcased the utility of sensor-equipped smart pens for realistic writing evaluation.

Conclusions:

• Deep learning with automatic feature extraction offers a more accurate and efficient approach to assessing handwriting skills.
• The use of smart pens enables evaluation in more naturalistic settings, enhancing diagnostic potential.
• This methodology facilitates earlier and more precise identification of dysgraphia, supporting targeted interventions.