Dysgraphia Detection in Children: Two Deep Learning Models Utilizing Handwritten Text Images

Dysgraphia is a disorder that affects children's ability to write legibly and can impact overall written expression, spelling, and composition skills. Traditionally, dysgraphia is assessed through various tests that measure working memory, cognitive ability, and orthographic processing. These assessments are time-consuming and may require significant effort and resources to administer. This study presents two novel frameworks based on Deep Learning (DL) architectures, such as Spatially Enhanced SegNet (SE-SegNet) and Self-Attention U-Net (SAU-Net), to identify children with dysgraphia using handwritten text images based on gender. The proposed frameworks were trained and tested on a collected dataset, comprising 1,853 text images, and their performance was evaluated using accuracy, recall, precision, and the Mathew Correlation Coefficient (MCC). According to performance analysis, the SAU-Net framework was superior, achieving maximum accuracy at 99%, recall at 99.1%, precision at 98.5%, and MCC at 98.3%. The proposed approach provides an efficient and accurate method for identifying dysgraphia in children, supporting early detection for effective interventions and improving children's academic progress.