Explainable AI in Handwriting Detection for Dyslexia Using Transfer Learning

TL;DR

This work tackles dyslexia screening through handwriting analysis by introducing an explainable AI framework that uses transfer learning and transformer-based models to identify dyslexia-linked handwriting features, with Grad-CAM visualizations ensuring interpretability. The proposed MobileNetV3-based pipeline, evaluated via 5-fold cross-validation, achieves $99.65\%$ test precision and demonstrates strong performance for both Small and Large variants, along with cross-linguistic applicability. The study shows that XAI can enhance trust and adoption in educational and clinical settings by providing transparent decision-making, while jointly delivering high diagnostic accuracy. The authors also outline future directions to broaden explainability (SHAP/LIME), enable real-time deployment, and extend the approach to other learning disabilities and personalized interventions.

Abstract

This study introduces an explainable AI (XAI) framework for the detection of dyslexia through handwriting analysis, achieving an impressive test precision of 99.65%. The framework integrates transfer learning and transformer-based models, identifying handwriting features associated with dyslexia while ensuring transparency in decision-making via Grad-CAM visualizations. Its adaptability to different languages and writing systems underscores its potential for global applicability. By surpassing the classification accuracy of state-of-the-art methods, this approach demonstrates the reliability of handwriting analysis as a diagnostic tool. The findings emphasize the framework's ability to support early detection, build stakeholder trust, and enable personalized educational strategies.

Figures (5)

• Figure 1: Proposed Model Architecture.
• Figure 2: Examples of the same letter across the three different classes: (a) Normal, (b) Reversed, and (c) Corrected.
• Figure 3: Methodology Flowchart
• Figure 4: Grad-CAM visualizations for MobileNet V3 Small, showing key handwriting regions the model focused on for classification. These visualizations highlight the model’s attention to critical handwriting features in both reversed and corrected samples.
• Figure 5: Grad-CAM visualizations for MobileNet V3 Large, showcasing its focus on handwriting features critical for distinguishing dyslexic writing traits, supporting its use in clinical and educational environments.