Scholarlink

This paper introduces a novel approach for early dyslexia detection through automated handwriting analysis, incorporating a hybrid convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architecture, enhanced with transfer learning. Our method analyzes children's handwritten text images in English, collected through carefully designed tests comprising word rewriting, sentence reconstruction, and paragraph composition. The custom CNN component, leveraging transfer learning, excels in feature extraction from handwriting images, while the BiLSTM layer captures sequential dependencies in writing patterns. The proposed model achieves 97% accuracy in dyslexia detection, outperforming existing methods. Key innovations included in current research are a custom CNN-BiLSTM architecture optimized for handwriting analysis, transfer learning to enhance feature extraction. This paper addresses gaps in current research, including improved model explainability, integration of multiple AI techniques, and focus on real-world applicability in educational settings. This research advances AI-assisted learning disability detection and provides a practical, non-invasive tool for early dyslexia screening. By enabling earlier and more accurate detection, this approach has the potential to improve educational outcomes for children at risk of dyslexia significantly. Future work will focus on expanding the dataset, exploring XAI techniques, and conducting longitudinal studies to assess long-term impact.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MedhaNetDys: A Novel Hybrid CNN-BiLSTM Architecture with Transfer Learning for Dyslexia Detection Through Handwriting Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shailesh Prabhakar Patil,
  • Ravindra Sadashivrao Apare,
  • Ravindra Honaji Borhade,
  • Parikshit Narendra Mahalle

摘要

This paper introduces a novel approach for early dyslexia detection through automated handwriting analysis, incorporating a hybrid convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architecture, enhanced with transfer learning. Our method analyzes children's handwritten text images in English, collected through carefully designed tests comprising word rewriting, sentence reconstruction, and paragraph composition. The custom CNN component, leveraging transfer learning, excels in feature extraction from handwriting images, while the BiLSTM layer captures sequential dependencies in writing patterns. The proposed model achieves 97% accuracy in dyslexia detection, outperforming existing methods. Key innovations included in current research are a custom CNN-BiLSTM architecture optimized for handwriting analysis, transfer learning to enhance feature extraction. This paper addresses gaps in current research, including improved model explainability, integration of multiple AI techniques, and focus on real-world applicability in educational settings. This research advances AI-assisted learning disability detection and provides a practical, non-invasive tool for early dyslexia screening. By enabling earlier and more accurate detection, this approach has the potential to improve educational outcomes for children at risk of dyslexia significantly. Future work will focus on expanding the dataset, exploring XAI techniques, and conducting longitudinal studies to assess long-term impact.