Automated Dyslexia Screening Using Children's Handwriting in English Language with Convolutional Neural Network and Bidirectional Long Short-Term Memory Model
Abstract
Dyslexia, a common learning disability, often goes undetected in early childhood, leading to delayed interventions. This study proposes a novel approach for early detection of dyslexia through automated analysis of children's handwritten text images written in the English language. Using computer vision techniques and machine learning models, handwriting patterns are analyzed to identify dyslexic tendencies based on unique characteristics such as spelling errors, letter shape, spacing, and consistency. The system incorporates a handwritten text recognition (HTR) model, trained on the IAM dataset, achieving an accuracy of 95.6% and a word error rate (WER) of 0.13% in recognizing text. For dyslexia prediction, convolutional neural network (CNN) architecture is utilized with bidirectional long short-term memory (Bi-LSTM) and connectionist temporal classification (CTC) loss, with the best-performing model demonstrating significant accuracy in distinguishing between dyslexic and non-dyslexic children. The handwritten images were collected from children through three written tests, ensuring various handwriting styles and content. The dyslexia prediction model is evaluated using performance metrics such as accuracy, F1 score, and WER. The results underscore the potential of using handwriting as a predictive tool for dyslexia, showing better accuracy and efficiency than previous models. This approach can be an early, non-invasive screening tool, enabling timely intervention and personalized learning strategies for affected children.