DxDetekt: A Dyslexia Detection Method from Handwriting using Ensemble Method

Millions of people throughout the world suffer from dyslexia, which is a learning impairment. Early detection of dyslexia is crucial for effective intervention, but traditional dyslexia screening methods such as psychometric tests and DSM - V criteria, have limitations in detecting the disorder in its early stages when symptoms may not be severe enough to be detected and also they are time-consuming and require specialized expertise. This can lead to problems that negatively affect academic and social functioning. In this context, the development of novel techniques for dyslexia identification is critical for early intervention and effective support. DxDetekt aims to explore the potential of deep learning techniques for dyslexia detection using handwriting analysis. Specifically, DxDetekt investigates the effectiveness of an ensemble of Convolutional Neural Network (CNN) models and a CNN - Long Short - Term Memory (CNN - LSTM) model for dyslexia identification, which combines the predictions of multiple models to improve accuracy and reliability. The system inputs handwriting image samples and produces a dyslexia score as output. The CNN model extracts image features and is trained to classify the input sample as dyslexic or non-dyslexic. The CNNLSTM model processes the input image sequence and captures the temporal dependencies in the handwriting sequence. The output of both models is combined using an ensemble approach to improve the accuracy of dyslexia detection. DxDetekt was implemented and evaluated using a dataset of handwriting samples from dyslexic and non–dyslexic individuals. The proposed approach has the potential to overcome the limitations of traditional diagnostic methods and provide a more accurate and effective tool for identifying dyslexia in its early stages