CNN-Based Handwriting Analysis for the Prediction of Autism Spectrum Disorder

Approximately 1 in 44 children worldwide has been identified as having Autism Spectrum Disorder (ASD), according to the Centers for Disease Control and Prevention (CDC). The term ‘ASD’ is used to characterize a collection of repetitive sensory-motor activities with strong hereditary foundations. Children with autism have a higher-than-average rate of motor impairments, which causes them to struggle with handwriting. Therefore, they generally perform worse on handwriting tasks compared to typically developing children of the same age. As a result, the purpose of this research is to identify autistic children by a comparison of their handwriting to that of typically developing children. Consequently, we investigated state-of-the-art methods for identifying ASD and evaluated whether or not handwriting might serve as bio-markers for ASD modeling. In this context, we presented a novel dataset comprised of the handwritten texts of children aged 7 to 10. Additionally, three pre-trained Transfer Learning frameworks: InceptionV3, VGG19, Xception were applied to achieve the best level of accuracy possible. We have evaluated the models on a number of quantitative performance evaluation metrics and demonstrated that Xception shows the best outcome with an accuracy of 98%.