A Novel Approach for Detection of Autism Using Neural Transformer

Autism which refers to autism spectrum disorder (ASD) is recognized as a neuro-developmental disorder where the person who is autistic may exhibit conditions of diverse groups. It can also be identified by some unique patterns such as the inability of a person to switch the activity that is being carried out, little eye contact while communicating, tending to repeat certain words or phrases, and reactions that are completely different and unusual to events or sensations. The characteristic that most of the autistic people exhibit is difficulty in communicating with other people. Researches are carried out to predict autism from family morbidity involving data from three generations, detecting the severity of autistic characteristics by analyzing brain patterns, using image data to predict autism, and speech recognition to monitor how well the patient communicates, or does the patient tend to repeat any phrases. Our solution to detect autism involves the use of deep learning techniques to extract features from the image dataset of autistic people which is labeled as autistic and non-autistic, and using the extracted features to train the classifiers for better prediction. The main motive is to provide a solution which could contribute in the early detection of autism.