ABSTRACT
Autism Spectrum Disorder (ASD) has become a growing concern globally, with increased incidence of neurodevelopmental disorders. With the onset autism, the development of brain is affected significantly, having direct repercussions on the psychomotor skills of the child/person. Existing modalities of diagnosis are qualitative and do not provide the comprehensive evaluation over a long period of time. Further, there is an increased risk of missing the detection at an early stage, having long-lasting health implications on the motor and cognitive skills. Deep learning algorithms have emerged as a potential tool for exploring possible association of the data with the symptoms that are clinically relevant for ASD diagnosis. Owing to higher accuracy in detecting the disorder, techniques such as CNN, RNN, DBN, LSTN, and AE have become very popular in assessing the level of risk in developing the disorder. However, due to non-corroboration of the morphological and behavioral specifics to the features extracted using deep learning techniques, effective diagnosis has been a challenge. This chapter delves into the reasons as to why the effective diagnosis are not possible and present author speculations that may direct future investigations toward improved accuracy using deep learning models.
