Artificial intelligence innovations have taken our society by storm. AI can be used for a variety of things. It can help you craft a title for your paper. It can also give you a recipe for the perfect chocolate chip cookies. But now, researchers are exploring using AI to diagnose individuals with autism. 

My Initial Thoughts:

Before I do my research, here’s my opinion on the matter. Every individual with ASD is different. The saying, “if you meet one person with autism, you meet one person with autism” reflects this reality. The characteristics that one individual has could vary significantly from the characteristics of others. Therefore, diagnostic criteria must account for such neurological variations which can be very difficult. In my opinion, I think AI is too broad to cover the niche characteristics that an individual may experience, especially since diagnostic criteria is controversial as it is. 

Research

AI has already established itself in the realm of medicine. AI has helped diagnosis conditions by analyzing medical images and by transcribing medical documents for better communication. Dr. Dennis Wall, a professor of pediatrics at Stanford responds, “ASD diagnostics is crying out for a solution that’s faster, more equitable, and more quantitative.” AI has started to transcend into the ASD space where researchers can utilize it to develop quantitative measures that they can use to diagnose young children with ASD.

In response, researchers have begun to explore how AI can develop quantitative tools to diagnose ASD. For example, there are already existing mobile screening applications for autism including ASDTests and Autism Test. However these tools have limitations:

• ASDTests uses existing conventional screening methods without incorporating advanced machine learning algorithms.
• Autism Test has limited reliability due to limited ratings and unclear mechanisms 

Other researchers have tried to develop a screening method using AI that would predict autistic traits and diagnoses based on past cases and controls. It would constantly learn from itself as the dataset updates. The proposed system is called Autism AI and its mission would be to provide a professionally designed interface that could provide results on autism traits. The system also will develop a report that parents or caregivers can give to physicians. The majority of reviews have been positive but there was one user who claimed that the system didn’t recognize he was autistic. 

My Analysis

In my opinion, I don’t think there are enough datasets for the AI system to learn from itself. I think the concept of using AI for autism diagnoses is fascinating and eventually will be used consistently but I don’t think it can until there is enough data for the system to use. There remains many challenges that would need to be resolved before AI can be fully integrated as a diagnosis source. 

I think AI is useful for interpreting neuroimaging. However, there also seems to be a lack of AI technology that can integrate not only neurological models but psych models for a more comprehensive diagnosis. AI is able to understand technique outcomes from Diffusion Tensor Imaging (DTI) and Functional Magnetic Resonance Imaging (fMRI) and make conclusions based on the images they produce. AI systems in this case show great promise but those neuroimaging techniques do have limitations in itself. Additionally, most times ASD can’t be attributed to a single brain region, proving that AI should not be considered a reliable source for a comprehensive diagnosis of ASD.

In summary, while AI does hold promise as a supplementary tool for understanding certain behavioral and neurological traits, it is not yet a reliable method for diagnosing ASD. Its greatest value may lie in enhancing our understanding of autism rather than providing definitive diagnostic outcomes.