Early autism detection leads to swift, effective treatment. For small children, weeks matter. The longer they wait for help, the more ground they lose. If experts choose the wrong treatment approach, the child waits even longer for effective care.
For some experts, computers hold the answer.
Artificial intelligence (AI) augments a doctor's ability to diagnose, treat, and understand disease. Medical experts all around the world use AI to address common conditions.
Some believe autism will be transformed by the AI revolution, while others believe more research is required before we can truly harness the power of computers to address autism.
How Does Artificial Intelligence Work?
We've all seen smart computers in movies and television programs. Actors bark out complex questions, and just moments later, the machine spits out the perfect answer. In real life, artificial intelligence works a bit differently.
AI is built on algorithms. These computer programs handle common and complex tasks, including:
- Pattern recognition.
- Image analysis.
- Decision making.
Computers can't do this work alone. People build the algorithms the machines use. That process involves:
- Data collection. Scientists gather a complete set of files for the computer to analyze. Some are healthy and normal, and some are not.
- Annotation. Scientists label each file for the computer.
- Learning. The computer moves through the files and the notes, and it develops an understanding of what separates healthy and unhealthy.
- Testing. The program labels its own files, and the scientists determine whether the results are accurate.
- Adjustment. If the tests highlight inaccuracies, the program may need more data, more labels, or some other type of fix.
Just as humans can be inaccurate, so can the programs that they build. If scientists feed computers incomplete data, the algorithm can learn the wrong lesson. And this means the results it gives will also be inaccurate.
Experts explain that algorithms can become oversensitive to bias and outliers. The computer makes more of the unusual files than required. That can lead to issues unless the experts spot the problems and fix them. That doesn't always happen.
Researchers are aware that their systems are prone to problems and inaccuracies. They do their best to build effective programs, and they use repeated rounds of testing to ensure that the systems work before they enter widespread use.
But it's critical to remember that humans, not celestial beings, build these systems. Mistakes are sometimes made, and that makes oversight even more important.
Where Is Artificial Intelligence Used in Medicine?
Artificial intelligence is remarkable, and it's reasonable to get excited about how it works and what it can do. Some people dream of a world where AI plays a part in almost every area of medicine.
The technology is relatively new, and that means use isn't widespread right now. In fact, AI is used in just a few different types of medicine.
Experts say AI plays a role in:
- Drug development. Sensors analyze the hands of doctors, nurses, and other medical professionals. The data feeds into a system that develops a medication and hygiene regimen to reduce infection rates in hospitals.
- Disease diagnosis. Computers compare scans and diagnostic tests with databases, and they identify samples that require a doctor's attention. This helps to recognize cases of disease that might otherwise go unnoticed.
- Analysis. Computers compare one health plan to another to determine how they match up. Sometimes, the results change the course of treatment.
- Monitoring. Patients wear sensors that pass data to computers. AI identifies traits that could lead to worsening disease, and that sparks doctors to take action.
- Consults. Patients complete online questionnaires, and artificial intelligence helps doctors understand what to focus on during a visit. This can make for more efficient consultations.
- Surgeries. AI tools help doctors make tiny incisions during major procedures. The analysis prompts the doctor to use a particular surgical approach and offers feedback based on data from other patients.
Artificial intelligence works best, experts say, in cases where the data is clear-cut. In dermatology, for example, a skin sample either shows a suspicious spot or healthy tissue. The computer just needs to know the difference between the two types of cells. When the target is harder to define, such as whether a nonverbal child is simply willful or unable to speak, computer systems can struggle.
Doctors will never be replaced by AI, experts say. Instead, these systems help doctors do their jobs better. They feed predictions and analysis to medical experts, and it's those humans that decide whether the information has merit or can be skipped. No matter how sophisticated artificial intelligence will become, doctors will always be in charge.
AI & Autism Diagnostics
Among all the challenges in the autism field, the diagnostic process remains supreme. It's critical for doctors to spot disease early, but the symptoms can be baffling and hard to detect. This means a lot of autism cases are missed or diagnosed later than they should be. Some hope artificial intelligence will assist.
At least one company offers an AI autism diagnostic tool. It uses:
- Behavior information. Parents complete online surveys and tests. They describe how their child regularly behaves.
- Data analysis. The computer compares the parents’ responses with a bank of other test results. It flags which symptoms might be concerning and which are in line with the development of neurotypical children.
- Diagnostics. The computer offers a tentative diagnosis to parents. This isn’t an official diagnosis since that will need to come from a doctor.
- Care options. The system determines what types of therapy will be best for the child. These recommendations can form the basis of a care plan that medical professionals build on.
It's unclear if the system works with a child's insurance company or school district. Sometimes, parents need an official diagnosis from a medical expert to tap into financial assistance programs. They need the same paperwork to discuss a child's needs with the school. But some parents might use this analysis to get a jump on care while they wait for in-person appointments with doctors.
Researchers say systems like this can be remarkably accurate. Sometimes, the programs are even more effective than humans in detecting autism. Since early detection is the key to best helping a child with autism, this is good news.
- Data libraries. Programs work best when computers have a vast amount of files to choose from. Sometimes, systems are built with a large number of normal files and few from autistic families.
- Symptoms. Programmers may simplify autism for the computer by leaving out some atypical symptoms. That could mean some people with mild autism get overlooked. When parents get an “all clear” from a system, they may not pursue diagnosis any further, and the child suffers as a result.
- A uniform technique. No one has decided how an AI diagnostic tool should work. Each product has different values, goals, and methods. In theory, families could get vastly conflicting information from one system to another.
Artificial Intelligence in Autism Prediction
Autism is a lifelong developmental disorder that can be treated but not cured. For some researchers, AI's help could be important in spotting children with the disorder who exhibit no signs quite yet.
- Age. Both biological parents provide information about their age when the child was conceived. When parents are older at birth, the child has an increased risk of autism.
- Socioeconomic status. Parents offer information about their salaries, inheritance, home equity, and other forms of wealth.
- Medication use. Mothers and fathers provide a list of drugs they are taking. Medications that may increase the risk of autism are flagged.
The AI system predicts whether a child born to this couple will develop autism, and experts say it’s right about a third of the time.
If artificial intelligence performance improves, potential parents could use the tool to make changes before pregnancy to lower certain risk factors. Or they could work with genetic counselors on custom medications to amend autism traits.
This is next-generation medicine, and it's likely years before any of these ideas come to market. When researchers think of autism and AI, solutions like this come to mind.
Other AI Hopes & Dreams
Artificial intelligence is a vast field, and its applications vary. Some people don't want to stop at autism identification and treatment. They have other solutions in mind.
AI could be used to address:
- Medication management. Researchers use artificial intelligence to help patients choose the depression medication that's right for them. A similar program could help someone with autism find the right combination of medications to soothe symptoms and ease distress while avoiding those that produce too many troublesome side effects.
- Diet changes. Plenty of theories about an autism diet exist, and researchers often say these amendments don't work. But artificial intelligence could help people to uncover secrets about the ways they eat and the symptoms they experience. Diet journals combined with symptom trackers could make patterns clear.
- Therapy. AI programs could help people make smart choices about effective autism treatment. Some AI programs could even deliver customized treatment, and machine learning could ensure that the techniques change based on a person’s needs.
Reading through these ideas is exciting, and some families find hope in the innovation happening all around them. But it's worth repeating that these therapies are new, and many years of work are required before they come to an office near you. The future for autism treatment is bright.
For now, the best autism treatments come from humans who work closely with you and your family to customize your care. These established therapies have an extensive body of research backing their use, and they produce real results in clients. Think about all the potential the future holds, but remember to listen to your doctors and follow their advice carefully.
Artificial Intelligence in Medicine: Applications, Implications, and Limitations. (June 2019). Harvard University.
Dealing With Confounders and Outliers in Classification Medical Studies: The Autism Spectrum Disorders Case Study. (July 2020). Artificial Intelligence in Medicine.
Overview of Artificial Intelligence in Medicine. (July 2019). Journal of Family Medicine and Primary Care.
Artificial Intelligence in Medicine: Current Trends and Future Possibilities. (2018). British Journal of General Practice.
Artificial Intelligence in Medicine: Today and Tomorrow. (February 2020). Frontiers in Medicine.
Artificial Intelligence in Medicine. (April 2017). Metabolism.
The Use of Artificial Intelligence in Screening and Diagnosis of Autism Spectrum Disorder: A Literature Review. (2019). Journal of the Korean Academy of Child and Adolescent Psychiatry.
3 Ways AI Is Already Changing Medicine. (March 2019). Vox.
High-Performance Medicine: The Convergence of Human and Artificial Intelligence. (January 2019). Nature Medicine.
How Autism Researchers Are Applying Machine-Learning Techniques. (March 2020). Spectrum.
The Practical Implementation of Artificial Intelligence Technologies in Medicine. (January 2019). Nature Medicine.
The Coming of Age of Artificial Intelligence in Medicine. (May 2019). Artificial Intelligence in Medicine.
Artificial Intelligence for Precision Medicine in Neurodevelopmental Disorders. (2019). NPJ Digital Magazine.
How Artificial Intelligence Can Help Pick the Best Depression Treatments for You. (February 2020). Time.
Gluten-Free Diet in Children With Autism Spectrum Disorders: A Randomized, Controlled, Single-Blinded Trial. (2020). Journal of Autism and Developmental Disorders.
Personalized "Deep Learning" Equips Robots for Autism Therapy. (June 2018). Massachusetts Institute of Technology.
Can AI Help With Autism Diagnosis? (March 2020). Forbes.
Autism Buddybot — An Artificial Intelligence Companion. Stanford Medicine.
Applying Machine Learning to Facilitate Autism Diagnostics: Pitfalls and Promises. (May 2016). Journal of Autism and Developmental Disorders.
Use of Artificial Intelligence to Shorten the Behavioral Diagnosis of Autism. (August 2012). PLOS ONE.
Providing an Independent Second Opinion for the Diagnosis of Autism Using Artificial Intelligence Over the Internet. (2009). Counselling, Psychotherapy, and Health.