Skip to main content


Brain MRI

Brain MRI

Using Machine Learning to Improve Autism Treatment

Nick Luckenbach ’22 thinks functional MRI scans could hold the key to the next big breakthrough in Autism Spectrum Disorder treatment.

Nick Luckenbach ’22 thinks functional MRI scans could hold the key to the next big breakthrough in Autism Specrtum Disorder treatment. 

If you look closely at the brain of a person suffering from Alzheimer’s or dementia, you can see the differences. Structurally, there is atrophy—shrinking in the cerebral cortex and hippocampus—and enlarged ventricles.

Doctors don’t diagnose Alzheimer’s or dementia solely through MRIs. There are tests and assessments, but the scan is a part of the puzzle that can help greatly in diagnosis and treatment.

For conditions like Autism Spectrum Disorder, however, there is no equivalent. While there are some physical differences in a brain with ASD, the exact relationship between the differences and the symptoms is unclear. As a result, the 75 million children with ASD worldwide are diagnosed solely by therapists.

That could change however. While researchers are still looking for clear structural biomarkers associated with autism, there are functional differences that can leave clues. For the past year, psychology and computer science and engineering major Nick Luckenbach ’22 has worked in Professor Lang Chen’s lab, using machine learning to analyze open-source functional MRI brain scans. By turning these scans into data, Luckenbach hopes artificial intelligence can teach us more about how brains with ASD function and, in turn, help doctors diagnose ASD earlier and potentially offer more effective treatment.

Luckenbach, who graduates this month, sat down to discuss the research he presented as part of his Senior Design project.

Using machine learning to assess functional MRI scans is a fairly new process. How might it help diagnose ASD?
Typically, ASD is diagnosed by a therapist because it’s very behaviorally driven. The therapist goes through a list of behaviors or difficulties a patient might be exhibiting and makes a diagnosis. With neuroimaging, we’re trying to see if we can come up with a brain-based understanding of ASD in addition to the behavioral one made by the therapist. Essentially, we’re taking a thousand fMRI brain scans—which measure brain activity by detecting changes associated with blood flow—and using high-performance computing available at SCU (WAVE) to analyze the correlations of brain networks.

Anna Riggs and Nick Luckenbach

Anna Riggs and Nick Luckenbach

The computers look at what regions of the brain are active at different points and try to identify how the brains of patients with ASD function differently. Some differences may be easy to identify, such as a single region having reduced activity, but if your goal is to find complicated patterns across the whole brain, machine learning can help identify a lot of correlations that might go unnoticed.

How could machine learning impact the way we treat ASD?
Biomarkers may allow us to better evaluate treatment methods or try experimental techniques like neurofeedback or direct brain stimulation. Machine learning could also help track and improve treatments. In the past few decades, the definition of Autism has expanded to Autism Spectrum Disorders to include other disorders such as Asperger’s. So instead of autism as a monolith, ASD is a spectrum of disorders with varying severity and actual behavioral differences. This makes it difficult in the short term to analyze the data, but as more becomes available and we track it, we might be able to find clusters of different ASD phenotypes, or observable characteristics. For example, there’s already some evidence showing that males and females present differently with ASD. If you were to do a longitudinal study and map out subcategories, you could understand what intervention works best and perhaps how certain interventions change the trajectory of outcomes entirely.

You’re double majoring in psychology and computer science, which aren’t typically associated with each other. How did you find this path?
I came to college thinking I’d get a computer engineering degree, but I had these other interests. I was kind of interested in psych. I was kind of interested in philosophy. I wanted to explore that so I took a couple classes early on. In Psych 51, Professor Lang Chen started discussing research methodology, basically researchers finding a really challenging question and investigating it in the most rigorous way they could. For example, scientists have been studying the brain for a long time, but there were still some things we didn’t understand. With these new, more complex models using machine learning, they were starting to understand things that previously were inaccessible via other methods. It was fascinating, so I decided to double major in psychology.

I looked at some possible research opportunities and some of the labs were doing cool stuff, but I wanted a project where I could use my skills as an engineer with a computer science background to bring value to the research right away. I talked with Professor Chen about his fMRI research and joined his lab.

It seems like you’ve been able to combine your two interests in a pretty organic way. How rare is that for an undergraduate student?
I feel pretty lucky. The faculty in both psychology and engineering have been supportive. Professor Chen helped me strike a nice balance between exploring on my own and having him there to help. When we do this research, we’re simplifying the brain model down quite a bit. There are so many steps required just to make the scans into usable data and I’m relying on existing setups for that. If I didn’t have his guidance, I wouldn’t have gotten anywhere.

I’ve already been fortunate to present some of my research at the Western Psychological Association conference and I continued it for my Senior Design project. Ideally, I’ll end up finding unique biomarkers that haven't been found before. From there, neuroscientists can determine if my findings are relevant and perhaps that can nudge the understanding along a bit.

What do you want to do after graduation?
I have two trains of thought on that. I can either continue to a Ph.D. or explore jobs related to machine learning. At this point, I feel pretty comfortable in the education realm, so I want to challenge myself to work more in the business and engineering world. If I come back to get my Ph.D. later, I want to have learned from other people first. As of right now, I’ve used machine learning to serve my curiosity and I really want to see what other people are exploring.