For most children, autism is not diagnosed until they're between three and five years old, when behavioral differences become more noticeable. But what if doctors could identify signs of autism much earlier — even during infancy?
That's the question rising senior Talithea Concepcion is exploring through her University of South Carolina Magellan Scholar research project.
The computer science major is studying infant heart rate patterns to see whether artificial intelligence can help identify early signs of autism risk before behavioral symptoms appear. Her research focuses on the autonomic nervous system, which controls automatic body functions such as breathing, digestion and blood pressure. Previous research has linked differences in how this system functions during infancy to autism spectrum disorder.
One method researchers use to measure autonomic nervous system activity is by looking at the timing between heartbeats, known as the interbeat interval. Small differences in these patterns can provide clues about how the nervous system is functioning, making it a promising, noninvasive tool for research.
Concepcion believes identifying those signs earlier could make a meaningful difference for children and families.
"If you're able to make an early diagnosis, you can get an intervention and allow for more help later on," Concepcion says. "A lot of people are not diagnosed with autism, while others get diagnosed but have already had to adjust or learn how to mask or hide themselves in certain ways."
As part of her project, Concepcion is analyzing infant electrocardiogram (ECG) data collected through USC psychology associate professor Jessica Bradshaw's Early Social Development Lab. She is comparing two different approaches: the traditional method looks at heart rate variability, while the newer approach uses pre-trained AI models that have learned patterns from millions of adult ECG recordings.
“It's more than specific ECG data; it's to see if the AI representations of the data can work better,” Concepcion says. “The question is whether the AI model sees things that humans may miss or whether the simpler design features are enough to determine if these infants will be more prone to being diagnosed with autism.”
Rather than building AI models from scratch, Concepcion is using three publicly available foundation models hosted on GitHub. Each was developed by other research teams using millions of adult heart recordings to recognize different cardiac patterns.
"These are different signals for our body, in the same way an Apple Watch reads data for heart rate," Concepcion says. "I download the finished version and then feed it my data from Dr. Bradshaw's lab. The model produces a learned representation of that signal that we can analyze."
If the AI-based approach proves effective, it could lead to a low-cost, noninvasive screening tool. ECGs are inexpensive and already commonly used in neonatal intensive care units.
"If a biomarker exists, it would be easy to scale it to people who don't have that kind of specialized assessment," Concepcion says. "It would be great for the field in general because having that type of clarity would lead to larger research because I feel some researchers don't know when to reach for big AI models or simpler methods."
Concepcion's path to her research project began last summer while she was changing her major to computer science. During that time, she discovered the lab of Computer Science and Engineering Assistant Professor Christian O'Reilly, whose research combined several of her interests, including computer science and neuroscience.
“I have watched Talithea grow from a novice Python programmer into a skilled machine learning researcher, applying foundation models to investigate what recordings of heart activity can reveal about infants with autism,” O’Reilly says. “She has successfully integrated her interests, demonstrating the value of bridging the fields of computer science, neuroscience and neurodevelopmental disorders to address important scientific questions.”
Concepcion plans to attend graduate school and pursue a career in research after earning her undergraduate degree in May 2027. She says the Magellan Scholar program has given her valuable experience that extends beyond the classroom.
“It has made me more aware of what I want to do and developing all the technical and methodological skills that I need for grad school,” Concepcion says. “This has been the biggest help for me getting that kind of experience and real results.”