From brain imaging and mental health to mineral extraction, several USC labs said they are optimistic about artificial intelligence use in research after it helped them achieve significant breakthroughs.

Ravi Bhatt, a postdoctoral scholar at the Keck School of Medicine, worked with a team at the Mark and Mary Stevens Neuroimaging and Informatics Institute to create an artificial intelligence-powered tool that mapped the brain’s corpus callosum — the region that connects the two hemispheres of the brain.

“The AI tool makes it extremely fast and accurate to extract the corpus callosum and develop these measures more so than any model that had existed to that point, even to now,” Bhatt said.

After creating large datasets composed of brain images and genetic data, Bhatt said the Stevens INI researchers used AI to conduct a meta-analysis to assess the heritability of corpus callosum traits. This was found by assessing whether genetic markers in the corpus callosum also affect the cerebral cortex, the brain’s outermost layer associated with reason, emotion and memory.

The findings gave more insight into how genetic variants influence corpus callosum development and how that’s related to disease and brain structure.

Bhatt said the AI tool developed was able to eliminate the “tedious, manual [and] error-prone process” of analyzing an MRI image due to the variability of individuals’ brains and the lines that separate the regions in it.

In the mental health sector, the Signal Analysis and Interpretation Lab is conducting an ongoing study to create a predictive AI model that identifies biological markers for mental health in young adults.

“[SAIL] has been using AI as a means to understand the human condition,” said Shrikanth Narayanan, professor of electrical and computer engineering, computer science, linguistics, psychology, pediatrics and otolaryngology. “Our vision is not to react to something that happened, but to be able to proactively think about things.”

The research project aims to understand why young adults react differently to similar stressors, including upcoming deadlines, exams and major life transitions. Analyzing data with tools like brain imaging, aura measurement and surveys, the researchers aim to apply the predictive properties of large language models to research electrical activity in the brain and heart.

“This provides us not only ways of capturing the patterns, but also being able to do generative things, explore things that we haven’t really imagined yet, and then look at the effect of interventions [and] treatments,” Narayanan said.

Craig Knoblock, research professor of computer science and spatial sciences, worked with the Viterbi Information Sciences Institute to develop an AI tool that accelerates the discovery process of critical minerals.

Finding new mineral sources previously required researchers to read through tens of thousands of journal articles to locate the necessary information and identify potential mining sites, Knoblock said.

The plan for the AI tool is to accelerate this labor-intensive process by processing as many journal articles as possible and creating a research resource, likely enabling mining companies to find information more quickly.

While Knoblock admitted that hallucinations — faulty information generated by AI — are a “big problem,” he said the data used to train the AI model is auditable, allowing users to trace back to the AI-generated information’s source and fact check it themselves.

Bhatt’s research team implemented fact-checking within the AI tool itself. After the brain-imaging model extracted an MRI scan, a separate AI tool composed of numerous algorithms acted as a “self-checking” tool to assess the accuracy of the corpus callosum image created.

When comparing it to other AI-based tools and software that extract the corpus callosum, Bhatt said their deep-learning-based model for segmentation could do so better than anyone else at this point.

Though AI tools have proven both efficient and accurate, ethical concerns about data collection, privacy and security still emphasize the importance of “human-centered AI.”

“It’s for the people, by the people,” Narayanan said. “AI can play a role in every aspect of research, but it is an enabler, not the end of it.”

Emily Zhou, a doctoral candidate in computer science and a research assistant at SAIL, said she sees the nuances of the mental health impacts of AI. While she said she worries about the lack of regulation around using AI chatbots for emotional support, Zhou also sees the potential for AI to provide critical assistance in the health sector.

“I see this as a very promising direction to not replace doctors or medical health professionals, but to really be able to complement their work in a very scalable way and reach and support more people,” Zhou said.

Members of all three labs voiced their expectation of new opportunities arising because of their respective studies as well as for AI in research in general.

“[AI] will really change research in that there’s this opportunity to greatly accelerate research and provide all kinds of access to knowledge that just wasn’t there before, [and] that suddenly we can really do things that weren’t possible,” Knoblock said.

Correction: A previous version of this article stated that Ravi Bhatt was a doctoral candidate. The article was updated on Sept. 14 at 12:45 p.m. to reflect that Bhatt has received his Ph.D. and is a postdoctoral scholar. The Daily Trojan regrets this error.