A new artificial intelligence model developed by researchers at USC can now detect cancer cells in blood samples in a fraction of the time it usually takes. Before the creation of the Rare Event Detector — an AI tool that automates the process of identifying cancer cells within blood samples — trained researchers manually examined each cell image, a process that could take several hours per sample.

The nine-year collaborative effort between the Viterbi School of Engineering and the Dornsife College of Letters, Arts and Sciences has reduced what was once a meticulous, up to 10-hour process to about 10 minutes. 

The research originated from a seminar that brought together Assad Oberai, professor of aerospace and mechanical engineering at Viterbi, and Peter Kuhn, director of the Dornsife Convergent Science Institute in Cancer. 

“Ever since [the seminar], we’ve been chatting off and on about working together,” Oberai said. “I understood the problems that he was dealing with better, and he understood the kinds of tools and algorithms that my group could develop better.”

With Assad’s expertise in the mechanical behavior of cancer and Kuhn’s background in cancer biology, the two professors launched a partnership that led to the creation of RED.

The project was mostly student-driven, according to Assad, and developed by Javier Murgoitio-Esandi, a recent Ph.D. graduate in mechanical engineering and current Google employee, along with Dean Tessone, a Ph.D. candidate studying molecular biology. 

Researchers involved in Kuhn’s lab, including Tessone, analyze blood samples from patients with and without cancer, searching for rare circulating tumor cells that can indicate the presence of the disease.

“The problem is easy to describe as a needle in the haystack,” Tessone said. “From each patient, we are assessing something like two and a half million cells from the blood, and we estimate that approximately one of those cells is cancer-associated.”

Before the model, individuals were trained by experts to sort through millions of pictures of cells and identify what was cancerous. By integrating deep learning techniques, the model now performs the same task in minutes, Tessone said.

So far, RED has been primarily applied to studies of breast and pancreatic cancer. These two diseases represent opposite ends of the early-detection spectrum, according to Tessone. He said breast cancer provides a valuable testing ground for AI-driven screening tools because established methods, such as mammograms, already produce large amounts of data.

“We actually already have really good screening for breast cancer, mammograms,” Tessone said. “They still have limitations … If you ever ask any woman that’s had a mammogram, it’s incredibly uncomfortable … One thing that I’m working on right now is seeing if we can replace that mammogram-based screening with a blood-based screening.”

For pancreatic cancer, Tessone said the need for early detection is even more critical. Pancreatic tumors are often diagnosed at later stages, when treatment options are limited. The team hopes that RED’s ability to identify rare circulating tumor cells could help catch signs of the disease sooner.

“Pretty much every single pancreatic cancer patient is detected when they already have extremely advanced disease,” Tessone said. “We can actually develop a method which will bring rates of late-stage pancreatic cancer way down because we’re able to catch it much earlier by using that blood-based screening approach.”

Oberai said the project is an example of how AI can enhance scientific research. He said that AI models are “useful tools” that are soon to be widely available to all. Tessone said that AI is intended to support, not replace, researchers.

“Our model is a medical tool, and it’s not acting to replace [us] — it’s really just benefiting us as scientists and making us better at our jobs,” Tessone said. 

Oberai said the team aims to refine its models to more accurately distinguish between potential and confirmed cancer cells. The researchers plan to expand their methods to other types of biological data, including those used for drug screening and treatment testing. 

“The real application for us is being able to leverage these kinds of tools to move really rapidly towards understanding cancer biology,” Tessone said. 

The RED project has already led to multiple published papers applying the tool to real-world cancer research, Oberai said. The most recent studies explore how the model can identify tumor cells and analyze how cancer spreads through the bloodstream.

Oberai said this work reflected the opportunities available at USC for students to collaborate and contribute to the future of meaningful discoveries. He encouraged students interested in research or interdisciplinary science to get involved now.

“Any student who’s interested in this type of work and finds this intriguing and rewarding should reach out to either Peter or me, and we’d be happy to chat with them. We’re always, always, always looking for bright students to work on this.”