Graduate student Sourya Sengupta works in Cancer Center at Illinois (CCIL) member Mark Anastasio’s research group, focusing on developing and training artificial intelligence (AI) algorithms to solve healthcare problems.

As AI continues to be adopted in our everyday lives, it remains a magic black box of fancy math for many people. Fortunately, researchers like Sourya can shed light upon the hidden, mysterious logic of AI. However, there’s catch. Developers of AI tools don’t necessarily know either.

While experts would agree that AI is not in fact magic, but rather complex computer algorithms trained with data to perform a certain task, AI doesn’t reveal the decision-making process that led to its output.

cancer center at illinois AI solutions

This Cancer Center at Illinois (CCIL) AI Solutions story features the work of a student researcher in a CCIL member lab.

Sourya’s work in the Anastasio group seeks to address this problem by developing more transparent AI models for healthcare. Recently, Sourya published work on the development of a self-interpretable AI tool for medical diagnosis tasks that can explain its decision.

“Cancer is an extremely complicated disease, so it is necessary to develop tools that help to make the diagnosis process faster and more efficient. AI tools can be important for this, especially in places where patients highly outnumber doctors,” says Sengupta.

Sourya Sengupta

Sourya Sengupta is an electrical and computer engineering Ph.D. student in Mark Anastasio’s research group.

Sourya’s novel algorithm, published in the Institute of Electrical and Electronics Engineers (IEEE) Transactions on Medical Imaging (TMI), incorporates a method for doctors to follow the AI logic to help determine prediction accuracy. He says, “IEEE TMI is arguably the highest-rated journal in our field of medical imaging and algorithms. Since I started working in this field, my dream was to publish in IEEE TMI. I hope AI tools like this will help to increase trust in AI models by both clinicians and patients.”

While Sourya is thriving in AI research now, with the icing on the cake being his recent selections as a Joan and Lalit Bahl Fellow and a Mavis Future Faculty Fellow, he faced a rocky start to graduate school.


In February 2020, Sourya visited the University of Illinois and cemented his decision to start a Ph.D. program in electrical and computer engineering that fall. At the time, he was finishing his master’s degree at the University of Waterloo in Canada and was excited to continue his academic research journey at Illinois – only for the world to shut down a month later due to the COVID-19 pandemic. With time running out on his student visa and apartment lease and uncertainty around his ability to move to the United States for graduate school, Sourya took the 30-hour journey home to Kolkata, India.

“It was a pretty messy experience to start graduate school during the pandemic, but the university was really supportive throughout,” says Sengupta, who was given a standalone tuition waiver to start remote coursework in India that fall semester. “My research advisor, Professor Mark Anastasio, also allowed me to join his group remotely in the spring of 2021 since our computational work was mainly performed online.”

When Sourya finally made it to campus, the pandemic only heightened the culture shock he already experienced during his time in Canada. “Here in Urbana, I don’t know anyone who lives on my apartment floor besides my roommates. In India, it is densely populated and you hang out with everyone in your neighborhood. The festivals are also very community centric,” he says. “I had to start living in a completely different, independent way – learning to cook and do everything on my own.”

Despite the feeling of isolation in his apartment building, Sourya has found his social footing by integrating himself into the university community. He says, “I have been to big U.S. cities including New York, and I find that the community presence is stronger here in Champaign-Urbana. I have gotten to know so many people through the university, and there is much more academic infrastructure and support than I have been used to in the past. It has really helped me develop socially, academically, and professionally.”

Through the support of the CCIL’s Tissue Microenvironment (TiME) Training Program, Sourya has built a strong academic and professional community. “I was the only computational student in my TiME cohort, which has really given me the opportunity to go beyond my expertise and develop fruitful collaborations. Sometimes I may be thinking about an algorithm that sounds really cool and has fancy math, but it doesn’t have a clear biological or clinical significance,” says Sengupta.

Like artists finding inspiration from the great works of art from before them, developing new algorithms often builds upon previous work. “We can fall prey to the popular algorithms developed for a very different purpose by trying to directly mimic them for medical imaging problems. But we need to be cautious because a successful method for a particular application might not be helpful for others,” says Sourya, emphasizing that this is an especially important consideration for medical applications where patient health is the highest priority.

“In my previous experiences, I mainly worked with engineers. It was an eye-opening experience to meet so many students from different academic backgrounds through the TiME program. I learned about their perspectives on developing advanced methods to solve research problems, which helped me to take a more interdisciplinary and comprehensive approach to my work.”

Drawing from his own experiences, Sourya recognizes that community and collaboration are a necessity to develop new and improved AI tools. “AI researchers working in the medical space must always collaborate with biomedical experts and clinicians so that the developed tools can be effective enough for real world problem-solving.”

Editor’s notes:

This story was written by Katie Brady, CCIL Communications Intern.

The first story in the CCIL AI Solutions series is available here.