Stony Brook University is advancing artificial intelligence (AI) research through its AI Innovation Institute (AI3), which recently awarded 13 faculty teams funding through its Seed Fund Program. With over $500,000 allocated across the projects, the program will further research areas such as education, data analysis, medical equipment and interpreting national literature.

The application process began in October of last year, with researchers expected to use the funding for their projects at the start of the spring 2025 semester. Nearly 50 applications were received and evaluated by a panel chosen by AI3 based on specific criteria. 

In an email to The Statesman, Steven Skiena, interim director of AI3 and distinguished teaching professor of computer science, described the process for selecting panel members.

“While [AI3] oversaw the selection process of reviewers, the administrative process was handled by the Office for Research and Innovation,” Skiena wrote. “These reviewers included faculty from a variety of disciplines who had the relevant subject matter expertise to fairly evaluate proposals based on a set of review criteria, which included the merit of the proposed project and its likelihood to attract external funding, among others.” 

Projects were also evaluated on their potential to advance knowledge across their fields of study, their impact on the Stony Brook and AI3 communities and the qualifications of each research team. 

When reviewing the proposals, Skiena shared that it was difficult selecting programs from a wide range of applications. 

“The proposals were competitive and held potential to help resolve some of the biggest challenges we’re facing today. We wish we could have funded many, many more of them,” he wrote.

Xin Qian, an assistant clinical professor and medical physicist in the Stony Brook University Hospital’s Department of Radiation Oncology, is among the grant recipients for a project dedicated to improving computed tomography (CT) image reconstruction. 

He works alongside three co-principal investigators, including Tiezhi Zhang, a professor in the Radiation Oncology Department, Ziyu Shu, a senior postdoctoral associate and Zhaozheng Yin, a SUNY Empire Innovation associate professor from the University’s Department of Computer Science.

The project focuses on the clinical applications of Deep Image Prior (DIP) technology to improve CT image reconstruction. Qian explained that current technology can result in inaccurate scans, which affects radiation therapy treatment plans. 

This is especially problematic in brachytherapy, a form of radiation therapy where an applicator is inserted into patients, typically women, to deliver localized treatment. The process requires moving patients from the treatment room to the CT simulator room, which can displace the applicator and cause misalignments in the scan, leading to inaccurate results. 

Qian emphasized that even slight misalignments of just a few millimeters can significantly impact the accuracy of the radiation dose, potentially making the treatment less effective and compromising patient outcomes. He stressed the importance of precise procedures to improve the quality of life for patients.

“When we treat the patient, the treatment outcome cannot be very ideal. It’s not where we want to treat. Maybe [it’s] a few millimeters off and the radiation dose may be a few percent off,” Qian said. “So that will affect the treatment outcome because we want to save patients’ lives [and] improve their life quality … But if you don’t have a very accurate set up, very accurate procedure workflow to achieve that goal, then that’s not fair for the patients.”

Qian and Shu highlighted that the DIP system would eliminate the need to transport patients between rooms. With DIP, image reconstruction can be performed in the same room, reducing the risk of miscalculations caused by patient movement. 

Shu also showed a comparison between the current CT system, which uses filtered back projection (FBP), and the DIP system. The DIP system produced clearer and more accurate images, which could help improve treatment precision. 

Another recipient is Giuseppe Gazzola, an associate professor in the Department of Languages and Cultural Studies who works with Jayesh Rathi, a graduate AI consultant, to investigate works of Italian, French and Spanish literature from circa 1733-1794.

Gazzola described the grant as a wonderful opportunity to use AI to develop a system that makes data from literature searchable, specifically from various tomes, which are volumes of literature that are compiled to make one larger work.

“Since these books are huge and all of them are like 20 tomes, I am rendering them searchable so that I can ask sophisticated questions [to] AI,” Gazzola said. 

He described some of the questions he would explore, such as how the authors of these works interacted with one another and how editions from different years compare in terms of content and meaning. 

A goal in Gazzola’s research is to document the ways in which these books indirectly created a national discourse in 18th-century Europe. He hopes to produce numerous articles on his findings and potentially write a book on the behavior and values of writers from different literary histories. 

AI3 also awarded David McKinnon, a professor in the Department of Neurobiology and Behavior and the principal investigator in his research, to study the use of machine learning to analyze single-cell multiomics data.

McKinnon works with Barbara Rosati, a assistant research professor in the Department of Physiology and Biophysics, Joshua Dubnau, a professor in the Department of Anesthesiology, Chi-Kuo Hu, an assistant professor in the Department of Biochemistry and Cell Biology and Grigori Enikolopov, a professor in the Department of Anesthesiology. 

With the grant, their team was able to bring on Deboparna Banerjee, a graduate AI consultant, to help them understand advanced machine learning.

“[Our department] generally lacks the expertise to [use machine learning],” McKinnon said. “What this grant does is [is that it] brings in a graduate student who has a lot of expertise in those areas, and starts a dialogue between us and the computer science department.”

McKinnon and Rosati explained their research is studying biological systems, which are highly complex, unlike physical systems, such as chemical, mechanical or planetary systems that can often be described with simple mathematical models.

“The biological systems have been developed in a very chaotic, contingent way, and they don’t lend themselves to simple descriptions,” McKinnon said. “What that means is that you need a vast data set to capture that complexity.”

He clarified that single-cell sequencing methods generate these large datasets by examining the gene expression of every cell in a tissue. 

Rosati added that what makes conducting the research difficult are numerous factors, such as the quantity and quality of the data, as well as the team’s ability to use the tools meant to interpret it. 

“The complexity does not only involve the amount of data, it also involves the type of the quality of the data and the quality of the tools,” Rosati said. “We need [Banerjee’s] guidance to understand all the tools that are out there and … to find what’s the best tool for a given system.” 

Rosati emphasized that the grant has successfully progressed their research and served its purpose in helping them comb through their data effectively.

“[Banerjee’s help has] completely fulfilled the format of [this] grant and the purpose to create a synergy between two different fields so that we can all grow exponentially,” Rosati said.