Why It Matters
Artificial intelligence systems used to diagnose cancer must perform flawlessly in clinical settings, where errors can have life-or-death consequences. A University of Rhode Island computer scientist is developing methods to ensure AI models correctly identify breast cancer by understanding—and correcting—the reasoning behind their diagnostic decisions.
The Rhode Island Foundation announced Monday it will distribute nearly $650,000 across 26 medical research projects statewide, with URI receiving more than half the funding for 14 projects.
What Happened
Alina Jade Barnett, an assistant professor of computer science and statistics at URI, received a $25,000 grant to advance her work on machine learning models for breast cancer diagnosis. Barnett specializes in deep learning and healthcare-focused AI systems.
The foundation’s awards will support research at multiple institutions including Rhode Island Hospital, The Miriam Hospital, Brown University, Johnson & Wales University, and Providence College. Six grants went to Rhode Island Hospital projects and three to Miriam Hospital initiatives.
Other funded research topics include adolescent concussion treatment, patient experiences with weight-loss medications, diabetes prevention programs, and antibiotic development using marine organisms.
By the Numbers
The Rhode Island Foundation distributed $650,000 total across the 26 projects. URI claimed 14 of the 26 awards, accounting for more than half the total funding. Most individual grants amounted to $25,000. Six projects at Rhode Island Hospital received funding, while three grants went to The Miriam Hospital.
The Technical Challenge
AI diagnostic models sometimes reach correct conclusions through flawed reasoning, Barnett explained in a May interview. A model may accurately detect cancer but focus on the wrong portion of a mammogram image—or flag images as cancerous based on irrelevant factors like which medical center produced them.
Training data quality remains critical. Models can develop biases based on variations in imaging equipment brands or protocols across different hospitals. An AI system might interpret mammograms from one manufacturer’s machine differently than images from another brand, introducing diagnostic inconsistencies.
Barnett noted that identifying reasoning errors is simpler than correcting them. Engineers cannot manually adjust individual decisions; they must develop systematic approaches to repair flawed reasoning without degrading the model’s overall performance.
Zoom Out
Healthcare AI faces heightened scrutiny as medical systems integrate machine learning into clinical workflows. Unlike low-stakes applications such as email summaries or grocery lists, diagnostic AI must perform consistently across all patients and imaging conditions.
The technology sector has made progress in recent years in understanding how AI models reach conclusions, but different systems vary in their ability to explain their reasoning processes. Balanced training data remains essential to prevent models from learning false correlations between non-medical factors and diagnostic outcomes.
What’s Next
Barnett’s grant will fund research into methods that steer AI reasoning toward accurate cancer identification. The foundation emphasized that while individual grants are modest, they can catalyze larger research investments and support healthier communities statewide, according to foundation president David Cicilline.
The researcher’s work focuses on ensuring AI models correctly identify cancerous tissue in mammogram images for the right reasons—a requirement for reliable deployment in clinical settings where diagnostic accuracy directly affects patient outcomes.