Highlights
-
Associate Professor Pegah Khosravi is leading a new project to develop an artificial intelligence platform that combines computer vision and clinical expertise to help physicians more efficiently evaluate kidney cancer and stones.
-
The AI platform is called Kidney-VLM meaning the vision language model draws from medical imaging and a patient鈥檚 medical history to give a comprehensive look at their health.
-
The project, funded by the Florida Department of Health, is an interdisciplinary collaboration between the College of Medicine, College of Engineering and Computer Science and AdventHealth.
W
hen imaging reveals an abnormality in a kidney 鈥� such as cancer or a stone 鈥� patients and their physicians face a critical decision about what to do next: Wait? Do an invasive procedure? Or run more tests?
A new UCF research project combines artificial intelligence (AI), computer vision and clinical expertise to help physicians make that decision with greater confidence and efficiency.
The project, called Kidney-VLM, is supported by a one-year $974,681 grant from the Florida Department of Health鈥檚 Florida Cancer Innovation Fund and is led by Pegah Khosravi, a medical AI expert and College of Medicine associate professor with a joint appointment at UCF鈥檚 College of Engineering and Computer Science.
VLM stands for vision-language model, a trainable AI platform that examines both images and data. The UCF system analyzes visual information such as CT scans, while the language component comes from radiology, bloodwork and pathology reports. Bringing together these sources allows the system to provide a more comprehensive view of a patient鈥檚 kidney health and support more informed treatment planning. A physician could also ask the system a question and receive a response tied specifically to the patient鈥檚 imaging data and pathology.
鈥淜idney-VLM is designed to connect pieces of a patient鈥檚 medical story that may often be reviewed separately,鈥� Khosravi says. 鈥淢edical AI should not merely produce an answer. It should provide evidence that allows clinicians to understand, evaluate and appropriately question that answer.鈥�
A Clearer Vision for Patient Care
Khosravi is leading the project through , a collaborative initiative that brings together more than two dozen AI-related faculty members, their students and industry partners across disciplines. AdventHealth, a UCF Pegasus Partner, is also partnering on the research.
Trustee Chair of Computer Science , IAI鈥檚 director and founding director 色花堂鈥檚 Center for Research in Computer Vision, brings decades of experience in computer vision, visual reasoning and machine learning to the project.
He says the new research addresses the challenges of integrating computer vision and medical AI into clinical settings.
鈥淜idney-VLM advances computer vision by moving beyond isolated detection or classification toward an integrated and clinically interpretable system,鈥� Shah says. 鈥淭his project offers an opportunity to translate advances in computer vision into a practical tool that may help clinicians characterize kidney tumors more confidently and reduce unnecessary procedures.鈥�
The project exemplifies UCF鈥檚 willingness to foster interdisciplinary research and continue responsibly exploring the potential of AI, he says.
鈥淏y bringing AI researchers, physicians, and clinical partners together from the beginning, the project creates a pathway from foundational research to meaningful patient impact,鈥� Shah says. 鈥淲ith UCF鈥檚 strength in computer vision, the College of Medicine鈥檚 clinical expertise, and our partnership with AdventHealth, Florida is well positioned to become a national leader in responsible and clinically impactful AI innovation. This is exactly the type of collaboration we seek to foster.鈥�
UCF computer science researchers Yu Tian and Chen Chen are helping develop the AI architecture. Tian is an assistant professor with joint appointments in the College of Engineering and Computer Science and the College of Medicine, while Chen is an associate professor of computer science.
鈥淪uccess will not be measured only by whether the model achieves a higher accuracy score,鈥� Khosravi says. 鈥淲e also want to determine whether it performs reliably across various patient groups and clinical settings, whether its explanations are medically sound and whether clinicians can recognize when the system is uncertain. Of course, a clinically useful AI system must be accurate, but it must also be transparent, calibrated and safe.鈥�
Making AI Work in the Clinic
Abdo Asmar, a professor of internal medicine at the College of Medicine and a nephrologist with UCF Health Faculty Physician Practice, is guiding the project from the clinical side. He cares for patients who may undergo imaging because of kidney disease, hypertension, kidney stones or another medical issue and then discover they have a renal mass or complex cyst.
鈥淭his project is important because it focuses on a very real clinical problem with direct implications for patients: how to make better decisions when kidney tumors or suspicious renal findings are discovered,鈥� Asmar says. 鈥淯ltimately, this could help patients avoid unnecessary surgery, reduce healthcare costs and ensure that those with more concerning cancers receive timely evaluation and treatment. To me, this project represents a practical and patient-centered application of AI.鈥�
He stresses that physicians must approach AI as a support, not a replacement, for medical judgment.
鈥淲e鈥檙e not replacing clinical judgment but strengthening it at the point where decisions are most consequential.鈥� 鈥� Abdo Asmar, professor
鈥淲e鈥檙e not replacing clinical judgment but strengthening it at the point where decisions are most consequential,鈥� he says. 鈥淲hile I am excited about the potential uses of AI in medicine, I remain grounded in the belief that physicians and medical researchers have an obligation to use AI carefully, ethically and always in service of better patient care.鈥�
AdventHealth Research Institute鈥檚 Michael McDonald, a urologic oncologist, surgeon and researcher is working with the team to ensure the research remains clinically relevant and scientifically accurate.
He envisions that Kidney-VLM can address existing diagnostic limitations.
鈥淥ur ultimate goal is to provide physicians with better information before treatment decisions are made,鈥� McDonald says. 鈥淔or example, in kidney cancer, one of the major challenges is that imaging can鈥檛 always reliably determine whether a renal mass is benign or malignant before surgery, and the final diagnosis may not be known until after the tumor is removed and examined by pathology.鈥�
Kidney-VLM is a focused tool to pinpoint a specific set of medical conditions, but Khosravi and her team are hopeful that the model will provide a foundation for broader advances in medical AI.
鈥淭he importance of this project extends beyond a single algorithm or disease,鈥� Khosravi says. 鈥淜idney-VLM provides an opportunity to develop a reproducible framework for trustworthy medical AI that can eventually be adapted to other cancers and clinical challenges.鈥�
Research reported in this article was funded by the State of Florida, Department of Health, Public Health Research, Biomedical Research Program, through the Casey DeSantis Florida Cancer Innovation Fund (grant number 26C36). This content is solely the responsibility of the authors and does not necessarily represent the official views of the State of Florida.
