Cut off from his lab during the early days of the pandemic, Eri Osta, M.D., MSc, then a medical student at Long School of Medicine at UT Health San Antonio, set aside his cancer immunology lab research and began working with patient data in a more computational role.
He taught himself to program and analyze data — a shift that allowed him to test hypotheses and explore clinical questions faster and more directly than in the wet lab. In the process, he discovered a new kind of problem-solving in medicine.
That pivot would eventually lead him to pursue a master’s degree in artificial intelligence and data science alongside his medical training, becoming one of the first physicians in the country to earn that dual credential.
Eri Osta, M.D., MSc
Now a diagnostic radiology resident at Dell Medical School at The University of Texas at Austin and Ascension Seton, and supported by Dell Med resources like the Texas Health Catalyst accelerator program, Osta is developing HepaticMD, an artificial intelligence model that can predict a patient’s risk of liver fibrosis — a dangerous scarring that can lead to irreversible liver failure — using routine lab tests.
His goal is to make early detection simple, affordable and available to every clinician, not just specialists with access to costly imaging or biopsies.
Diagnostic radiology stands to gain the most from early AI implementation in medicine. What I hope is going to be the new trend is a lot more clinician input… We can be more than just key stakeholders.
Eri Osta, M.D., MSc
Turning everyday labs into early warnings
The idea for HepaticMD started with a simple question: Could a computer learn to spot the signs of liver disease using only the labs that most patients already have on file?
For Osta, the focus was metabolic dysfunction–associated steatotic liver disease, or MASLD, a condition once known as nonalcoholic fatty liver disease and now one of the most common causes of liver scarring. Often linked to diabetes and obesity, MASLD can progress quietly for years before symptoms appear.
What makes Osta’s model distinct isn’t just its accuracy but its transparency.
“Most calculators and AI models just give you an output: yes or no, fibrosis or no fibrosis,” Osta says. “We show the specifics and break down what it is about the data that has been input into the prediction model that is driving the prediction.”
Drawing on publicly available data from the National Health and Nutrition Examination Survey, Osta built and trained a model to predict fibrosis risk from routine lab values such as liver enzymes, platelets, hemoglobin A1C, and creatinine, and demographics such as age and gender. To test how well his model would perform beyond that initial dataset, he validated the findings using an independent cohort with mortality outcomes. The results held, and HepaticMD performed on par with recommended noninvasive tools for predicting fibrosis. A second independent external validation using data from the U.S. National Institute of Diabetes and Digestive and Kidney Diseases, one of the world’s largest data repositories of MASLD, is currently underway.
That clarity can help clinicians see what’s influencing a patient’s risk and, in turn, help patients understand their own results. A person with normal liver enzymes but a high A1C and low platelet count, for example, might still be flagged as high risk — and HepaticMD makes it clear why.
“AI models usually have a black box algorithm in the back,” Osta says. “As a clinician, it’s important for the tools that we use to be transparent and explainable. It helps us make better decisions about the kind of tools we want to integrate into our workflows.”
From code to clinic
Now, Osta is translating HepaticMD from research to real-world use. With funding and mentorship from translational research accelerator Texas Health Catalyst — including strategic guidance from expert advisors during the program’s mentorship phase —he’s building an application that integrates with electronic health records to automatically analyze lab results and flag patients at risk for liver fibrosis.
“HepaticMD exemplifies the type of innovation Texas Health Catalyst seeks to support by leveraging cutting-edge AI to deliver a noninvasive, cost-saving diagnostic solution with significant potential to improve health outcomes,” says Nicole Clark, MBA, director of programs and partnerships. “Its ability to predict long-term mortality and reduce unnecessary liver biopsies aligns with the program’s mission to accelerate impactful, early-stage ideas into practical health products.”
Jack Virostko, Ph.D., associate professor of diagnostic medicine and Osta’s principal investigator, says HepaticMD reflects the kind of creativity and cross-disciplinary thinking that defines clinician-led innovation.
“As science becomes increasingly compartmentalized, there is a need for people who can bridge across domains,” he says. “Eri has the knowledge and skill set to identify clinical problems and solve them with data-driven approaches.”
For Virostko, that blend of curiosity and technical fluency represents more than a single project — it signals where medicine is headed.
“AI is going to change all aspects of life,” he says. “We need future leaders like Eri who can harness its power to advance and improve medicine.”
Innovation, rooted in Austin
HepaticMD’s evolution reflects Austin’s growing role as a hub for health innovation. Anchored by Dell Medical School and The University of Texas Medical Center, the region is becoming a place where engineers, physicians and data scientists collaborate to define the future of health.
That convergence is part of what drew Osta to continue his training at UT. The University’s strength in computational science and engineering — from research leadership to infrastructure like the Texas Advanced Computing Center — offered the foundation he was looking for to keep building. With a dedicated research rotation within his residency program, faculty and entrepreneurial mentorship, and access to state-of-the-art tools, he’s now expanding his portfolio of AI projects, including a tool that uses large language models to summarize radiology reports across time.
“Diagnostic radiology stands to gain the most from early AI implementation in medicine,” he says. “What I hope is going to be the new trend is a lot more clinician input — or the clinicians themselves driving the development. We can be more than just key stakeholders.”
For Osta, MASLD is exactly the kind of condition that calls for that approach. “MASLD matters to me because it affects a lot of underserved populations,” he says. As HepaticMD moves closer to real-world care, it reflects not just a new tool for early detection, but a broader effort to bring better care to the patients who need it most — and a clear expression of The University of Texas Medical Center’s mission to revolutionize how people get and stay healthy, one question, one algorithm and one patient at a time.
Graduate medical education, or GME, refers to the period of education in a particular specialty or subspecialty following completion of medical school. This continuation of training through residency and fellowship programs provides the clinical and educational experience needed for physicians to achieve autonomy, deliver high-quality patient care, and prepare for challenges in an evolving health care landscape.
Dell Med serves as the academic home and Ascension Seton as the clinical home for 494 resident and fellow physicians within more than 45 residency and fellowship programs ranging from family medicine and neurology to pediatric emergency medicine and cardiovascular disease.