Able to absorb, examine and analyze staggering amounts of data, artificial intelligence (AI) helps medical researchers identify disease patterns and predict patient outcomes.
AI tools also improve researchers’ understanding of how cellular mechanisms influence and interact with one another, how tumors grow and much, much more. Five researchers in the University of Wisconsin School of Medicine and Public Health are leveraging AI for insights that could lead to more personalized care for patients.
Dr. Majid Afshar, an associate professor in the Department of Medicine, developed an AI screening tool that helps refer patients for treatment of opioid use disorder. The tool analyzes information within patients’ electronic health records, such as clinical notes and medical history, to identify features and patterns associated with opioid use disorder. Upon recognition, the system issues repeated alerts to providers when they open the patient’s medical chart. In a recent study, the tool was as effective as a health provider at successfully identifying hospitalized adults at risk for opioid use disorder and recommending referrals to inpatient addiction specialists. What’s more, patients identified by AI screening for referrals who then received consultations had 47% lower odds of being readmitted to the hospital within 30 days of their initial discharge, compared to patients who received provider-initiated consultations. The study suggests that under the right circumstances, AI may help health care systems increase access to addiction treatment, allowing patients to get the help they need faster and cost-effectively.
More men than women get a lethal form of brain cancer called glioblastoma, and their tumors are often more aggressive than women’s tumors. Pallavi Tiwari, PhD, an assistant professor in the Departments of Radiology, Biomedical Engineering and Medical Physics, turned to digital images of pathology slides — thin slices of tumor samples — in search of patterns that might forecast how quickly a tumor could grow and thus how long a patient might expect to survive. Tiwari and team built an AI model that can identify subtle patterns in pathology slides. Using data from more than 250 studies of glioblastoma patients, they trained the model to recognize tumors’ unique characteristics, such as the abundance of certain cell types, and to identify patterns between these characteristics and patients’ survival time while accounting for their sex. The result was an AI model capable of identifying risk factors for more aggressive tumors strongly associated with each sex, which could help lead to more individualized care for glioblastoma patients.
Daifeng Wang, PhD, associate professor of biostatistics and medical informatics, helped create a powerful new machine learning algorithm to help researchers untangle a complex puzzle: the cellular and molecular mechanisms underlying developmental disabilities and neurodegenerative diseases. Traditional machine learning algorithms struggle when working with more than one large, complicated dataset to tease out relationships between various molecular activities, such as those that occur with Alzheimer’s disease, autism and other neurodevelopmental disabilities. To better understand how the mechanisms interact and influence one another to cause disease or clinical phenotypes, which are the observable hallmarks of disease, members of Wang’s lab designed COSIME (Cooperative Multi-view Integration with Scalable and Interpretable Model Explainer), a nimble machine learning algorithm designed to analyze two large datasets simultaneously. The tool’s unique two-part structure — the first component makes predictions, while the second component analyzes how different genes and cell types work together— makes it particularly powerful. By pinpointing parts of the dataset that influence disease outcomes, COSIME could significantly improve researchers’ capacity to study a variety of complex conditions.
Machine learning and generative AI are reshaping how researchers like Dr. Jeff Nirschl, assistant professor of pathology and laboratory medicine, study Alzheimer’s disease and related dementias. The Nirschl Lab develops advanced machine learning, computer vision, and generative AI approaches to analyze large-scale datasets derived from the study of Alzheimer’s disease at the microscopic level. The models capture the presence, distribution and interaction of diseased cells and molecules within their native tissue environment, a process known as deep phenotyping. Many patients with Alzheimer’s or Parkinson’s disease often have multiple co-pathologies, such as amyloid plaques, tau tangles, Lewy bodies, and vascular lesions. By quantifying these proteins on histological stains, AI models may enable doctors to more confidently link brain tissue findings to other biomarkers emerging from blood, fluid, and imaging tests, improving the accuracy of disease stage diagnosis. A major focus of the lab’s recent work is building AI systems that can flag when AI may be unreliable, helping to ensure safe clinical use. In 2025, Dr. Nirschl received a Developmental Project award from the Wisconsin Alzheimer’s Disease Research Center to integrate these AI-driven tools into modern neuropathology pipelines.
Most women with ovarian cancer are not diagnosed until stage 3, when cancer has spread outside the pelvis, or stage 4, when cancer has spread outside of the abdominal cavity to distant organs like lungs or brain, contributing to poor patient outcomes. This is because ovarian cancer symptoms often mimic other common, less serious conditions. Irene Ong, associate professor of biostatistics and medical informatics and obstetrics and gynecology, is leveraging AI to develop a decision-support system that primary care doctors and specialists can use to make earlier, more accurate diagnoses of this deadly disease. Ong and her team use machine learning methods to comb through electronic health records (EHR), training models to better discriminate between patients with ovarian cancer and control groups with similar symptoms and behavior. Their dataset includes physician notes, labs and any other information collected when patients visit their doctor. The algorithm has identified certain symptoms and data points that can distinguish non-cancer patients and women with undetected cancer. Ong also hopes to include genomic information to potentially identify women at risk for ovarian cancer so that they, and their doctors, can be proactive in screening for the disease.
