Rising STARs Award Launches DIVA-AI Lab at UTMB

Vibhuti Gupta, PhD, associate professor in the Department of Biostatistics and Data Science at the School of Public and Population Health, received the University of Texas System Rising STARs award on April 28. The award funds the establishment of his Discovery and Innovation through Visual Analytics-AI (DIVA-AI) Laboratory at UTMB, where he will lead the development of multimodal artificial intelligence methods designed to integrate medical images, clinical notes, genomics, wearables, and other data into a single unified view of a patient.

The lab takes on two long-standing challenges that have slowed the clinical adoption of AI. The first is data complexity, the difficulty of combining heterogeneous, often incomplete biomedical data into a model that learns anything useful. The second is the black box problem, the tendency of high-performing AI models to produce predictions that clinicians cannot interpret or trust.

A computer scientist's path into healthcare research

Dr. Gupta's training began in computer science. He completed his PhD at Texas Tech University in 2019 with a dissertation on big data stream analytics. During those years, he worked outside his dissertation on a series of smaller healthcare-related projects, and the experience reshaped his sense of what computer science could be used for.

"I chose to focus on healthcare research over industry because of its potential for real-world impact," Dr. Gupta said. "I was motivated by the possibility of applying artificial intelligence and data science to some of society's most pressing health challenges, where even small advances can make a meaningful difference in patients' lives."

That conviction guided his next move. Dr. Gupta took a postdoctoral fellowship at the University of Michigan in the Department of Pediatrics, Hematology/Oncology Division, where he worked on a project asking whether continuous physiological data from Fitbits and smartphones, paired with survey responses from a mobile health app, could identify post-transplant risks early in patients undergoing hematopoietic cell transplantation.

The same project examined the strain on family caregivers, who support transplant patients around the clock and often carry their own physical and mental health burden in the process. Dr. Gupta's contribution focused on linking the wearable signals with the survey data, and the Michigan collaboration has continued for five years.

He moved to Meharry Medical College in January 2021 as the founding assistant professor in the newly launched School of Applied Computational Sciences, the fourth school in the institution's history. He was one of the first three faculty members appointed in the new school. He also directed the school's mHealth Wearable Sensors Lab from its founding, continuing the research thread he had started at Michigan.

The early years at Meharry were spent learning the mechanics of academic research funding. Dr. Gupta served as a peer reviewer for the National Science Foundation, contributed to grants as a co-investigator, and studied the patterns that separated funded proposals from rejected ones. His first grant as principal investigator came in 2021, a pilot award from the American Cancer Society to develop deep learning methods for detecting virus integration sites in tumor genomes.

In 2023, Dr. Gupta was awarded an NIH AIM-AHEAD consortium development award of roughly $1.27 million to build a multimodal framework for prostate cancer risk prediction. That award also introduced him to Dr. Suresh Bhavnani, who held a parallel AIM-AHEAD award at UTMB. Their conversations through the AIM-AHEAD investigator network are what eventually brought Dr. Gupta to Galveston in September 2025.

How the Rising STARs award works

The Rising STARs award is one of two categories within the University of Texas System Science and Technology Acquisition and Retention program, funded by Permanent University Fund bond proceeds. Awards in this category provide up to $300,000 for tenure-track faculty at any academic rank, restricted to equipment and laboratory renovations with a useful life beyond one year. UT System distributes the funds annually to each academic institution, and each institution administers its own selection process.

At UTMB, applications come from newly joined faculty across all schools and are evaluated collectively by the deans.

Inside the DIVA-AI Lab

The DIVA-AI Lab represents a methodological expansion of an existing UTMB facility. Dr. Suresh Bhavnani received his own Rising STARs award years earlier and used it to establish the original DIVA Lab, focused on visual analytics methods that translate complex statistical models into forms interpretable to clinicians and researchers. Dr. Gupta's lab will add multimodal foundation modeling to the work already underway. The renaming to DIVA-AI signals the new scope.

The central technical contribution is a pipeline the team calls Multimodal Hybrid AI, or MoHAI. The pipeline uses deep learning to handle the complexity of biomedical data, including heterogeneous formats, missing values, and per-patient variation in which signals carry the most information, then uses machine learning to deliver the interpretability that clinical applications require. The two methods do different jobs at different points in the pipeline, and the architecture reflects what Dr. Gupta has identified as a gap in much of the current clinical AI literature.

On the deep learning side, the lab will draw from national repositories including All of Us, TriNetX, the Medical Information Mart for Intensive Care, The Cancer Genome Atlas, The Cancer Imaging Archive, the Surveillance, Epidemiology, and End Results program, and the UK Biobank, alongside local resources such as the UTMB Moody Brain Health Initiative and partnerships with the Houston Health Department.

Each data type passes through a modality-specific encoder, itself a foundation model trained for that kind of data, whether the input is an image, a clinical note, a time-series stream, or a tabular record. The encoder outputs are fused into a unified patient embedding, and graph neural networks with attention-based message passing capture relationships across patients while handling missing data. The resulting model is what the field calls a disease-agnostic multimodal foundation model, with the potential to provide a general patient representation that can be adapted to many different conditions. The lab will evaluate this foundation model for bias and completeness across patient cohorts and healthcare settings.

On the machine learning side, the foundation model becomes a feature extractor that disease-specific models can fine-tune for tasks like patient subtyping, stratified risk prediction, and causal discovery. These downstream models are the ones clinicians and researchers would actually use in applied settings. They are simpler, more interpretable, and grounded in the rich representations the foundation model has already learned.

The third pillar is dissemination. The MoHAI pipeline, the foundation model, the curated datasets, and the supporting documentation will be released as open-source software for use by researchers across the UT System and beyond. The team plans tutorials, workshops co-developed with the UTMB AI Center, and an online course that will give students and practitioners hands-on experience with multimodal modeling.

From artificial intelligence to multimodal models

Hospitals, research universities, and the general public are all hearing the same vocabulary right now. Artificial intelligence, machine learning, deep learning, large language models, foundation models. The terms get used interchangeably, but they describe distinct things that nest inside one another. Part of what Dr. Gupta brings to UTMB is the ability to explain these concepts in plain terms to audiences without a computer science background. Here is how the pieces fit together.

Artificial intelligence is the broadest term. The phrase was coined in 1955 by John McCarthy, who organized the 1956 Dartmouth Summer Research Project widely credited as the birthplace of AI as a formal academic discipline. The intellectual roots trace further back to a 1950 paper by Alan Turing that asked whether machines could think. The original goal of the field was to build machines that can sense, reason, act, and adapt the way humans do. The field today extends well beyond clinical applications into robotics, autonomous vehicles, and other areas that have little to do with healthcare.

Machine learning is a subset of artificial intelligence. Where the broader AI field has historically included rule-based systems and symbolic reasoning, machine learning takes a different approach. It feeds large amounts of data into algorithms that learn patterns directly from that data and apply those patterns to new cases without further human intervention. The vast majority of what is described today as AI in healthcare is, more precisely, machine learning.

Deep learning is a subset of machine learning, specifically the methods built on neural networks with many layers. The depth of those layers gives the field its name. Deep learning is what makes it possible to analyze unstructured data like medical images, audio, and free-text clinical notes, the kinds of data that earlier machine learning methods could not handle directly. Generative AI and the large language models behind tools like ChatGPT sit inside this category as well.

Multimodal AI is where the DIVA-AI Lab does most of its work. The idea is straightforward in principle. Build models that can take in many different kinds of data at once, such as images, text, time-series, and tabular records, and produce a single integrated representation of the patient. The technical challenge is in the fusion. Each modality has its own structure, its own missing-data patterns, and its own information content, and combining them well requires careful methods.

Foundation models are the most recent layer of the picture. A foundation model is pre-trained once on enormous volumes of data and then fine-tuned for many downstream tasks. The advantage is that researchers no longer need to train a fresh model for every new disease, dataset, or hospital. A model trained on imaging data across many different conditions can be adapted to a new condition with a much smaller specialized dataset. This is what makes the DIVA-AI pipeline disease-agnostic and scalable.

The same lab work touches different groups at UTMB in different ways.

Where the lab is heading

The work ahead falls into three broad phases. The first centers on building and evaluating the multimodal foundation model itself, including the high-performance computing infrastructure to support large-scale data analysis and the curation of the national-level datasets that will train it. The second brings the disease-specific machine learning models online and tests their performance against unimodal baselines, quantifying how much each data modality contributes to a given clinical task. The third phase focuses on getting the methods into other people's hands through workshops, tutorials, an online course, and open-source software releases.

Dr. Gupta has identified collaborations across UTMB and beyond as a priority from the start. The lab will work with the UTMB AI Center, the Sealy Center on Aging, Information Technology Services, and faculty at SPPH and the John Sealy School of Medicine, alongside continuing partnerships with the University of Michigan, Meharry Medical College, the Houston Health Department, and Galveston County.

"I believe multimodal AI will redefine the future of medicine by seamlessly integrating vast and heterogeneous health data, learning holistic representations of patient health, and revealing complex biological and clinical patterns that are difficult for humans to detect alone," Dr. Gupta said. "Through the DIVA-AI Lab, we aim to bridge the gap between AI innovation and clinical impact by developing robust, interpretable, and equitable multimodal AI models that support trustworthy decision-making can be effectively translated into routine healthcare practice and capable of improving patient care at scale."