The 2026 Institute Seed Funding Awardees
Introduction
We are pleased to announce that the following four teams have each been awarded $50,000 each through the 2026 Institute for Health Data Science Seed Funding program.
All team members participated in a day-long Institute incubator workshop focused on projects related to artificial intelligence (AI) and heart, lung, blood, and sleep disorders. The overarching goal of this initiative is to support the development and submission of a competitive NIH National Heart, Lung, and Blood Institute (NHLBI) AI Center proposal in 2027. The seed funding proposals have been reviewed by the Institute directors.
Natural Language Processing to Evaluate Irregular Rhythm Notifications and Healthcare Engagement
David Sanders (College of Medicine), Barbara Di Eugenio (College of Engineering)
A multitude of consumer-grade wearable devices capable of detecting atrial fibrillation are available, but the downstream impact of an irregular rhythm alert is unknown. We intend to develop natural language processing methods to retrospectively identify patients who have clinical encounters related to an irregular rhythm alert, from free text clinical notes. We will subsequently analyze the patient characteristics, pattern of healthcare engagement, and consequent health utilization, that we will also identify from structured and unstructured data in the patient’s Electronic Health Record.
Smart Echo: Multimodal AI for Decisive Aortic Valve Assessment
Hema Krishna (College of Medicine), Sathya Ravi(College of Engineering)
This project will develop artificial intelligence (AI) tools to improve the diagnosis of Aortic Stenosis, a dangerous narrowing of a heart valve. Aortic stenosis can be difficult to detect with standard heart ultrasound, especially when blood flow through the heart is reduced, causing severe disease to appear milder than it really is. We are developing transformer based models that can help ultrasound technicians and doctors identify these cases more accurately so that patients who need a lifesaving valve replacement are not missed. Our tools will ultimately guide ultrasound technicians in real time toward the measurements needed for an accurate diagnosis.
Platform for Integration of Biospecimens-Clinical Data and Validation of Algorithms: Case of Lung and Atherosclerotic Cardiovascular Disease Respectively
Kal Pasupathy (College of Applied Health Sciences), Yue Huang (Research Resources Center), and Bharati Prasad (College of Medicine)
Models developed in one institution do not transfer over to another institution, thereby being restricted to the patient population that they are trained on. There is documented evidence of differences by age, sex and race/ethnicity. This leads to algorithmic bias in the models being developed. Further, there are challenges in retrieving specimen information and linking to the corresponding clinical records, thereby hampering research & discovery due to a silo’d approach. This project proposes to address these issues by developing an integrated specimen-clinical data platform that helps address algorithmic bias and furthers clinical relevance and translation.
Leveraging AI for Obstructive Sleep Apnea Screening with Photoplethysmography
Bharati Prasad (College of Medicine), Wei Tang (College of Engineering), and Kal Pasupathy (College of Applied Health Sciences)
Obstructive sleep apnea (OSA) has a global prevalence of 1 billion and disproportionately affects minority populations served at UI Health. Artificial Intelligence (AI)- aided software and care pathways are rapidly changing healthcare delivery in sleep medicine, particularly for OSA. Our research program aims to leverage Artificial Intelligence (AI) to improve the efficiency of Obstructive Sleep Apnea (OSA) screening. This pilot project will develop a Deep Learning (DL) model for OSA screening using photoplethysmography (PPG), a physiologic signal recorded in clinical polysomnography. We will explore the utility of Fitbit-derived sleep, activity, and heart rate variables for screening for OSA using standard statistical and machine-learning models. Both aims will implement demographic-aware training and evaluation.