Doctoral students Tonushree Dutta and Sourav Kumar are among winners at health data competition

Graduate students Tonushree Dutta and Sourav Kumar Ghosh (both of ECE) and Ingrid Rodriguez Aragon of the University of Minnesota Medical School won second place at the 6th Annual Interdisciplinary Health Data Competition (IHDC). The theme of the competition this year was “Healthcare Provider Deep Dive: Connecting Quality, Payment, and Social Context.” The competition is hosted by the Business Advancement Center for Health.

Guided by the goal to promote interdisciplinary collaboration, the IHDC requires that each participating team comprise students from at least two different colleges or schools. The teams have to  address real world healthcare challenges through data analysis. This year participants were given curated datasets by competition data partner mimilabs. The goal was to analyze and identify issues that can be addressed through data-driven solutions. 

The second place team’s presentation was titled, “Geographic Disparities in Rheumatoid Arthritis Prevalence, Costs, and Quality of Care Among US Adult Medicare Beneficiaries.” We had the opportunity to catch up with Dutta and Ghosh and learn about their research interests, how they approached the competition, and the solutions they suggested that made them a winning team in the competition. 

Tell us a little bit about yourself.
Tonushree Dutta: I am originally from Bangladesh, a small, beautiful country in south-east Asia. Before joining the University of Minnesota, I completed my bachelor’s degree at Bangladesh University of Engineering and Technology. Currently, I am pursuing my doctoral degree under the supervision of Professor John Sartori.

Sourav Kumar Ghosh: I am a Ph.D. candidate doing my research under Professor John Sartori’s supervision. I completed my master’s degree in electrical and computer engineering from the University of Minnesota Twin Cities. Prior to that, I earned my bachelor’s degree in industrial and production engineering from Bangladesh University of Engineering and Technology. I have been actively engaged in research and teaching roles as a graduate research assistant and graduate teaching assistant at the University since 2022, focusing on machine learning applications in healthcare and high-impact computational problems.

What are your academic and research interests?
Tonushree Dutta: My research interests include machine learning in healthcare, processing physiological signals for machine learning models, and healthcare devices. My focus is on developing machine learning models for predicting pain levels in patients with chronic pain using physiological signals such as heart rate variability, temperature, and changes in circulation and skin conductance. This work is part of the PhysiAroma project, a collaboration with PhysiAroma LLC, which aims to integrate AI-driven scent delivery systems for personalized pain management. Currently, I am working on individualized models, but the long-term goal is to develop a global model by normalizing pain levels across different individuals. The potential impact of my work lies in improving pain assessment and management through objective, data-driven approaches, ultimately enhancing patient care. It could be groundbreaking in the field of pain management since quantifying pain has not yet been feasible because of its inherent subjective quality. The project leads are my advisor Professor John Sartori and Beth Groenke who is an assistant research professor with Minnesota Dental Research Center for Biomaterials and Biomechanics at the University.

Sourav Kumar Ghosh: My research focuses on machine learning (ML), Deep Learning, and generative artificial intelligence (AI) with a strong emphasis on healthcare applications. I aim to develop AI-driven healthcare solutions, such as an Omnidoc, a digital AI assistant to automate clinical workflows and enhance real-time decision-making using multimodal medical data (electronic health records, images, clinical notes, etc). I am passionate about leveraging large language models (LLMs) and reinforcement learning to improve diagnostic accuracy and healthcare efficiency. For instance, my projects include sleep stage analysis via clustering algorithms, convolutional neural network-based X-ray classification, and gene expression data analysis. The potential impact of my work lies in reducing clinician workload, improving patient outcomes, and advancing AI’s role in precision medicine. Professor Sartori supervises my research.

What are your plans for the future? 
Tonushree Dutta: In the near term, I am focused on completing my Ph.D. and refining machine learning techniques for healthcare applications. In the long term, I am interested in working in the healthcare industry, particularly in developing AI-powered healthcare devices and leveraging healthcare data to improve patient outcomes. I enjoy working with healthcare-related data, whether it’s provider data, EHR data, medical imaging, or physiological signals. My goal is to process and analyze these data sources to develop impactful machine learning models that enhance patient care and clinical decision-making.

Sourav Kumar Ghosh: Right now, I plan to complete my Ph.D. while continuing to publish research on ML applications in healthcare and optimization. In the long-term, I aspire to join the healthcare industry as a data analytics professional, where I can apply my expertise in machine learning, LLMs, and generative AI to develop scalable solutions for clinical decision support, medical diagnostics, and personalized care. I aim to bridge the gap between cutting-edge AI research and real-world healthcare challenges, contributing to innovations that enhance patient care and operational efficiency in medical systems.

We understand that this is the second time that you have participated in the IHDC. What is it about the competition that first got you interested? 
We first learned about the IHDC in the "Machine Learning for Healthcare" class taught by Professor Yogatheesan Varatharajah in the Computer Science department. He announced the competition in class, and after we won second place last year, he shared our work with the class. Given our backgrounds in machine learning, healthcare analytics, and medical research, we saw the competition as an exciting opportunity to apply our skills to real-world public health challenges.

We returned this year with refined techniques and strategies, securing second place again. The competition’s focus on health equity, rural-urban disparities, and data-driven policy recommendations directly aligned with our research making it the perfect platform to showcase our expertise.

How did your own research help you prepare for the IHDC? 
Our research backgrounds in healthcare data analysis have been crucial in preparing us for IHDC. Sourav’s work on generative AI, LLMs, and statistical analysis for healthcare data paired with my expertise in predictive modeling and health data analysis provided us with the necessary tools to tackle complex datasets and extract meaningful insights. This experience allowed us to refine our analytical approaches and apply advanced techniques, making us well-prepared to compete in the challenge.

What issues did you identify and what recommendations did you offer at the IHDC? 
We analyzed disparities in Medicare costs for rheumatoid arthritis (RA) patients and found that rural beneficiaries incur higher healthcare expenses. A key reason is that most providers serving rural beneficiaries are located in urban areas, leading to increased costs due to differences in reimbursement rates, higher operational expenses, and travel burdens. Limited healthcare resources in rural areas also result in delayed care, causing disease progression that requires more intensive and costly treatments.

To address these issues, we proposed an ML-based solution to identify rural providers who are more likely to serve high-risk patients, allowing them to receive better preparation and resources. Additionally, we recommended expanding rural healthcare infrastructure, increasing the use of telemedicine, implementing accountable care organizations (ACOs) in rural areas to improve care coordination and reduce hospitalizations, enhancing non-emergency medical transportation (NEMT) services, and providing early intervention programs to detect and manage RA at an earlier stage to reduce long-term treatment costs. 

How did the three of you coordinate your skills/what strengths did you bring to the table?
Our team brought a diverse and complementary skill set to the competition: Ingrid R. Aragon, being a Ph.D. candidate in Integrative Biology and Physiology program, provided medical expertise and healthcare policy knowledge, ensuring that our interpretations and recommendations were clinically relevant and aligned with real-world challenges.

Sourav and I focused on data analysis, model development, and solution strategies, with Sourav handling complex data preprocessing, statistical modeling, and leveraging machine learning models, while I worked on extracting meaningful insights from healthcare datasets, predictive modeling, and feature engineering.

What makes our team unique and efficient is that all three of us are truly passionate about the work and open to constructive criticism. We welcome suggestions from each other and use them to improve our approach. This collaborative environment allowed us to refine our workflow, delegate tasks effectively, and present our findings in a more impactful way. And also that is why we could work as a team two times in a row and win second place. 

The competition was evaluated by a panel comprising both faculty members and industry experts, with finalists presenting their findings to a distinguished judging panel. The final round judges included Heather Britt, Executive Director at Wilder Research; Hayley Borck, Managing Director at the Data Science Initiative at the University of Minnesota; Dr. Yubin Park, CEO Mimilabs and Data Partner; and Dr. Benjamin Lynch, Director of the Minnesota Supercomputing Institute at the University of Minnesota.

The event was made possible by the generous support of the University of Minnesota Research & Innovation Office, Research Computing, the University of Minnesota Data Science Initiative, and the Office of Academic Clinical Affairs.