IU-led study uses AI to predict breast cancer risk while addressing health inequalities: IU News

INDIANAPOLIS — Researchers at Indiana University School of Medicine are leading a multi-site study using a privacy-preserving artificial intelligence approach called federated learning to improve breast cancer risk prediction and reduce health disparities in cancer screening. The research is funded by a new five-year, $3.7 million grant from the National Cancer Institute of the National Institutes of Health.

Researchers in the joint study will use federated learning to analyze and learn from data collected at participating sites. Their goal is to predict breast cancer risk while gaining insights across diverse patient populations. Photo courtesy of IU School of Medicine“We will have an AI methodology that can contribute to the future of women’s health,” said Spyridon Bakas, director of the Division of Computational Pathology at the IU School of Medicine and principal investigator on the project. “Federated learning is a novel paradigm for multi-site collaborations like this one because it allows access to large-scale and, more importantly, diverse data essential for developing robust models without the need to share patient data across sites. This grant will enable us to leverage federated learning technology and develop an improved breast cancer risk assessment model aimed at predicting breast cancer and transferable to multiple patient populations.”

Federated learning is a mechanism for collaboratively training complex AI models using data that remains decentralized, meaning it never leaves the relevant institution, which increases data privacy. Bakas said this creates more trust and mitigates patient privacy concerns.

The other institutions involved in this joint study are the Mayo Clinic, Washington University in St. Louis, the University of Pennsylvania and Columbia University. Each site will provide anonymized data from patients undergoing 3D digital breast tomosynthesis, a breast cancer screening method that is becoming more widely used than traditional 2D digital mammography. The researchers will use federated learning to analyze and learn from the data collected at all participating sites. They will then build an open-source AI model with the goal of predicting breast cancer risk while gaining insights across different patient populations.

Breast cancer is the second leading cause of cancer-related deaths in women. The data used in the project come from patients who undergo breast cancer screenings. Some of them develop cancer over time, others do not.

The project’s objectives for the next five years include:

• Develop breast cancer risk assessment models that use data from women undergoing breast cancer screening from multiple sites and from different ethnic groups.
• Improving these initial models by including additional geographically diverse locations.
• Generate realistic synthetic image data that matches the characteristics of each site’s local patient population and use it for data augmentation and privacy protection.
• Creating an automated mechanism to quantitatively and interpretably determine optimal data protection in healthcare AI models.

Spyridon Bakas. Photo courtesy of IU School of Medicine

“The goal of our models will be to predict much earlier when and if a woman will develop breast cancer and to estimate her risk of developing breast cancer in the future,” Bakas said. “We are focusing more on prediction than diagnosis and are proactive rather than reactive.”

The researchers are also focusing on developing AI models that take health disparities and inequities into account, as many patients do not have access to a comprehensive healthcare system. “These models typically cannot be trained in a community hospital because they lack the resources,” Bakas said. “With federated learning, we gain this knowledge from different populations and then distribute the AI ​​model to other community facilities for application.”

“Our overarching goal for this study is to create an easy-to-use, transferable, and trusted federated learning framework that lowers the barrier for underserved populations to participate in large-scale federated learning studies and benefit from such technological advances, paving the way to eliminating health disparities.”

Other study co-principals include Despina Kontos of Columbia University, Celine Vachon of the Mayo Clinic, Aimilia Gastounioti of Washington University, Anne Marie McCarthy of the University of Pennsylvania, and Prashant Shah of Intel.

What they say:

“The era of digital transformation in healthcare has been driven by open-source software tools developed by the scientific community. These tools have not only democratized AI by making it accessible, but have also encouraged healthcare researchers to explore how reproducibility and robustness can positively impact treatment outcomes. By leveraging open-source tools and making our trained AI models publicly available, we are fostering a culture of collaboration and transparency essential to innovation while building a better future together.” — Sarthak PatiSoftware architect at Indiana University

“Federated learning enables models to learn from limited data from numerous collaborators while overcoming concerns about data ownership, privacy, and regulations. It enables the collection of meaningful data sets for rare diseases that can help build robust machine learning models that work across diverse populations, thereby reducing health disparities and inequities.” — Prashant ShahHead of Artificial Intelligence at Intel Health and Life Sciences