Suffolk Reporter

A team of researchers at Stony Brook University is developing a new method to analyze breast cancer imaging using mathematical modeling and deep learning. The project, led by Chao Chen, PhD, Associate Professor, and Prateek Prasanna, PhD, Assistant Professor in the Department of Biomedical Informatics, aims to create more interpretable and robust predictive disease models. Their goal is to improve diagnosis and develop treatment plans specific to biomarker imaging and modeling findings.

The research focuses on understanding breast tissue architecture and its changes over time. Breast tissue complexity often makes subtle changes difficult for clinicians to detect with standard imaging. To address these challenges, the researchers are developing "TopoQuant," a suite of informatics tools built on advanced mathematical modeling and machine learning. TopoQuant will help analyze the structural complexity of breast parenchyma.

Supported by a four-year $1.2 million grant from the National Cancer Institute (NCI), this project will run through August 2028. Both Chen and Prasanna are affiliated with the Stony Brook Cancer Center’s Imaging, Biomarker and Discovery and Engineering Sciences Research Division.

“This research will offer new insights into how structural changes in breast tissue can influence cancer screening and treatment outcomes,” says Chen. “Topology is the area of mathematics that studies structures. By incorporating topology with deep learning in a seamless fashion, we can develop novel algorithms to capture structural changes in ways that were previously difficult with traditional techniques such as textural radiomics, potentially leading to better predictive models and treatment strategies.”

Existing machine learning-driven tools used by cancer imaging researchers lack interpretability or explanation capacity. In contrast, TopoQuant aims to provide clinicians with quantitative evidence of changes in breast tissue architecture related to cancer risk and treatment response.

In preliminary findings published in 2021, the team demonstrated their approach's efficacy using one of their informatics tools in predicting patient responses to neoadjuvant chemotherapy for breast cancer. The study suggested differential topological behavior between patients who responded favorably to therapy and those who did not.

“Our prediction models will be unique in that they do not rely on traditional post-hoc interpretation but ensure interpretability by design,” explains Prasanna. “The research is intended to not only benefit breast cancer diagnosis and treatment but will also have broader applications in fields like neuroscience. Therefore, we are excited about the cross-disciplinary collaborations this project will foster and the new avenues it will open for medical imaging research.”

Other collaborators include Alexander Stessin from the Department of Radiation Oncology; Wei Zhao from the Department of Radiology; and Haibin Ling from the Department of Computer Science within CEAS.