SCH: Topological Methods for Breast Tissue Quantification - To better understand breast cancer and its response to treatment, the key is to understand breast tissue architecture. Modifications to tissue architecture are a direct consequence of the rearrangement of fine- grained structures such as the fibroglandular tissue and vessels, brought about by events including, but not limited to angiogenesis, and treatments such as radiation therapy. Changes in breast structural topology has the potential to influence cancer risk, prognosis, and treatment response; however, this has not been extensively studied nor quantified. Existing work is limited only to analysis of features such as radiomics. This project, undertaken by a multi-disciplinary team comprising topologists, computer scientists, imaging informatics experts, and clinicians, aims to develop advanced methods for topological modeling and reasoning with the primary hypothesis that these algorithms can ascertain weakening of tissue architecture on imaging. This can help in identifying high-risk cases that are prone to cancer manifestation and cancer recurrence. Specifically, the project proposes to develop TopoQuant, a suite of Topology Data Analysis (TDA)-driven techniques for extracting and interpreting fine-grained topological information from breast parenchyma, based on both 2D and 3D breast imaging. TopoQuant will generate high-quality annotations of breast tissue, produce advanced topological descriptors to characterize breast tissue complex, learn topology-informed prediction models using these topological features, as well as provide clinically intuitive visualization of relevant topological features. The proposed work will advance both TDA and cancer research by create new TDA methodologies to extract and analyze rich structural information from breast imaging. This will be achieved through: (1) Developing novel topological algorithms to effectively capture the structural diversity in breast parenchyma; (2) Addressing the challenge of limited data availability by devising 2D to 3D mapping methodologies that yield highly actionable topological information even when the available data is sparse or restricted; (3) Developing algorithms to co-harness the powerful learning ability of ‘black-box’ deep networks with biologically-grounded ‘glass-box’ topological descriptors; and (4) Constructing interpretable TDA frameworks specifically tailored for assessing cancer risk and radiation treatment response in breast tissue. Our algorithms will reveal topological insights from breast tissue structures, enabling clinicians and researchers to better comprehend, plan, and evaluate the effectiveness of treatments in breast cancer patients. RELEVANCE (See instructions): The research will enhance our understanding of breast cancer, potentially improving early detection and treatment, thereby benefiting women’s health and quality of life. The novel methods for analyzing radiology scans will have significant implications across multiple disciplines, including neuroscience and biology, and promote cross-disciplinary collaboration. Algorithms developed will be shared with the scientific community, and integrated into different platforms.
