Contextual Regularization-Based Energy Optimization for Segmenting Breast Tumor in DCE-MRI

Accurate breast tumor segmentation is crucial for precise diagnosis, effective treatment planning, and the development of automated decision-support systems in clinical practice. The imprecision of trending segmentation models in differentiating tumors from their surrounding tissues, particularly in weighing the boundary pixels across tumor regions poses a significant challenge in precise tumor delineation. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) effectively captures tumor vascularity and perfusion dynamics and is a reliable modality for extracting the region of interest (ROI). Nevertheless, the intricate intensity variations in DCE-MRI owing to heterogeneous tumor morphology pose considerable challenges in tumor delineation, necessitating a highly adaptive and robust model for precise tumor segmentation. Accordingly, this manuscript presents a Contextual Regularization-Based Energy Optimization (CRBEO) model that effectively captures these intensity variations in the form of energies contributed by data fidelity and regularization terms. The formulated non-linear energy-based convex optimizer is adaptively tuned by a variational Minimax principle to achieve the desired solution. An iterative gradient descent algorithm is engaged to minimize the energy-based cost function, obtaining stable convergence towards the optimal solution. The extensive relative analysis of CRBEO on complex breast DCE-MRI datasets including QIN breast DCE-MRI, TCGA-BRCA, BreastDM, RIDER, and ISPY1 has recorded significant dice improvements of 30.16%, 11.48%, 20.66%, 1.012%, and 28.107%, respectively on par with trending SOTA methods. The complexity analysis of CRBEO with time and space has justified its extension to real-time clinical diagnosis.