Automated 3D Tumor Segmentation from Breast DCE-MRI using Energy-Tuned Minimax Optimization

• Published in: IEEE access Vol. 12; p. 1
• Main Authors: Babu, Priyadharshini, Asaithambi, Mythili, Suriyakumar, Sudhakar Mogappair
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Analytical models; BEFVBTS; Breast cancer; Complexity theory; Datasets; DCE-MRI; Energy Functionals; Image segmentation; line search; Magnetic resonance imaging; Mathematical models; Metric space; Minimax technique; Optimization; Real time; Regularization; Search algorithms; Three-dimensional displays; Tumors; variational minimax optimization
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3417488

Breast cancer (BC) is a multifaceted genetic malignancy that accounts for the majority of cancer fatalities in women. Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is predominant in evaluating perfusion, extravascular-extracellular volume fraction, and microvascular vessel wall permeability in breast cancer patients. Precise tumor segmentation using DCE-MRI is a key component of assessing diagnosis and treatment planning. However, the slice-wise analysis of DCE-MRI fails to preserve 3D surface continuity and is insufficient for evaluating the invasion depth of the tumor. Hence, this work proposes an analytical model labeled as Bezier-tuned Energy Functionals optimized via variational minimax for Volumetric Breast Tumor Segmentation (BEFVBTS). The formulated energy functionals consist of non-linear convex edge-sensitive data and regularization terms. Also, the variational minimax technique adopts gradient descent with an exact line search algorithm for obtaining a global minimax solution. The self-analysis of BEFVBTS on the Duke- Breast-Cancer-MRI dataset registered remarkable performance in segmenting tumors with different grades (Grade 1,2 & 3). Likewise, the relative analysis on QIN Breast DCE-MRI and TCGA-BRCA datasets revealed improvements of 8%, 22%, 8.7%, 4%, 0.120%, and 68.17% in Dice, Jaccard, Precision, Sensitivity, Specificity, and Hausdorff distance (HD) respectively over the recent competitors. At last, the complexity analysis of the model demonstrated simplicity and amicability for its extension to real-time clinical applications.