Three-Dimensional Whole Breast Segmentation in Sagittal and Axial Breast MRI With Dense Depth Field Modeling and...

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate whole breast segmentation in MRI is crucial for quantitative analysis and cancer risk assessment.
• Locating the chest-wall line (CWL) is a significant challenge due to its variable appearance and surrounding structures.
• Existing methods struggle with the spatial variations and artifacts that affect CWL identification.

Purpose of the Study:

• To introduce DeepSeA, a novel automatic 3D segmentation method for whole breast MRI.
• To address the challenges of CWL localization in breast MRI segmentation.
• To provide a robust and adaptable solution for breast MRI analysis.

Main Methods:

• DeepSeA reformulates CWL localization as a smooth depth field optimization problem, leveraging 3D continuity.
• A localized self-adapting algorithm is employed to handle spatial variations in the CWL.
• The method is designed to work equally well on both sagittal and axial MRI orientations without requiring specific training.

Main Results:

• DeepSeA achieved high accuracy in segmenting whole breasts across 99 diverse MRI scans.
• Average performance metrics include Dice similarity (96.04%), sensitivity (97.27%), specificity (98.77%), and CWL deviation error (1.63 mm).
• The method's configuration was generalized for potential application to broader datasets.

Conclusions:

• DeepSeA offers a fully automatic and accurate solution for whole breast segmentation in both sagittal and axial breast MRI.
• The developed method demonstrates significant potential to enhance quantitative analysis for breast cancer risk assessment using MRI.