Automated deep learning method for whole-breast segmentation in contrast-free quantitative MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Developing automated segmentation for breast MRI is crucial for efficient clinical analysis.
• Diffusion-weighted imaging (DWI) and synthetic MRI (SyMRI) offer valuable quantitative data.
• Deep learning architectures show promise for medical image segmentation tasks.

Purpose of the Study:

• To develop and evaluate the nnU-Net deep learning architecture for fully automated whole-breast segmentation.
• To compare the performance of nnU-Net against the U-Net architecture.
• To assess segmentation accuracy using both DWI and SyMRI data.

Main Methods:

• nnU-Net and U-Net deep learning algorithms were applied to segment 196 breasts from 98 patients.
• Data included 3.0T MRI scans with DWI and SyMRI sequences.
• Performance was quantified using Dice Similarity Coefficient (DSC), accuracy, and Pearson's correlation coefficient.

Main Results:

• nnU-Net significantly outperformed U-Net in whole-breast segmentation for both DWI and SyMRI (PD).
• SyMRI (PD) demonstrated superior performance over DWI, achieving the highest DSC and accuracy.
• nnU-Net achieved excellent correlation coefficients (R² 0.99–1.00) for DWI and SyMRI (PD).

Conclusions:

• nnU-Net provides exceptional performance for automated whole-breast segmentation using contrast-free quantitative MRI.
• This automated method is effective for processing large clinical datasets.
• The approach represents a significant advancement in computer-aided quantitative analysis of breast DWI and SyMRI.