Automated fibroglandular tissue segmentation in breast MRI using generative adversarial networks

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate segmentation of fibroglandular tissue (FGT) in breast MRI is essential for quantifying background parenchymal enhancement (BPE).
• BPE analysis aids in breast cancer risk assessment.
• Existing segmentation methods require improvement for clinical applications.

Purpose of the Study:

• To develop and evaluate an automated deep learning method using a generative adversarial network (GAN) for FGT segmentation in 3D breast MRI.
• To assess the impact of the proposed GAN-based segmentation on BPE quantification.

Main Methods:

• A GAN comprising an improved U-Net generator and a patch deep convolutional neural network (DCNN) discriminator was developed.
• The model was trained and tested on 100 3D bilateral breast MRI scans.
• Segmentation accuracy was evaluated using Dice Similarity Coefficient (DSC) and Jaccard Index (JI), compared against a baseline U-Net method.

Main Results:

• The proposed GAN method achieved superior segmentation accuracy with DSC of 87.0 ± 7.0% and JI of 77.6 ± 10.1%, compared to the baseline U-Net (DSC: 81.1 ± 8.7%, JI: 69.0 ± 11.3%).
• GAN-based FGT segmentation led to improved correlation coefficients (0.46 ± 0.15) between quantified BPE and radiologist-provided BI-RADS BPE categories.
• The baseline U-Net resulted in lower correlation coefficients (0.41 ± 0.16).

Conclusions:

• The developed GAN-based method significantly outperforms the traditional U-Net for FGT segmentation in breast MRI.
• Improved FGT segmentation using the GAN model enhances the accuracy of background parenchymal enhancement quantification.
• This advancement holds promise for more reliable breast cancer risk assessment.