Full 3D Microwave Breast Imaging Using a Deep-Learning Technique

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Microwave imaging offers a non-ionizing method for breast cancer detection.
• Contrast-Source Inversion (CSI) is a reconstruction technique used in microwave imaging.
• Artifacts in CSI reconstructions can hinder accurate tumor detection.

Purpose of the Study:

• To develop and evaluate a deep learning technique for enhancing 3D breast permittivity images obtained via microwave imaging.
• To improve the accuracy of tumor detection by reducing artifacts in Contrast-Source Inversion (CSI) reconstructions.
• To extend a previously developed 2D image enhancement technique to 3D.

Main Methods:

• A 3D Convolutional Neural Network (CNN) based on the U-Net architecture was employed.
• The U-Net model was trained using synthetic 3D CSI images and corresponding ground truth numerical phantoms.
• The network was tested on both synthetic and experimental 3D microwave imaging data.

Main Results:

• The 3D U-Net effectively reduced artifacts typical of CSI reconstructions.
• Tumor detectability was significantly enhanced in the processed images.
• The model demonstrated good performance on both synthetic and experimental data, despite being trained solely on synthetic data.

Conclusions:

• Deep learning, specifically a 3D U-Net, is a powerful tool for enhancing 3D microwave breast imaging.
• The developed technique improves image quality and aids in the detection of breast tumors.
• The approach shows potential for clinical application in breast cancer diagnostics.