Machine Learning Approach to Quadratic Programming-Based Microwave Imaging for Breast Cancer Detection.

Area of Science:

• Medical Imaging
• Computational Electromagnetics
• Artificial Intelligence in Healthcare

Background:

• Microwave imaging offers a non-ionizing approach for breast cancer detection.
• Solving the inverse scattering problem in microwave imaging is computationally intensive and ill-posed.
• Accurate reconstruction of dielectric properties is crucial for differentiating cancerous tissue.

Purpose of the Study:

• To develop a novel, accelerated technique for microwave imaging reconstruction.
• To accurately recover the dielectric permittivity of breast phantoms from scattering data.
• To enhance the clinical applicability of microwave imaging for breast cancer detection.

Main Methods:

• Integration of the Born iterative method (BIM) with a quadratic programming (QP) approach.
• Application of convolutional neural networks (CNNs) to accelerate the inverse problem solution.
• Validation using simulated breast phantoms in a circular imaging configuration.

Main Results:

• The combined BIM-CNN approach significantly reduces image reconstruction time.
• High accuracy (exceeding 90%) was achieved in recovering breast phantom permittivity.
• The method demonstrates robust performance across various validation scenarios.

Conclusions:

• The proposed BIM-CNN technique offers a fast and accurate solution for microwave imaging reconstruction.
• CNNs can substantially decrease the computational burden in microwave breast cancer detection.
• This advancement holds promise for improving the efficiency and accuracy of clinical microwave imaging systems.