A convolutional neural network algorithm for breast tumor detection with magnetic detection electrical impedance...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Biology

Background:

• Early detection of breast cancer is crucial for patient survival.
• Current methods for breast tumor detection have limitations.
• Non-invasive, high-resolution imaging techniques are needed.

Purpose of the Study:

• To propose and evaluate a novel Magnetic Detection Electrical Impedance Tomography (MDEIT) method for non-invasive breast tumor detection.
• To utilize Convolutional Neural Networks (CNN) for enhanced image reconstruction and analysis in MDEIT.
• To assess the feasibility and performance of the CNN-MDEIT approach in simulated and experimental settings.

Main Methods:

• Simulated MDEIT forward problems using the Finite Element Method to generate breast lesion datasets.
• Developed and trained a CNN to predict conductivity distribution from MDEIT data.
• Conducted noise and phantom experiments to validate the CNN algorithm's anti-noise performance and detection capabilities.

Main Results:

• CNN algorithm reduced reconstruction relative error to 10%, outperforming traditional methods.
• Demonstrated superior stability and anti-noise performance; tumor shape was recoverable even with 60 dB noise.
• Phantom experiments confirmed legible conductivity distribution images and accurate tumor localization.

Conclusions:

• The proposed CNN-based MDEIT reconstruction method is effective for early and non-invasive breast tumor detection.
• This technique offers practical importance for improving breast cancer diagnosis.
• The CNN-MDEIT approach shows promise for real-world clinical applications.