A Teacher-Student Transfer Learning Framework for Breast Tumor Detection Using Time-Domain Microwave Sensing Data

View at publisher Summarise with VIRAT
Authors: Greeshma Lingam, Naina Narang
Cited by: 0
References: 29

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Microwave imaging for breast tumor detection faces challenges due to complex biological structures, mutual coupling, and signal-processing constraints, leading to a large number of unknowns and an inherently ill-posed problem. Moreover, training deep learning models from scratch may lead to overfitting. To address these challenging issues, we propose a deep teacher-student transfer learning model with distilled knowledge for breast tumor detection. The proposed model utilizes distilled knowledge from the ResNet18-UNet teacher model to the Genetic Liquid Neural Network (GLNN) student model, which prevents overfitting. Specifically, our proposed ResNet18-UNet teacher model incorporates pixel-wise segmentation capability of U-Net with the effective feature extraction of the ResNet18 model, which captures precise frequency-dependent scattering feature representations in microwave sensing data. Moreover, our proposed GLNN student model adapts a genetic algorithm to optimize feature selection by preserving the most relevant microwave sensing features. Moreover, LNN utilizes a neural ordinary differential equation to identify the evolving dielectric contrast patterns within the tissue. Extensive experimental results illustrate that the proposed teacher-student model achieves 4-10% higher efficacy than existing algorithms.

Fields of Study & Subjects

Microwave Imaging and Scattering Analysis Electrical and Bioimpedance Tomography AI in cancer detection

Publication details

Journal / source
IEEE Journal of Electromagnetics RF and Microwaves in Medicine and Biology
Volume
10
Issue
2
Pages
253–262
Publication date
20 March 2026
Language
EN