Updated: Jul 7, 2026

Terahertz Imaging and Characterization Protocol for Freshly Excised Breast Cancer Tumors
Published on: April 5, 2020
Shakti K Davis1, Barry D Van Veen, Susan C Hagness
1Department of Electrical and Computer Engineering, University of Wisconsin, Madison 53706, USA. shaktid@ieee.org
You might also read
Articles linked to this work by shared authors, journal, and citation graph.
This study explores using microwave radar signals to identify the size and shape of breast tumors. By analyzing how these waves bounce off targets, researchers can classify lesions as smooth, lobulated, or spiculated, potentially improving cancer detection without extra hardware.
Area of Science:
Background:
No prior work had resolved whether microwave signals could reliably distinguish subtle architectural features of breast lesions. Traditional imaging often requires secondary assessments to confirm the nature of detected abnormalities. That uncertainty drove researchers to explore alternative diagnostic modalities. Prior research has shown that dielectric properties vary significantly between healthy and malignant tissues. This gap motivated the investigation of high-frequency electromagnetic waves for non-invasive characterization. Existing radar systems primarily focus on detection rather than detailed morphological classification. The potential for extracting shape and size data from scattered signals remains largely untapped in clinical settings. This study addresses the feasibility of utilizing specific frequency bands for enhanced diagnostic precision.
Purpose Of The Study:
The aim of this study is to investigate the feasibility of using multichannel microwave backscatter for classifying architectural features of dielectric targets. Researchers seek to determine if signal analysis can identify the shape and size of suspicious lesions. This work addresses the need for improved diagnostic information in conjunction with medical imaging. The authors focus on the 1-11 GHz frequency band to extract salient morphological data. They hypothesize that specific structural characteristics, such as spiculation or lobulation, leave distinct signatures in the backscattered waves. By testing various size categories, the team evaluates the sensitivity of their classification models. This investigation is motivated by the potential to enhance breast cancer detection without requiring extra hardware. The study provides a numerical assessment of how well these features can be categorized under different noise conditions.
The researchers propose using multichannel microwave backscatter within the 1-11 GHz range. This mechanism allows for the extraction of morphological data from dielectric targets by analyzing reflected electromagnetic waves, which helps distinguish between smooth, microlobulated, and spiculated lesion architectures.
The authors employ local discriminant bases and principal component analysis to construct linear classifiers. These mathematical tools process expansion vectors derived from the backscatter data to categorize the physical dimensions and structural irregularities of the simulated targets.
A 10 dB signal-to-noise ratio is necessary to achieve the reported classification performance. This threshold ensures that the reflected signal remains distinct enough from background interference to allow for the accurate identification of target size and shape.
Main Methods:
Review approach involves numerical simulations of dielectric targets across a broad frequency spectrum. Investigators utilize Gaussian random spheres to represent various lesion morphologies, including smooth, microlobulated, and spiculated structures. The team evaluates performance across four distinct size categories ranging from 0.5 to 2 cm. Classification relies on two specific mathematical frameworks: local discriminant bases and principal component analysis. These techniques facilitate the construction of linear classifiers using subset expansion vectors as input features. The study assesses the average rate of correct classification as the primary performance metric. Researchers systematically vary the signal-to-noise ratios to determine the stability of the classification algorithms. This computational design allows for controlled testing of signal processing efficacy in a simulated environment.
Main Results:
Key findings from the literature demonstrate that target size is reliably classified with over 97% accuracy at a 10 dB signal-to-noise ratio. This performance metric is averaged across 360 distinct numerical targets. Shape classification achieves over 70% accuracy under identical noise conditions. The data reveal a clear relationship between the signal-to-noise ratio of the test inputs and the resulting classifier performance. Higher noise levels generally correspond to a reduction in the precision of morphological identification. The results confirm the feasibility of extracting architectural details directly from the reflected electromagnetic signals. These findings suggest that both size and shape characteristics are detectable within the 1-11 GHz band. The evidence supports the potential for enhancing radar-based diagnostic systems through advanced signal processing.
Conclusions:
The authors propose that dielectric targets can be classified directly from their electromagnetic signatures. Synthesis and implications suggest that size estimation achieves high reliability under moderate noise conditions. Shape classification remains feasible, though it demonstrates lower performance metrics than size determination. These findings indicate that morphological assessment integrates seamlessly with existing radar-based detection protocols. No specialized hardware modifications appear necessary for implementing these classification techniques. The researchers suggest that this approach enhances the diagnostic value of current radar systems. Future clinical utility relies on the robustness of these algorithms against biological variability. This work provides a foundation for improving non-invasive breast cancer screening through signal analysis.
The researchers utilize Gaussian random spheres to simulate target constructs. These models allow for the introduction of moderate shape irregularities, which are essential for testing the sensitivity of the classification algorithms against varied architectural features.
The study measures the average rate of correct classification for targets ranging from 0.5 to 2 cm in diameter. Results indicate that size is classified with over 97% accuracy, while shape identification reaches over 70% accuracy at a 10 dB signal-to-noise ratio.
The authors suggest that this characterization method integrates directly into existing radar-based breast cancer detection systems. This implies that clinicians could obtain detailed lesion information without requiring additional hardware or supplementary data collection procedures.