Clinical Imaging of Microwave Mammography

Summary

This protocol describes the imaging procedure of a microwave imaging device for breast imaging.

Abstract

Breast cancer is the most common malignancy among women, and early detection and treatment are critical for improving clinical outcomes. X-ray mammography remains the standard screening modality; however, it has several limitations, including radiation exposure, patient discomfort, and reduced sensitivity in women with dense breast tissue. Microwave imaging, a non-ionizing technique, has emerged as a promising alternative. We have developed a device that reconstructs breast tissue structures by solving the inverse scattering problem and is currently undergoing clinical trials. This system generates three-dimensional tomographic images without the use of contrast agents and without causing pain during examination. To date, 24 breast cancer patients have been imaged, with an accuracy of 86% for tumors ≥ 1 cm in diameter, and 58% when tumors < 1 cm are included. In this article, we present a detailed protocol for device preparation, clinical imaging, and data processing, along with representative imaging results from selected patients.

Introduction

Breast cancer is the most common malignancy among women, with approximately 95,000 new cases diagnosed annually in Japan¹. Early detection and treatment are essential for improving patient outcomes.

X-ray mammography is currently the standard screening tool. However, its sensitivity is limited in women with dense breast tissue, making small tumor detection difficult². To address this limitation, ultrasound is often used as an adjunctive diagnostic method. Ultrasound examinations require a high level of operator skill, and the diagnostic accuracy of mammography or ultrasound alone is approximately 60%. Contrast-enhanced magnetic resonance imaging (MRI) achieves accuracy exceeding 90%; nevertheless, due to high equipment and examination costs, MRI is not suitable for routine screening and is mainly reserved for surgical decision-making and treatment planning³.

Recently, microwave imaging has attracted increasing attention as a potential alternative capable of overcoming the limitations of mammography and ultrasound⁴. When microwave radiation interacts with biological tissue, strong reflections occur at dielectric boundaries. Reflectivity is more than an order of magnitude higher than that of X-rays or ultrasound, enabling the detection of small lesions. Research on breast cancer detection using microwave imaging has been ongoing for over 25 years, with several clinical studies reported⁵.

Microwave imaging can be broadly classified into two categories: confocal imaging, which reconstructs the distribution of scattered power, and scattering tomography, which reconstructs the distribution of complex permittivity (dielectric constant and electrical conductivity)^6,7. Confocal imaging is conceptually similar to ultrasound diagnostics; however, due to multiple reflections within tissue, internal structures cannot be accurately reconstructed. In contrast, scattering tomography provides accurate reconstructions even in the presence of multiple reflections but requires precise numerical modeling of electromagnetic scattering. Current computational electromagnetic techniques still face challenges in fully replicating these phenomena. Therefore, calibration that aligns measured data with numerical models is critical for practical implementation⁸. Moreover, because lesion responses are less than 1/1000 of the skin reflection⁹, highly sensitive antennas and receivers with excellent signal-to-noise performance are indispensable.

Our system addresses these challenges by introducing innovative approaches to solving the inverse scattering problem using an iterative distorted Born approximation (IDBA)¹⁰. Several key technologies have been developed, including a dual-polarized dielectric-loaded horn antenna¹¹, a calibration method using two homogeneous phantoms¹², breast fixation to the sensor via suction¹³, multi-polarized wave excitation¹⁴, and exploitation of the linear relationship between tissue permittivity and conductivity¹⁵. A prototype device incorporating these technologies was completed in September 2023, and clinical trials commenced in November 2023.

Although microwave imaging has been extensively investigated, there is still no widely accepted clinical protocol that guarantees reproducible data acquisition, robust calibration, and stable image reconstruction across patients. Establishing such a protocol is essential for the translation of microwave imaging from experimental research to practical clinical application. The protocol presented in this study, incorporating suction-based breast fixation, dual-phantom calibration, and multi-polarized wave excitation, addresses these requirements and provides a standardized framework for reliable clinical evaluation of microwave tomography.

Figure 1A illustrates the principle of microwave imaging¹⁶. Multiple antennas are arranged around the breast, with one antenna transmitting while the others receive scattered waves. A vector network analyzer (VNA) functions as both transmitter and receiver. If only a single port is available, transmission and reception are alternated sequentially, and scattered signals are acquired for all possible antenna pairs. The collected data are then used to reconstruct tomographic images of the breast.

The hardware configuration of the prototype is summarized in Table 1 and shown in Figure 1B, while the external appearance of the prototype is presented in Figure 2A. The device consists of a ceramic container for the breast, 16 antennas attached to the container, an antenna switch for selecting transmitting and receiving antennas, a VNA for measuring transmission characteristics between antenna pairs, a laptop computer with measurement software, a control unit for coordinating the laptop and antenna switch, and an aspirator for suction-based breast fixation (Figure 2C).

The ceramic container with attached antennas (referred to as the imaging sensor, Figure 2B) and the antenna switch are housed within the main body of the device. Although there are 16 physical antennas, 12 of them are dual-polarized, resulting in an effective antenna count of 28. The device is positioned beside the bed, and the patient lies prone, placing her breast into the imaging sensor (Figure 1C). To minimize imaging artifacts caused by body movement, the breast is suction-fixed to the sensor using the aspirator.

During measurement, a transmitting antenna is selected, and microwaves at 1.9 GHz with an output power of -5 dBm are irradiated into the breast. Scattering parameters (S-parameters) are measured by the VNA while sequentially switching transmitting and receiving antennas, and the data are recorded on the laptop. This process is repeated for all antenna combinations. Acquisition of one breast takes approximately 14 min. The relatively long measurement time is due to the serial acquisition process and additional waiting intervals implemented to prevent communication errors. Measurement duration is consistent across patients.

The recorded measurement files are transferred to a dedicated workstation for image reconstruction. The core control program is implemented in Excel VBA, which coordinates data processing and interfaces with the electromagnetic field simulator FEMTET and the numerical analysis platform MATLAB, as illustrated in Figure 3^17,18. Reconstruction of a single breast image requires approximately 7 h on a workstation equipped with an Intel Core i7-14700K CPU and 96 GB of RAM. Since the reconstruction algorithm is based on IDBA¹⁰, large-scale three-dimensional full-wave electromagnetic simulations must be executed repeatedly. Each forward electromagnetic field analysis requires approximately 10 min, and retrieving the computed field distribution from FEMTET takes an additional 30 min. This cycle is repeated ten times per reconstruction, yielding a cumulative processing time of roughly 400 min.

The software environment is summarized in Table 2. The laptop is configured with the Analyzer Control program to manage communication with the VNA and control unit for the antenna switch. Hardware connectivity requires a digital input/output (DIO) interface and a general-purpose interface bus (GPIB)-USB conversion cable to ensure reliable command transfer. All other reconstruction programs are installed on the workstation, which must support Microsoft Excel, FEMTET, and MATLAB. FEMTET is used to solve Maxwell's equations in the imaging domain via the finite element method, while MATLAB executes numerical routines required for the IDBA inversion.

A rigorous calibration procedure is required to align measured scattering data with the numerical forward model. Calibration is performed using two homogeneous dielectric phantoms (Figure 2D) designed to approximate tissue-mimicking media. At 1.9 GHz, phantom 1 exhibits a relative permittivity of 6.65 and a dielectric loss tangent of 0.438, whereas phantom 2 has a relative permittivity of 4.66 and a loss tangent of 0.289. A calibration file, cal_2314_94_ver2.mat, is pre-generated, containing correction coefficients that compensate for systematic discrepancies between measured and simulated S-parameters. These coefficients are derived from FEMTET-based simulations of the phantoms' transmission characteristics and cross-validated against experimental measurements obtained with the prototype¹². This calibration ensures consistency between numerical modeling and experimental data, stabilizing the convergence of the iterative reconstruction process.

Protocol

All clinical procedures were approved by the Institutional Review Board of Aichi Medical University (Certification Number of Specific Clinical Research: CRB4200004).

1. Device setup

1. Prepare a room sufficiently large to accommodate the main unit, office automation (OA) rack, and patient bed.
   NOTE: To ensure patient privacy, it is recommended to separate the main unit and bed using a curtain or partition (Figure 4A). A power supply of at least AC 100 V, 10 A is required.
2. Place the VNA, control unit, and laptop on the OA rack (Figure 4B).
3. Connect the electronic calibration module (Figure 4C) to the VNA via a USB cable.
4. Connect the antenna switch to the control unit using five D-sub cables.
5. Connect the cable linking the digital input/output (DIO) interface and the D-sub connector to the control unit (Figure 4D).
6. Connect the DIO interface to the laptop using a USB cable.
7. Connect the input/output ports of the VNA to the antenna switch using RF cables (Figure 4B,E).
8. Connect the GPIB port of the VNA to a GPIB-USB conversion cable (Figure 4F), and attach the USB connector to the laptop.
9. Connect the intake tube from the imaging sensor to the aspirator (Figure 4E).
10. Plug in the power cables of the VNA, antenna switch, laptop, and aspirator into AC outlets and turn on the power.

2. Input of initial parameters

1. Launch the measurement program Analyzer Control on the laptop.
   NOTE: The start window appears (Figure 5A).
2. Enter the following parameters: start frequency = 1000 MHz, stop frequency = 6000 MHz, number of data points = 101, and response waiting time = 200 ms.
   NOTE: Default values should be used for all other parameters.
3. Press the DIO button, followed by the GPIB Confirmation button.
   NOTE: If all connections are correct, the CAL Wizard, CAL Load, Auto Measurement, and Manual Measurement buttons become active (Figure 5B).

3. Calibration of the Vector Network Analyzer (VNA)

1. Disconnect the cables connected to antennas 4 and 6 of the imaging sensor, and connect them to the input/output terminals of the electronic calibration module (Figure 6A).
2. Set both AUTO/MANUAL switches on the control unit to MANUAL, and configure the Tx toggle switch to binary 5 and the Rx toggle switch to binary 9 (Figure 6B).
3. Press the CAL Wizard button.
   NOTE: The CAL Wizard window appears (Figure 6C).
4. Press the Set Wizard button (Figure 6C).
5. Perform two-port calibration over the 1-6 GHz frequency band using the electronic calibration module¹⁹.
6. Upon completion, press CAL Wizard Exit to close the window (Figure 6C).
7. Press the CAL Save button.
   NOTE: The data save window appears (Figure 6D).
8. Enter a name for the calibrated data (e.g., 20250717.sta) and press Save.
   NOTE Calibration data are saved in the VNA. If calibration has already been performed, repeating it is not necessary. In that case, press CAL Load, select the appropriate file (Figure 6E: e.g., 20240829.sta), and click Apply to load the calibration data into the VNA.
9. Set the AUTO/MANUAL switch on the control unit back to AUTO.

4. Imaging procedure

1. Position the patient prone on the bed (Figure 7A).
2. Pull up the patient's top and place one breast into the imaging sensor, ensuring no clothing obstructs the sensor.
3. Turn on the aspirator to initiate suction.
4. Adjust the vacuum pressure to 0.04 - 0.06 MPa using the CTRL knob (Figure 7B). If the pressure cannot be reduced, reposition the patient until proper suction is achieved.
5. Stop imaging if the aspirator pressure remains at 0 MPa.
   NOTE: A gap between the sensor and the breast prevents accurate image acquisition.
6. Press the Auto Measurement button (Figure 5B).
   NOTE: The measurement status window appears (Figure 7C).
7. Set the measurement range to 1-28 and enable Skip Reciprocity.
8. Press Start Measurement to begin data acquisition.
   NOTE: The transmission and reception combinations are displayed during measurement (Figure 7D). Upon completion, results are saved in the VNA_Results folder in CSV format (Figure 7E).

5. Image reconstruction

1. Open the data_read program in MATLAB.
2. Set the path (e.g., C:\Users\yoshihiko\Desktop\VNA_Results) and specify the measurement files (e.g., P31_L.csv and P31_R.csv) along with the output data name (Figure 8A).
3. Click the button in the editor to generate a .mat file for equipment calibration (Figure 8B).
4. Open the initialize_for_clinic program in MATLAB.
5. Enter the following parameters (Figure 8C): Reconstruction frequency: f = 1.9 × 10⁹ Hz; Path and names of the measurement files (e.g., C:\Users\kuwahara\OneDrive\patient_data\xxxxxxxx); Designation of the left and right breast (L(:,:,fn)); Path and name of the patient worksheet ('xxxxxxxx.xlsx', sheet='sheet1')
6. Click to generate a patient worksheet for use in the main Excel VBA program.
   NOTE: A directory for storing reconstruction results is also created.
7. Open the image_reconstruction_# MATLAB file (# indicates processing number).
8. Specify the patient worksheet path and name (e.g., G:\patient_sheet\xxxxxxxx_L.xlsx), the reconstruction frequency (1.9 × 10⁹ Hz), and the calibration file (cal_2314_ver2), then save the settings (Figure 8D).
9. Open the main image reconstruction program multi_person in Excel.
10. Enter the patient worksheet names created in step 6 (e.g., xxxxxxxx_L to vvvvvvv_R) (Figure 9A).
    NOTE: Multiple patients' images can be reconstructed simultaneously by entering the series of worksheet names.
11. Open the VBA program from the Excel add-in.
12. Specify the range of worksheets to process (np = 1 to 8).
13. Confirm that the directories for worksheets are correctly set in the Data_read and Data_write standard modules.
14. Select getPDTdata and click in the editor to start the image reconstruction process (Figure 9B).
15. After processing, verify the reconstructed images (3D tomographic longitudinal and transverse sections), which are saved in the specified directory with filenames such as mwsdbim_20250617-14h36m04s, alongside the patient worksheets.

Results

The imaging results of three representative patients are presented below:

Patient 1
A 78-year-old woman with a 34 mm × 21 mm cancer located at the 2 o'clock position of the left breast. The mammary glands are predominantly fatty. Figure 10A and Figure 10B show X-ray mammography images, whereas Figure 10C-F present imaging results obtained with the pro...

Discussion

Calibration in this context has two distinct meanings. First, it refers to the correction of measured scattering data so that the results can be accurately incorporated into numerical models. Second, it refers to the calibration of the vector network analyzer (VNA) itself.

Prior to any calibration or measurement, the VNA should be powered on at least 15 min in advance, and the READY indicator on the electronic calibration module must be green. All device connections should be ...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Aspirator	AS ONE	DAS-01
digital IO	National Instruments	USB6501
Elecreonic Calibration Module	Agilent	85093C	Discontinued. Replace with 85093D.
Electromagnetic simulator	Murata Software	FEMTET
Excel VBA	Microsoft	Office
GPIB USB Conversion Cable	National Instruments	GPIB-USB-HS
Numerical Software	Mathworks	Matlab
VNA	Keysight	E5071C