Microwave breast imaging: A review of clinical potential and technological advances

Abstract

Breast cancer is a leading cause of morbidity and mortality among women worldwide, and the limitations of conventional imaging, especially in women with dense breast tissue have sparked interest in safer, more accessible diagnostic alternatives. Microwave breast imaging (MBI) is a promising, non-ionizing, and potentially cost-effective modality that leverages dielectric contrasts between malignant and healthy tissues. This review presents MBI technologies, focusing on microwave tomography and radar-based techniques such as confocal microwave imaging and tissue-sensing adaptive radar. We evaluate system designs, antenna configurations, image reconstruction algorithms, and clinical performance from leading research groups. Additionally, we highlight the growing role of machine learning in enhancing image resolution, segmentation, and lesion classification. Despite encouraging results from early-phase clinical trials, challenges such as limited spatial resolution, anatomical variability, and high computational demands continue to hinder broader adoption. We outline future directions, including real-time motion correction, improved antenna sensitivity, and stronger clinical validation. With continued innovation and regulatory support, MBI has strong potential as an adjunct or alternative to current breast imaging modalities, especially for dense breast tissue and in resource-limited settings.

Key Words: Microwave breast imaging; Microwave tomography; Radar-based imaging; Confocal microwave imaging; Breast cancer detection; Dielectric properties; Non-ionizing imaging; Antenna design; Deep learning; Image reconstruction

Core Tip: Microwave breast imaging (MBI) offers a non-ionizing, cost-effective alternative to traditional breast imaging, particularly beneficial for women with dense breast tissue. By exploiting dielectric contrasts between malignant and healthy tissues, MBI-through techniques like microwave tomography and confocal radar imaging-shows promise in early detection. Despite challenges like limited spatial resolution and anatomical variability, advancements in antenna design, machine learning, and real-time imaging continued to push MBI closer to clinical integration, especially in low-resource or high-risk settings.

INTRODUCTION

Breast cancer remains the most commonly diagnosed cancer and a leading cause of cancer-related death among women worldwide. In 2022, approximately 2.3 million women were diagnosed with breast cancer globally, resulting in 670000 deaths[1]. In the United States, an estimated 316950 new cases of invasive breast cancer are expected to be diagnosed in women in 2025, along with 59080 cases of ductal carcinoma in situ. Approximately 42170 women are projected to die from breast cancer in the United States in 2025. Early detection significantly improves outcomes, with five-year survival rates exceeding 90% when diagnosed at an early stage[2].

CONVENTIONAL IMAGING MODALITIES

Conventional imaging modalities used in breast cancer screening include X-ray mammography, ultrasound, magnetic resonance imaging (MRI), and positron emission tomography. Among these, mammography remains the cornerstone of population-based screening due to its widespread availability and ability to detect early-stage lesions[3]. However, mammography has drawbacks, including the use of ionizing radiation, discomfort due to breast compression, and significantly reduced sensitivity in women with dense breast tissue, where both dense fibroglandular tissue and tumors appear similarly radiopaque[4]. Additionally, the risk of false-positive findings is higher in younger women, contributing to unnecessary biopsies, patient anxiety, and overtreatment[5].

Ultrasound and MRI are often used as adjunct tools. While ultrasound is accessible and radiation-free, its diagnostic accuracy is heavily operator-dependent and limited by relatively poor spatial resolution. MRI offers excellent sensitivity, particularly in high-risk and dense-breast populations, but its high cost, limited availability, and contraindications in certain patients restrict its routine use. These limitations have prompted the exploration of alternative, safer, and more accessible imaging technologies[6,7].

MICROWAVE BREAST IMAGING

Microwave breast imaging (MBI) has emerged as a promising, non-ionizing, and potentially cost-effective technique for breast cancer detection (Table 1)[8]. This review provides a comprehensive overview of MBI technologies, focusing on two principal approaches: Microwave tomography (MT) and radar-based imaging, while also discussing advances in antenna design, integration with deep learning (DL), and the current landscape of clinical evidence supporting their diagnostic value.

Table 1 Comparison of different imaging modalities by smallest visible tumor size.

Modality	Smallest visible tumor	Best use case
Mammography	2-5 mm (ideal), about 5-10 mm avg	General screening, calcifications
Ultrasound	3-5 mm	Dense breasts, targeted follow-up
Magnetic resonance imaging	1-2 mm	High-risk screening, dense breasts
Microwave imaging	4-10 mm (experimental)	Radiation-free, low-cost adjunct (early phase)

MICROWAVE IMAGING MODALITIES AND MECHANISMS

Microwave-based breast imaging offers several advantages over conventional diagnostic methods. It is non-invasive, free from ionizing radiation, relatively inexpensive, and generally more comfortable for patients. These techniques rely on the fact that malignant breast tissues have higher permittivity and conductivity compared to healthy tissues when exposed to microwaves. This dielectric contrast enables tumor detection by measuring how breast tissues absorb and scatter microwave signals[9,10].

Microwave imaging systems can be broadly categorized into three types: Passive, hybrid, and active modalities (Figure 1). In passive imaging, the system detects natural thermal emissions from the breast. Since tumor tissue tends to be warmer due to increased vascularization and metabolic activity, these methods identify cancerous regions by measuring temperature differences, a good example would be radiometry techniques[11,12]. Hybrid techniques, such as thermoacoustic imaging, combine microwave energy with ultrasound. Microwaves are absorbed preferentially by malignant tissues, which then emit acoustic waves. These waves are captured by ultrasound transducers and used to reconstruct images that reflect tissue composition based on both electrical and mechanical properties[13,14]. Active microwave imaging is the most extensively developed category. It involves illuminating the breast with low-power microwave signals and analyzing the backscattered waves to map tissue structure[15]. Active methods include two primary techniques: MT and radar-based imaging.

Figure 1 Classification of different modalities that have been explored for microwave-based breast cancer detection. Citation: Aldhaeebi MA, Alzoubi K, Almoneef TS, Bamatraf SM, Attia H, M Ramahi O. Review of Microwaves Techniques for Breast Cancer Detection. Sensors (Basel) 2020; 20: 2390. Copyright© The Authors 2024. Published by MDPI. The authors have obtained the permission (Supplementary material).

MT

MT reconstructs cross-sectional images of the breast by measuring how microwave signals are transmitted through and reflected by tissues. Its strength lies in its ability to quantify electrical properties, such as permittivity and conductivity, which can help differentiate benign from malignant lesions[16,17]. To generate these images, MT solves what is known as an inverse problem-working backward from the detected signals to estimate the internal structure of the breast. This is challenging because many different tissue configurations can produce similar signals, and the relationship between the signals and the tissue properties is complex.

To address this, researchers use computational algorithms. Some are gradient-based methods, like Landweber iteration and conjugate gradient least squares, which gradually adjust the image estimate to better match the measured data. Others use global optimization techniques, which aim to search a broad range of possible solutions rather than getting stuck on a single, potentially suboptimal one. For example, genetic algorithms simulate the process of natural selection by evolving a population of possible solutions over time, while particle swarm optimization mimics the collective behavior of swarming animals, where each potential solution learns from its own experience and that of others. These global approaches are particularly helpful in MT because of the complexity and ambiguity inherent in reconstructing images from scattered microwave signals[18].

A seminal development in MT came from Meaney et al[15], who designed a 2D clinical prototype using 16 monopole antennas operating between 300 MHz and 1000 MHz. While innovative, this system required long processing times. A more advanced 3D system followed, based on finite element modeling, which significantly reduced processing time-achieving sub-centimeter resolution in under two minutes. Tumor detectability has been further improved by incorporating magnetic nanoparticles as contrast agents, which enhance the difference in microwave response between malignant and healthy tissues. Another MT system, developed by Jeon et al[19], operates at higher frequencies (3-6 GHz) and demonstrated the ability to detect tumors as small as 25 mm in women aged 40-68. Despite these promising advances, MT still faces challenges, including lower spatial resolution and limited depth penetration compared to more established imaging modalities like MRI or ultrasound (Table 2)[20-34].

Table 2 Summary of clinical and perfromance metrics of different tomography-based microwave radiometric imaging systems.

System	Scan time	Breast size	Accuracy (true positive rate)	Sensitivity (true negative rate)	Specificity	Patients tested
Dartmouth college[20,21]	2.2 minutes	14.2 cm diameter	Detected a 4-cm tumor in a trial test with one patient	Not reported	Not reported	1
McMaster university[22]	5 hours	4.8 cm diameter	Reported a 247 mm resolution in simulation	No clinical trials	No clinical trials	0
Istanbul Technical University (SAFE)[23,24]	20 minutes	Small and large	Detected lesions as small as 4 mm	79%	77%	113
University of Bristol (MARIA)[25]	5 minutes	32A and 32DD	Smallest recorded lesion was 5 mm	76%	Not reported	8
University of Calgary (TSAR)[26]	< 30 minutes	B and C size	10 mm lesion detected with SCR ~5.0 dB	No clinical trials	No clinical trials	1
McGill University[27]	6 minutes	Amid A to D size	10 mm spherical phantom tumor detected	No clinical trials	No clinical trials	4
Galway University (Wavelia)[28]	6 minutes	Greater than 32B	Did not detect tumors < 10 mm in trials	75%	Not reported	5
Chalmers University[29,30]	Not reported	55.0-75.0 cm diameter	Used oil cups mimicking tumors (55-75 mm)	No clinical trials	No clinical trials	0
University of Perugia (Mammowave)[31]	8 minutes	Various cup sizes	73%	82%	353
Hiroshima University[32]	15 minutes	Small (Japanese patients)	Minimum size 40 mm in trials (Japanese only)	49%	55%	9
University of São Paulo[33]	28 minutes	34B (imperial system)	1 mm tumor in 150 mm breast phantom with SCR > 7.0 dB	No clinical trials	No clinical trials	3
University of Queensland[34]	45 minutes	12.4 cm	20 mm × 30 mm water target in phantom	No clinical trials	No clinical trials	1

SAFE: Scanning aperture focused examination; MARIA: Multistatic array processing for radio wave image acquisition; TSAR: Tissue sensing adaptive radar; SCR: Signal-to-clutter ratio.

RADAR-BASED MBI

Radar-based MBI utilizes ultra-wideband signals and antenna arrays positioned around the breast to detect dielectric contrasts associated with tumor boundaries. Major techniques include confocal microwave imaging (CMI), tissue-sensing adaptive radar (TSAR), microwave imaging via space-time (MIST), and holographic microwave imaging (HMI), among others. These methods vary in imaging resolution, detection depth, beamforming strategies, and antenna configurations, see Table 3[35-43].

Table 3 Comparison of radar-based microwave breast imaging techniques.

Technique	Detection range (mm)	Resolution	Image type	No. of antennas	Ref.
CMI	2–4 mm	High	2D/3D	Varies	[35,36]
TSAR	> 4 mm	Moderate	2D	Multiple	[37-39]
MIST	About 4 mm	High	3D	16	[40]
HMI (near-field)	About 5 mm	Moderate	2D	Varies	[41]
HMI (far-field)	4–6 mm	Moderate	2D	Varies	[42,43]

CMI: Confocal microwave imaging; TSAR: Tissue sensing adaptive radar; MIST: Microwave imaging via space-time; HMI: Holographic microwave imaging.

CMI and MIST offer high-resolution imaging, with the ability to detect tumors as small as 2-4 mm, while TSAR and HMI typically provide moderate resolution with slightly deeper or broader detection ranges. Some approaches reconstruct 3D volumes using time-domain signal acquisition (e.g., MIST), while others like TSAR and HMI rely on 2D reconstructions. The number of antennas ranges from compact 2-element systems to large arrays with 60 elements, as seen in the Micrima radio-wave radar breast imaging (MARIA) system developed in Bristol to reduce motion artifacts and improve imaging consistency[25,26,35-43].

Institutional implementations differ significantly in frequency range, coupling medium, patient positioning (e.g., prone vs supine), and signal processing methods such as delay-and-sum, delay-multiply-and-sum, or time-reversal techniques like TR-MUSIC. While some systems, like MARIA and TSAR, are approaching commercial readiness or undergoing validation, others remain in early feasibility stages[25,26,35-43]. Although advances in compressive sensing and dielectric contrast agents have enhanced both resolution and acquisition speed, broad clinical adoption is still limited by computational demands and the need for multi-center clinical validation.

SUMMARY OF TOMOGRAPHY AND RADAR-BASED MWI SYSTEMS

Tables 4 and 5[20-23,25,26,44-49] provide summaries of tomography and radar-based MWI systems reviewed in this study, highlighting that about half of the current research is at the clinical investigation stage. Nonetheless, radar-based systems are regarded as being close to gaining clinical acceptance.

Table 4 Summary of the technical specifications in different radar-based microwave imaging systems.

Origin	Antennas and arrangement	Frequency range (GHz)	Patient position	Coupling	Hardware	Imaging algorithm	Development stage
Dartmouth College[20,21]	16-monopole array	0.5-2.3	Prone	Glycerin/water	Stationary array PC controlled function generator and Rx circuitry	Log-phase formulation	Concept and feasibility
	16-monopole array	0.5-2.5	Prone	Glycerin/water	Stationary array Ettus B210 SDR	Log-phase formulation	Concept and feasibility
McMaster University[22]	1-horn (Tx), 9-bowtie array (Rx)	3.0-8.0	Not specified	Not specified	Stationary array planar scanning	Scattered power mapping, quantitative microwave holography	Concept and feasibility
Istanbul Technical University (SAFE)[23]	1 (Tx), 36-antenna array (Rx)	1.4-8.0	Prone	Liquid coupling medium	Stationary array VNA usage	Linear sampling method, factorization method	Verification and validation

Tx: Transmitter; Rx: Receiver; VNA: Vector network analyzer; SDR: Software defined radio; SAFE: Scanning aperture focused examination.

Table 5 System specifications and clinical readiness of institution-based radar-based microwave breast imaging systems.

Institution	Antenna setup	Frequency range (GHz)	Patient orientation	Coupling medium	System configuration	Imaging technique	Development phase
Bristol (United Kingdom)-MARIA[25]	60 wide-slot antennas	3.0-8.0	Prone	Paraffin oil, wax, aqueous media	Stationary, 16-port Keysight VNA	Modified delay-and-sum	Near product launch
Calgary (Canada)-TSAR[26]	Monostatic antipodal Vivaldi	0.05-15.0	Prone	Canola oil	PC-controlled synthetic array, laser	Confocal DAS, skin artifact correction	Verification and validation
McGill (Canada)[44]	16 resistively-loaded traveling wave antennas	2.0-4.0	Prone	Ultrasound gel	Oscilloscope and pulse generator array	Delay-multiply-and-sum	Early feasibility
Galway (Ireland)-Wavelia[45]	18-element Vivaldi array	1.0-4.0	Prone	Liquid	Stationary array with VNA	TR-MUSIC (time-reversal method)	Verification and validation
Chalmers (Sweden)[46]	20 monopole antennas	0.2-3.0/0.5-6.0	Prone	Air or oil-filled phantom	Off-the-shelf SDR components	Confocal DAS or no imaging	Feasibility
Perugia (Italy)–Mammowave[47]	1 horn + 1 microstrip monopole	1.0-9.0	Prone	Air	Cobalt C1209 and Copper M. VNA	Huygens-based approach	Verification and validation
Hiroshima (Japan)[48]	4 × 4 dome-shaped array	3.1-10.6	Supine	Glycerine	UWB CMOS synthetic array	Confocal DAS	Verification and validation
São Paulo (Brazil)[49]	Dual-patch bistatic array	5.0-7.0	Fowler	Silicone rubber interface	UWB transceiver setup	Enhanced confocal DAS	Feasibility

VNA: Vector network analyzer; DAS: Delay-and-sum; TR-MUSIC: Time-reversal multiple signal classification; SDR: Software defined radio; UWB: Ultra-wideband; CMOS: Complementary metal-oxide-semiconductor; MBI: Microwave breast imaging.

Machine learning and DL in MBI

Artificial intelligence is becoming a cornerstone of modern MBI, with machine learning (ML) and DL methods increasingly used to improve image quality, automate tumor detection, and classify tissue types[50]. Techniques such as Convolutional Neural Networks (CNNs), Deep Neural Networks, and U-Net architectures have shown significant promise. For instance, Shah and Moghaddam[51] used a two-stage CNN to improve spatial resolution in reconstructed images, while Yahya et al[52] achieved a 100% early breast cancer detection rate by combining wavelet transforms with neural networks.

A particularly active area of research is the use of DL to denoise and segment images, often using architectures like U-Net. Khoshdel et al[53], for example, trained a U-Net model on synthetic 3D contrast-source inversion data to enhance image clarity and tumor visibility. Although many models rely on simulated data, some, like TAT-Net and transfer learning-based AlexNet, have been tested in experimental or mixed settings with promising results.

While ML and DL techniques, such as CNNs and U-Nets, are playing an increasingly essential role in advancing MBI, there is still no consensus on the “best” approach. DL methods have demonstrated clear improvements over traditional algorithms like delay-and-sum and linear sampling by enhancing spatial resolution, tumor visibility, and denoising efficiency, leading to greater automation and improved diagnostic accuracy. Traditional methods, however, tend to offer greater interpretability and lower computational demands, often at the cost of lower image quality and more manual analysis[54-71].

Researchers are actively exploring which DL architectures perform best across different imaging scenarios, particularly when transitioning from simulation to clinical settings. Promising directions include physics-informed neural networks that incorporate imaging physics into DL models and hybrid approaches combining traditional inversion algorithms with DL post-processing. Despite their advantages, DL models face significant challenges such as overfitting to simulated data, limited generalizability to anatomically diverse patient cohorts, reliance on large annotated datasets, and high computational requirements. Models trained purely on simulations often struggle with clinical validation, and the scalability of data-intensive methods remains limited. Addressing these hurdles will be crucial for DL-enhanced MBI to evolve from proof-of-concept research into practical, reliable clinical tools[72].

CLINICAL EVIDENCE AND VALIDATION

While early-phase clinical studies with systems such as MARIA^®, MammoWave, and Wavelia have demonstrated safety and feasibility, MBI remains in a pre-commercial stage with important technical and clinical hurdles yet to be resolved before widespread adoption. From a clinical standpoint, the most pressing challenge is validation. While some comparative studies have been conducted, large-scale trials directly benchmarking MBI against gold standards (e.g., MRI and digital breast tomosynthesis) are sparse. Without such comparative evidence, it remains uncertain whether MBI can achieve the sensitivity and specificity necessary for widespread diagnostic use[73].

REGULATORY CONSIDERATIONS AND MARKET READINESS

Regulatory pathways and cost-effectiveness analyses must be addressed early, particularly considering the variability in healthcare infrastructure across regions[54-69] (Table 6). Another key limitation is scalability. Although several private companies are pushing MBI toward commercialization, practical deployment in gynecological and radiology clinics will require resolving logistical and economic constraints, such as device size, maintenance, training requirements, and reimbursement policies.

Table 6 Summary of clinical evaluation of microwave breast imaging systems.

Application	Description	Advantages	Challenges/limitations	Ref.
Primary breast cancer detection	Detects malignant tumors based on dielectric contrast	Non-ionizing, safe, repeatable; good for dense breasts	Low spatial resolution; noise sensitivity	[54,55]
Differentiating benign vs malignant lesions	Uses dielectric differences for classification	Good contrast between tissues; non-invasive	Needs clinical validation	[56]
Post-treatment monitoring	Tracks tissue changes after surgery or radiation	Can detect permittivity changes over time	Limited scans at longer follow-ups	[57]
Lesion size and localization	Assesses size/location in real-time	Complements mammography where sensitivity is limited	Dense breasts can obscure signals	[58]
Lymph node metastasis detection	Detects axillary lymph nodes using radar MWI	Helps with TNM staging; reduces unnecessary surgeries	Difficult with overlapping tissues	[59]
Breast tissue classification with ML	Uses AI/ML to classify lesion-containing vs healthy scans	Improves sensitivity; useful for screening	Signal variability; system training needs	[60,61]
Early detection in dense breasts	MBI outperforms mammography in dense tissues	No compression, non-ionizing, suitable for frequent use	Still requires more data from large trials	[62]
Adjunct to conventional imaging	Used in combination with ultrasound/MRI	Improves diagnostic confidence	Requires data integration methods	[56,63]
ML-based diagnostic models	CNNs and U-Nets used for tumor classification & segmentation	Enhances accuracy and image reconstruction	Needs diverse, well-annotated datasets	[64,65]
Detection of treatment-related changes	Tracks dielectric changes post-radiotherapy	Shows significant permittivity differences	Limited follow-up data	[57]
Clinical feasibility of portable systems	Portable devices tested for in-clinic use	Cost-effective, accessible, repeatable	Small patient numbers so far	[66]
Synthetic breast phantoms and simulations	Used to validate algorithms and device configurations	Allows modeling of dielectric variability	Phantom data may not generalize to real breasts	[67,68]
Contrast-agent enhanced MWI	Use of nanoparticles (e.g., ZnO) for dielectric contrast	Enhances visibility of tumors	Needs further safety studies	[69]

MBI: Microwave breast imaging; MWI: Microwave imaging; TNM: Tumor-node-metastasis staging system; ML: Machine learning; AI: Artificial Intelligence; CNN: Convolutional neural network; UWB: Ultra-wideband; ZnO: Zinc oxide.

REAL-TIME IMPLEMENTATION CHALLENGES AND FUTURE DIRECTIONS

A realistic path forward requires a coordinated, interdisciplinary effort. From an engineering perspective, improving spatial resolution remains a top priority. This will depend on the development of more sensitive, broadband antennas with better coupling to biological tissue, as well as optimized array configurations that can image deep structures without compromising signal fidelity. In parallel, computational scientists must refine image reconstruction algorithms-balancing speed, accuracy, and robustness to noise. ML models show promise in lesion detection and tissue characterization, but they must become more interpretable, standardized, and computationally efficient to be useful in real-time clinical environments[74].

CONCLUSION

In conclusion, MBI holds clear promise as a radiation-free, comfortable, and potentially cost-effective breast imaging method. Although not yet ready to replace established modalities, ongoing innovation, rigorous clinical validation, and efforts to improve integration and affordability could eventually make MBI a key part of a personalized, accessible breast cancer screening.