Radar Technologies in Motion-Adaptive Cancer Radiotherapy

Abstract

Intra-fractional respiratory management represents one of the greatest challenges of modern cancer radiotherapy (RT), as significant breathing-induced lesion motion might affect target coverage and organs at risk (OARs) sparing, jeopardizing oncological and toxicity outcomes. The detrimental effects on dosage of uncompensated organ motion are exacerbated in RT with charged particles (e.g., protons and carbon ions), due to their higher ballistic selectivity. The simplest strategies to counteract this phenomenon are the use of larger treatment margins and reductions in or control of respiration (e.g., by means of compression belts, breath hold). Gating and tracking, which synchronize beam delivery with the respiratory signal, also represent widely adopted solutions. When tracking the tumor itself or surrogates, invasive procedures (e.g., marker implantation), an unnecessary imaging dose (e.g., in X-ray-based fluoroscopy), or expensive equipment (e.g., magnetic resonance imaging, MRI) is usually required. When chest and abdomen excursions are measured to infer internal tumor displacement, the additional devices needed to perform this task, such as pressure sensors or surface cameras, present inherent limitations that can impair the procedure itself. In this context, radars have intrigued the radiation oncology community, being inexpensive, non-invasive, contactless, and insensitive to obstacles. Even if real-world clinical implementation is still lagging behind, there is a growing body of research unraveling the potential of these devices in this field. The purpose of this narrative review is to provide an overview of the studies that have delved into the potential of radar-based technologies for motion-adaptive photon and particle RT applications.

1. Introduction

Radiotherapy (RT) plays a central role in oncology, being indicated in approximately half of cancer patients [1]. The primary aim of RT is to deliver therapeutic radiation to tumor cells while minimizing the dosage to the surrounding healthy tissues [2]. Most advanced RT technologies (e.g., intensity-modulated RT (IMRT), stereotactic body RT (SBRT), particle therapy (PT)) allow the delivery of highly conformal radiation doses and enable the treatment margins to be tightened [3]. However, motion-induced displacement of organs due to physiological movements (i.e., respiration, cardiac activity, peristalsis, etc.), also known as intra-fractional motion, may alter the dose distribution and impact target coverage and radiosensitive organs at risk (OARs) sparing [4,5]. As these uncertainties might thwart the benefits of RT if uncompensated, respiration-induced motion should be estimated, monitored, and accounted for [3,6]. Specifically, the American Association of Physicists in Medicine (AAPM) Task Group 76 (TG-76) recommends motion management in cases of respiratory excursion greater than 5 mm in any direction [7,8]. A more recent survey from the AAPM Task Group 324 [9] reports no consensus on the threshold for respiratory motion management for SBRT (38% threshold depending on tumor position and clinical factors, 31% no threshold, 31% threshold from 1 to 8 mm). For thoracic and abdominal cancer districts, including the liver, lungs, and pancreas, respiration-induced target displacement of several centimetres is reported [10]; therefore, motion management strategies must be considered. On this regard, the above-mentioned survey [9] reports that 95% of respondents use some form of motion management for thoracic and abdominal malignancies. Limiting this to lung cancer, there have been 2.5 million new cases worldwide in 2022 [11]; considering that over 60% of non-small-cell lung cancer patients (approximately 85% of the total) will undergo RT at some point [12], it becomes clear that the proportion of the problem is considerable.

There exist several methods for counteracting intra-fraction target displacement; the simplest is to use wider margins to cover the full range of tumor motion, while others consist of restricting respiration through abdominal compression or by asking the patients to hold their breath while the beam is on (i.e., breath hold) [3]. However, the use of large margins leads to increased normal tissue complications and hinders dose escalation [6]; compression belts are often uncomfortable and breath holding may prolong the treatment time and be impracticable for certain patient categories [13]. On the other hand, there exist other methods that instead of restricting respiration synchronize with it. Specifically, respiratory gating limits radiation exposure to a specific portion of the breathing cycle, rescanning irradiates the target volume multiple times to smooth out dose errors, and tumor tracking follows the tumor position and delivers continuous radiation [4,10,14].

Regardless of the technique employed, additional devices are usually required to capture the patient’s respiratory wave [6,12]. Respiratory-induced tumor displacement can be inferred either using external or internal surrogates. The external surrogate signal can be derived measuring the abdominal pressure variations (e.g., Anzai respiratory gating system, Anzai Medical Co., Ltd., Nishi-Shinagawa, Tokyo), chest wall motion (e.g., infrared-based or optical monitoring, Varian RPM system; Abches, University of Yamanashi, Japan), or air volume fluctuations (e.g., spirometry, Elekta ABC system) [2,6,13]. However, pressure sensors are invasive and sensitive to positioning, while optical-based systems are limited by the presence of obstacles. Additionally, studies have shown that often there is a non-constant phase shift between external and internal motion, which is difficult to be compensated for [6]. Indeed, all these systems rely on the hypothesis that external surface motion correlates well with internal tumor motion, which is not always the case [13]. On the other hand, internal marker tracking with fluoroscopy directly measures the trajectory of fiducials in close proximity of the tumor. Despite being a diffused approach, it is associated with serious clinical risks (e.g., pneumothorax and intra-alveolar bleeding) [15], poses concerns regarding the additional dose to patient, and can create inhomogeneity in the dose distribution. To overcome the problem of marker implantation and exposure to ionising radiation, markerless tumor tracking and electromagnetic transponders have been explored [6]. In this regard, ultrasound- [16,17] and magnetic resonance (MR)-guided RT approaches [18,19,20,21] are promising, due to their high spatial and temporal resolution, as well as the absence of ionising radiation. However, ultrasounds are limited by the presence of air and bones [17], while MR equipment is expensive and poses several technical challenges when used with particles [22,23].

In this scenario, radars represent a novel and promising approach for dealing with respiration-induced intra-fractional motion in cancer RT [4], as they are contact-less, non-invasive, and able to pass through obstacles. Their high temporal resolution makes them the perfect candidates to complement existing technology for managing motion, especially in RT with charged particles. The use of microwaves for detecting tiny physiological movements, including respiration and heartbeat, has been an active research topic since the early 1970s [13,23]; however, the first documented implementation attempts in the RT setting date back to 2011 [24]. There is a growing body of research on the potential of radars for monitoring respiration [25,26,27,28,29,30], including in RT settings [24,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48], although their use in clinical environments still lags behind. The scope of this narrative review is to survey the experiences of different research groups worldwide that have envisioned radar-based technologies in motion-adapted RT scenarios, focusing on limitations and innovations that could facilitate their utilization in real-world RT clinical settings.

2. Materials and Methods

Search Strategy

All fully retrievable original articles and editorials written in English on the use of radars for measuring displacement or motion in the context of RT applications were included in this study. Review papers were also cited but not analyzed in depth. Papers dealing with motion detection but not explicitly designed for RT applications or papers dealing with cancer diagnosis or treatment applications differing from motion monitoring (e.g., diagnostic support tool [49]) were not included. No additional inclusion or exclusion criteria were adopted. A search was conducted up to August 2025 and no constraints on publication date were set. Appropriate combinations of the terms “radar”, “millimeter wave”, “radiotherapy”, “radiation therapy”, “radiation oncology” and “particle therapy” were queried on PubMed, Web of Science, Embase, and Scopus databases.

3. Results

As many as 18 papers [24,31,32,33,34,35,36,37,38,39,40,42,43,44,45,46,47,48] met the inclusion criteria and were included in the study. Papers authored by the same research team were grouped together to avoid redundancy and better describe ideas and progress made within the same labs. Research experiences have been organized in the results section in a pseudo-chronological order, to highlight the milestones achieved. As a result, six research groups working with radars in motion-adaptive RT settings were identified. The different studies within the same research group were categorized with letters in alphabetical order and with a title representative of key findings to facilitate the reading.

The paragraph ends up with a section dedicated to the use of radars for motion management in medical imaging applications, being a pillar of RT workflow. The following subparagraphs highlight differences in experimental setups, in the choice of objects to be tracked, in the measures performed, and in the gold standards used, and report main results towards clinical implementation in a real RT environment.

Table 1. Summary of the research experiences dealing with radar in motion-adaptive RT applications.

Research Team	Papers	Sensing Infrastructure	Object Tracked	Experimental Setup/Tests	Gold Standard (GS)	Main Findings
Group 1 USA, 2011–2017 Texas Tech University (Lubbock, TX) University of California (San Diego, CA)	Study (a) Li et al., 2011 [24] Gu et al., 2011 [31]	Radar 2 patch antennas f = 5.8 GHz Single tone signal DC offset calibration and arctangent demodulation to extract respiratory signal	Actuator moving back and forth (A = 0.5 cm, f = 0.3 Hz) Healthy subject	The radar is placed in front of the actuator/subject (fixed distance 1 m) Arbitrary breathing rhythm and coached respiration (subject) Cough and breath hold were also simulated (subject)	MATLAB-simulated sinusoidal signal (phantom) No GS for measures on healthy subject	Human breathing motion was accurately tracked Radar is sensitive enough to identify disruptions (i.e., cough, breath hold) in RT process
	Study (b) Gu et al., 2011 [32] Gu et al., 2011 [33]	2 miniature Doppler radars (chest/abdomen) 2 pairs of patch antennas f = 2.4 GHz, P = 0 dBm Wireless transmission (ZigBee) Single tone signal DC offset calibration and arctangent demodulation to extract respiratory signal	Healthy subject (free/coached breathing) Lung cancer patient	Two radars are placed in front of abdomen and chest, respectively (fixed distance 0.5 m) Arbitrary breathing rhythm and regular (coached) respiration (subject) Impact of antenna array size has been investigated Tumor motion from simulated radar output is compared to that measured by X-ray (fluoroscopy) for one lung cancer patient	X-ray fluoroscopy (lung cancer patient) No GS for measures on healthy subject	The 2 radar configuration is feasible for respiratory tracking As antenna array size increases, radiated area decreases, spatial resolution increases Lung cancer patient: absolute tracking error 0.12 mm
	Study (c) Gu et al., 2012 [34] Fung et al., 2016 1 [40]	Miniature DC-coupled radar sensor f = 2.4 GHz	Actuator (sinusoidal motion simulating also physiological stationary moment) Moving phantom (sinusoidal motion, T = 5 s) Healthy subject (coached breathing) Respiratory gating platform (RGP) (A = 5, 10, 20 mm; T = 4, 5, 6 s)	DC-coupled vs. AC-coupled radar measuring actuator motion Radar vs. Varian RPM on moving phantom (sinusoidal motion, T = 5 s), beam on Radar only on healthy subject Radar on RPG	Programmed actuator movement (DC vs. AC) RPM (moving phantom) No GS for measures on healthy subject Actual motion pattern of the respiratory gating platform (RPG)	Unlike the AC-coupled one, the DC-coupled radar preserves stationary information Phantom motion is well reproduced by radar when compared to RPM Compatibility with beam on was demonstrated Reproducible respiratory signals from healthy subject allows for gating Radar RMSE for RGP submillimetric scale (0.091–0.211 mm)
	Study (d) Gu et al., 2012 [35] Salmani et al., 2012 [37] Salmani et al., 2012 [38] Ren et al., 2012 [39]	Radar sensor Beam scanning highly directional Fermi antenna array (decrease size of system/improvement of SNR) Omnidirectional monopole antenna (for comparison) f = 5.8 GHz	Actuator (f = 0.2 Hz, A = 6 mm)	Fermi vs. monopole antennas placed in front of the actuator (fixed distance 0.5 m)		SNR of the Fermi wrt monopole antenna was ~53 vs. ~33 dB The proposed antenna array serves well for a radar sensor to monitor respiration at multiple locations
	Study (e) Gu et al., 2013 [36]	DC-coupled radar sensor	Actuator moving sinusoidally Healthy subject (breathing normally but with stationary moment) Shaker (used only for distortion analysis of AC-coupled radar)	DC- and AC-coupled radars to track actuator motion DC- and AC-coupled radars to track healthy subject respiratory motion AC-coupled radar to measure shaker motion at different vibration frequencies	Programmed actuator motion	AC-coupled radar: distortion when tracking stationary moment of actuator and of subject (generation of false gating signals) DC-coupled radar: accurately follow the actual motion of actuator/subject, preserving stationary information It is possible to use AC-coupled radars for distortion-free measurement at the extra cost of settling time and hardware cost
Group 2 USA, 2013–2021 John Hopkins University School of Medicine (Baltimore, MD) Arizona State University (Tempe, AZ)	Study (a) Han-Oh et al., 2013 1 [42]	UWB radar f = 3.1–10.6 GHz	Aluminum plate + a series of biological (or equivalent) tissues: pig skins, cow muscles, bones, fat, and wet sponges (lung substitute) Miniature version of human thorax (skin–fat–muscle–wet sponge stacked in series), with a 5 cm balloon behind to mimic lung tumor	Signals from aluminum plate under tissue samples and tissue only are recorded. The 2 signals are subtracted to extract the signal from aluminum only Signal from water balloon behind miniature thorax phantom was collected	No GS employed, strength of reflected signal with and without tissue layers was used as a measure of penetration capability	Penetration capability of radar signal through biological tissues for tumor detection was demonstrated
	Study (b) Han-Oh et al., 2021 [43]	VNA Wideband directional horn antenna (f = 2–6 GHz, P = 3 dBm) Super-resolution MUSIC algorithm to localize markers (1.2 × 10 mm)	Human-like (i.e., made of skin-, fat-, muscle-equivalent layers) phantom with markers embedded in modular Styrofoam blocks at different depths (0 to 50.8 mm at 12.7 mm steps)	Static case: two experiments, one with one single marker (to assess object detectability) and the other with two markers (to assess capability of separating two markers close in space, put 50.8 mm apart) behind the tissue layers; both numerical simulations and real experiments were performed Dynamic case: a microwave MIMO radar (18 antenna pairs, f = 3–10 GHz) measures 3D trajectory of a moving fiducial marker	Layers of known thickness	Feasibility of detecting markers inside patient’s body and to distinguish two markers close in space Average localization accuracy rates were 1.0 ± 0.4 mm and 3.5 ± 1.3 mm for simulation and experiment, respectively Returned power decreases with increasing distance of marker from the antenna Capability of depicting 3D trajectory of a moving marker
Group 3 Japan, 2016 University of Hyogo (Hyogo) Kyoto University (Kyoto)	Muragaki et al., 2016 [44]	UWB array radar with adaptive beamforming technique f = 60.5 GHz, BW = 1.25 GHz 4 TX, 4 RX antennas Optical Kinect sensor to construct body surface model DOA technique (Capon method) + DCMP algorithm + phase rotation to separate abdominal and chest motions	Healthy subject	Simulation study Radar placed at fixed distance (600 mm) 3 scattering points on the body surface were identified	Computed reflection points position	Separation of abdominal and chest motions was successful RMSE for the 3 points was submillimetric (<0.26 mm)
Group 4 Canada, 2022 University of Ottawa (Ottawa, ON) The Ottawa Hospital Research Institute (Ottawa, ON) Carleton University (Ottawa, ON)	Fallatah et al., 2022 [45]	Single, low-cost, off-the-shelf UWB radar One patch antenna (65 degrees aperture) f = 5.9–10.3 GHz ANN to filter out gantry noise k-means algorithm to classify breathing patterns	Respiratory motion phantom (RMP)	RMP was placed at 0.7 m from radar Simulation of noise introduced by LINAC and MLC motion No-breathing, breath hold, free breathing, and deep inspiration were investigated	Known breathing pattern	Feasibility to recover breathing signal in presence of gantry noise Feasibility to classify breathing patterns with accuracy >0.85 and >0.70 without and with noise, respectively
Group 5 USA, 2019–2024 Washington University in St. Louis (St. Louis, MO) California Institute of Oncology (Pasadena, CA)	Study (a) Olick-Gibson et al., 2019 [46]	mmWave radar 2 TX, 4 RX Chirp signal f = 77–81 GHz FFT to extract surface displacement FOV: ±60° horizontal, ±20° elevation	Brass plate as a reflective marker put on two slabs of solid water on treatment couch Human volunteer (face mask/chest displacement)	Radar installed on the top extremity of gantry bore Table moving in vertical direction by different distances (1–10 mm) at different heights below isocenter (100, 150, and 200 mm) Fine displacement (0.1–0.7 mm) at 200 mm was also investigated Presence of obstacles was also investigated (phantom: gown, alpha cradle; human volunteer: face mask) Chest displacement of the subject was measured to extract respiratory and cardiac waveforms	Known displacements of the brass plate Vernier respiration monitor belt (respiratory signal of subject) ECG (cardiac signal of subject)	Table displacement was tracked with <0.1 mm accuracy (same as table) With obstructions, object distance is slightly underestimated (around 1 mm accuracy) The device is able to detect respiratory and cardiac signals simultaneously (correlation with gold standard respiratory/cardiac signal: 0.9156/0.7895)
	Study (b) Bressler et al., 2024 [47]	FMCW mmWave radar Chirp signal f = 77 GHz, BW = 4 GHz CZT-phase algorithm	Solid water slabs (calibration) for displacement measures Human-shaped phantom Typical RT obstruction objects (Styrofoam, sponge, thermo-plastic mask)	Static measurement (slab) Gantry angle rotating by 360°	Tape measures (slab) CBCT (phantom)	Submillimetric displacement accuracy (0.71–0.92 mm) Lower accuracy for angles posterior to the couch
Group 6 Japan, 2025 Kindai University (Osakasayama, Osaka, Japan) Kyoto University of Medical Sciences (Kyoto, Japan)	Kosaka et al., 2025 [48]	Doppler mmWave radar f = 24 GHz	QUASAR respiratory motion platform 20 healthy volunteers (18 adults, 1 child, 1 infant)	Experiments on phantom to test repeatability and reproducibility Adults lying supine 40 cm and standing 180 cm from the system in free breathing and breath hold; child and infant lying supine 40 cm from system	No GS for measures on healthy subjects	Waveforms from heathy subjects were correctly reproduced, both in supine and standing positions, both in free breathing and breath hold

¹ conference abstract. Abbreviations. A: amplitude; AC: alternating current; ANN: artificial neural network; BW: bandwidth; CBCT: cone beam computed tomography; CZT: chirp z transform; dBm: Decibel-milliwatt; DC: direct current; DCMP: directionally constrained minimization of power; DOA: direction of arrival; ECG: electrocardiogram; FFT: fast Fourier transform; FMCW: frequency-modulated continuous wave; FOV: field of view; GS: gold standard; LINAC: linear accelerator; MIMO: multiple input multiple output; MLC: multi-leaf collimator; mmWave: millimetric wave; P: power; RGP: respiratory gating platform; RMSE: root mean square error; RPM: real-time position management; RT: radiotherapy; RX: receiver; SNR: signal-to-noise ratio; T: period; TX: transmitter; UWB: ultra-wideband; VNA: vector network analyzer; wrt: with respect to.

summarizes all these findings, while Figure 1 highlights milestones achieved.

3.1. Group 1: USA, 2011–2017

This research group published 10 experimental papers [24,31,32,33,34,35,36,37,38,39,40], 1 patent [41], and three reviews [4,12,50] on the topic between 2011 and 2017. They can be considered pioneers and early adopters, as they were the first ones investigating the feasibility of using radars for RT applications. In the following, five different macro-studies were identified and discussed.

3.1.1. Study (a): First Feasibility Study

In this first series of studies [24,31], the authors used a 5.8 GHz radar with quadrature demodulation architecture featuring two patch antennas, one used for transmitting (TX) and the other one for receiving (RX). From the TX antenna, a single tone microwave signal is radiated towards a moving surface. The periodic motion of the surface modulates the phase of the signal, which is reflected back towards the RX antenna. The respiration signal is obtained after proper direct current (DC) offset calibration and arctangent demodulation. Firstly, the radar was employed to track the periodic motion of an actuator to assess the reliability of the sensing infrastructure. Subsequently, it was used to track free and coached breathing of a sitting healthy volunteer placed at one meter distance from the antennas. Finally, to account for some possible off-nominals that could occur during an actual RT session, breath holding and coughing were also simulated. In [24], the authors also presented a mathematical model of lung motion, based on 4D computed tomography (4DCT) of a patient, to correlate the external chest movement with the internal tumor motion. Being a feasibility study, only descriptive results were reported, and no specific accuracy values can be retrieved.

3.1.2. Study (b): Tumor Motion Estimation

In these two works [32,33], the authors used a pair of 2.4 GHz Doppler radars mounted on a fixation frame anchored to the RT treatment couch to track the chest and abdomen motions of a healthy volunteer, during free and coached breathing. The respiratory signal extracted from radar measurements was benchmarked against the Varian real-time position management system (RPM), tracking the motion of an infrared marker placed on the phantom. Similarly to the previous set of studies, DC offset removal and arctangent demodulation were performed. Collected data were transmitted outside the treatment room wirelessly using ZigBee technology. The authors also highlighted the safety of the system, transmitting a power of 0 Decibel-milliwatt (dBm), i.e., 1000 times lower than cellular phone peak power at the time of the study. They showed that the signal extracted from coached breathing allows for both amplitude or phase gating, depending on the clinical case. They also demonstrated that as the array size increases, the radiation pattern narrows, the radiated area decreases, chest and abdomen signals are better separated, and the respiration signal is closer to the ground truth. Finally, they validated the tracking capability of the system by comparing tumor motion obtained from simulated radar output with that measured by X-ray imaging, considering a series of 41 s fluoroscopy of a real lung cancer patient. They obtained an average tracking error of 0.15 mm.

3.1.3. Study (c): Tests with RT Beam on

In these experiments [34,40], they programmed an actuator to perform sinusoidal trajectories, including stationary movements between adjacent cycles, to reproduce the static phase characterizing the last portion of a standard end-exhale breathing pattern. The DC coupled radar outperformed its alternating current (AC) counterpart, as it preserved the stationary information and overall signal integrity. After completing these tests, they validated the DC radar in a real RT environment. The DC-coupled radar sensor was fixed on the treatment couch, and the accuracy in tracking the surface of a moving phantom and a respiratory gating platform (RGP) was benchmarked against the Varian RPM. Measures on a healthy volunteer were also performed. They showed that phantom motion was well reproduced, reliable gating signals from the healthy subject were derived, and the radar root mean square error (RMSE) for the RGP was submillimetric. Finally, to better simulate its use in a real RT environment, they performed some tests with the beam on, demonstrating the compatibility of the radar with photon radiation.

3.1.4. Study (d): Multi-Point Measurements

In this set of studies [36,37,38,39,40], the innovation is the introduction of an antenna array featuring beam scanning; instead of using multiple radars, one single unit is employed to measure respirations at multiple points. Not only does this have a positive impact on the size and cost of the system, but also the single antennas feature higher gain that results in a higher signal-to-noise ratio (SNR). The antenna array chosen for the experiments was a Fermi antenna array, and its performances were evaluated in comparison to an omnidirectional monopole antenna in measuring the motion of an actuator placed 0.5 m away from the radar. They proved that the directivity and SNR of the antenna array were significantly higher than those of the monopole.

3.1.5. Study (e): Stationary Information

Here [37], the authors compared the performances of a DC- and an AC-coupled radar to track the sinusoidal and stationary motion of an actuator, breathing of a subject, and oscillations of a shaker vibrating at different frequencies. For both the actuator and healthy subject experiments, DC-coupled radar was able to preserve the stationary information, unlike the AC counterpart. Regarding the measures with the shaker, they showed that distortion-free measurements using AC-coupled radar is possible, as long as all harmonic characteristics are preserved, at the cost of the augmented settling time and hardware cost.

For the sake of completeness, the same research group also published a review [12] and a book chapter [4] summarizing the main findings of all their previously published studies. In 2017, they published another review on microwave radars, where they gave a name to the 2.4 GHz radar-based system they developed for motion management, namely “iMotion2” [50]. Outside of RT applications, in 2015, they presented a 2.4 GHz multi-system radar to classify physical and emotional respiration patterns of a subject [51].

3.2. Group 2: USA, 2013–2021

This was the first research group to investigate the feasibility of tracking a fiducial marker below a human-like surface (study a).

3.2.1. Study (a): Inner Tumor Detection

Here [42], the authors investigated the theoretical capability a ultra-wideband (UWB) radar to penetrate biological tissues and detect tumors. Two experimental setups were employed. In the first one, some biological (or equivalent) tissues were piled up upon an aluminum plate to be detected by the radar. In the second one, a phantom simulating a miniature thorax embedding a water ballon used as a tumor surrogate was employed. In both cases, the radar was able to succesfully penetrate tissues and register a signal from the aluminum plate and from the tumor surrogate.

3.2.2. Study (b): Fiducial Marker Localization

In this study [43], the authors performed a series of experiments with a vector network analyzer (VNA) and a wideband directional horn antenna, transmitting signals with frequencies ranging from 2 to 6 GHz and output power of 3 dBm. The experiments were in line with those of their previous study. However, instead of using real biological tissues from animals or surrogates (e.g., sponge for lungs), they carefully selected synthetic tissue-equivalent materials in terms of dielectric properties. Accordingly, they built up a phantom with skin-, fat-, and muscle-equivalent layers of different thicknesses, and below the surface inserted a cylyndrical gold marker (1.2 × 10 mm) embedded in Styrofoam blocks. They performed two experiments to prove the ability of the sensing infrastructure to (1) localize a moving fiducial marker and (2) isolate the signal coming from two fiducials placed at a certain distance (around 5 cm). The average localization accuracy was found to be 3.5 ± 1.3 mm (mean ± standard deviation). At the end, they demonstrated that fiducial marker localizazion was possible and that two markers close in space can be discerned. Additionally, they showed that the 3D trajectory of a moving marker can be tracked.

3.3. Group 3: Japan, 2016

Fine Separation of Points

This research group [44] proposed a new method to separate signals from the chest and abdomen using a UWB array radar with the so-called adaptive beamforming technique. They validated the method with a numerical simulation, using a Kinect sensor to model the body surface. They were able to identify at the same time displacement of three scattering points with submillimetric accuracy (0.138, 0.123, 0.259 mm). They also published another paper on the use of radar for monitoring the respiration of multiple people at once [52], which is not detailed here as it is not specifically related to RT settings.

3.4. Group 4: Canada, 2022

LINAC Motion Noise

As the authors point out [45], the novelty of this work is the use of an artificial neural network (ANN) to extract breathing signals using a radar in the presence of interference from gantry motion, to simulate a real clinical volumetric-modulated arc therapy (VMAT) environment. They used a single UWB radar in the range of 5.9–10.3 GHz and a patch antenna (65 degrees aperture) placed at approximately 0.7 m from a respiratory motion phantom (RPM). A regular sinusoidal breathing pattern and a deep inspiration breath hold (DIBH) sequence were explored. They also used a k-means algorithm to classify different breathing patterns. They were able to recover breathing signals effectively and to carry out the classification task with acceptable accuracy (>0.85 and >0.70 without and with noise, respectively). Among future directions, Fallatah and colleagues are willing to construct a multisensor device integrating a radar, microphone arrays, and cameras to boost performance [45].

3.5. Group 5: USA, 2019–2024

3.5.1. Study (a): Presence of Obstacles

In this publication [46], the authors investigated millimetric wave (mmWave) technology to monitor patient surface displacement, as well as respiratory and cardiac cycles. They installed a radar inside a gantry-based RT linear accelerator (LINAC), measuring the displacement of a reflective surface placed on slabs of solid water on the couch. They performed measurements at different table heights, varying the vertical position between 1 and 10 mm. Submillimetric variations (0.1–0.7 mm) were also tested. Two TX and four RX antennas were used. Displacement measurements were performed also with the presence of obstacles (i.e., gown, alpha cradle and mask) in the radar line of sight (LOS). They used chirp signals at 77–81 GHz and employed the fast Fourier transform (FFT) to detect the surface. They estimated the distance from frequency and phase information of the signal obtained as the difference between the transmitted and received one. Respiratory and cardiac waveforms were obtained with proper bandpass filtering. Such sensing infrastructure was capable of monitoring surface displacement with an accuracy better than 0.1 mm without obstructions and 1 mm with obstructions, sufficient for RT applications. They also demonstrated the capability of the system to monitor cardiac and breathing cycles.

3.5.2. Study (b): Submillimetric Accuracy

In this second publication [47], the authors employed a 77 GHz frequency-modulated continuous wave radar and a chirp z transform-based (CZT) algorithm to process the signal. As a first experiment, they measured the absolute distance of a metal plate position below a slab of solid water, with and without the presence of obstacles (Styrofoam, sponge, and thermoplastic mask). Subsequently, they compared radar source to skin distance (SSD) measurements to those of cone beam computed tomography (CBCT), considered as the ground truth. The gantry was rotated 360 degrees at 10 degree steps and each time measure was acquired by the radar. They discovered that for angles posterior to the couch, the phantom signal overlapped with that of the couch, which was composed of conductive carbon fiber material. With and without obstructions, submillimetric displacement accuracy of a flat surface was achieved using the CZT algorithm. However, the same measurements performed on a human-shaped phantom were as large as 8 mm at some angles. Regarding future directions, Bressler and colleagues suggest embedding in the treatment couch a small radiofrequency (RF) reflector, which is transparent to the treatment beam but not to mmWaves (e.g., aluminum foils or graphene sheets), to get a stronger reflection signal. They also observed that the fiberglass table top could represent a valuable alternative to carbon fibre ones, especially in the presence of posterior beams [47].

3.6. Group 6: Japan, 2025

Towards Clinical Implementation

This research group from Japan published the latest paper on the topic [48], envisioning the use of radars for clinical management of respiratory motion during imaging and RT. They employed a 24 GHz mmWave Doppler sensor (MWS) capable of measuring the velocity of objects by detecting the frequency change between transmitted and reflected mmWaves. The directivity properties of the antenna were analyzed in both azimuth and elevation directions. A QUASAR respiratory platform (Modus Medical Devices, Ontario, Canada) was employed to test the repeatability and reproducibility of the MWS. The experiments were performed on 20 healthy volunteers. Eighteen of them were adults, and their respiratory waveforms were recorded both laying 40 cm and standing 180 cm distant from the MWS, during both free respiration and breath-hold. The respiratory motion patterns of a child (age: 4 years) and an infant (age: 6 months), while sleeping in the supine position at 40 cm distance from the system, were also evaluated. The performance of the system was mainly evaluated qualitatively (i.e., the shape of the waveforms was deemed to be plausible). They offered valuable insights into some conditions that facilitate the use of radars and that are automatically met in imaging and RT scenarios. First, they observed that when using the system, no one should be in close proximity to the patient (about 1 m) to avoid any possibile interference; this is a condition that is automatically fulfilled in X-ray-based imaging and RT applications, as it is generally forbidden for the operators to be inside the diagnostic room or RT bunker during image acquisition and treatment delivery due to radiation protection reasons. In addition, they emphasized that the patient should be in contact with stationary objects to limit motion unrelated to respiration, which is true in this setting.

3.7. Applications in Imaging

There were two other research groups, both from Germany, who in 2023 investigated the feasibility of radars to be used for tracking respiration in medical imaging, namely computed tomography (CT) and MR [53,54]. These investigations will not be analyzed in detail, as they are not strictly related to motion management in RT. However, as imaging is central in the RT pipeline, including in diagnosis, planning, in-room guidance, and follow-up, the main findings from this studies are herein also reported for completeness. Pfanner and colleagues [53] investigated the capability of a radar unit working at 860 MHz equipped with four TX antennas and one RX antenna placed on the surface of the couch of a CT scanner to measure motions of a phantom and respiration of ten volunteers. The radar infrastructure was able to assess the respiratory motion quite well, even if tests with operating CT were not performed. However, the actual effect of noise due to CT gantry rotation was deemed negligible, as (1) the patient’s body partially shields antennas from this source of noise, and (2) the gantry rotating frequency is significantly higher than the breathing frequency and can be easiliy filtered out. Neumann and colleagues [54] proved the capability a 1–5 GHz UWB radar to predict displacements in the liver due to respiration in MR applications and prospectively correct for them. At every tissue interface, part of the radar signal is reflected back to the receiver and part is transmitted through the layers. Variations over time of the signal embedding all reflections are representative of changes in internal anatomy displacement and allow the extraction of a surrogate of respiration. To guarantee compatibilty with the MR system, they used an MR-compatible antenna described elsewhere [55]. They evaluated this approach in simulations, phantom experiments, and in vivo scans of two healthy volunteers and demonstrated that it is accurate, suitable for different motion types, and stable over time.

4. Discussion

To the best of our knowledge, this work represents the first comprehensive overview of studies investigating the use of radars for motion management in radiation oncology. In the following, the main results of the above-mentioned studies will be critically discussed, highlighting strengths and weaknesses of radar technologies, envisioning their application in real-world radiation oncology settings.

4.1. Intra-Fraction Motion in RT: Can Radars Solve the Unmet Needs?

Accurate setup and motion monitoring are fundamental to deliver safe and effective RT [47]. This is particularly true for most advanced photon-based RT (e.g., SBRT) approaches, where ablative doses are delivered to the tumor and tight margins are needed to spare healthy tissues, as well as for PT, whose steep dose gradients require submillimetric precision [10]. Regardless the technique employed, in the era of high-precision cancer RT, effective motion management strategies have, therefore, become mandatory, and accurate respiration tracking is crucial [34]. To tackle this issue, a large variety of methods and solutions have been developed over the last two decades [10]. However, differing from image guidance to counteract inter-fraction motion (e.g., variations occurring between fractions due to discrepancies in setup and patient anatomy), a standard procedure for effectively addressing intra-fractional motion does not exist yet, especially for PT. Indeed, gating, mid ventilation, tracking, breath hold, and 4DCT represent only some of the available motion management strategies [9], which have their own strengths and weaknesses. This lack of consensus on the best technique to employ is due to the fact that several challenges still remain unaddressed. For this reason, there is space and interest in the research community for new solutions, and radars are perfect candidates.

4.2. Radars: An Established Technology for a Novel Application

Although radars represent a niche in the intra-fractional motion management panorama, the latest studies published on the topic have received renewed interest in this regard. Radars have attracted attention as inexpensive, compact, non-contact, non-invasive, and harmless systems. Pre-clinical tests proved the feasibility of such devices to accurately measure static displacement, to register chest and abdomen excursions to derive respiratory signal surrogates, and even to track tumor motion in RT environments. The analyzed studies put significant efforts into better modeling of real-world RT scenarios, including fiducial marker tracking using a phantom with human-like tissues [43], a simulation of the presence of thermoplastic mask [46,47], and a study of the interference due to LINAC motion [45]. In addition, most recent studies introduced artificial intelligence (AI)-based processing [45].

4.3. How Far Are We from Real-World Clinical Applications?

As anticipated in the previous sections, clinical implementation of radars in RT still lags behind research, as outstanding challenges remain:

• Radar and antenna technology: A variety of radar units, working frequencies, and antenna types have been explored. The best trade-off between the radar unit size, antenna gain and emitted power, spatial and temporal resolution, and motion tracking accuracy should be pursued. In practice, as the signal is attenuated as it travels through tissues, it is important to evaluate the effective attenuation in the back and forth path of transmitted and received pulses, respectively, to assess the optimal radar working parameters [56]. It may help to perform numerical and empirical simulations with human models of different sizes and tissue thicknesses to accurately determine the frequency and power ranges that should be valid for most patient categories.
• Real-world accuracy: For most advanced RT techniques, a high level of accuracy is required. For instance, in PT, the sharp dose gradients are vulnerable to even small variations in patient geometry [2]. Experiments show that in simulation scenarios or in very controlled environments, this goal can be achieved [34,44,47], but in real clinical scenarios, with the presence of thermoplastic mask and gantry motions, the accuracy often dramatically decreases below the requested needs. In addition, motion perpendicular to radar LOS is difficult to track and can undermine the overall system performance [54]. In this regard, it could be interesting to test the systems against all the complexities characterizing a real-world clinical environment at once.
• RT integration: Most available commercial devices for respiration monitoring in RT are well integrated with hardware and software components of modern LINACs. Proper systems for anchoring radar units to LINACs and thorough analyses of the best layout in the room are warranted to make the most of this technology. Integration with RT software should be pursued as well. Wired and wireless connection options from the treatment bunker to local control room should be also investigated.
• RT compatibility: Some authors have investigated compatibility with beam and gantry motions. However, studies to assess any possible interference with different types and energies of radiation (e.g., not only photons but also charged particles), with different accelerator architectures, and with other technologies populating RT bunkers should be also assessed.
• Safety: Although several authors claim that the system is harmless [32,53], radiation thresholds that ensure acceptable SNRs, which at the same time are safe for the patient, should be identified.
• Compatibility with cardiac electronic implantable devices (CIEDs): To monitor respiration, thorax motion is usually tracked. This means that the electromagnetic waves are directed towards mediastinal organs including the heart and possible CIEDs. The presence of CIEDs is generally allowed for RT treatments, as long as accurate monitoring of the device is performed [57]. In the use of radars in patients with CIED, an accurate evaluation of the power of the transmitted wave is, therefore, mandatory to assess the electromagnetic compatibility. As a reference, for high-frequency signals (i.e., >150 kHz), 141 V/m is set as the limit for safe use of Medtronic CIEDs (Medtronic PLC, Minneapolis, MN), according to the standards ANSI/AAMI/ISO 14117 [58], EN 45502-1 [59], EN 45502-2-1 [60], EN 45502-2-2 [61], ISO 14708-1 [62], ISO 14708-2 [63], ISO 14708-6 [64], ANSI C95.6 [65], ICNIRP guidelines [66].
• Presence of metallic implants. Patients undergoing RT sometimes present with metallic implants. While this does not represent a contraindication per se to the use of radars, the implants may distort the reflected signal, if placed along the beam path. A careful analysis of the reflected signal is, therefore, useful to detect any anomalies.

4.4. Radars in Particle Therapy: A Niche in a Niche

The long-term goal of the RT community has been to “see what we treat, as we treat” [10]. Accordingly, accelerator technology is moving towards real-time treatment adaptation, and additional hardware and software equipment has been developed to serve this purpose. However, marked differences and evident unbalances between photons and particles exist. As an example, MRI guidance has gained popularity in photon RT, given its excellence spatial and temporal resolution [67], while MRI-guided PT has been hitherto limited to research experiences and is in its infancy due to several technical challenges still to be fully addressed [22]. This means that besides adapting existing technologies developed for conventional photon RT, there is the need to find new solutions for efficient motion-adaptive PT. Admittedly, the potential benefits of using radars to monitor respiration in conventional photon RT are further enhanced in PT scenarios. As a matter of fact, it is well-known that precise coverage of the target volume is even more challenging in PT than in conventional photon-based RT, as particles are more sensitive to anatomical variations [22]. However, although for moving target volumes real-time image guidance is of even greater importance in proton than for photon RT [22], motion monitoring technologies are paradoxically far more advanced in more conventional RT settings [21], as transferring motion monitoring technologies from photons to particles presents several challenges [10]. Therefore, it is plausible that radar could help PT to catch up with photon RT in this regard [68].

5. Conclusions

Although traditionally associated with defense and security domains, radar sensors have been successfully used for decades for medical applications. In motion-adaptive cancer RT, they have gained increasing attention due to their contactless technology and small form factor. All the studies reported in this review proved the feasibility of employing radar sensing infrastructure in this setting. They hold promise for addressing the shortcomings and unmet needs of the currently implemented technologies, towards the standardization of motion management techniques. In principle, the same research findings could be easily extended to the PT world, for which the benefits of such technologies could be even more visible, although no such studies exist to date. The efforts of the radar and cancer RT research communities should be directed towards providing more solid and robust data for clinical implementation. To accomplish this purpose, a reliable benchmark against established gold standard systems and safe integration with existing technologies in a multi-sensor fusion approach should be pursued.