---
title: 'FLASH Radiotherapy: Advances & Challenges'
url: https://www.emergentmind.com/topics/flash-radiotherapy-flash-rt
type: topic
---

# FLASH Radiotherapy: Advances & Challenges

Searching arXiv for recent and foundational papers on FLASH radiotherapy to ground the article in current literature.
arXiv query: FLASH radiotherapy proton review dose rate optimization commissioning dosimetry mechanisms
FLASH radiotherapy (FLASH-RT) is the ultra-high dose-rate delivery of a therapeutic radiation dose in a fraction of a second, with a commonly used operational threshold of $\ge 40\ \mathrm{Gy/s}$ and, in several formulations, single large doses delivered in sub-second time windows [2203.11047]. In preclinical literature, this delivery regime is associated with the “FLASH effect,” namely reduced normal-tissue toxicity with apparently preserved tumor control in many models, although the exact physical and biological conditions that trigger the effect remain incompletely characterized and the underlying mechanism is still unclear [2405.10219]. Contemporary FLASH-RT research therefore spans radiobiology, accelerator physics, dosimetry, treatment planning, quality assurance, and translational beam-delivery engineering.

## 1. Definition, dose-rate criteria, and scope

FLASH-RT is conventionally distinguished from conventional dose-rate radiotherapy by both average and instantaneous dose-rate scales. One formulation gives conventional RT as $0.01\text{–}0.1\ \mathrm{Gy/s}$ and FLASH-RT as $\ge 40\ \mathrm{Gy/s}$, with instantaneous dose rates $\ge 10^5\ \mathrm{Gy/s}$, total dose per fraction $D\ge 10\ \mathrm{Gy}$, and total delivery time $T_{\mathrm{tot}}<0.2\ \mathrm{s}$ [2203.11047]. In proton and scanned-particle work, a voxelwise FLASH criterion is often defined by a joint dose and dose-rate condition. For example, one proton Bragg peak formulation treats a voxel $v$ as FLASH-qualified only if both $D(v)\ge 5\ \mathrm{Gy}$ and $\dot{D}(v)\ge 40\ \mathrm{Gy/s}$ [2505.06223].

Several dose-rate metrics coexist. In pencil-beam-scanning proton therapy, the literature distinguishes instantaneous dose rate, average dose rate, dose-averaged dose rate (DADR), average dose rate (ADR), and dose-threshold dose rate (DTDR), and explicitly notes the need for standardization of PBS dose-rate definitions [2206.11722]. This directly bears on the interpretation of “FLASH coverage,” because different metrics can report materially different irradiated volumes as FLASH-compliant.

A common misconception is that FLASH-RT is defined by a single universal threshold. The literature instead presents thresholds as modality- and assay-dependent operational criteria. In vitro studies summarized in proton SFRT-FLASH planning report tissue-sparing beginning at $40\ \mathrm{Gy/s}$ and $\ge 5\ \mathrm{Gy}$ per fraction, whereas accelerator-oriented overviews describe current estimates in which large doses of $10\ \mathrm{Gy}$ or more delivered in $200\ \mathrm{ms}$ or less produce normal-tissue sparing effects [2505.06223]. This suggests that “FLASH conditions” should be treated as an experimentally grounded multidimensional parameter space rather than a single scalar cutoff.

## 2. Physicochemical and biological hypotheses

Current mechanistic work organizes the FLASH effect around oxygen depletion, radical chemistry, DNA-damage signaling, mitochondrial pathways, and immune modulation. A minimal oxygen-depletion model writes
$$
\frac{d[\mathrm{O}_2]}{dt}=-k_1[\dot{D}(t)],
$$
with empirical depletion constants reported as $\approx 0.17\ \mathrm{mmHg\ Gy^{-1}}$ in vitro and $\approx 0.11\ \mathrm{mmHg\ Gy^{-1}}$ in vivo in one review [2405.10219]. However, the same review emphasizes that in many FLASH experiments the net $\Delta p\mathrm{O}_2$ is only a few mmHg, too small to explain the full normal-tissue sparing by a simple oxygen-fixation argument alone [2405.10219].

Radical-recombination hypotheses address this limitation by making the relevant state variable the transient concentration of radiolysis products rather than only mean oxygen level. A representative model writes
$$
\frac{d[R^\bullet]}{dt}=k_{\mathrm{prod}}\dot{D}(t)-k_{\mathrm{rec}}[R^\bullet]^2,
$$
with analogous kinetics for peroxyl radicals [2405.10219]. A more explicit 2025 study adopts a two-compartment radiochemical model for peroxyl radicals and reports that under CONV-RT at $0.05\ \mathrm{Gy/s}$ recombination is negligible ($<0.001\%$) even at $15\ \mathrm{Gy}$, whereas under FLASH-RT at $900\ \mathrm{Gy/s}$ recombination efficiency reaches $\sim 8\text{–}10\%$ at $15\ \mathrm{Gy}$ delivered in $<20\ \mathrm{ms}$; the recombination-versus-delivery-time curve yields a $50\%$-maximal recombination at $\approx 300\ \mathrm{ms}$, corresponding to a mean dose-rate threshold of $\approx 100\ \mathrm{Gy/s}$ [2508.20510].

Temporal microstructure also matters. Monte Carlo IRT modeling and physicochemical measurements indicate that compared with multi-pulse irradiation, single-pulse mode with a width less than $1/10$ of the radical lifetime, a repetition interval longer than the radical lifetime, and a dose exceeding $1\ \mathrm{Gy/pulse}$ can lead to rapid radical consumption, reducing residual content; when the single-pulse dose exceeded $1\ \mathrm{Gy}$, the overlap probability approached $100\%$ [2504.19927]. This suggests that average dose rate alone is insufficient to specify the relevant biological regime.

At the biological level, the literature describes reduced clustering of double-strand breaks, lower cytosolic dsDNA release, suppression of cGAS–STING signaling in normal intestine, altered mitochondrial signaling, modulation of immune response, and circulating lymphocyte sparing as candidate components of the downstream response cascade [2405.10219]. In the intestinal-injury model that links radical kinetics to transcriptomics, FLASH-RT activated the NRF2 antioxidant pathway, suppressed ERK signaling, increased the GSH/GSSG ratio from $3.8 \pm 0.2$ to $5.3 \pm 0.3$, increased CAT activity from $4.1 \pm 0.3\ \mathrm{U/mg}$ to $6.5 \pm 0.4\ \mathrm{U/mg}$, reduced cleaved caspase-3 to $0.43 \pm 0.04$ relative to CONV, and abolished the protection after NRF2 knockdown [2508.20510].

Mechanistic controversy persists. In silico tumor-control modeling based on radiolytic oxygen depletion predicts systematically lower tumor control probability for FLASH-RT than for conventional RT when extrapolated to TCP curves, even though preclinical tumor-volume curves may appear iso-effective [2310.01281]. A plausible implication is that any clinically robust account of the FLASH effect likely requires multiple coupled mechanisms, not ROD alone.

## 3. Modalities, beam delivery, and platform-specific implementations

FLASH-RT is not a single beam modality. Electron systems remain the most mature preclinical and early translational platform, but proton, ion, synchrotron, and laser-driven implementations are under active development. Accelerator-oriented summaries describe electrons at $4\text{–}25\ \mathrm{MeV}$, very-high-energy electrons at $150\text{–}250\ \mathrm{MeV}$, photons requiring high beam power to reach $>100\ \mathrm{Gy/s}$ at isocenter, and proton shoot-through beams from $230\text{–}250\ \mathrm{MeV}$ iso-cyclotrons with average beam current on the order of $1\ \mu\mathrm{A}$ for FLASH intensities [2203.11047].

In PBS proton therapy, delivery constraints are dominated by beam current, minimum monitor units per spot, spot-switch overhead, and energy-layer switching. Reported room beam currents include values below $350\ \mathrm{nA}$ and up to approximately $680\ \mathrm{nA}$, minimum spot weights around $100\text{–}500\ \mathrm{MU/spot}$ in Varian ProBeam FLASH modes, transverse switch times of about $1\text{–}2\ \mathrm{ms/spot}$, and energy switching around $200\ \mathrm{ms}$ per layer for energy-degrading cyclotrons and greater than $1\ \mathrm{s}$ for synchrotrons [2206.11722]. These machine constraints explain the strong interest in single-energy transmission beams, pullback systems with universal range shifter and range compensator, and other strategies that avoid slow multi-layer delivery [2206.11722].

An important recent proton development is the explicit combination of FLASH-RT with spatially fractionated radiation therapy (SFRT). One treatment-planning study proposes “SFRT-FLASH” in two forms: pGRID-FLASH implemented as scissor-beam FLASH (“SB-FLASH”) and proton minibeam-FLASH (“MB-FLASH”). The rationale is depth complementarity: FLASH-RT achieves ultra-high dose-rate sparing mostly in deep tissue near the Bragg peak, whereas SFRT with protons achieves high peak-to-valley dose ratio sparing in shallow-to-intermediate entrance regions [2505.06223]. Across four anatomical sites, MB-FLASH and SB-FLASH achieved high FLASH effect coverage of approximately $60\text{–}80\%$ in the CTV1cm ring while preserving PVDR values of about $2.5\text{–}7$ at shallow-to-intermediate depths [2505.06223].

Research beamlines have also been adapted specifically for mechanistic and preclinical FLASH studies. At the Bern Medical Cyclotron, an 18 MeV proton beamline was modified to support both conventional and FLASH regimes spanning dose rates from $0.01$ to $100\ \mathrm{Gy/s}$ in passive-scattered configuration, with minibeam collimators supporting patterns such as $50\ \mu\mathrm{m}$ slits with $200\ \mu\mathrm{m}$ pitch and PVDR for the $100/400\ \mu\mathrm{m}$ grid falling from approximately $11$ at $0\ \mathrm{mm}$ separation to approximately $4$ at $4\ \mathrm{mm}$ [2605.05441]. This kind of platform is central for controlled exploration of dose-rate effects, LET interplay, and SFRT parameters.

At the extreme end of temporal compression, laser-driven proton accelerators deliver ultra-high instantaneous dose rates in ultrashort bunches. In the first in vivo investigation of normal tissue response to laser-driven proton irradiation, each proton bunch delivered $2.0 \pm 0.4\ \mathrm{Gy}$ in an $11\ \mathrm{ns}$ full width at $1/e$ amplitude, giving an instantaneous dose rate of approximately $1.3\times 10^8\ \mathrm{Gy/s}$ when accounting for $72\%$ of the pulse dose in that interval, while the mean dose rate remained approximately $0.1\ \mathrm{Gy/s}$ because one bunch was delivered every $\sim 20\ \mathrm{s}$ [2602.20460]. This sharp separation of instantaneous and mean dose-rate scales is relevant to ongoing debates over which temporal variable is biologically determinant.

## 4. Treatment planning and optimization

Because FLASH-RT introduces dose-rate objectives in addition to dose objectives, treatment planning departs from standard DVH-only optimization. A proton planning formulation for SFRT-FLASH defines fraction-wise spot weights $x_k$, dose-influence matrices $A_k$, fraction doses $d_k=A_kx_k$, and total dose $d=\sum_{k=1}^K d_k$, then solves
$$
\min_{x_1,\dots,x_K}\; f(d)+\sum_{k=1}^K g(d_k)
$$
subject to minimum-monitor-unit constraints and fieldwise FLASH coverage in a CTV1cm ring [2505.06223]. The CTV1cm structure is a $1\ \mathrm{cm}$ isotropic expansion of the CTV excluding the CTV interior, used as a surrogate organ-at-risk to enforce FLASH-rate objectives [2505.06223].

Clinical trade-offs are explicit. In the same study, conformity index to CTV decreased under MB-FLASH and SB-FLASH; an example given for head-and-neck is CI changing from $0.79$ in conventional planning to $0.47$ in MB-FLASH, reflecting dose-rate and MMU constraints [2505.06223]. However, the effective FLASH-modified dose using a dose-modifying factor for FLASH restored plan conformity, with $\mathrm{CI}_{\mathrm{eff}}\approx 0.98\text{–}1.00$ [2505.06223]. This is not a proof of biological equivalence, but it shows how biological-effect modeling can materially alter the interpretation of physical-dose conformity.

For scanned proton delivery, spot order is itself an optimization variable because voxelwise dose rate depends on the temporal ordering of spot delivery. A TSP-based heuristic represents each proton spot as a node in a complete graph and optimizes the delivery permutation to maximize FLASH-rate voxels in a region of interest. In a 26-patient prostate cohort, the approximate TSP heuristic required $1\text{–}15\ \mathrm{s}$ per beam, compared with an average of $1\ \mathrm{h}\ 49\ \mathrm{min}\ 7\ \mathrm{s}$ for the simulated-annealing global optimizer, while recovering essentially the same FLASH volume: $550\ \mathrm{mL}$ versus $557\ \mathrm{mL}$ for ROI = patient minus prostate [2409.11794]. Since the method does not alter spot weights or the nominal dose distribution, it can be applied as a post-processing step to an existing PBS plan [2409.11794].

FLASH optimization has also motivated new nonconvex and nonsmooth algorithms. A stochastic three-operator splitting formulation rewrites FLASH planning as minimization of
$$
F(x)+G(x)+H(x),
$$
where $H(x)=\tfrac12\|Ax-y\|^2$, $F$ is the indicator of the nonconvex zero-or-$\ge\alpha$ minimum-MU set, and $G$ is the indicator of the dose/dose-rate feasibility set [2311.14292]. In reported lung and brain patient cases, STOS improved conformity index and dose-rate coverage relative to a convex-relaxation ADMM baseline; for the brain case, $P_{dr}$ increased from $96.2\%$ to $97.8\%$ [2311.14292].

A further planning complication is that the toxicity gain from FLASH may depend on organ architecture when dose distributions are heterogeneous. An in silico NTCP study using both a radiolytic oxygen depletion model and a phenomenological logistic model found that sparing increased with decreasing LKB volume-effect parameter $n$, meaning more important sparing for serial organs. In one specific calculation with conventional NTCP fixed at $0.2$, the corresponding FLASH NTCP ranged from $0.14$ for $n=1$ to $0.11$ for $n=0.1$ [2604.25364]. This suggests that local high-dose voxels may be disproportionately relevant for FLASH benefit in serial structures.

## 5. Dosimetry, commissioning, and quality assurance

FLASH dosimetry is difficult because detectors developed for conventional radiotherapy can saturate, become nonlinear, or lose temporal fidelity under high dose per pulse and high instantaneous dose rate. A review of FLASH dosimetry describes ion chambers as suffering recombination when $D_p\gtrsim 1\ \mathrm{Gy/pulse}$, with charge-collection efficiency losses greater than $10\text{–}20\%$ unless high bias and complex corrections are applied [2006.03755]. The same review emphasizes the appeal of scintillation and Cherenkov methods because of nanosecond-scale response, sub-millimeter spatial resolution, and dose-rate independence up to at least $10^5\ \mathrm{Gy/s}$ [2006.03755].

Commissioning frameworks for electron FLASH units therefore include non-standard beam parameters such as pulse width, pulse repetition frequency, dose per pulse, and instantaneous dose rate. A general protocol for acceptance testing, commissioning, and routine QA specifies, among other criteria, reproducibility of $10$ consecutive irradiations at $\pm 0.5\%$, dose-monitoring proportionality versus pulse number at $\pm 2\%$, $D_{\mathrm{pulse}}$ independence versus PRF at $\pm 2\%$, $D_{\mathrm{pulse}}$ linearity versus pulse width at $\pm 2\%$, angular output stability at $\pm 3\%$, and PDD parameters with $d_{\max}$ within $1\ \mathrm{mm}$ and $R80$ within $2\ \mathrm{mm}$ or $3\%$ [2405.15146]. The same framework advocates redundant dosimetry using ion chambers, beam current transformers, and passive dose-rate-independent dosimeters such as EBT3 film, TLD, and OSLD [2405.15146].

Monte Carlo beam models are emerging as the practical basis for electron FLASH treatment planning because commercial treatment-planning systems for electron FLASH radiotherapy are unavailable. For the Mobetron UHDR system at $9\ \mathrm{MeV}$, pulse-width-specific phase-space modeling showed that mean energy decreased exponentially from $9.58$ to $9.04\ \mathrm{MeV}$ as pulse width increased from $1.2$ to $4.0\ \mu\mathrm{s}$, while energy spread increased quadratically, both with $R^2=0.99$ [2605.05491]. A universal reference pulse width of $2.28\ \mu\mathrm{s}$ yielded $9.32\ \mathrm{MeV}$ mean energy and reduced computational burden by $75\%$ while maintaining clinical accuracy within AAPM TG-106 tolerances [2605.05491].

Real-time beam monitoring is another central requirement. A point scintillator detector coupled to a gated amplifier and FPGA controller was shown to be linear with mean dose rate from $40$ to $380\ \mathrm{Gy/s}$ and dose per pulse from $0.3$ to $1.3\ \mathrm{Gy/pulse}$ to within $\pm 3\%$, enabling pulse-resolved dose integration and dose-based gating of a modified LINAC [2111.08769]. The same system revealed a ramp-up of $4\text{–}5$ pulses in vivo, during which average dose per pulse was only about $0.045 \pm 0.004\ \mathrm{Gy/pulse}$ before stabilizing at $0.65 \pm 0.01\ \mathrm{Gy/pulse}$ [2111.08769]. This is important because early underdose during ramp-up may compromise the intended FLASH regime for low total doses.

Large-area two-dimensional scintillator monitors address complementary QA needs. A prototype FLASH Beam Scintillator Monitor using a proprietary inorganic hybrid scintillator achieved a small $-0.02\%/\mathrm{kGy}$ signal decrease after a cumulative dose of $212\ \mathrm{kGy}$, dose-per-pulse linearity with $R^2>0.998$, spatial-resolution residual RMS of $37\ \mu\mathrm{m}$ in one configuration, and FPGA beam-parameter computation in less than $1\ \mu\mathrm{s}$ after the end of frame at $20\ \mathrm{kfps}$ [2305.15306]. Flexible 2D scintillating coatings further extend this approach to non-homogeneous preclinical electron beams and curved ex vivo surfaces, with linear response up to at least $90$ pulses and setup robustness within $5\%$, though signal becomes noisy for field sizes at or below about $1\times 1\ \mathrm{cm}^2$ [2504.15824].

## 6. Experimental evidence, controversies, and translational outlook

The empirical basis for FLASH-RT remains strongest for normal-tissue sparing in preclinical systems. A white-paper synthesis cites reduced lung fibrosis, improved neurocognitive outcomes, and reduced skin toxicity in several electron-beam models, with normal tissues “universally spared” in preclinical studies while tumors are not [2203.11047]. More recent mechanistic work sharpens the dependence of this sparing on dose and delivery conditions. In a murine whole-abdomen model, a total-dose series at fixed mean dose rate of $900\ \mathrm{Gy/s}$ found that only at $15\ \mathrm{Gy}$ did FLASH-RT significantly reduce histological scores compared to CONV-RT, and a mean dose-rate series at fixed dose of $15\ \mathrm{Gy}$ found the FLASH effect emerging at $\ge 100\ \mathrm{Gy/s}$, deepening up to $200\ \mathrm{Gy/s}$ and saturating thereafter [2508.20510].

The same study also demonstrated that hybrid schedules can preserve or abolish sparing depending on the FLASH component: at total dose $15\ \mathrm{Gy}$, FLASH component $>6\ \mathrm{Gy}$ retained sparing, whereas $3\ \mathrm{Gy}$ FLASH plus $12\ \mathrm{Gy}$ conventional did not [2508.20510]. Objective injury markers tracked these schedule dependencies. For example, tissue MDA at $72\ \mathrm{h}$ after $15\ \mathrm{Gy}$ was $1.75 \pm 0.12\ \mathrm{nmol/mg}$ for C15 and $1.32 \pm 0.10\ \mathrm{nmol/mg}$ for F15, a $24\%$ reduction; TUNEL-positive crypt cells at $24\ \mathrm{h}$ were reduced from $48 \pm 5\%$ to $22 \pm 4\%$ [2508.20510].

Proton evidence is expanding but remains heterogeneous across delivery platforms. The Bern 18 MeV beamline supports systematic pre-clinical proton radiobiology studies over five decades of dose rate, while the first in vivo laser-driven proton study reported reduced tissue swelling compared with conventional-dose-rate X-rays and distinct RNA-sequencing signatures after ultra-high instantaneous dose-rate proton irradiation [2605.05441]. In the laser-driven mouse ear model, peak swelling after $36\ \mathrm{Gy}$ protons was $286 \pm 28\ \mu\mathrm{m}$ versus $372 \pm 66\ \mu\mathrm{m}$ for $40.8\ \mathrm{Gy}$ X-rays, approximately $23\%$ lower [2602.20460]. However, the authors also note that definitive attribution of a FLASH effect in that regime would require conventional-dose-rate protons with matching LET as the true dosimetric control [2602.20460].

Several controversies follow from these data. First, the exact molecular targets and the weighting of oxygen depletion, radical recombination, immune modulation, and mitochondrial or nuclear signaling remain unresolved [2405.10219]. Second, the clinically relevant dose-rate descriptor is unsettled, especially in pulsed and scanned beams where mean dose rate, instantaneous dose rate, dose per pulse, and temporal microstructure can dissociate strongly [2504.19927]. Third, the assumption of tumor iso-effectiveness is not mechanistically secure: radiolytic oxygen depletion models predict lower TCP for FLASH-RT than for conventional RT unless compensating mechanisms not included in the model contribute in real tumors [2310.01281].

The translational agenda implied by current work is therefore technically specific. It includes collimator-free minibeam generation to reduce monitor-unit burden in proton SFRT-FLASH, preclinical organ-specific validation of combined SFRT-FLASH biological effects, prospective clinical trials evaluating toxicity reduction and tumor control, verification of beam-model regressions on additional Mobetron installations, incorporation of universal and pulse-width-specific phase space files into Monte Carlo or GPU-accelerated dose engines, continued standardization of commissioning and QA for electron UHDR units, and end-to-end treatment-planning studies that jointly evaluate dose, dose rate, delivery accuracy, and normal-tissue sparing [2505.06223].

FLASH-RT thus occupies a distinctive position in radiotherapy research. Its central claim is not merely geometric dose conformity, but biologically favorable temporal delivery. The field’s present state is defined by that promise, by substantial preclinical support for normal-tissue sparing, and by equally substantial uncertainty regarding the precise conditions under which the effect is robust, safe, and clinically generalizable.

Source: https://www.emergentmind.com/topics/flash-radiotherapy-flash-rt