Volume 50, Issue 1 p. 570-581
Open Access

First experimental measurements of 2D microdosimetry maps in proton therapy

Consuelo Guardiola

Corresponding Author

Consuelo Guardiola

Université Paris-Saclay, CNRS/IN2P3, IJCLab, Orsay, France

Université de Paris, IJCLab, Orsay, France

Centro Nacional de Microelectrónica (IMB-CNM, CSIC), Bellaterra, Spain


Consuelo Guardiola, Université Paris-Saclay, CNRS/IN2P3, IJCLab, 91405 Orsay, France.

Email: [email protected]

Search for more papers by this author
Diana Bachiller-Perea

Diana Bachiller-Perea

Université Paris-Saclay, CNRS/IN2P3, IJCLab, Orsay, France

Université de Paris, IJCLab, Orsay, France

Search for more papers by this author
Emmanuel M. Mate Kole

Emmanuel M. Mate Kole

Université Paris-Saclay, CNRS/IN2P3, IJCLab, Orsay, France

Search for more papers by this author
Celeste Fleta

Celeste Fleta

Centro Nacional de Microelectrónica (IMB-CNM, CSIC), Bellaterra, Spain

Search for more papers by this author
David Quirion

David Quirion

Centro Nacional de Microelectrónica (IMB-CNM, CSIC), Bellaterra, Spain

Search for more papers by this author
Ludovic De Marzi

Ludovic De Marzi

Department of Radiation Oncology, Institut Curie, PSL Research University, Centre de protonthérapie d'Orsay, Campus Universitaire, bâtiment 101, Orsay, France

Institut Curie, PSL Research University, Université Paris-Saclay, INSERM LITO, Campus Universitaire, Orsay, France

Search for more papers by this author
Faustino Gómez

Faustino Gómez

Departamento de Física de Partículas, Universidad de Santiago de Compostela, Santiago de Compostela, Spain

Search for more papers by this author
First published: 06 September 2022
Citations: 1



Empirical data in proton therapy indicate that relative biological effectiveness (RBE) is not constant, and it is directly related to the linear energy transfer (LET). The experimental assessment of LET with high resolution would be a powerful tool for minimizing the LET hot spots in intensity-modulated proton therapy, RBE- or LET-guided evaluation and optimization to achieve biologically optimized proton plans, verifying the theoretical predictions of variable proton RBE models, and so on. This could impact clinical outcomes by reducing toxicities in organs at risk.


The present work shows the first 2D LET maps obtained at a proton therapy facility using the double scattering delivery mode in clinical conditions by means of new silicon 3D-cylindrical microdetectors.


The device consists of a matrix of 121 independent silicon-based detectors that have 3D-cylindrical electrodes of 25-µm diameter and 20-µm depth, resulting each one of them in a well-defined micrometric radiation sensitive volume etched inside the silicon. They have been specifically designed for a hadron therapy, improving the performance of current silicon-based microdosimeters. Microdosimetry spectra were obtained at different positions of the Bragg curve by using a water-equivalent phantom along an 89-MeV pristine proton beam generated in the Y1 proton passive scattering beamline of the Orsay Proton Therapy Centre (Institut Curie, France).


Microdosimetry 2D-maps showing the variation of the lineal energy with depth in the three dimensions were obtained in situ during irradiation at clinical fluence rates (∼108 s−1 cm−2) for the first time with a spatial resolution of 200 µm, the highest achieved in the transverse plane so far. The experimental results were cross-checked with Monte Carlo simulations and a good agreement between the spectra shapes was found. The experimental frequency-mean lineal energy values in silicon were 1.858 ± 0.019 keV µm−1 at the entrance, 2.61 ± 0.03 keV µm−1 at the proximal distance, 4.97 ± 0.05 keV µm−1 close to the Bragg peak, and 8.6 ± 0.1 keV µm−1 at the distal edge. They are in good agreement with the expected trends in the literature in clinical proton beams.


We present the first 2D microdosimetry maps obtained in situ during irradiation at clinical fluence rates in proton therapy. Our results show that the arrays of 3D-cylindrical microdetectors are a reliable microdosimeter to evaluate LET maps not only in the longitudinal axis of the beam, but also in the transverse plane allowing for LET characterization in three dimensions. This work is a proof of principle showing the capacity of our system to deliver LET 2D maps. This kind of experimental data is needed to validate variable proton RBE models and to optimize LET-guided plans.


Some normal tissue toxicities are beginning to be reported in proton therapy,1-7 which may lead to severe side effects as influencing cognitive function in children,8 functionality of organs, or even secondary cancers.9 This is particularly important in pediatric and young patients as they are prone to carcinogenic effects, and their life expectancies are longer than the latent period of the secondary effects. It is hypothesized that these toxicities can be due to the fact that ions deliver more energy per unit of track length (linear energy transfer, LET) than conventional radiotherapy sources.10 It means that if high-LET spots are not properly located, collateral damages can be generated, for example, acute and late effects. Therefore, the LET is one of the physical descriptors (among others, e.g., dose and dose rate) of the biological damage at the macroscopic level. At cellular level, the stochastic nature of LET comes up and the corresponding lineal energy (y) distribution turns to be the appropriate parameter of the microdosimetry characterization of the radiation quality.11-13 Lineal energy (y) is defined as the ratio between the deposited energy by a single event into a given microscopic volume (ε) and the corresponding mean chord length ( l ¯ $\bar{l}$ ) of that irradiated volume.14 As radiation interactions occur stochastically, y is described by its probability density distribution, f(y). Once the f(y) is obtained, it is possible to calculate the dose-weighted mean lineal energy ( y ¯ D ${\bar{y}}_D$ ), which is the equivalent to the macroscopic dose-weighted LET (LETd). Direct measurements of the LET with microscopic resolution would allow us to optimize the treatments either removing high-LET spots from critical structures or focalizing high-LET regions into the target.15-17 This physical quantity is related to the relative biological effectiveness (RBE).18

In this context, microdosimetry distributions can be obtained either with Monte Carlo (MC) simulations or through experimental measurements with microdosimeters. However, the accuracy of simulations depends on the precision of the modeling of the clinical beamline and on the uncertainties in the physics models implemented. Simulations also require a large time-consuming process. In contrast, direct experimental measurements will yield the microdosimetry spectra from the acquired data for beam characterization. Quantifying the stochastic nature of the interactions of ionizing radiation with matter to characterize the radiation quality remains an important issue in clinical practice. Indeed, in the last years, due to the increasing number of hadron therapy facilities, microdosimetry has reassumed a relevant role in the field.19 In particular, measurements are not only relevant for hadron therapy, but also in mixed or complex radiation fields as those in aviation or space altitudes.20 As microdosimetry spectra vary markedly in short distances in the distal edge due to the rise of the stopping power of ions, high accuracy microdosimetry measurements are required.21 Furthermore, microdosimetry data can reveal small variations of radiation quality with an impact in the dose-equivalent values. For decades, the gold standard microdosimeter has been the tissue-equivalent proportional counter (TEPC).13 Nevertheless, they have some shortcomings and specially suffer pile-up effects under therapeutic fluence rates. New mini-TEPCs have improved their performance in the last years,22, 23 but they are still isolated (one spot) sensors. In contrast, silicon-based radiation detectors are feasible microdosimeters due to the well-established microfabrication processes that provide radiation-sensitive micro-volumes with high spatial resolution. However, they have other limitations; for example, they are not water equivalent, and corrections considering both material conversion and charge collection efficiency (CCE) are mandatory. Even so, they have contributed significantly to microdosimetry verification in the last years. For example, Rosenfeld et al. have worked extensively in the field in the last two decades and developed several microdosimeter generations based on planar PN junctions etched around the sensitive volumes.24-26 However, there is still room for improving several solid-state microdosimeter features; for example, the CCE, the spatial resolution along the transverse plane, the data acquisition for further adaptation in clinical conditions to make it practical for daily quality assurance (QA) measurements of y. Alternative solid-state microdosimeters are based on diamond as they are close to tissue equivalent and have a high radiation hardness.27, 28 To improve the performance of silicon-based microdosimeters and overcome some of those limitations, in 2013–2015, we proposed and created a new 3D-cylindrical architecture.29, 30 The sensors have already shown a good performance in microdosimetry in both carbon therapy31 and low-energy proton32 facilities. In these proof-of-concept works, we had only one individual 3D-cylindrical sensor (one pixel) connected to a single-channel readout electronics. In our first work,31 the in-house electronics consisted of a charge-sensitive preamplifier, a spectroscopy shaping amplifier, and an Amptek MCA8000D to digitize the pulse height. In a second version,32 another electronics was created for increasing the signal-to-noise ratio, which allowed us to reduce the low-level discrimination one order of magnitude, from 36.4 to 2.5 keV µm−1 (in silicon). In both cases, the single microdetector was irradiated at average fluence rates of 107 s−1 cm−2. The silicon sensors belonged to the first31 and second32 generations of 3D-cylindrical microdetectors manufactured at the Instituto de Microelectronica de Barcelona (IMB-CNM (CSIC)). In the first prototype generation, the CCE was not fully optimized, which affected the reconstruction of the probability distributions and momenta of the stochastic quantities related to microdosimetry, as discussed in Ref. [33]. In the second generation, the CCE was significantly improved by reducing the overall thermal budget of the microfabrication process in order to obtain shallower and steeper dopant profiles.34 The 11 × 11 arrays of microdetectors used in this work belong to the second improved generation. The calibration must be done for each setup as the readout electronic is different in each case. Additionally, in the 11 × 11 array, the calibration has been verified for the 128 channels of the corresponding ASIC (application-specific integrated circuit).35

In this work, we face the challenge of extending the single-dimension to a two-dimensional measurement from one 3D-cylindrical microdetector to an 11 × 11 array, scaling the radiation-sensitive area to several centimeters, which we have performed recently.35 It allows us to obtain LET 2D-maps with high spatial resolution along the transversal plane relative to the beam direction. It has been a challenging technological development due to the complexity associated to create (i) a readout electronics specifically customized for the multichannel signal analysis of the microdetector array, (ii) large and fragile pitch adapters designed and manufactured to connect the sensor arrays with the ASICs without introduction extra electronic noise, (iii) in-house data analysis codes specifically developed for the multichannel readout. The final goal has been enabling the assessment of the microdosimetry distributions in relevant clinical volumes and critical areas, for example, organs-at-risk and penumbras. It is of upmost importance to further RBE optimization,36 and minimizing dose/LET hot spots in intensity-modulated proton therapy (IMPT).37 The importance of the LET-guided optimized plans has gained relevance in the last years38, 39 to reduce LET at the same dose level or redistributing high LET to low-dose regions. In this context, accurate microdosimeters covering large surfaces that are capable of quantifying experimentally the LET maps in those regions are crucial for LET-guided plan evaluation and optimization. So far, such evaluations have been done either with analytical methods or MC simulations. We propose a microdosimetry tool for experimental verification of LET maps.

We report on the first microdosimetry 2D maps with an array of 121 individual 3D-cylindrical microdetectors covering a 2 mm × 2 mm–sensitive area at clinical-equivalent fluence rates in a clinical proton beam with the highest spatial resolution so far in the transverse plane. This microdosimetry characterization in both transverse and longitudinal directions, with 20- and 200-µm resolutions, respectively, allows us to evaluate the microdosimetry variations in proton therapy. The two-dimensional microdosimetry measurements at various depths, compared to single-point measurements, can measure directly heterogenous LET distributions in just one shot, saving a considerable measurement time (proportional to the surface covered and the spatial resolution required). This is particularly interesting for optimizing IMPT or proton arc therapy where the LETd evaluation would be an asset to verify the current simulations considered in those fields.

Measurements were performed in double scattering (DS) delivery mode along a pristine Bragg peak of 89 MeV. The lineal energy distributions were measured with the 3D-cylindrical microdetector array. The experimental results were compared with MC simulations.


2.1 Array of silicon 3D-cylindrical microdetectors

The second generation of 3D-cylindrical microdetectors was manufactured on 4-in. silicon-on-insulator wafers, silicon with 〈1 0 0〉 orientation, n-type doped with phosphorus, with a nominal resistivity greater than 3.5 kΩ cm, and with a thickness of 20.0 ± 0.5 µm. An SiO2/Si3N4 passivation bilayer of 2.4 ± 0.1 µm was deposited over the whole surface. Further details can be found elsewhere.40-43 An array of 11 × 11-independent 3D-microdetectors was designed with a 25-µm diameter and a distance between them of 200 µm (pitch). After the manufacturing process, the final microdetector thickness was 19.86 ± 0.04 µm. The entire array covers a radiation-sensitive surface of 2 mm × 2 mm. Figure 1 shows images of the 11 × 11 3D-microdetector array (left) and of the top-view of a few of detectors (right), and the metal-strips connected to the readout-electronic systems can also be seen.

Details are in the caption following the image
Left: Optical image of one 11 × 11 matrix of the second generation of 3D-cylindrical microdetectors (25-µm diameter, 19.86-µm thickness, and 200-µm pitch). Each microdetector is connected to the corresponding electronics-readout channel by wire-bonding (displayed on the top and bottom of the image). Right: Scanning electron microscopy (SEM) image of a detail of the front-face of the microdetector array with the corresponding metal strips for the readout

Devices have leakage currents in the order of 50 fA/cell and capacitances of 140 fF/cell at 5 V, measured at 10 kHz with a Probe station MPI-TS2000SE and HP4155B Semiconductor Parameter Analyzer (at 21°C) and Agilent 4284A LCR Meter, respectively. The devices showed a good diode behavior and a depletion voltage of 5 V.

Likewise, we have improved the CCE of the devices in this second generation,34 which plays a relevant role as an inhomogeneous CCE makes that events from different regions in the volume generate different pulse heights, which biases the reconstruction of the energy imparted per event.31 This CCE correction factor was included in the corresponding MC simulations to take this effect into account and to do a precise comparison with experimental results.

2.2 Energy calibration

Due to the small size of the microdetectors (<10−5 mm−3), the extremely low statistics collected with conventional radioactive calibration sources, for example, 241Am, makes a proper calibration unfeasible with them. Therefore, we performed the energy calibration of the system at the Accélérateur Linéaire et Tandem à Orsay (ALTO) facility using an 14.5-MV Van de Graaff tandem. Monoenergetic proton beams with energies ranging between 6 and 20 MeV (±0.1 MeV) were used with beam currents on the order of 100–200 pA. We adjusted the areas of irradiation to have ion fluxes of ∼108 cm−2 s−1 (clinical values). The sensor array was placed in air at a distance of (5.1 ± 0.1) cm from the beam exit (200-µm-thick Kapton). Then, first, we performed the energy calibration by fitting the experimental values obtained in analog-to-digital converter units (ADC channels) to the energy values calculated with Stopping and Range of Ions in Matter (SRIM) MC simulations.44 The corresponding energies deposited were simulated with the SRIM code considering the detector geometry described above as well as the irradiation conditions, and the most probable imparted energies were matched with the corresponding most probable ADC channel values. We assume a linear proportionality between the energy deposited in the silicon microdetectors and the corresponding electronics output voltage pulse height.45

2.3 Experimental setup

Irradiations were performed in the Y1-passive beamline at the Proton Therapy Centre of Orsay (ICPO, Institut Curie, France), which uses a DS delivery system. The ICPO facility is equipped with a C230 isochronous cyclotron that delivers 230-MeV protons at the accelerator exit. For our experiments, the energy beam was degraded to 89 MeV in the isocenter by a range shifter, and we considered a pristine Bragg peak. A brass collimator of 2 cm × 2 cm was placed at the exit of the beamline to obtain a homogeneous beam profile. The measurements were carried out with the proton beam in perpendicular incidence to the front-face of the microdetectors. Detectors were placed in the isocenter with an uncertainty positioning depth estimated ≤0.1 mm. The proton beam covered the full detection area and the average flux was ∼108 s−1 cm−2 in each measurement. Different thicknesses of solid–water blocks (30 cm × 30 cm) of SP34 (RW3 type, 98% polystyrene and 2% TiO2, with a density of 1.045 g cm−3) were interposed between the beam and the daughterboard (circuit board that plugs with the data acquisition board) to place progressively the microdetectors along the dose profile.

For a fast data analysis in situ, we have developed an in-house python code that displays the microdosimetry quantities following the experimental methodology detailed by Knoll45 as well as the microdosimetry principles described by Rossi and Zaider.13 The python code was optimized to obtain the microdosimetry quantities from the raw data of the multichannel readout electronics in ≤2 min right after the data acquisition.

Additionally, we have customized a multichannel readout electronics of low noise to assess their performance at therapeutic fluence rates.

2.4 Monte Carlo simulations

MC simulations were used to cross-check the pulse-height spectrum measurements. The complete ICPO beamline was previously modeled and benchmarked against experimental data.46 In particular, we used MC simulations based on the GATE v8.1 code47 with the binary cascade (BIC) model for the hadronic interactions and the G4EmStandardPhysics_option3 to describe electromagnetic interactions. Ionization potentials for water and air were set to 78 and 85.7 eV, respectively. We implemented range cuts of 10 mm, 1 µm, 0.5 µm, 0.2 µm, and 2 mm (one fifth or less of the size of each simulated volume) for all the particles in the “world” (air), SV (silicon), passivation and TEOS (SiO2/Si3N4) layers geometries, and SP34 solid–water blocks (98% polystyrene, 2% TiO2, and 1.045 g cm−3 density), respectively. The initial simulated geometries consisted of (i) the virtual source simulating the beamline46 and (ii) the full 11 × 11 3D-cylindrical microdetector array considering the geometrical and material description detailed in Sections 7 and 9. Geometries were embedded in air.

In a first approach, the complete beamline was modeled and benchmarked against experimental data for another application.46 In a second phase, for saving computational time, we used the virtual source modeled of that work to gain one order of magnitude in computational efficiency. For generating the virtual source, we simulated the full-beam characteristics at the nozzle exit, beam energy spread, spot size, and angular distribution, and the outcomes were stored in phase-space files. Then, an iterative comparison of simulated and measured dose distributions was performed for fine-tuning of the parameters of the virtual source. The simulation results were validated with experimental dosimetry data obtained with a microDiamond detector and EBT3 Gafchromic films in a water tank (BluePhantom IBA) and RW3 solid–water phantoms, respectively. Further details can be found in Ref. [46].

The number of simulated primary particles was 1011, resulting in an uncertainty below 1% at the plateau. The simulations were run into a Tier-1 computation cluster (CC-IN2P3) to speed the calculations by using the parallel computing platform developed in GATE with a hundred of servers with X86_64 processors in the Linux system.

The simulations were used to assess the energy spectra obtained in the 3D-microdetectors (SVs) along the Bragg peak. The total energies deposited into the SV of the 3D-cylindrical microdetectors were recorded in binary output files containing the energy deposited (keV) into each sensor. Then, they were treated to account for the CCE dependence on the entry point of the particle trajectory to the SV as follows: The energy spectra were converted into a list of events, a random position in a circle of a 12.5-µm radius centered in the sensitive volume was assigned for each event. The CCE correction factor for each point was applied by using the measured CCE described by Bachiller-Perea34 as a function of the distance to the center of the SV. Finally, the energy spectra were reconstructed and compared to the experimental data. Consequently, the lineal energy distributions were calculated by dividing the energies spectrum histograms by the mean path length of the particles in the SV, that is, 19.86 µm, which corresponds to the silicon thickness, as we are irradiating perpendicularly to the front face of the microdetectors.

The proton source was modeled as a so-called general particle source, which is a GATE-predefined option for particle source generation. The final beam energy was characterized as a Gaussian distribution centered at 89 MeV.


In this section, we report on the preliminary tests required, namely, the energy calibration and the modeling of the simulated virtual source, and on the microdosimetry data acquired experimentally as well as on the comparison with the simulations.

3.1 Calibration and virtual source modeling

Equation (1) shows the energy calibration, where a linear regression fit was used to correlate both ADC and energy values, a very good correlation value (R2 ≥ 0.98) was obtained. This energy calibration was used to analyze the microdosimetry measurements:
E keV = 0.272 × ADC + 12.13 $$\begin{equation}E\left( {{\rm{keV}}} \right) = 0.272{\rm{ }} \times {\rm{ ADC}} + 12.13\end{equation}$$ (1)

Figure 2(left) shows the comparison that we obtained of the energy deposited into the microdetectors with both SRIM and GATE simulations from 6 to 20 MeV. Small differences between 9 and 12 keV were observed for the extreme beam energies values, which may explain some discrepancies between the measured and simulated spectra, as it is detailed later.

Details are in the caption following the image
Left: Energy deposited in silicon as a function of the ALTO (Accélérateur Linéaire et Tandem à Orsay [France]) beam proton beam energy simulated with Stopping and Range of Ions in Matter (SRIM) and GATE. Right: Experimental and simulated dose profiles with the virtual proton source at 89 MeV in SP34 water equivalent

Figure 2(right) shows the comparison of both experimental and simulated percentage depth doses in water. There is a globally satisfactory agreement between the simulated and measured dose profile.

3.2 Microdosimetry 2D maps

As a figure of merit, Figure 3 shows the first experimental microdosimetry maps in two dimensions of both the frequency-mean lineal energy ( y ¯ F ${\bar{y}}_F$ , left), and the dose-averaged lineal energy ( y ¯ D ${\bar{y}}_D$ , right) for four representative depths in the entrance, proximal distance, Bragg peak, and distal edge, namely, 0, 4, 5.6, and 6.3 cm. Each quadrant corresponds to one of these four representative SP34 solid–water thicknesses. It depicts both values for each one of the 11 × 11 microdetectors of the array (represented as pixels). It is a 2D distribution covering a surface of 2 mm × 2 mm with a spatial resolution of 200 µm in both dimensions (distance between adjacent microdetectors).

Details are in the caption following the image
Experimental frequency-mean lineal energy ( y ¯ F ${\bar{y}}_F$ , left), and dose-averaged lineal energy ( y ¯ D ${\bar{y}}_D$ , right) for each of the 11 × 11 microdetectors for four representative SP34 solid–water thicknesses, namely, 0, 4, 5.6, and 6.3 cm. Average values of both and for each array are depicted at the bottom of each figure.

We obtain increasing values of y ¯ F ${\bar{y}}_F$ and y ¯ D ${\bar{y}}_D$ for raising water thicknesses as the energy deposited increases along the beam depth. The average values are detailed in Section 15.

3.3 Average pulse-height spectra

The energy threshold (in silicon) was fixed at 20 keV to discard low pick-up noise, which is roughly equivalent to 0.6 keV µm−1 in water. It is worth noting that the energy threshold during clinical measurements can be an issue as the starting value in proton beams ranges from 1 to 2 keV µm−1 in water, which may be considerably higher than the detectable y values. Figure 4(left) shows the energy spectra recorded by all the 11 × 11 3D-cylindrical microdetectors for the different solid–water thickness along the Bragg curve. The spectra have been normalized, so the area below the curves is equal to 1. The relative shapes of these distributions show the energy fluctuations of the data set, where it is clearly observed that in the distal edge (i.e., for thicknesses higher than 5.5 cm), a wide distribution, which denotes a high fluctuation of the energy deposited by the proton beam at the micrometer level.

Details are in the caption following the image
Top: Pulse-height spectra measured by second generation 3D-cylindrical microdetectors as a function of different SP34 solid–water phantom thicknesses. Bottom: Experimental microdosimetric spectra in the microdetectors at the different solid–water depths

Figure 5 shows the comparison between four of these experimental pulse-height spectra (in blue) and the corresponding simulated ones before (gray) and after (red) applying the CCE correction factor, for a representative set of solid–water thicknesses. As is expected, the spectra shifted to higher deposited energies with depth, as, at higher depths, the protons have lower energies and, therefore, higher stopping powers. The agreement between the spectra shapes of both experimental and simulated data was very good, but we found an energy shift ≤10 keV from the entrance to Bragg peak and 18 keV in the distal edge between the experimental and the simulated peak maxima. We hypothesize that these shifts can be due to one or several of these inaccuracies: Either to uncertainties in the geometrical modeling of the beamline, or to an imperfect multiple Coulomb scattering model in the MC code, or to experimental misalignments as the daughterboard was placed manually. Another plausible explication is the fact that the energy calibration was performed by using the SRIM code (Section 12), but the simulations were developed with GATE code to have an independent comparison. However, as it is shown in Figure 2(left), in the extreme beam energies used, there were small differences in the energy deposited in silicon, which could be even larger for lower and higher proton beam energies. This could partially explain the energy shift that we found when we compared both the measured and simulated spectra. The insets in Figure 5 show the good agreement in the spectral shapes of both experimental and simulated spectra when a shift offset in the maximum peaks is applied for the sake of clarity.

Details are in the caption following the image
Experimental spectra (blue), Monte Carlo simulations with (red) and without (gray) the charge collection efficiency (CCE) correction in the entrance (0 and 2 cm), proximal distance (4 cm), and distal edge (6 cm). The experimental energy thresholds were fixed at 20 keV. The inset plots show the superposition of the experimental and simulated spectra when the peaks are shifted for additional spectral shapes comparison.

Spectra shifts were also found in other microdosimetry works, for example, Debrot and Bertolet et al.48, 49 Interestingly, this discrepancy is generally unnoticed in the literature as it is not the pulse-height spectra that is usually shown, but the microdosimetry spectra, which are in logarithm scale and where the shifts of the peaks are less evident.

3.4 Frequency-mean lineal energy and dose-averaged lineal energy

Figure 6(left) shows the experimental frequency-mean lineal energy, in silicon, for all the SP34 solid–water thicknesses. The experimental values in silicon were 1.858 ± 0.019 keV µm−1 at the entrance, 2.61 ± 0.03 keV µm−1 at the proximal distance, 4.97 ± 0.05 keV µm−1 close to the Bragg peak, and 8.6 ± 0.1 keV µm−1 at the distal edge.

Details are in the caption following the image
Experimental values of y ¯ F ${\bar{y}}_F$ and y ¯ D ${\bar{y}}_D$ in silicon calculated as the average value of the 11 × 11 microdetectors

Likewise, Figure 6(right) shows the experimental dose-averaged lineal energy, y D ${y}_D$ , being 2.17 ± 0.05 keV µm−1 in the entrance, 3.08 ± 0.06 keV µm−1 in the proximal zone, 6.69 ± 0.11 keV µm−1 close to the Bragg peak, and 11.60 ± 0.13 keV µm−1 in the distal edge.


Both the experimental y ¯ F ${\bar{y}}_F$ and y ¯ D ${\bar{y}}_D$ are in good agreement with the expected trends in the literature in clinical proton beams.25, 28, 48

On one hand, Anderson et al.25 found values from 2 to 11 keV µm−1 between 2 and 4 cm of water-equivalent depth (WED) for a 71-MeV pristine Bragg peak and from 2.2 to 9 keV µm−1 between 5- and 17.6-cm WED for a 160-MeV pristine Bragg. Although their system was based on a single silicon microdosimeter (one-dimensional measurement/one spot) with planar PN junctions, the boundaries of which were etched in parallelepiped shapes into the silicon-bulk, and their results along the beam are consistent with our findings.

On the other hand, Loto et al.28 carried out microdosimetry measurements with an scCVD diamond detector under similar conditions as the ones employed in our tests, which allows us to compare both results. They calculated the values in water-equivalent depth by using a diamond-to-water constant conversion factor of 0.32. The obtained values were 1.75 keV µm−1 in the entrance, and from 3 to 8 keV µm−1 between the proximal and distal regions (at 6.3-cm WED, which is our last experimental depth measured). If we apply a variable silicon-to-water conversion factor by using the ratio of the stopping power in both materials50 to our values of Figure 6, we would obtain values ranging between 1.7 and 7.5 keV µm−1. This result is in very good agreement with Loto's data. However, there are some small discrepancies can be explained by different causes. First, as it is explained in Loto et al., their spectra showed a low-energy tail due to an incomplete CCE. Inhomogeneous CCE modifies the reconstruction of the energy imparted per event,33, 34 producing counts at lower energies, which is a general problem for solid-state structures. In contrast, we have included a CCE correction factor (Section 10). To the best of our knowledge, this is the first time that such CCE correction factor is fully included in the reconstruction of the imparted energy into the detector active volume. Each event pulse height here is considered the convolution of the actual energy deposition along the silicon detector with the effective CCE map. It is worth noting that the construction of buried silicon structures for microdosimetry using any microfabrication technique would yield similar effects on the CCE whenever the volumes considered are on the range of micrometers. We explained the importance of this issue elsewhere.33 Second, Loto et al. applied cutoffs in the spectrum tails for four representative values, but it was not clarified if the energy cutoff considered was the same one for all the cases or variable depending on the spectra. Using different cut-off values can give rise to an under/overestimation of the y ¯ D ${\bar{y}}_D$ , which makes difficult to compare devices at the same depths. In our case, we fixed an equal energy cutoff to all our spectra of 20 keV in silicon in order to be consistent in the intercomparison studies. Third, even if the energy calibration for protons or light ions (e.g., alpha particles) can be performed with both types of particle beams or sources, ideally, an accurate calibration requires to use the same type of particle that those involved in the measurement.45 Besides, conventional calibration alpha sources as the one used by Loto et al. (241Am) usually have a protection layer in their encapsulation that slows down the alphas and, additionally, those particles lose a non-negligible energy in air if the calibration is not performed in vacuum. This could have an impact in the calibration curve precision and, consequently, in the spectra. In contrast, we managed this issue by performing an energy calibration with a monoenergetic proton beam (from 6 to 20 MeV) without any intervening material between it and the detector, except for the Kapton layer (200 µm) in the exit window and the passivation layer (2.4 µm). Finally, it is worth noting that our readout system was assembled with an ASIC that has an energy detection limit of ∼550 keV. This means that we are not able to measure lineal energy values higher than ∼28 keV µm−1 in silicon (∼18 keV µm−1 in water), which limits its use in extreme positions at the distal edge. This could produce a light underestimation of the y ¯ D ${\bar{y}}_D$ values as the events with energy higher than 550 keV are not taken into account. We are currently working on a new electronic system to extend the energy range limit of the readout electronics to be used with heavy ions.

Regarding the overall designs as solid-state microdosimeters, in particular comparing with the more conventional PN planar junctions, the novel 3D-cylindrical architecture has a truly isolated and well-defined convex SV that contains the full charge collection, without charge sharing between adjacent electrodes, and the highest spatial resolution so far. To the best of our knowledge, this configuration allows us to generate for first time a modular structure to scale the sensitive areas in order to cover several centimeters. In this work, we extended the individual 3D-cylindrical microdetector configuration to an 11 × 11 array with the final aim of scaling the radiation sensitive area (2 × 2 mm2) and enabling the experimental evaluation of LET 2D-maps (in just a single shot) along the transversal plane relative to the beam direction. It has been not only a technological advancement, but also it has required challenging adaptations of the multichannel readout electronics, the design and manufacture of special pitch-adapters, the development of in-house codes for in situ data analysis, and the channel-by-channel calibration of the readout system. Indeed, we have recently successfully performed the first extension of the array to a multi-array covering 12 cm of length.35 The access limitation to clinical proton beamtime restricted this work to be performed with a single-beam energy and a broad beam, which is the simplest configuration for evaluating the behavior of the device as a proof of principle. Under this simple condition, we did not observe LET variations. Further tests with more complex irradiation conditions to be able to quantify 2D measurement of varying LET will be performed in the future. These arrays would save a considerable experimental time and could be a practical tool for evaluating the LET distributions in three dimensions for guided optimized treatment plans.

Regarding the MC simulations, it was observed that the general spectra shapes of the experimental results were reproduced by those simulated with the MC used. However, systematic energy shifts (from 4 to 30 keV) are found between the experimental and the simulated peak maxima. Nevertheless, when a shift offset is properly applied in energy range, there is an overall good agreement between the experimental and simulated spectra.

Considering the main limitations of our system, first, to cover square centimeters in the XY plane, new sensor layouts have to be created to minimize the impact of the restrictions due to the (i) complexity associated with the readout electronics and (ii) the design and fabrication of the corresponding pitch-adapters to connect the sensors with the electronic channels. This is a technological challenge that will be addressed in a future redesign of our system. Second, it is worth noting that our system is not able to directly quantify neutrons. It means that even if the microdosimetry system can be used for both passive scattering and pencil beam scanning delivering methods, without potential limitations nor saturation effects, we cannot assess the secondary neutron contributions that are generated in passive scattering mode (when the proton beam hits the scatter foil, the compensators, and collimators). Third, solid-state microdosimeters have some intrinsic limitations; for example, they are not water equivalent, the micro-technology processes associated with the manufacturing are complex, and their performance can be deteriorated due to radiation damage over time. Moreover, corrections considering both material conversion and CCE are mandatory. Some of these limitations have been previously discussed in Prieto-Pena et al.33 In particular, we concluded that the CCE and electronic noise pose limits to their performance, as they can affect the fidelity of the microdosimetric distribution. Considering this issue, we have already improved the CCE in the second and third generations of our devices thanks to an optimized microfabrication process.34 Regarding the radiation-hardness of our devices, the first high-dose tests are currently being performed to evaluate the usable lifetime of the devices under realistic clinical irradiation conditions, which is key for a future clinical implementation.

Taking into account these limitation and the state of the art, there is still room to create new microdosimeters aimed for QA; for example, by (i) tailoring the spatial resolution along the transversal plane relative to the beam direction for obtaining LET 2D-maps adapted to the application requirements, (ii) adapting the readout electronics to deal with therapeutic fluence rates without saturation effects, (iii) developing large microdosimeters (∼centimeters) to assess microdosimetry distributions in relevant clinical volumes and critical areas, for example, organs-at-risk and penumbras, (iv) developing acquisition tools for real-time analysis, and so on. In response to some of these challenges and in pursuit of improving the existing silicon-based microdosimeters, we have already designed, manufactured, and characterized the first multi-arrays of microdosimeters.35


We have obtained the first 2D microdosimetry maps in a proton therapy center under clinical conditions with an array of 3D-cylindrical silicon-based microdetectors. Measurements at various depths along the Bragg curve of a pristine 89-MeV proton beam were performed at therapeutic-equivalent fluence rates without pile-up and saturation effects. We customized readout electronics with a multichannel data-acquisition system for spectroscopy and pitch-adapters that allowed us to connect the individual microdetectors to the readout. Additionally, we created an in-house Python code with firmware control that allows one to display microdosimetry maps in situ. Microdosimetry spectra were obtained and cross-checked with MC simulations finding a reasonable agreement.

Our results show that our arrays of 3D-cylindrical microdetectors are reliable microdosimeters to evaluate LET maps not only in the longitudinal axis of the beam, but also in the transverse plane allowing for LET characterization in three dimensions. To be best of our knowledge, this is the first solid-state microdosimeter able to quantify microdosimetry maps in one single shot with the highest spatial resolution in the transverse plane so far. This kind of experimental data is needed to validate variable proton RBE models and to optimize LET-guided plans. This configuration allows us to set the path to cover larger radiation sensitive areas, by scaling multi-arrays of microdetectors, for further voxel-by-voxel RBE optimization in critical areas of clinical relevance where we may have heterogeneous LET distributions, for example, organs-at-risk, penumbras, and out-of-field zones.

Finally, we have customized a low-noise multichannel readout-electronics system to work at therapeutic fluence rates, reaching energy thresholds (in silicon) of 20 keV (∼0.6 keV µm−1 in water), which is the order of the lowest threshold in the literature in solid-state detectors.25, 28

Thanks to these promising outcomes, additional studies are planned under more complex clinical scenarios to consolidate the capacity of the 3D-cylindrical arrays as QA tool for proton therapy.


The 3D-cylindrical microdetectors fabrication was funded from the H2020 project AIDA-2020, GA no. 654168. This work made use of the Spanish ICTS Network MICRONANOFABS partially supported by MEINCOM. The readout electronics was funded from the European Union's Horizon research and innovation program under the Marie Sklodowska-Curie grant agreement no. 745109. D. Bachiller-Perea and C. Guardiola thank the funding from the CNRS-Momentum fellow and the ALTO facility staff for the help to perform the calibration tests. C. Guardiola acknowledges the computational access from the CNRS/IN2P3 Computing Center (Lyon, France). Beamtime at ICPO was funded by the European Union's Horizon 2020 Research and Innovation program under grant agreement no. 730983 (INSPIRE).


    The authors have no relevant conflicts of interest to disclose.