Acoustic Time-of-Flight for Proton Range Verification in Water Kevin C. Jones1, François Vander Stappen2, Chandra M. Sehgal3, Stephen Avery1,a) 1 Department of Radiation Oncology, University of Pennsylvania, Philadelphia, Pennsylvania Ion Beam Applications SA, Louvain-la-Neuve, Belgium 3 Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania 2 a) Author to whom correspondence should be addressed. Electronic mail: [email protected]; Telephone: 215-615-5631; Fax: 215-746-1727. SUPPLEMENTAL INFORMATION Experiment The electronic pulse used to modulate the cyclotron proton source was created by a function generator (HP8116A) that has a faster rise time (7 ns) and larger pulse-width range (10 ns - 999 ms) than the home-built chopper used previously1 (rise time 5-6 µs, FWHM 30 µs). For the pulse generation method used here, the achievable rise times are limited by the cyclotron: shorter pulses can be created at greatly reduced instantaneous currents (a decrease in rise time from 11 to 8 µs came at the expense of a ~4x drop in proton current). The protons per pulse were calculated by subtracting the dark ionization chamber current, Idark (in the absence of the modulating electronic pulses), from the average ionization chamber current, I (electronic pulse on), dividing by the pulse repetition frequency (PRF [Hz]), and converting the ionization chamber current to proton current: protons I I dark 1 pulse PRF qe K (S1) where qe is the charge of an electron and K is the number of ionization chamber electrons produced per irradiating proton. K is calculated as 25.8 e-/p+ for 230 MeV protons, and it was measured as 24 e-/p+ for 226 MeV protons for the ionization chamber used here. 1 The instantaneous proton current was calculated based on the protons per pulse and the scintillator/PMT-measured shape of the proton spill. To maximize the proton current reaching the treatment room, the energy selection slit width openings were maximized (the energy spread was 0.2% of the average energy), and no energy degrader was inserted into the 230 MeV beam emitted from the cyclotron. Pulse Repetition Frequency Figure S1: The protoacoustic pressure measured with varying PRF by a detector at (s,z) = (9.6, 300 mm). The data is presented as a waterfall plot (incrementing offsets of 30 mPa were added to each data set). The dashed line indicates the 20 µs electronic function generator pulse used to modulate the proton ion source. To determine the proper proton pulse repetition frequency (PRF) for the experiments, the protoacoustic signal was collected at various function generator (and, thus, proton pulse) repetition rates. The collected traces are shown in Fig. S1. In these averaged protoacoustic traces, additional acoustic peaks at negative and positive times (preceding and following the proton pulse) were observed at proton pulse repetition frequencies ≥ 200 Hz (see asterisk for negative 2 time peak appearing at PRF = 200 Hz). The pressure traces measured at proton pulse repetition frequencies ≤ 100 Hz were consistent, and did not display these additional peaks. The appearance of these additional acoustic peaks in averaged pressure traces collected with proton pulse repetition frequencies greater than or equal to 200 Hz indicates that nonnegligible reflected protoacoustic waves persist within the experimental setup for > 5 ms. Due to the large number of reflecting surfaces (5 tank walls and 1 water surface) and the interference between reflections, assigning reflection peaks after the first pass of reflections (which arrive at < ~400 µs) is complex. Increasing the delay between repeated proton pulses to 10 ms (100 Hz pulse repetition frequency), however, allowed a sufficient period for decay of these reflections. Experiments were performed at ≤ 100 Hz to prevent this pile-up of echoes from persisting to the next proton pulse irradiation, a phenomenon that may have influenced the arrival time accuracy and shape of previous results1. Given the low attenuation of water at these frequencies (water attenuates ~2 dB/km at these temperatures2, in 10 ms the protoacoustic waves travel 15 m), the observed decay in echo intensity is most likely due to destructive interference created by reflections from the air/tank wall and tank wall/water surfaces. Following the leading and trailing edge, the function generator pulse appears to induce electronic noise in the protoacoustic traces via proximal BNC cables at the oscilloscope. This noise is of small amplitude, and it is delayed relative to arrival of the protoacoustic signal. Random Noise The random noise, N, was calculated as the standard deviation of the pressure trace, p, over the time range 0.205 - 10 ms following the proton pulse irradiation. The noise was calculated as N = 27 mPa when high-pass and low-pass frequency filters of 0.5 and 100 kHz were applied to p during data processing. As the number of averages was increased, the standard 3 deviation of p decreased with the square root of the number of averages, n1/2, as is shown in Figure S2. Figure S2: The standard deviation of p decreases with increasing averages. Additional Arrival Times In addition to the three arrival time metrics analyzed in the main text ( comp , comp,p , max(P ) ) three additional metrics were also assessed: rare , the arrival time of the rarefaction peak from p; ( comp + rare )/2, the mean of the compression and rarefaction peak arrival times from p, which was previously proposed3; and rare,p , the analgous rarefaction peak from pδ. In addition, for the infrequent case where detectors were placed at z < RBP and the α and γ wave were separably distinguishable, the rare and rare,p arrival times were also extracted (in addition to the higher amplitude α wave features), but the associated positive compression peaks were too weak 4 to consistently measure comp and comp,p . In theory, the measurement of α and γ arrival times allows for measurement of both s and l. The arrival time metrics are indicated in Figure S3. Figure S3 (a) The experimentally measured proton pulse, simulated protoacoustic wave ( p sim ), and experimentally measured protoacoustic wave ( p exp ) collected after 1024 averages for a hydrophone positioned at (s,z) = (45, 62 mm). Metrics corresponding to the arrival time of the experimental compression ( comp ) and rarefaction ( rare , rare ) peaks are labeled. The same metrics (not shown) are also measured from the simulated trace. t = 0 is set to the midpoint between the 50% rise and fall times of the proton pulse. (b) From the data shown in (a), deconvolution of p exp with the excitation proton pulse results in pdecon . The corresponding simulated pressure trace calculated for an impulsive proton pulse, psim , is also plotted. Metrics corresponding to the deconvoluted arrival times are labeled ( comp,p , rare,p , rare,p ). (c) Cumulative integration of the traces shown in (c) results in Pdecon and Psim , from which the metric max(P ) is measured. 5 Arrival Time Analysis: cΔτ To compare simulation to experiment, the differences in observed and simulated arrival times (cΔτ = cτexp- cτsim) extracted from all data traces (excluding data with proton pulse repetition frequency >100 Hz and averages < 512) were calculated and plotted in Fig. S4. The six described arrival time metrics are plotted (from p: comp , rare , ( comp + rare )/2; from pδ: comp,p , rare,p ; from Pδ: max(P ) ). The data is partitioned into three non-overlapping subsets: variable proton width (function generator ≠ 20 µs), α wave (from detectors with z < RBP), and γ wave (from detectors with z > RBP). For the infrequent case where detectors were placed at z < RBP Figure S4: Histograms of the difference between simulated and experimentally determined protoacoustic arrival times, cΔτ = cτexp- cτsim, are plotted for the considered arrival time metrics (see horizontal axis labels). The top row (a-c) metrics are drawn directly from the measured pressure trace, while the bottom row (d-f) displays metrics that are extracted after deconvolution of the proton pulse. The data is partitioned into 4 subsets: the arrival times of the α wave (teal), γ wave (blue, z > RBP), variable pulse width (vpw, red; γ wave at [s,z] = [10,300] ), and γ wave (black, z < RBP). The mean and standard deviation is given for each data subset and the union of the α and γ wave data. 6 and the α and γ wave were separably distinguishable, the rare and rare,p arrival times were also extracted (and plotted in Fig. S4b+e). Consistent with the analysis in the main text, measurement of the arrival time based on the compression peak ( comp and comp,p ) is more robust (lower standard deviation) compared to measurement from the rarefaction peak or cumulatively integrated pressure trace ( rare , rare,p , max(P ) ). This can be explained by the fact that the compression peak is the first arriving and most intense feature of the pressure wave. The rarefaction peak is generally broader, skewed towards later times, and has a lower magnitude amplitude than the compression peak, which may shift the distribution of measured (3.5 mm) and c rare rare to higher values, which may explain the differences between c comp (6.7 mm). As a sum of comp and rare , Δ( comp + rare )/2 reflects a mean and standard deviation intermediate between Δ comp and Δ rare . For pressure traces emitted in response to variable proton pulse-shapes, deconvolution ( comp,p , rare,p , max(P ) ) provided a lower variation compared to the original data ( comp , rare ). For detectors placed at the appropriate pre-Bragg peak position (z < RBP, s ≠ l), the α and γ wave can be simultaneously measured to provide information on s and l. The γ wave has weaker amplitude than the α wave in these cases (the Bragg peak is further away, s < l), and the γ wave baseline is generally lowered by the persisting α wave negative rarefaction peak (see Fig. S3a). Therefore, it is difficult to measure comp for these detector positions. The rare was measured, although the decreased amplitude and interference from the α wave result in a broad 7 distribution (Fig. S4b). Therefore, given the proton pulses used here, calculation of the Bragg peak position based on this rare (detectors z < RBP, s ≠ l) is not tenable. Use of sharper proton pulses (faster rise times) will generate narrower protoacoustic peaks which would allow for these calculations4. Deconvolution results in oscillations that can obscure the simultaneous measurement of this γ wave (see Fig. S3b). Arrival Time Analysis: εTOF To assess the range verification measurement, the difference between experimental protoacoustic arrival time and relevant distance, εTOF = τexpc - (l or s) are plotted in Fig. S5. The standard deviations observed in εTOF reflect the variations observed in cΔτ. Deconvolution resulted in smaller standard deviations, particularly for the variable pulse width data. The compression peak metrics also exhibited smaller standard deviations than the rarefaction or intermediate metrics (( comp + rare )/2 + comp,p ). Figure S5: Histograms of the difference between experimental protoacoustic arrival time and relevant distance, εTOF = τexpc - (l or s), are plotted for the considered arrival time metrics (see horizontal labels). The data is partitioned as described in Fig. S4. 8 None of the analyzed εTOF had distributions (means) centered at 0, which would be ideal. TOF TOF Because the compression peak precedes the rarefaction peak, the means of comp and comp, p are TOF TOF less than rare and rare, p , while the other metrics based on ( comp + rare )/2 and max(P ) are intermediate. Although simulations and/or intuition predict that ( comp + rare )/2 and max(P ) should provide lower average, absolute error in the case of impulsive proton pulses3, the experimental TOF TOF data indicate otherwise for the case of longer proton pulses: the mean of comp and comp, p are closer to zero than those displayed in Fig. S5c+f. None of the considered metrics have a εTOF = 0. But, for eventual use, we propose that the standard deviation is a more important measure for assessing the usefulness of a particular metric: simulations and/or experimental calibration can Figure S6: The arrival times measured from a 5 cm radius circle of detector positions at fixed depth were used to calculate the x and y position of the beam axis. xoff, yoff, and S were fit to minimize the error in observed and calculated τcomp with 2 2 calc. The comp x xoff y yoff c S . comp calculated with this equation are shown for the acoustic data best fit (xoff,yoff) = (-2.0, 0.5 mm) and the film-measured beam propagation axis positions (xoff,yoff) = (0,0). A systematic error of S = -2.9 µs was used in both cases. 9 measure the systematic offset mean of εTOF to correct experimental results. Even if one of the metrics had displayed a low absolute mean εTOF (close to 0), this correlation would not have justified universal adoption of the metric as the ideal because the mean εTOF is expected to be specific to the detector (due to Sdet) and other experimental variables (represented by Sinher). Lateral Beam Position Measurement As described in the main text, the lateral position of the proton beam was measured based on the arrival time of the protoacoustic pressure collected at 8 positions forming a circle at fixed depth surrounding the beam. The angle-dependent arrival times and the results of the fitting are plotted in Figure S6. Amplitude Dependence on Proton Pulse Rise Time As the pulse-width of the function generator rectangular pulse increased, the resulting emitted proton pulse rise time increased. To understand how the pressure amplitude scales with proton pulse rise time, the maximum observed pressure scaled by the maximum of the proton pulse (E(t)) is plotted in Fig. S7. The protoacoustic signal amplitude depends on the shape of the proton pulse, and it scales linearly with proton current (or dose). By normalizing the pressure amplitude by the maximum instantaneous proton current, the dependence on proton pulse shape (ie proton pulse rise time) can be identified. Consistent with previous3 and presented simulation, the protoacoustic amplitude per maximum proton current decreases with lengthening rise time, which indicates that sharper, faster rising proton pulses will increase the protoacoustic signal generated per deposited dose. The non-monotonic, increased ratio observed in Fig. S7 at 33-34 µs may be due to saturation of the PMT detector and/or increased recombination of the ionization chamber at longer proton pulse lengths. 10 Figure S7: The maximum protoacoustic pressure divided by the maximum of the instantaneous proton pulse power plotted as a function of proton pulse rise time. The variable proton pulse width data presented here was collected at a hydrophone position of (s,z) = (10,300) mm. Projected Total Dose versus σ'noise BP The total dose at the Bragg peak ( Dtotal ) and the standard deviation of the arrival time measurement (σ'noise) can both be related to the signal-to-noise-ratio (SNR). The SNR is given by the noise, N (measured to be 27 mPa), the number of averages, n, and the maximum protoacoustic pressure, pmax (measured to be 5.2 mPa / 1 x 107 protons at 5 cm) : SNR n pmax N (S2) The total dose at the Bragg peak is related to the number of averages and the dose from a single BP proton pulse, Dsingle (measured to be 6.1 mGy / 1 x 107 protons) by: BP BP Dtotal nDsingle (S3) The relation between SNR and σ'noise was empirically fit to an exponential function in Figure 3 as: 11 'noise A exp SNR B (S4) Combining eqs. (S2), (S3), and (S4) gives: 'noise A exp BP Dtotal pmax BP Dsingle NB (S5) BP where both pmax and Dsingle scale linearly with the protons / pulse. BP Eq. (S5) was used to project σ'noise as a function of Dtotal in Figure 4 in the main text. References 1. 2. 3. 4. K. C. Jones, F. Vander Stappen, C. R. Bawiec, G. Janssens, P. A. Lewin, D. Prieels, T. D. Solberg, C. M. Sehgal and S. Avery, "Experimental Observation of Acoustic Emissions Generated by a Pulsed Proton Beam from a Hospital-Based Clinical Cyclotron," Medical Physics 42, 7090-7097 (2015). R. J. Urick, Principles of Underwater Sound, 3rd Edition, 3rd ed. (McGraw-Hill, New York, 1983). K. C. Jones, C. M. Sehgal and S. Avery, "How proton pulse characteristics influence protoacoustic determination of proton-beam range: Simulation studies " Phys. Med. Biol. 61, 2213-2242 (2016). K. C. Jones, A. Witztum, C. M. Sehgal and S. Avery, "Proton beam characterization by proton-induced acoustic emission: simulation studies," Physics in Medicine and Biology 59, 6549 (2014). 12
© Copyright 2026 Paperzz