A proof-of-concept study of an in-situ partial-ring time-of-flight PET scanner for proton beam verification

Abstract

Development of a PET system capable of in-situ imaging requires a design that can accommodate the proton treatment beam nozzle. Among the several PET instrumentation approaches developed thus far, the dual-panel PET scanner is often used as it is simpler to develop and integrate within the proton therapy gantry. Partial-angle coverage of these systems can however lead to limited-angle artefacts in the reconstructed PET image. We have previously demonstrated via simulations that time-of-flight (TOF) reconstruction reduces the artifacts accompanying limited-angle data, and permits proton range measurement with 1–2 mm accuracy and precision. In this work we show measured results from a small proof-of-concept dual-panel PET system that uses TOF information to reconstruct PET data acquired after proton irradiation. The PET scanner comprises of two detector modules, each comprised of an array of 4×4×30 mm³ lanthanum bromide scintillator. Measurements are performed with an oxygen-rich gel-water, an adipose tissue equivalent material, and in vitro tissue phantoms. For each phantom measurement, 2 Gy dose was deposited using 54 – 100 MeV proton beams. For each phantom, a Monte Carlo simulation generating the expected distribution of PET isotope from the corresponding proton irradiation was also performed. Proton range was calculated by drawing multiple depth-profiles over a central region encompassing the proton dose deposition. For each profile, proton range was calculated using two techniques (a) 50% pick-off from the distal edge of the profile, and (b) comparing the measured and Monte Carlo profile to minimize the absolute sum of differences over the entire profile. A 10 min PET acquisition acquired with minimal delay post proton-irradiation is compared with a 10 min PET scan acquired after a 20 min delay. Measurements show that PET acquisition with minimal delay is necessary to collect ¹⁵O signal, and maximize ¹¹C signal collection with a short PET acquisition. In comparison with the 50% pick-off technique, the shift technique is more robust and offers better precision in measuring the proton range for the different phantoms. Range measurements from PET images acquired with minimal delay, and the shift technique demonstrate the ability to achieve <1.5 mm accuracy and precision in estimating proton range.

I. Introduction

Proton beam therapy (PBT) provides highly conformal and precise dose deposition relative to conventional photon radio-therapy (XRT), and has been gaining popularity as one of the more effective forms of cancer therapy with fewer side effects. PBT treatment plans are therefore optimized to either deliver increased dose to the tumor relative to XRT with similar total energy deposited in tissue adjacent to the tumor, or to deliver similar dose to the tumor with reduced total energy deposited in normal tissue. Hence, PBT promises lower patient morbidity, reduced radiation damage to critical organs, and/or lower risk of radiation induced cancer in the future, while providing an effective treatment to treat cancer patients [1–3]. However, PBT treatment plans are currently sub-optimal due to uncertainties in estimating the proton beam range [4–7], and so larger than needed safety margins (as much as 1 cm) are used during proton treatment planning [8] leading to reduced efficacy of this promising technique to spare healthy tissue and side-effects. In the worst-case scenario, range estimate errors can lead to the maximum proton dose being incorrectly deposited in healthy tissue or even some critical organs.

Imaging is routinely used in XRT to monitor dose delivery in patients by imaging the photons exiting the patient [9]. In PBT all the energy is deposited within the patient and hence the only promising signal for imaging is from the secondary radiation produced within the patient due to inelastic and nuclear interactions of protons with tissue nuclei along the proton irradiation path. Nuclear interactions of protons with tissue nuclei leads to the production of β⁺ emitters, while inelastic interactions with tissue nuclei result in the production of short-lived prompt gamma rays. The β⁺ emitters primarily produce ¹⁵O and ¹¹C isotopes, with half-lives of 122s and 1218s respectively. Range monitoring in PBT has been previously proposed by imaging the β⁺ emitters using PET [10–12], or, by capturing the prompt gamma-rays using an arrangement of collimated gamma-ray detectors [13–17]. While both of these strategies have their merits and challenges, they both need to collect fast-decaying or short-lived emissions that are crucial to perform range monitoring in PBT.

For proton beam verification with PET, some research sites have performed investigational studies with a commercial PET/CT located near, or, in the PBT room where the patient is imaged ~10–30 mins after the end of irradiation (delayed PET). Studies performed with rigid, well-defined phantoms show < 1 mm accuracy can be achieved with delayed PET/CT [18]. However, even this delay can significantly hamper range monitoring as a significant fraction of the ¹⁵O signal is lost with a delay as short as 10 mins. In highly perfused or non-rigid regions such as those present in the abdominopelvic (A/P) area, the range measurement using delayed PET is thus significantly compromised [19]. A PET scanner that will allow us to image a patient on the treatment bed during treatment (in-beam or in-situ PET) or with minimal delay (< 1 min) and short scan duration (< 10 mins) after treatment (in-situ PET), will help reduce biological washout and patient motion effects while collecting maximum signal for a high SNR PET image.

Development of a PET system capable of in-beam or in-situ imaging requires a design that can accommodate the proton treatment beam nozzle. The idea of using dual-detector PET for this purpose has been previously explored by several researchers (GSI in Germany [20], HIMAC [21] and Kashiwa [22, 23] in Japan, and Pisa, Italy [24, 25]), and has shown good ability to accurately measure the range of the hadron beam [26]. Besides recent advances in detector technology for particle therapy and beam monitoring [27], Yamaya et al., [28], Shao et al., [29, 30], Bisogni et al., [31, 32], D’Ascenzo et al., [33] and Ferrero et al., [34] have also recently investigated techniques or developed dedicated PET instrumentation to facilitate accurate in-situ proton range measurements. Amongst the different PET scanner designs, the dual-detector system design is most commonly employed as it is easier to develop and integrate within the proton gantry. However, partial angular coverage of these systems can lead to limited-angle effects that cause artefacts in the reconstructed PET image. Previously [35], we have performed system design simulations to compare a full angular coverage scanner with a partial angular coverage system for this purpose. Our results showed that for a realistic number of β⁺ decays and detected 511keV coincidences to reconstruct the PET image, time-of-flight (TOF) reconstruction reduces the artifacts inherent with limited-angle data as opposed to Non-TOF PET. The study also demonstrated that proton range estimate with <1.5 mm uncertainty is achievable. While TOF PET technology has been developed in the diagnostic imaging realm, its use in PBT to generate accurate images with partial angular coverage is relatively new.

In this work we use an experimental setup of a dual detector PET system that uses TOF information to reconstruct accurate 3D images in an integrated PET design that can also accommodate the proton gantry treatment nozzle. As instrumented here, the detector size and imaging field-of-view (FOV) are too small for imaging patients. However, the aim of this work is to model appropriate angular coverage and imaging statistics that will be representative of a clinical scale-up of such a design. Hence, the experimental results shown here will be demonstrative of the performance of larger system meant for clinical imaging. Measurements are performed with an oxygen-rich gel-water phantom, an adipose tissue equivalent material phantom which has elemental carbon and oxygen, and a soft-tissue rich in vitro tissue phantom. Proton range were calculated using data that modeled minimal delay present in in-situ measurements as well as data that modeled longer delays (20 mins) as typically seen in delayed imaging.

II. Materials & Methods

A. Proof-of-concept dual-detector PET system

Based on previous simulation results [35], we developed a small-FOV tabletop PET system using two PET detector modules. Each detector module is 12 × 25 cm² in size and composed of an array of lanthanum bromide (LaBr₃) crystals similar to detector technology developed by us for a research whole-body PET scanner [36–38]. Each LaBr₃ crystal is 4×4×30 mm³ in size and provides adequate spatial resolution and scintillator stopping power for this application [35]. The light-sharing PMT-based detector design is also cost-effective and provides system coincidence timing resolution (CTR) of 450 ps FWHM. The CTR measured in this study is lower than the 375 ps (FWHM) measured with the research whole-body scanner [38] due to the lack of dedicated front-end timing electronics and data acquisition which slightly lowered the CTR. The 450 ps is however, expected to be sufficient for a partial-ring design with limited angle reconstruction as demonstrated in our system design simulation studies [35]. The two detector modules were arranged with their long side aligned along the proton beam irradiation path. The detector separation was set to 12 cm, leading to ~125° angular coverage and a 12 × 12 cm² FOV for this proof-of-concept system. The data acquisition system uses off-the-shelf NIM and CAMAC electronics. Figure 1 shows the small-FOV, tabletop system as currently assembled and placed in the proton beam room. Each PET detector is enclosed in a light-tight aluminum box and the detectors are placed on a horizontal gantry. For the proof-of-concept imaging studies [39], the detectors were placed near the treatment bed for imaging immediately after proton beam irradiation.

Fig. 1.

Photo of the proof-of-concept PET system. The system consists of two detector modules, each containing an array of 4×4×30 mm³ lanthanum bromide scintillation crystals readout by an array of 51 mm diameter PMTs in a light-sharing PET detector design. The two detector modules are separated by 12 cm, and provide a 12 (Y) × 12 (Z) × 25 (X) cm³ FOV. The entire scanner assembly is portable and was placed adjacent to the proton beam irradiation table during experiments.

B. Experimental measurements

Measurements were first performed with two types of rectangular phantom blocks measuring 10–20 cm (X) × 6–7.5 cm (Y) × 10–20 cm (Z) in size. The first phantom (gel-water) had an abundance of elemental oxygen (88.9%), and was prepared by mixing measured quantity of gelatin [40] in distilled water. The second phantom is an adipose tissue equivalent (ATE) material [41] which approaches elemental carbon (73.5%) and oxygen (14.86%) composition in human adipose tissue. Table. 1 summarizes elemental composition for both the phantoms used in this study. A central 4 × 4 cm² area (YZ plane) of each phantom was irradiated with a monoenergetic proton beam. The gel-water phantom was irradiated with 54 MeV and the ATE phantom was irradiated with 92 MeV proton beam. A total dose of 2 Gy was deposited within ~3 seconds for each of the irradiations. The phantom was immediately transferred to the PET gantry and imaged for 60 mins (resulting in 1 min delay between end of irradiation and start of PET scan). PET data were acquired in list-mode and reconstructed using blob-basis functions for regularization [42] and our reconstruction algorithm that is implemented for clinical studies. The list-mode TOF OSEM algorithm uses a 450 ps Gaussian TOF kernel, 25 subsets, and normalization, attenuation, and scatter corrections built into the system model [43]. Scanner calibrations were performed with a phantom uniformly filled with ¹⁸FDG activity. Attenuation correction was performed using an analytical model of the phantom, and its positioning within the scanner FOV was determined using a separate short sequential acquisition with fiducial markers (²²Na point source) placed on the phantom. The list mode data was used to generate images with varying scan start times and durations, and PET images were reconstructed with 2 mm cubic voxels. Delayed imaging was modeled by using list data that were collected 20 mins after irradiation, consistent with prior delayed PET measurements performed by Parodi et al., with a commercial PET/CT scanner [19] and a typical 20 min delay after irradiation. Measurements with the ATE phantom were also repeated with 90 – 92 MeV proton beams used to construct a spread-out Bragg peak (SOBP) depositing uniform dose over a 1 cm region.

TABLE I.

Elemental composition of Phantoms evaluated in this study and comparison with common tissue types in human body

Phantoms evaluated	Oxygen	Carbon
Gel-water	88.9%	NIL
Adipose Tissue equivalent	14.9%	73.5%
Common tissue types in human body
Soft-tissue	63%	23.2%
Adipose-tissue	23.2%	63.7%

We also constructed an in vitro animal tissue phantom by placing organs from a sheep abdomen within an acrylic container (7.5 (Y) × 15.2 (Z) × 15.2 (X) cm³ with 9 mm thick walls) and freezing it. The phantom was irradiated with a ~100 MeV proton beam for a total of 2 Gy deposited over a 3 × 3 cm² (YZ plane) region. List-mode data was acquired after a 1 min delay for a total of 40 mins followed by image reconstruction for varying scan times and durations. For all the above phantoms, the proton irradiation plan was developed by performing a CT scan and importing the CT scan to a treatment planning software Eclipse (Varian, Palo Alto, CA) that is routinely used in the clinic. Fig. 2 shows a schematic of the single-material and in vitro phantoms placed between the two PET detectors, as well as highlights the proton beam irradiation region.

Fig. 2.

(a) Schematic of the phantom block placed between the two PET detectors and the direction of the proton beam. The central region colored in pink highlights the 4 × 4 cm² proton irradiation region. (b) Single slice (YZ plane) of CT image of the in vitro tissue phantom made primarily from abdominal organs placed in a rectangular acrylic container with 9 mm thick walls. The phantom measured 7.5 (Y) × 15.2 (Z) × 15.2 (X) cm³ and the pink region highlights the proton beam irradiation region.

C. Monte Carlo simulations

Nuclear interactions produce β+ emitters along the path of the proton beam travel. These positron-emitter productions occur through a number of distinct nuclear interactions that depends on the incident proton beam energy and elemental composition of the tissue nuclei along the path of the proton beam [44]. Thus, while the distribution of PET activity production is well correlated with the proton range, the endpoint determined from the PET activity distribution differs relative to the Bragg peak and true proton range. The expected distribution of β+ emitters generated by the proton beam, as well as true estimate of the proton range within each of the materials evaluated in this paper was determined using GEANT4 Monte Carlo simulations.

The GEANT4 simulations for the single-material phantoms were performed using TOPAS [45], which provides a user friendly wrapper to use the GEANT4 toolkit specifically for performing proton beam simulations. Rectangular phantom blocks comprised of material with matched elemental composition as the physical phantoms tested in this paper (Section II.B) were simulated. A uniform (flat) proton beam profile to match the experimental conditions was also simulated. The proton beam was tracked through the phantom material and phase space files with information related to all β+ generation within the phantom were saved. For the in vitro phantom, the entire CT DICOM image stack of the phantom was imported, and image data was converted from CT Hounsfield space to material composition and density space [46]. The proton beam irradiation plan corresponding to the 5 mm spaced proton pencil beam scanning used for irradiating the in vitro phantom was also emulated in GEANT4. Due to elemental abundance in human tissue, ¹⁵O, ¹¹C, ¹³N, ³⁰P and ³⁸K represent majority of the β+ generation [44], and information related to these five PET radioisotopes was simulated from known interaction cross-sections [47, 48] at each of the proton step lengths within the in vitro phantom. To match the different PET-acquisition start times and scan times evaluated in this study, decay correction was applied to each of the individual PET isotopes generated. A 5.1 mm isotropic Gaussian smoothing was also applied to the volumetric distribution of the PET radioisotopes generated within each of the phantoms to match the expected spatial resolution from a full-ring PET scanner designed using detector blocks used in this study [38].

D. Image Analysis

For quantitative analysis, several single-voxel (2-mm) profiles were drawn along the proton beam direction (X) in a central region-of-interest (ROI) of the PET image as well as the equivalent image from the Monte Carlo simulation of the positron emitter distribution generated by the proton beam. Bias (i.e. accuracy) in measured range (Δ_i for the i-th profile) was calculated as the difference in the 50% distal fall-off value (position at which the profile counts are 50% of the peak counts) of the simulated and measured profiles. Since each profile is measured over a single voxel (YZ plane) of the PET image, to reduce sensitivity to statistical image noise, the peak value of each individual profile was calculated by averaging all voxels over 90% of the maximum voxel value in that particular profile. In addition, a shift technique [49] that compares the entire fall-off region of the simulated and measured depth profiles was also used. In this technique, the bias in measured range is determined by shifting the entire normalized fall-off region of the measured profile until the minimum of the sum of absolute differences in the activity values is found:

$${\Delta }_i=\underset{\delta }{\text{min}}\left(\sum \left|P_{sim}\left(x_i\right)-P_{\mathit{\text{meas}}}\left(x_i\pm \delta \right)\right|\right)$$ (1)

where P_meas and P_sim are the fall-off regions of measured and simulated profiles along the x direction, δ represents small shifts, and i is the profile index. For the single-material phantoms, a single simulated profile was estimated by averaging over a ~2.5 cm² central area of the phantom, while for the in vitro phantom, corresponding single-voxel profiles were compared. Mean bias and its standard deviation for both the techniques outlined above were calculated over the multiple profiles measured in the entire ROI.

III. Results

A. Gel-water phantom

Figure 3 shows PET images from reconstructing the first 10 mins of the list-mode PET data acquisition after irradiating the gel-water phantom with a monoenergetic beam of 54 MeV protons. The TOF-assisted reconstructions demonstrate good image quality even with a limited-angle PET scanner design. Since the gel-water phantom is primarily composed of oxygen, proton irradiations primarily generate ¹⁵O isotope, which rapidly decays within the first 10 mins. Thus, a large fraction of PET coincidences is generated immediately after irradiation, and a delayed PET scan (e.g. 20 min delay evaluated in this study) is unable to collect ¹⁵O signal from the oxygen rich gel-water phantom. Fig. 3 also highlights the central 3.6 × 3.6 cm² ROI (red box in left image) over which 324 single-voxel depth profiles were generated to measure the proton range using the two range estimation techniques discussed in Section II.D. Measurements show that capturing the ¹⁵O signal with minimal delay post proton irradiation helps in measuring the proton range with accuracy and precision <1.5 mm. Table II summarizes measurements from the experiment.

Fig. 3.

Central slices of PET image reconstructed from a 10 min PET acquisition of the gel-water phantom immediately after irradiation with a 4 × 4 cm² monoenergetic beam of 54 MeV protons. PET reconstructions have 2 mm³ image voxels, and used a TOF OSEM algorithm with 450 ps Gaussian TOF kernel. Even with a partial-ring scanner design, time-of-flight helps maintain image quality by minimizing limited-angle artifacts. The red box (left image) outlines the central 3.6 × 3.6 cm² ROI over which range measurements were performed.

TABLE II.

Proton Range Measurements in Gel-Water Phantom

Delay (mins)	Scan time (mins)	Coincidences collected (million)	Proton range measurement (mm)
			50 % pickoff		Shift technique
			Δ	Std dev	Δ	Std dev
1	6	0.45	−0.6	1.1	1.2	0.9
	10	0.72	−0.1	1.0	0.9	0.5

B. Adipose tissue phantom

Figure 4 shows the PET coincidence rate measured using a NIM event counter that was placed in-line with the scanner data acquisition system and used to monitor the coincidence event rate when acquiring data with the ATE phantom. As the ATE phantom is primarily composed of carbon and oxygen (Table I), the data was fit with a biexponential fit. The biexponential fit shows very good agreement with ¹¹C and ¹⁵O production (see Section I for half-life). PET data collected with the ATE phantom was reconstructed under the same conditions as the gel-water phantom. However, since the ATE phantom has significantly larger carbon composition, PET images were also reconstructed with a scan time of 60 mins.

Fig. 4.

PET coincidence rate measured during the 60 min PET acquisition of the adipose tissue equivalent phantom immediately after irradiation with a monoenergetic 92 MeV proton beam. The PET coincidence event rate (red dots) was fit with a biexponential decay function (red dashed line). There is very good agreement between the measured and expected half-life for the two isotopes, confirming ¹⁵O (green dashed line) and ¹¹C (blue dashed line) production. Also highlighted are the varying scan start time and scan duration over which PET image reconstruction and image analysis to estimate proton range within the phantom.

Figure 5 shows the PET images from the first 10 minutes of the PET acquisition reconstructed with (top row) and without (bottom row) time-of-flight information. Figure 6 also highlights and compares the central sagittal slices of the TOF (left) and corresponding non-TOF (right) reconstructions. Proton range measurements performed over a central 3.6 × 3.6 cm² ROI using both range estimation methods are summarized in Table III for both TOF and non-TOF reconstructions. Identical data acquisition and analysis was also performed with a SOBP proton beam irradiation of the ATE phantom, and Table IV summarizes those results. Even with relatively higher carbon content, higher fraction of coincidence events is collected with a 10 min scan which commences shortly after proton irradiation. Consistent with earlier measurements, proton range with accuracy and precision <1.5 mm can be measured with both monoenergetic and SOBP proton irradiations.

Fig. 5.

Central slices of PET image reconstructed from a 10 min PET acquisition of the adipose tissue equivalent phantom immediately after irradiation with monoenergetic beam of 92 MeV proton, and reconstructed with (top row) and without (bottom row) time-of-flight.

Fig. 6.

Focus on the sagittal slices of the reconstructed PET images from a 10 min PET acquisition of the adipose tissue equivalent phantom immediately after irradiation with monoenergetic beam of 92 MeV proton. The TOF (left) and non-TOF (right) reconstructions can be compared. TOF reconstructions are better at preserving the true-shape of the PET image in a scanner utilizing partial-ring design.

TABLE III.

Proton range measurements in Adipose Tissue Equivalent material Phantom irradiated with Monoenergetic 92 MeV proton beam – Both TOF and Non-TOF reconstructions are compared

Delay (mins)	Scan time (mins)	Coincidences collected (million)	Recon	Proton range measurement (mm)
				50 % pickoff		Shift technique
				Δ	Std dev	Δ	Std dev
1	60	3.4	TOF	1.1	0.9	0.7	0.6
			Non-TOF	0.3	1.2	1.2	0.7
	10	1.12	TOF	−0.4	1.1	0.9	0.7
			Non-TOF	−0.1	1.3	1.5	0.7
21	10	0.59	TOF	−0.4	1.7	0.6	0.8
			Non-TOF	−0.3	1.7	1.2	1.0

TABLE IV.

Proton range measurements from TOF reconstructions of Adipose Tissue Equivalent material Phantom irradiated with 92 MeV 1-CM Spread-Out Bragg Peak proton beam

Delay (mins)	Scan time (mins)	Coincidences collected (million)	Proton range measurement (mm)
			50 % pickoff		Shift technique
			Δ	Std dev	Δ	Std dev
1	60	3.73	−0.3	1.0	0.4	0.3
	10	1.14	−0.8	1.4	0.4	0.4
21	10	0.65	−0.8	1.4	0.4	0.5

C. In vitro animal tissue phantom

Figure 7 shows a central slice from CT images of the in vitro phantom with the ~3 × 3 cm² proton irradiation region highlighted (red box). PET data with the phantom was acquired for 40 mins commencing 1 min after irradiation with a monoenergetic 100 MeV proton beam. Figure 8 shows PET images of the phantom obtained with a 10 min scan. The images acquired immediately after irradiation (top row) can be compared with images acquired for the same scan time, but with a 20 min delay in acquisition (bottom row). Abdominal organs are primarily comprised of soft tissue and subsequently have larger concentration of elemental oxygen. The irradiations are thus expected to generate higher ¹⁵O activity which can only be efficiently captured if PET imaging commences without significant delay.

Fig. 7.

Single CT slice of the in vitro phantom shown along the standard imaging planes. CT scan was used to develop the proton irradiation plan using a ~100 MeV proton beam to deposit 2 Gy dose in a 3 × 3 cm² region that is highlighted by the overlaid red box.

Fig. 8.

PET images from the in vitro animal tissue measurements. Shown above are TOF-assisted reconstructions (along the three imaging planes) from a 10 min PET scan – with almost no delay (top row), and after a 20 min delay (bottom row) following proton irradiation. The delayed PET scan fails to capture endogenous production of ¹⁵O signal from soft-tissue typically found in the abdomen. Note that for the delayed PET scan, the bright activation region near the entrance face of the proton beam (indicated by the orange arrows) is from ¹¹C generation in the carbon-rich acrylic container that holds the in vitro tissue sample and not the tissue sample itself.

Inability to collect ¹⁵O signal not only reduces the overall signal collected, but can also generate an inaccurate representation of PET tracer generation and subsequent assessment of proton dose deposition. Figure 9 includes sample single-voxel profiles obtained from the PET images. Reconstructions using the first 6 and 10 min of the acquisition (top plot), and 10 and 20 min of acquisition after a 20 min delay (bottom plot) are compared with the expected PET tracer profile obtained from a Monte Carlo simulation modeling a 20 min PET scan commencing with a 1 min delay to match the experimental measurements. There is very good agreement between the expected and measured PET activity distribution profiles, especially for acquisitions that begin without delay. For the delayed PET acquisition however, there is a mismatch between the expected (as described above) and measured PET profiles. The inability to collect ¹⁵O signal from the tissue sample, and ¹¹C production in the acrylic container holding the tissue sample artificially exaggerate the measured PET profiles at the entrance region (indicated by orange arrow in Fig. 8 bottom). Sixty-four single-voxel profiles were also generated over a central 1.6 × 1.6 cm² ROI and used to estimate the proton range using both the range estimation techniques. Table V summarizes data from the in vitro phantom measurements. Range measurements from TOF-assisted PET images acquired with minimal delay, and using the shift technique demonstrate ability to achieve <1.5 mm accuracy and precision in estimating proton range in the in vitro phantom.

Fig. 9.

Sample proton range profile generated by measuring the PET activity from a single 2-mm voxel ROI drawn over the TOF PET images reconstructed from measurements with the in vitro phantom. Acquisitions immediately after proton irradiation (top) and after a 20 min delay post proton irradiation (bottom) are included for comparison. Each of the profile is normalized to the total counts within that profile. Also included for comparison with in-situ and delayed PET measurements is the expected PET activity profile generated from Monte Carlo simulations assuming a 20 min scan commencing 1 min after proton irradiation (red dashed line in top and bottom plots).

TABLE V.

Proton Range Measurements in In-vitro Phantom measured from TOF Reconstructions

Delay (mins)	Scan time (mins)	Coincidences collected (million)	Proton range measurement (mm)
			50 % pickoff		Shift technique
			Δ	Std dev	Δ	Std dev
1	10	0.71	3.0	1.4	0.6	1.1
	6	0.55	2.7	1.4	0.1	0.9
21	20	0.26	−5.4	3.9	3.8	0.9
	10	0.15	−4.9	4.6	4.0	1.1

IV. Discussion & Conclusion

Time-of-flight has been previously shown to be beneficial in alleviating limited-angle artifacts in PET scanners employing partial-ring design [50, 51]. A partial-ring PET scanner would also be helpful for proton range verification as the design permits accommodating the proton beam nozzle necessary to perform in-situ measurements. In addition to reducing image artifacts, TOF has also been shown to be beneficial in improving PET image quality, and thereby improve accuracy of proton range measurements [39, 52]. This paper presents experimental results from a small dual-panel PET scanner developed specifically to demonstrate benefits of a TOF capable in-situ PET scanner for proton range verification.

Measurements are performed with an oxygen-rich gel-water, an adipose tissue equivalent material, and in vitro tissue phantoms. The gel-water and ATE phantom measurements demonstrate ability to successfully collect fast-decaying ¹⁵O and ¹¹C PET tracers that are most likely to be generated from nuclear interactions of the proton beam with tissue nuclei in the human body. Measurements demonstrate that PET acquisition with minimal delay is necessary to collect ¹⁵O signal, and maximize ¹¹C signal collection with a short PET acquisition. The ability to collect ¹⁵O using a short scan time is particularly significant for maximizing signal collection required to obtain PET image with sufficient SNR necessary to perform accurate range measurements. Collecting ¹⁵O also helps with measuring an accurate map of the true radio-tracer production, and makes the technique less susceptible to biological washout when measuring signals from highly perfused regions within the body.

Obtaining PET dynamic information about the activity measured from the PET scan has been shown to be beneficial for modeling and correcting for biological washout [53, 54] which is an additional factor that can limit range measurements performed with PET imaging. While the phantoms studied in this paper did not have any biological washout, list-mode data format from the scanner can easily accommodate it. This is however limited in this study by the limited count-rate performance from using an off-the-shelf NIM-CAMAC data acquisition system. Thus, while we see excellent agreement for production of ¹⁵O and ¹¹C (see Figure 4 for ATE phantom data), we are limited in our ability to effectively utilize this information.

Proton range estimate from the PET measurements were obtained by comparing the 50% distal-edge fall-off with corresponding profiles obtained via Monte Carlo simulations. While the technique is fairly simple and has been often used in the past [18, 49, 55], studies have shown that a single distal-edge fall-off fraction is not optimal for all materials, or, for the different isotopes that are generated immediately after irradiation. In addition, the method is susceptible to image (statistical) noise and thus is not robust. This could explain some of the minor inconsistencies when comparing the accuracy and precision in range measurements across the different phantoms using this technique. The shift method [49] on the other hand compares the entire normalized measured profile with the simulated profile. By minimizing the absolute sum of differences over the entire profile, it is less sensitive to image noise and more robust. Overall, for all the phantoms evaluated in this study, the shift technique demonstrates lower standard deviation, suggesting more precise measurements. For the in vitro phantom data, accuracy (bias) is also lowered when estimated using the 50% fall-off technique, and <1.5 mm accuracy and precision is only achievable with the shift technique.

The study also measures the larger PET dataset (coincidence events, or, PET signal) collected when imaged with minimal delay. This is true for all phantoms studied in this paper, with the difference rising with increasing elemental oxygen composition. While the longer 60 min PET acquisition results in large PET signal (i.e. collected coincidence events), it is non-practical. These evaluations were merely performed to study the influence from statistical noise present in the PET data for shorter acquisition times. Measurements demonstrate good PET image SNR for the 10 min PET acquisitions. In fact, the 10 min PET scan times chosen in study is only representative of the short imaging times that would benefit clinical measurements. The insignificant impact of lowering the scan time to 6 min on both measured profile, and range measurements illustrates this. With PET scanner dead-time of 40–50% at the highest coincidence rates immediately after irradiation, we are confident of matching or improving upon the ~5 min scan time evaluations with in-room PET [56] when using a DAQ with minimal or no deadtime.

Results from this study are also consistent with recent in-situ studies [32, 52, 57]. Lopes et al., [52] used 4-mm wide lutetium-yttrium oxyorthosilicate (LYSO) crystals coupled to digital silicon photomultiplier (PDPC), while Topi et al., [32] utilized PET detector using 2-mm wide LYSO scintillation crystals to perform range measurements in polymethyl methacrylate and polyethylene phantoms.

When comparing the non-TOF and TOF measurements in this study, we observe that TOF-assisted reconstructions help alleviate limited-angle image artifacts arising from a partial-ring scanner design. Its impact is evident when comparing the accuracy between measurements from TOF and non-TOF reconstructions (Δ in Table III). The higher image fidelity from using TOF results in profiles closer to Monte Carlo expectation, and slightly lowers the accuracy in measuring proton range. However, we see minimal impact on the precision (std dev in Table III). Further, under clinical conditions, there are numerous other factors that will impact range measurements (e.g. treatment plan, proton beam energy, irradiation volume), and thus, it is likely that the small gain observed in this study could be clinically irrelevant. This could potentially explain why Ferrero et al., [57] report similar accuracy when reporting first clinical measurements with non-TOF reconstruction using data from a 1–1 coupled, silicon photomultiplier (SiPM) based dual-panel PET scanner.

Prior simulation study [35] performed by our group found range measurements using a partial-ring PET scanner to be directly correlated with scanner TOF performance, but fairly insensitive to crystal size when comparing 2–4 mm wide LYSO and LaBr₃ scintillation crystals. The simulation study also suggested accuracy and precision of 1 – 2 mm is achievable when using a scanner with 300 – 600 ps (FWHM) timing resolution. The proof-of-concept scanner in this paper uses 4-mm wide LaBr₃ scintillator in a light-sharing PMT-based PET detector design with 450 ps (FWHM) TOF resolution. Experimental measurements presented in this study are thus also in agreement with the previous simulation study. The experimental setup in this work uses only two detector modules, and thus has limited FOV. However, a scaled up in-situ PET design can be developed to provide the FOV necessary for clinical use with similar imaging performance. For example, a system using a total of 36 detectors similar to what we used here can be arranged in two heads (each head having an array of 3 transverse × 6 axial detectors). Detector head separation of about 65 cm will have similar transverse angular coverage (125 degrees) and geometric sensitivity (Fig. 10). Such a large FOV partial-ring PET scanner can not only be integrated within the proton beam nozzle, but also eliminates the need to transport the patient for PET imaging-based range measurements. As proton beam irradiations are typically fractioned over multiple sessions, accurate proton range measurement after each fraction can improve confidence in developing more accurate proton treatment plans.

Fig. 10.

Conceptual translation of the proof-of-concept PET scanner as used in this study into a larger system suitable for in-situ clinical measurements. Each block represents a 12 cm × 25 cm detector. The transverse and angular coverage is similar in both designs. Besides accommodating the proton beam nozzle, a partial-ring design provides the option to temporarily increase the detector separation for any specific application that necessitates use of proton beams with larger transverse beam width.

Recent developments in PET instrumentation [58], specifically, the push to using silicon photomultiplier (SiPM) has also demonstrated substantial improvements in TOF resolution [59] e.g. state-of-the-art commercial whole-body PET scanner utilizes SiPMs and has TOF performance of 210 ps (FWHM) [60]. Thus, these measurements also suggest that scanner with improved system sensitivity and TOF performance could potentially further improve the overall accuracy and precision of these proof-of-concept PET-based proton range measurements performed with a partial-ring PET scanner.