Detection system, its optimization and characterization
The designed and built radiation detection system was optimized for single ion radiography. The chosen radiation detection technology Timepix exhibits sufficient geometrical segmentation and temporal resolution to register single therapeutic ions. Moreover, it provides a high level of freedom concerning the building of the system, as well as the data acquisition parameters.
The imaging method is based on the measurement of the energy deposition in the rising part of the Bragg curve [
20]. The energy detector is complemented by a tracker composed of the front and rear part. Therefore, the system is applicable for both helium and proton imaging, and thus it enables their direct comparison [
21].
While the majority of the published ion imaging systems are, at least partially, based on scintillation detectors [
51], fully pixelized semiconductor based detection systems are scarce [
48]. Although the electronics for pixelized detectors is significantly more complex than for 1D detectors, this approach has several advantages. While scintillating fibers or silicon strip detectors provide only one coordinate of the particle hit per layer, a tracker based on pixelated detectors provides both coordinates of a particle hit in each layer. The WET of a single used Timepix layer with a thinned readout is about 1 mm, which minimizes the scattering of the ions in the imaging system. Moreover, the pixel technology allows to lower the occupancy of the tracker and enables an improved disentangling of situations where multiple particles are detected in the same time window, e.g. for multiple nuclear fragments originating from the same primary ion.
Our concept is unique in using a single technology for the measurement of the energy deposition, tracking and ion identification [
21]. This allows e.g. a straight forward investigation of different order of tracking and energy deposition modules, which is difficult, and often even impossible, with the existing detection systems. The developed detector alignment procedure enables us to reach subpixel accuracy of the position of the detector layers with respect to each other.
Optimal settings of the detection system, like the acquisition time duration and bias voltage, were found in dedicated studies by maximizing the CNR and SR [
5,
19]. For the measurement of the energy deposition, a fully depleted detector was found to produce too high signals that exceed the linear regime of the detector. A partially depleted sensor provides a lower signal, that leads to a larger effective dynamic range and thus an improved image contrast.
The unique positioning of the rear tracker behind the energy deposition detector enabled us to minimize the deterioration of the energy deposition information by the interactions of the ions with the tracker. A comparison with MC simulations have shown that the accuracy of the measured energy deposition, with the developed recalibration procedure, is below 7% for energy depositions between 0.2 and 17 MeV in 300 μm silicon [
19]. A potential for a further increase in accuracy was found in the systematic trend of the found differences.
In a study about radiation hardness, we observed that for the investigated detector there can be relevant and time-dependent changes of the response due to radiation above 7 Gy [
5]. This shows that a monitoring of the detector response, and performing a recalibration if necessary, is important for high quality radiographies.
Data processing method
A dedicated data processing method was developed. It includes an identification and removal of radiation background and detector artifacts, homogenization of the detector response, single ion identification and tracking [
20]. Finally, all signals in the 5 detectors, which originate from a single particle, were matched [
21]. The improvement of the images by different data processing steps was evaluated in detail. In the energy deposition spectra we successfully identified sources of background – secondary electron and photon radiation, image artifacts due to temporally and spatially incomplete signal readout, overlapping signals and overshoot signals. Their removal improved the CNR by 40% in comparison to the raw data.
When ions heavier than protons are used, a challenge is represented by the nuclear fragments of the primary ions which have a different energy deposition than the primary ions. This leads to an increased image noise and thus limits the WET resolution in the direction along the beam. That issue was addressed by including an ion identification capability, which is based on pattern recognition of the signal measured in the energy detector (Gallas et al. 2017). The removal of identified hydrogen ions improved the CNR by further 110%.
Finally, the consideration of the measured entering and exit position of single ions improved the CNR by additional 60%. The total improvement of the image quality in terms of CNR achieved by the developed image processing method reached 350%.
As expected, the largest improvement of the SR was gained by the consideration of the entrance and exit position of single ions. It was found to be as high as 150%.
The criteria for a clinically applicable ion imaging system include SR, density / thickness resolution for clinically applicable doses, the size of the field of view, imaging time, image reconstruction speed, radiation hardness, issues of patient safety and dimensions of the device making it feasible for implementation in the treatment rooms.
With helium ions the CNR, which quantifies the resolution in tissue thickness or density, was found to be high enough to visualize the 1 mm step (or 0.6% WET difference) in a head-sized phantom at a diagnostic dose of approximately 350 μGy. For protons the CNR was comparable at the same dose level. In case of carbon ions the image quality was found to be limited by the low number of carbon ions per pixel [
21].
Publications on helium ion beam imaging, which we could compare our results to, are rare. Approaches based on passive detectors [
7] have low applicability for the current high throughput facilities. The reported active systems for helium imaging are mainly tomographic. The first system based on scintillating paddles and an MWPC tracker was published already in 1975 [
13]. It was capable to visualize a density difference below 2% in a head sized phantom at a clinically feasible dose. Since its advantages over a clinical CT (status at that time) were demonstrated, it was even approved for a trial with humans.
In [
45] another helium CT system was presented. It was based on a plastic scintillator calorimeter and a scintillating fiber tracker. A WET resolution of 1.5% was found for cylindrical phantoms significantly smaller than an adult head. The imaging dose is not explicitly given. That system was tested also for carbon and neon ion imaging. A helium imaging study with a system designed for proton imaging is reported in [
67]. The system consists of two silicon strip trackers and an energy/range detector based on a plastic scintillator. The relative stopping power accuracy was found to be 2.5% or better in a helium CT of the used phantom.
For comparison, in proton imaging [
57] reports a WET resolution of 0.6 mm for 100 protons per pixel for a proton CT system evaluated with head sized phantoms. [
3] reports a WET resolution of 3.05 ± 0.3 mm per proton at the maximal thickness of the cylindrical phantom of 20 cm WET. A range resolution of 8.4% with a systematic deviation from the expected range of about the same size is reported in [
48] for proton imaging with a digital tracking calorimeter. In that work the deposited energy was determined indirectly from the cluster size.
The high CNR found in the present study makes the developed method promising for a direct visualization of targets with a small WET difference to the surrounding tissue, at clinically feasible doses. In this way the use of fiducial markers, whose placement is invasive, could be avoided.
Additional contrasts due to fluence attenuation, cluster size, particle angle and spread of the measured energy loss in a pixel were investigated. All of them were found to be lower than the contrast due to energy deposition, which is used in the final method.
With the novel ion imaging system a spatial resolution of 0.56 ± 0.04 lp/mm at the MTF
10%, was reached for imaging of a 1 mm step in a head-sized PMMA phantom with helium ions. As expected, due to the increased multiple Coulomb scattering, the SR for protons was found to be lower – only 0.37 ± 0.02 lp/mm. These values were obtained for the inhomogeneity position in the middle of the phantom, which has the maximal distance from both tracker parts. The superior spatial resolution of the helium radiography was found at a comparable thickness resolution (CNR) and imaging dose [
21]. Possible further improvements of the spatial resolution with this system, in particular the performance of different image reconstruction algorithms, were studied in our further research [
21]. Spatial resolution in terms of MTF
10% was found to be 0.61 lp/mm for helium and 0.34 lp/mm for protons in [
67]. Due to the different sizes of the phantoms (10 cm vs. 18.6 cm WET), these values are not directly comparable to our findings.
For comparison, to proton CT systems evaluated with head-sized phantoms, [
57] reports SR of 3.53 mm FWHM for the worst-case scenario. Plautz et al. [
49] found the radial SR to be 0.511 ± 0.061 lp/mm at MTF
10% at the maximal phantom thickness of 20 cm WET.
The system also exhibits further properties important for a clinical application. In contrast to systems with trackers based on multiwire proportional chambers, it does not require any high voltage and gas filling. This increases the patient safety and keeps the size of the system small. With the weight below 0.5 kg, the current prototype is light enough to be mounted on gantries. Its flexibility is important with respect to further developments.
The imaging time was largely dominated by the dead time of the detector (see “
The Timepix detectors” section). However, there are technologies to overcome this in the near future (see “
Outlook” section).
With this kind of system, interfractional imaging of the patient could be performed directly before the treatment start. While the patient is in the treatment position, the detectors could be positioned in front and behind him. After the imaging, the detectors would be removed in order to not impair the quality of the treatment beam. For intrafractional imaging the treatment would have to be paused during the imaging, since for both the ion beam is needed, however with different energies. In contrast to ion computed imaging, no rotation of the beam or the patient is needed for ion radiography, what makes it faster and less complicated, and thus more suitable for first clinical applications.