The developed BPD method demonstrated a clear benefit for the image quality achiev-able with the considered integration-mode range telescope by resolving ambiguities in the
Figure 5.14: Box plot comparison of the RSPrelative error distributions obtained with a simu-lated detector configuration (integration-mode) of 3 mm and 1 mm absorber thickness for the rod phantom. The same representation as in figure5.12 is used.
maximum signal identification in tissue heterogeneities and interfaces [Meyer et al., 2017].
Without post-processing, lateral interfaces increase the integrated detector signal of more proximal channels due to an overlap of the signal components (cf. figure 5.7). This general bias toward larger WET values potentially causes distortions of the actual object geom-etry and size within the iRADs. Hence, the BPD also helps to preserve the accurate object shape. The reduced amount of inconsistencies in the projection data substantially enhanced the RSP accuracy in tomographic image reconstruction (cf. table 5.3).
The implemented simulation model demonstrated generally very good agreement with experimental results, due to an accurate modeling of the beam properties as well as phan-tom and detector composition, RSPand geometry. Variations in the obtainedWET accu-racy between experimental and simulated data for lung tissue originate from neglecting the porous material structure within the MCsimulations, thus, not considering the additional Bragg peak degradation [Titt et al.,2015; Magallanes et al., 2019]. The slightly increased WETerror observed in the stepped wedge phantom (configuration 2) for the experimental cRAD image compared to the heRAD image, is only attributed to an unfavorable carbon ion Bragg peak position within the range telescope for the WET value of the phantom.
(a) (b)
Figure 5.15: (a) Exemplary reconstructed transaxial cCT slices in comparison to the ground truth images. (b) RSP relative error for the reconstructed images as a function of the transaxial slice location. The corresponding anatomical regions for selected slices are shown above the plot.
Since this is the only WET value within the image other than air, it dominates the er-ror, even though cRAD yields a better geometrical agreement with the reference. For the presented range telescope design, the general WET resolution is limited by the detector granularity of 3 mm PMMA-equivalent thickness. This value exceeds the intrinsic range straggling limit of around 2 mm, 0.9 mm and 0.5 mm (water-equivalent) for the used ener-gies of the simulated clinical-like proton, helium and carbon ion beams, respectively. Since the individual ion beams also have approximately the same range (within less than 2 mm in water), a comparableWETaccuracy for thepRADs,heRADsandcRADswas observed.
Due to the comparableWETaccuracy at the same imaging dose for all investigated ion species, the superior image quality obtained for carbon ions reflects the reducedMCSand beam spot size (e.g., 3.8 mmFWHM in air at isocenter for around 18.2 cm range in water compared to 6.1 mm and 10.4 mm for the corresponding helium ion and proton beams, respectively). Potential differences in the biological damage between the investigated ions are discussed in chapter 7. Remaining uncertainties at material boundaries or defective (i.e., consistently under- or over-responsive) channels can cause inaccuracies in the
projec-tion data. Those effects are translated into centered ring artifacts in the reconstructediCT images (cf. figure 5.11) due to the rotational symmetry [Boas and Fleischmann,2012].
In order to reduce the observed immanent noise issue in experimental acquisitions, the detector signal strength needs to be further improved, either by increasing the thickness of the PPICs active volume or by enhancing the particle collection efficiency. The latter can be potentially achieved by using electronics with adaptive gated trigger capability.
This would enable a matching between the integration period and the fluctuating dwell times of individual RPs. Preliminary investigations of adaptive gating using an I4000 multi-channel electrometer (Pyramid Technical Consultants, Inc., Lexington, MA, USA) connected to four detector channels indicated a potentially substantial improvement in the measured integrated current [Magallanes et al.,2019].
The proposed BPDmethod also entails the possibility of supporting the experimental data reconstruction with prior information [Magallanes,2017]. Using an accurate and ex-perimentally validated MC framework for iCT simulations like the one developed in this thesis and available prior knowledge of the object, e.g., from xCT data, the decomposed simulated detector signal can be used to facilitate the BPD for experimental data. By accounting for a potential misalignment between experimental scenario and MC simula-tion, this method can further enhance the image quality even in the presence of channel malfunctioning.
While simulated iCTsdo not considerably benefited from full angular coverage [Meyer et al., 2017], experimental acquisitions can yield further improved image quality [Magal-lanes et al., 2019]. The reason is that the additional information from more projections can be used to compensate the noise inherent in the individual radiographies, however, at the cost of an increased dose exposure to the imaged object of interest. Moreover, full an-gular coverage can be of value for strongly asymmetric cases, i.e., whereMCS and nuclear reactions are considerably different for two opposite projections.
In comparison to the currently in-use prototype detector, the performance of a setup with a reduced absorber thickness of 1 mm demonstrated a clear benefit for the RSP accuracy ofcCT, due to an improvedWET resolution. This quantitatively underlines the usefulness of the envisioned detector upgrade [Meyer et al.,2017]. However, the achievable image quality for the presented integration-mode system lacks of sufficient spatial resolution to be useful for ion therapy treatment planning, as indicated by the the poor image quality for clinical-like data (cf. figure5.15). While the developedBPDstrategy and the proposed system upgrade present promising approaches to enhance the system performance, further
improvement is required.
One potential approach is to fully embed each individual signal component from the BPD into the tomographic image reconstruction [Seller Oria et al., 2018]. Therefore, the integration-lines have to be substituted by cones representing the initial beam spot size convolved with a pre-calculated scattering contribution for the corresponding WETvalue.
Moreover, the iterative image update formula has to be modified accordingly.
Another possibility is to optimize the images in radiographic domain before the tomo-graphic image reconstruction. The inferior spatial resolution limited by theRPspacing can be potentially restored by exploiting the redundant information encoded in the overlap of neighboring beam spots. Such enhancement or reassignment strategies have already been proposed by using additional prior information [Krah et al.,2015] and also without [Gianoli et al., 2016]. These methods yielded an improved image quality for the more challenging proton imaging and could possibly also enable an increased RP spacing and consequently imaging dose reduction.
Due to the statistical distribution of ions within a pencil beam, the presented integration-mode configuration requires certain statistics for obtaining a meaningful repre-sentation of theWETdistribution for eachRPposition [Meyer et al.,2017]. This limitation can be eluded with single-particle tracking, since the information available for each particle is processed individually. Therefore, this configuration enables substantially better spatial resolution [Krah et al., 2018a] and lower imaging doses. However, these benefits come at the cost of a more complex detector instrumentation and increased computational time due to the enormous amount of data and the required path estimation.
trivial.”
Sir Ernest Rutherford
6
Single-Particle Tracking Ion CT for Clinical Treatment Planning
Heavier ions can yield improved image quality by benefiting from reducedMCSin the path estimation for single-particle tracking. However, they also exhibit an increased physical dose (per particle) and suffer from an elevated loss of primaries due to fragmentation, ultimately affecting the noise present in reconstructed images. In this chapter the first comparison of different ion species for iCT, including image quality and range accuracy for proton therapy treatment planning in realistic scenarios is presented. The quality of simulated pCT, heCT and cCT images was assessed for clinical cases of cranial locations.
Moreover, the associated range prediction and dose calculation accuracy was quantitatively evaluated using the RayStation TPS. All investigated scenarios were compared to the established clinical practice of deriving RSP maps via a stoichiometric calibration from single energyxCTdata. Parts of the results presented in this chapter have been published inMeyer et al. [2019].