A comparative study of machine-learning approaches in proton radiography using energy-resolved dose function.

PURPOSE: The feasibility of machine learning (ML) techniques and their performance compared to the conventional χ2-minimization technique in the context of the proton energy-resolved dose imaging method are presented. MATERIALS AND METHOD: Various geometries resembling a wedge and varying gradients are simulated in GATE to obtain energy-resolved dose functions (ERDF) from proton beams of different energies. These ERDFs are used to predict the WEPL using a conventional technique and other ML-based methods. The results are compared to gain an understanding of the performance of ML models in proton radiography. RESULTS: The results obtained from the χ2-minimization technique indicate that it is robust and more reliable compared to the ML-based techniques. It is also observed that the ML-based techniques did not mitigate the effect of range-mixing but seem to be more affected by it compared to the χ2-minimization technique. Substantial data reduction was required in order to make the results of ML-based methods comparable to that of χ2-minimization. We also note that such data reduction might not be possible in a clinical setting. The only advantage in using the ML-based technique is the computational time required to generate a WEPL map which, in our case study, is 10-30 times shorter than the time required for the conventional χ2-minimization technique. CONCLUSIONS: The first results from this preliminary study indicate that the ML techniques failed to be on par with the conventional χ2-minimization technique in terms of the achievable accuracy in the predictions of WEPL and in the mitigation of range-mixing effects in the WEPL image. Modern strategies like the GAN-based models may be suitable for such applications.

Feasibility of energy-resolved dose imaging technique in pencil beam scanning mode.

Purpose:Proton energy-resolved dose imaging (pERDI) is a recently proposed technique to generate water equivalent path length (WEPL) images using a single detector. Owing to its simplicity in instrumentation, analysis and the possibility of using the in-room x-ray flat panels as detectors, this technique offers a promising avenue towards a clinically usable imaging system for proton therapy using scanned beams. The purpose of this study is to estimate the achievable accuracy in WEPL and Relative Stopping Power (RSP) using the pERDI technique and to assess the minimum dose required to achieve such accuracy. The novelty of this study is the first demonstration of the feasibility of pERDI technique in the pencil beam scanning (PBS) mode.Methods:A solid water wedge was placed in front of a 2D detector (Lynx). A library of energy-resolved dose functions (ERDF) was generated from the dose deposited in the detector by 50 PBS layers of energy varying from 100 MeV to 225 MeV. This set-up is further used to image the following configurations using the pERDI technique: stair-case shaped solid water phantom (configuration 1), electron density phantom (configuration 2) and head phantom (configuration 3). The result from configuration 1 was used to determine the achievable WEPL accuracy. The result from configuration 2 was used to estimate the relative uncertainty in RSP. Configuration 3 was used to evaluate the effect of range mixing on the WEPL. In all three cases, the variation of the accuracy with respect to dose, by varying the number of scanning layers, was also studied.Results:An accuracy of 1 mm in WEPL was achieved using the Lynx detector with an imaging field of 10 PBS layers or more, which is equivalent to a total dose of 5 cGy. The RSP is measured with a precision better than 2% for all homogeneous inserts of tissue surrogates. The pERDI technique failed for tissues surrogates with total WEPL outside the calibration window (WEPL < 70 mm) like in the case of lung exhale and lung inhale. The imaging of an anthropomorphic head phantom, in the same condition, produced a WEPL radiograph and compared to the WEPL derived from CT using gamma index analysis. The gamma index failed in the heterogeneous areas due to range mixing.Conclusions:The pERDI technique is a promising clinically usable imaging modality for reducing range uncertainties and set-up errors in proton therapy. The first results have demonstrated that WEPL and RSP can be estimated with clinically acceptable accuracy using the Lynx detector. Similar accuracy is also expected with in-room flat-panel detectors but at significantly reduced imaging dose. Though the issue of range mixing is still to be addressed, we expect that a statistical moment analysis of the ERDFs can be implemented to filter out the regions with high gradient of range mixing.

Concept of proton radiography using energy resolved dose measurement.

Energy resolved dosimetry offers a potential path to single detector based proton imaging using scanned proton beams. This is because energy resolved dose functions encrypt the radiological depth at which the measurements are made. When a set of predetermined proton beams 'proton imaging field' are used to deliver a well determined dose distribution in a specific volume, then, at any given depth x of this volume, the behavior of the dose against the energies of the proton imaging field is unique and characterizes the depth x. This concept applies directly to proton therapy scanning delivery methods (pencil beam scanning and uniform scanning) and it can be extended to the proton therapy passive delivery methods (single and double scattering) if the delivery of the irradiation is time-controlled with a known time-energy relationship. To derive the water equivalent path length (WEPL) from the energy resolved dose measurement, one may proceed in two different ways. A first method is by matching the measured energy resolved dose function to a pre-established calibration database of the behavior of the energy resolved dose in water, measured over the entire range of radiological depths with at least 1 mm spatial resolution. This calibration database can also be made specific to the patient if computed using the patient x-CT data. A second method to determine the WEPL is by using the empirical relationships between the WEPL and the integral dose or the depth at 80% of the proximal fall off of the energy resolved dose functions in water. In this note, we establish the evidence of the fundamental relationship between the energy resolved dose and the WEPL at the depth of the measurement. Then, we illustrate this relationship with experimental data and discuss its imaging dynamic range for 230 MeV protons.

Validation of an in-vivo proton beam range check method in an anthropomorphic pelvic phantom using dose measurements.

PURPOSE: In-vivo dosimetry and beam range verification in proton therapy could play significant role in proton treatment validation and improvements. In-vivo beam range verification, in particular, could enable new treatment techniques one of which could be the use of anterior fields for prostate treatment instead of opposed lateral fields as in current practice. This paper reports validation study of an in-vivo range verification method which can reduce the range uncertainty to submillimeter levels and potentially allow for in-vivo dosimetry. METHODS: An anthropomorphic pelvic phantom is used to validate the clinical potential of the time-resolved dose method for range verification in the case of prostrate treatment using range modulated anterior proton beams. The method uses a 3 × 4 matrix of 1 mm diodes mounted in water balloon which are read by an ADC system at 100 kHz. The method is first validated against beam range measurements by dose extinction measurements. The validation is first completed in water phantom and then in pelvic phantom for both open field and treatment field configurations. Later, the beam range results are compared with the water equivalent path length (WEPL) values computed from the treatment planning system XIO. RESULTS: Beam range measurements from both time-resolved dose method and the dose extinction method agree with submillimeter precision in water phantom. For the pelvic phantom, when discarding two of the diodes that show sign of significant range mixing, the two methods agree with ±1 mm. Only a dose of 7 mGy is sufficient to achieve this result. The comparison to the computed WEPL by the treatment planning system (XIO) shows that XIO underestimates the protons beam range. Quantifying the exact XIO range underestimation depends on the strategy used to evaluate the WEPL results. To our best evaluation, XIO underestimates the treatment beam range between a minimum of 1.7% and maximum of 4.1%. CONCLUSIONS: Time-resolved dose measurement method satisfies the two basic requirements, WEPL accuracy and minimum dose, necessary for clinical use, thus, its potential for in-vivo protons range verification. Further development is needed, namely, devising a workflow that takes into account the limits imposed by proton range mixing and the susceptibility of the comparison of measured and expected WEPLs to errors on the detector positions. The methods may also be used for in-vivo dosimetry and could benefit various proton therapy treatments.