Reconstructed and Real Proton Radiographs for Image-Guidance in Proton Beam Therapy

Abstract

One of the major challenges to proton beam therapy at this time is the uncertainty of the true range of a clinical treatment proton beam as it traverses the various tissues and organs in a human body. This uncertainty necessitates the addition of greater “margins” to the planning target volume along the direction of the beam to ensure safety and tumor target coverage.

Proton radiography holds promise as both an image-guidance method for proton beam therapy and as a means of estimating particle beam range in the clinic. In this brief report, we present some of the first real and reconstructed proton radiographs using our particular system. Our qualitative review of these images indicates that this method has excellent potential as a proton radiography-based image guidance system. Based on the encouraging results of our preliminary work, more rigorous and quantitative analyses will be performed shortly and we shall continue to explore the potential of this approach for addressing the particle beam range uncertainty issue.

Proton beam therapy potentially offers significant dosimetric advantages through exploitation of the unique physical characteristics of the Bragg curve. The Bragg curve can be described as an “inverted” depth-dose function with a modest entrance/plateau dose followed by a peak region (the Bragg peak) and a relatively abrupt termination of dose after this Bragg peak. The key advantage of proton beam therapy is the stoppage of dose after a certain distance. Compared to photon –based treatment, this abrupt termination of the proton beam should allow proton beam therapy to better spare healthy tissues beyond the Bragg peak while maintaining or increasing dose to the targeted region (which is typically placed along a spread-out Bragg peak).

However, capitalizing on the Bragg curve demands high precision. This precision is presently hindered by difficulties in determining the range of a clinical proton beam to within several millimeters. Proton beam range uncertainties thus remain a significant challenge.

Our group has focused on proton radiography and proton CT to address proton beam range uncertainties. Proton CT is expected to provide better anatomical maps of proton stopping power.^1,2 These improved proton stopping power maps could then be used in treatment planning with better estimates of proton bream range. Using a similar imaging technology, our approach could provide be adapted to provide daily proton radiographs, which could offer image-guidance for proton beam therapy. Such proton radiographs could serve “double-duty” for both image-guidance and for providing proton beam range information. In this qualitative study, we present the first results of our technique of proton radiography as a possible means of image-guidance.

RATIONALE

In routine clinical practice of proton beam therapy, we first obtain an x-ray CT scan for proton therapy treatment planning. Treatment planning is performed using this x-ray CT data set. However, this approach is accompanied by inherent range uncertainties due to an “apples-to-oranges” stoichiometric conversion method that translates CT data (quantified in Hounsfield units) into proton stopping power.¹ One aim of our ongoing project is to address this weakness of conventional x-ray based CT simulation by using proton CT.

Data from the x-ray CT simulation is also used to create digitally reconstructed radiographs (DRRs). These DRRs are used by the radiation therapy treatment team for properly setting patients up for daily treatment (i.e. for daily image-guidance). This form of image-guidance is achieved by comparing DRRs to actual x-rays taken of the patient just before treatment and adjusting patient position based on the comparison.

In our planned approach, rather than starting with an x-ray treatment planning CT scan, we will first obtain a high-energy, low intensity proton CT scan (pCT) which will be used for treatment planning. The pCT must have sufficient energy to completely penetrate through the patient (rather than depositing the Bragg peak within the patient). This data set will provide an “apples-to-apples” comparison in proton stopping power calculations, which is expected to significantly reduce proton beam range uncertainties.

Additionally, in analogy to what is done with x-ray CT simulation and the creation of DRRs, we will generate “reconstructed proton radiographs” (RPRs) also known as “proton digitally reconstructed radiographs” (pDRRs) from the proton CT data set. Again, in analogy to what is done with x-rays, these pDRRs could theoretically be used for daily image-guidance by comparing them with actual proton radiographs taken just before treatment.

The proton radiographs would be obtained immediately before treatment in a manner exactly analogous to present day x-ray-based image-guidance. With the patient in the treatment position, a high-energy, ultra-low intensity proton beam would be directed through the anatomical area to be treated. Such an approach would address both image-guidance and range uncertainty challenges.

Our underlying hypothesis is that proton radiography could someday be used instead of x-rays for high-precision patient alignment in the daily image-guidance process, and the information obtained should also be useful for estimating the range of the treatment beam about to be delivered. In such a manner, image-guidance would be obtained via proton radiography and the information regarding beam range might be used in a “go/no go” decision tree. Of note, if this technique proves adequate, the estimated radiation dose to the patient is only around 1% of what an x-ray approach might deliver.³

In this brief report, we present some of our first reconstructed and real images along with our subjective interpretations.

MATERIALS and METHODS

We have developed a technique that employs a very compact, monolithic set up with plastic fast scintillation 2D tracking detectors that detect individual protons in two dimensions (i.e. in the xy -plane) and a photomultiplier tube (PMT) residual range detector.³ The tracking detectors and photosensors are placed both proximal and distal to the patient, while the PMT range detector is located beyond the distal tracking detector. (Figure 1). In the future we expect to use this type of detector technology for both proton radiography and for proton CT.

Figure 1:

A diagram of our proton radiography method showing the two 2D proton detectors (one proximal and one distal to the patient from the beam perspective) along with a residual energy detector. The residual energy detector carries information about both residual range and anatomical information. The latter information can be used to reconstruct an anatomical image that resembles a megavoltage plain x-ray.

In this preliminary investigation, we present for the first time a series of proton digitally reconstructed radiographs (pDRRs ) from actual patient x-ray CT data. All patients from which the CT data was obtained were treated at the Northwestern Chicago Proton Center and were enrolled in an IRB-approved registry and provided informed consent for use of their anonymized images for research purposes.

The pDRRs are analogous to x-ray digital reconstructed radiographs (xDRRs) that we routinely use to align patients prior to treatment. To create these pDRRs we used GEANT4 (Geometry and Tracking version 4), a Monte Carlo-based algorithm for the simulation of the passage of particles through matter.

After generating these idealized pDRRs, three physicians subjectively evaluated them for image quality (e.g. by determining whether we could see key structures such as bones, air cavities, and soft-tissue landmarks such as the carina. We also compared the pDRRs to xDRRs with two goals: 1. determining if such proton radiographs are adequate for clinical image-guidance (and if they could someday substitute for xDRRs for patient alignment) and 2. reviewing the pDRRs DRRs for any interesting medical or anatomical details.

We also obtained the first actual proton radiographs of biological materials with our imaging system. In this study we chose a frozen tilapia cichlid fish as our first subject. The fish (which was purchased as a frozen food item), was imaged while still bagged and frozen solid at an estimated temperature of −5° C.

RESULTS

In the subjective analysis, all three physicians involved in the study felt that the pDRRs were adequate for patient alignment purposes. Sufficient anatomical detail is visible for alignment of patients for proton beam radiation therapy. The pDRRs appeared reminiscent of megavoltage photon “port films” that were more commonly used in the past.

On a non-objective comparison of our pDRRs with xDRRs, we discovered some very interesting differences between the two image sets. For example, in Figure 2, a reconstructed anterior/posterior (AP) chest image, one can see the carina (a frequently used anatomical landmark for image-guidance) quite clearly on the pDRR.

Figure 2:

Reconstructed proton radiograph of the chest. Note that the carina, a commonly used anatomical landmark for set-up purposes is seen clearly on this image.

In Figure 3, a reconstructed right lateral head & neck image, one can see air cavities within the paranasal sinuses better on the pDRR than on most xDRRs. In fact, one can clearly observe a past surgical procedure (a modified Caldwell-Luc antrectomy) in the hard palate region on the RPR.

Figure 3:

Right lateral RPR of the head. Note that the air cavities in the paranasal sinuses are readily seen along with relatively easy identification of the sphenoid sinus, epiglottis and uvula. Note also the relative ease of identification of a previous surgical procedure (a Caldwell-Luc antrostomy) draining the maxillary sinus.

Figure 4 is the first actual (i.e. not digitally reconstructed) image obtained of a biological specimen using our system. This frozen tilapia cichlid fish image reveals internal bony anatomy relatively well. It also exposes a dark gas bubble, which is likely not the fish’s swim bladder (air bladder), but rather is probably an artifact due to the preparation of this food item.

Figure 4.

Actual proton radiograph of a frozen tilapia fish. There is a large gas bubble (dark area) seen in the body due to processing and packaging but bones are clearly visible. In future clinical applications, images such as this would be compared to reconstructed x-rays (DRRs) or proton radiographs (RPRs) to set patients up on the treatment table. The quality of this first effort with our method suggests that such future efforts have real potential.

DISCUSSION

In this preliminary, non-quantitative study, we have generated a series of idealized reconstructed proton radiographs or proton digital reconstructed radiographs (pDRRs) from patient CT data sets using GEANT4. The resulting pDRRs appear adequate for image guidance and subjectively they resemble linac-based, megavoltage x-rays that are still occasionally used for image guidance.

It is not surprising that our idealized pDRRs are reminiscent of megavoltage x-ray images. At megavoltage photon energies, the relative predominance of the Compton effect over the photoelectric effect becomes more pronounced. Thus, images obtained with megavoltage photons are almost independent of atomic number (Z) but remain directly proportional to electron density. At megavoltage energies, the Compton effect only varies by about 20% from the lightest to the heaviest elements found in animal tissue (except for hydrogen). In a similar fashion, proton beams interact with tissue in manner that, like Compton scattering, is relatively Z-independent but highly dependent on electron density. Therefore, it is no surprise that our proton radiographs (both real and reconstructed) look somewhat like megavoltage photon radiographs. Given that radiation oncologists have used megavoltage photon radiographs to align patients for decades, it might be relatively easy to switch over to similar appearing proton radiographs for image-guidance.

In addition to creating reconstructed proton radiographs from human patient CT data sets, we obtained the first real images of biological materials using our proton radiography system. The images of a frozen osteichthyan teleost (Tilapia sp.) illustrated the bony anatomy with surprising detail, perhaps thanks to our single proton counting method. Such exquisite detail suggests that human bony anatomy will similarly be evident using our proton radiography system. Not evident in Figure 4 is the fact that we found that we could further digitally enhance the fish image (just as plain x-ray windows/levels can now be digitally manipulated to bring out desired contrast). This suggests to us that such processing of human proton radiographs will yield images that will prove adequate for daily image-guidance.

At this time, many proton therapy facilities use X-ray DRRs and daily orthogonal x-rays for patient alignment. While this approach is satisfactory for image-guidance, uncertainties in proton range remain unanswered by this method. In contrast, using daily proton radiographs for image guidance could potentially “kill two birds with one stone” and offer both image-guidance as well as information on proton beam range. In this manner, the physician may determine if both the anatomical alignment is acceptable as well as if the beam range uncertainty is within an acceptable limit for the treatment about to be delivered. Thus, we hope that our pilot study is a step in the direction of eventually using proton reconstructed radiographs and daily orthogonal proton radiographs for both routine image-guidance and proton range determination.

Of importance, we estimate that the radiation dose to the patient with our method is approximately 1% of the absorbed dose using x-rays for the same purpose.^4,5

One purpose of this preliminary study was to determine whether or not proton DRRs might feasibly replace or supplement x-ray DRRs for the purposes of image-guidance. If our pDRRs failed to appear adequate for routine image guidance, then any plans for clinical image-guidance based on pDRRs and pretreatment proton radiographs would be futile. Based on our encouraging results, we plan to continue to explore the potential of pDRRs and actual proton radiographs for image-guidance.

Additionally, our study aimed to obtain and inspect actual proton radiographs using our proton radiography system. The first teleost fish image obtained in this manner suggests to us that this approach will prove useful in the proton therapy clinic. Based on the reconstructed and actual images obtained in this study, we are now confident that clinical proton radiography is viable.

CONCLUSIONS

This preliminary study suggests that daily proton radiography should be feasible for routine image-guidance and patient set-up. This would be performed through comparison of proton DRR with an actual pretreatment proton radiograph (in exactly the same manner as we currently perform x-ray-based image guidance – that is, an x-ray DRR compared to a pretreatment x-ray). Based on our encouraging early results, we will continue our research into proton radiography as a practical and potentially very important method of both image-guidance and reduction of proton beam range uncertainty in the clinic. A more rigorous, objective, and quantitative analysis of proton radiography compared to x-ray DRRs has been initiated.