Proton CT for Improved Stopping Power Determination in Proton Therapy, invited

INTRODUCTION

Following the first suggestion that the Bragg peak of protons provides distinct advantages in radiation therapy by Harvard physicist Robert Wilson in 1946, it took more than 40 years until the first hospital-based proton treatment center opened in 1990. Many factors may have contributed to this relatively slow development, but the lack of appropriate 3D imaging and accurate planning techniques were among the major reasons. The situation for proton therapy improved in the early 1970s, when X-ray computed tomography developed rapidly. For the first time, one could obtain proton stopping power maps, allowing scaling inhomogeneous tissues to water-equivalent density, which then made simulation of proton treatment plans possible. It was soon recognized, however, that due to the use of X-ray CT systematic errors entered the planning process leading to either under- or overestimation of proton range. In order not to under-dose tumors, physicians had to add an extra margin for range uncertainty, which prevented them from stopping the Bragg peak of proton beams in front of critical normal structures, giving up one major advantage of proton therapy. This situation is not different today. The purpose of this paper is to give a brief historical overview of the development of proton radiography and proton CT, i.e., modalities addressing the issue of proton range uncertainty, and to illustrate how advancements in particle detectors and readout electronics lead to the first preclinical prototype of a proton CT scanner.

HISTORICAL DEVELOPMENT OF PROTON RADIOGRAPHY AND PROTON CT

Proton Radiography

Using principles of 2D radiographic X-ray imaging, the first proton imaging studies were proton radiographs using the 160 MeV Harvard Cyclotron. Koehler showed that by using a stack of parallel sided aluminum plates of a thickness just less than the proton range, radiographs of much greater contrast sensitivity could be recorded with protons compared to X-rays [1]. Steward and Koehler [2–4] and others [5, 6] demonstrated that the high contrast images obtained by proton radiography provided improved resolution of low-contrast lesions in human specimens over conventional X-ray techniques. The high contrast obtained in this energy-loss form of radiography is a consequence of the high sensitivity of proton fluence on the distal Bragg peak to small variations in integral water-equivalent thickness of the object. It was not until the mid-1990s that interest in proton imaging was revived when Schneider and colleagues at the Paul Scherrer Institute in Switzerland used proton radiography as a QA tool to measure the accuracy of proton therapy range prediction with X-ray CT methods [7–9].

Proton Computed Tomography

Soon after methods for reconstructing tomographic images from X-ray projections had been developed based on the landmark contributions by Cormack and Hounsfield during the 1960s, first tomographic image reconstructions were also performed with alpha-particle tomography at LBL [10, 11]. Cormack and Hounsfield [12] and a few years later Hanson et al. [13–16] performed first experimental studies with proton CT (pCT). It was found that pCT had a dose advantage when compared to X-ray CT, and multiple Coulomb scattering (MCS) was identified as one of the principal limitations of pCT in terms of spatial resolution.

In 2000, Zygmanski and colleagues presented a cone-beam pCT system for improving the relative proton stopping power (RSP) estimation needed for proton treatment planning [17]. Their principle was based on a fluence modulated proton beam, where the fluence of the exit beam from any projection was proportional to the traversed water-equivalent thickness (WET) of the object. Their fluence detector consisted of a solid state intensifying screen viewed by a cooled CCD camera, allowing for fast acquisition and reconstruction of 3D proton RSP maps using the Feldkamp-Davis-Kress (FDK) [18] cone-beam reconstruction algorithm. However, due to the significant lack of spatial resolution and a large degree of noise, this concept was not pursued further.

Driven by continued clinical need for more proton range accuracy, a diverse group of scientists, among them the first author of this contribution, met at the Brookhaven National Laboratory (BNL) in 2003 and formed a pCT collaboration. In the following years, physicists from INFN in Italy did joint work with the pCT, resulting in similar types of CT detectors being developed [19, 20] as in the U.S. A series of publications related to pCT appeared during the period of 2003–2008 [21–28], documenting progress made in the development of new pCT technology during these years.

The conceptual development of the pCT scanner, described below, has mostly become possible by application of the latest detector technology from High Energy Physics (HEP). Silicon microstrip trackers and crystal calorimetry, commonly applied in HEP, allow for the achievement of excellent spatial and energy resolution in pCT. These detector elements, which can be considered the standard in any HEP experiment, are slated for state-of-the-art application in pCT.

Our approach, outlined in some detail in the results section, can be viewed as innovative, as it deviates from the original pCT approach by Hanson and the more recent work by Zygmanski. The main difference is that it is based on single-proton detection and uses the most likely path (MLP) concept [29–31] for reconstruction, thus taking into account the curved proton path to produce tomographic reconstructions with sufficient spatial resolution, despite MCS. In this paper, an overview of the design concepts developed since 2003, the approach to pCT image reconstruction, and the first prototype built based on these concepts are presented. A more detailed description of the first prototype pCT scanner (Phase 1), the design of the next scanner (Phase 2), and a report on new developments in pCT reconstruction are presented elsewhere in these proceedings.

RESULTS

Conceptual pCT Detector Approach

Proton CT can directly obtain the RSP map from proton energy loss measurements and, therefore, removes the errors associated with conversion of X-ray CT Hounsfield units. With currently available detector technology, we initially developed the pCT design shown in Fig. 1 [23], where individual protons are tracked before entering and after exiting the patient with 2D sensitive silicon strip detectors (SSDs), providing information about proton position and direction at the boundaries of the image space. This allows accounting for the effects of multiple Coulomb scattering within the object by estimating the MLP.

Fig. 1.

Schematic illustration of a single-proton-tracking pCT scanner. Protons with known incident energy E_in are individually recorded by the four planes of position-sensitive silicon detectors and the exit energy E_out is recorded with a segmented crystal calorimeter.

In addition to tracking the position of individual protons, the energy lost by each proton after traversal of the image space is recorded. Using these measurements in conjunction with a careful calibration of the calorimeter response against known integral RSP with plates of constant water-equivalent thickness, one can reconstruct the integral of RSP along the MLP for each proton.

Image Reconstruction

The MLP concept allows the use of reconstruction algorithms based on iterative projection methods. One can precede this reconstruction by a fast, but less accurate reconstruction based on filtered back projection (FPB), which also results in knowledge of the outer contour of the object in reconstruction space and provides entry and exit points for the MLP calculation. With this knowledge, and using the FPB solution as the initial iterate, the image reconstruction proceeds as follows (Fig. 2).

Fig. 2.

Formation of the matrix A using the reconstructed MLP of each proton.

Assume the image space is digitized, forming an m-dimensional vector x of initially unknown stopping power values of the object. As mentioned, the FBP vector can serve as the initial estimate. A total number of n proton histories traversing the object is collected and the MLP of each history is calculated from tracker information. In addition, one estimates the integral of RSP along the MLP, i.e., the water equivalent pathlength (WEPL) of each proton, from the calorimeter response. The set of all WEPL values forms the n-dimensional vector b. The MLPs are also digitized and expressed as a matrix row vectors {a_ij} where i is the index of the proton (i = 1 …n) and j is the object voxel index (j = 1…m). The n × m matrix A composed of these vectors is the “design matrix” of the linear equation system

$$A\mathit{x}=\mathit{b},$$ (1)

, where the elements a_ij correspond to the length of intersection (chord length) of the i-th proton MLP with the j-th voxel. In realistic pCT reconstructions, the system of equations will be inconsistent and the matrix A is very large (~10⁸ × 10⁷) and sparse.

The reconstruction problem is then reduced to finding a solution to the linear system (1). After initially using the algebraic reconstruction technique (ART) [26], which is a sequential projection method, we have tested faster, parallelizable algorithms, which are suitable for implementation on fast graphics processing units (GPUs). Using Geant4 simulations generating realistic pCT data sets, it has been shown that pCT reconstructions of good quality can be obtained [32, 33]. More recently, we have investigated to the use of superiorization methods for pCT reconstruction leading to improved image quality [34] and potentially faster convergence [35].

Phase 1 pCT Scanner

The first preclinical pCT prototype capable of scanning a head size object was built between 2008 and 2010 by a collaboration between the Department of Physics at Northern Illinois University, the Department of Radiation Medicine at Loma Linda University Medical Center (LLUMC), and the Santa Cruz Institute of Particle Physics (SCIPP) at UC Santa Cruz. The design of this system followed the layout shown in Fig. 1. It is comprised of front and rear silicon tracker modules, each consisting of 4 XY planes for full coordinate and direction data. An array of 18 CsI (T1) crystals provides the energy detector, which is integrated with the rear tracker modules. The scanner is mounted on a rail system to bring the detectors close to the phantom object that rotates on the horizontal proton beam axis in the proton research room at LLUMC. The system, which is currently being used to gain experience with pCT, is described in more detail in the paper by Hurley et al. in these proceedings [36].

Fig. 3.

The Phase 1 pCT scanner consisting of a front module (right) containing the front tracker and a rear module (left) containing the rear tracker and the crystal-calorimeter.