Defining a radiotherapy target with positron emission tomography

Purpose

F-18 fluorodeoxyglucose positron emission tomography (FDG-PET) imaging is now considered the most accurate clinical staging study for non–small-cell lung cancer (NSCLC) and is also important in the staging of multiple other malignancies. Gross tumor volume (GTV) definition for radiotherapy, however, is typically based entirely on computed tomographic data. We performed a series of phantom studies to determine an accurate and uniformly applicable method for defining a GTV with FDG-PET.

Methods and materials

A model-based method was tested by a phantom study to determine a threshold, or unique cutoff of standardized uptake value based on body weight (standardized uptake value [SUV]) for FDG-PET based GTV definition. The degree to which mean target SUV, background FDG concentration, and target volume influenced that GTV definition were evaluated. A phantom was constructed consisting of a 9.0-L cylindrical tank. Glass spheres with volumes ranging from 12.2 to 291.0 cc were suspended within the tank, with a minimum separation of 4 cm between the edges of the spheres. The sphere volumes were selected based on the range of NSCLC patient tumor volumes seen in our clinic. The tank and spheres were filled with a variety of known concentrations of FDG in several experiments and then scanned using a General Electric Advance PET scanner. In the initial experiment, six spheres with identical volumes were filled with varying concentrations of FDG (mean SUV = 1.85 ∼ 9.68) and suspended within a background bath of FDG at a similar concentration to that used in clinical practice (0.144 μCi/mL). The second experiment was identical to the first, but was performed at 0.144 and 0.036 μCi/mL background concentrations to determine the effect of background FDG concentration on sphere definition. In the third experiment, six spheres with volumes of 12.2 to 291.0 cc were filled with equal concentrations of FDG and suspended in a standard background FDG concentration of 0.144 μCi/mL. Sphere images in each experiment were auto-contoured (simulating a GTV) using the threshold SUV that yielded a volume matching that of the known sphere volume. A regressive function was constructed to represent the relationship between the threshold SUV and the mean target SUV. This function was then applied to define the GTV of 15 NSCLC patients. The GTV volumes were compared to those determined by a fixed image intensity threshold proposed by other investigators.

Results

There was a strong linear relationship between the threshold SUV and the mean target SUV. The linear regressive function derived was: threshold SUV = 0.307 × (mean target SUV) + 0.588. The background concentration and target volume indirectly affect the threshold SUV by way of their influence on the mean target SUV. We applied the linear regressive function, as well as a fixed image intensity threshold (42% of maximum intensity) to the sphere phantoms and 15 patients with NSCLC. The results indicated that a much smaller deviation occurred when the threshold SUV regressive function was utilized to estimate the phantom volume as compared to the fixed image intensity threshold. The average absolute difference between the two methods was 21% with respect to the true phantom volume. The deviation became even more pronounced when applied to true patient GTV volumes, with a mean difference between the two methods of 67%. This was largely due to a greater degree of heterogeneity in the SUV of tumors over phantoms.

Conclusions

An FDG-PET-based GTV can be systematically defined using a threshold SUV according to the regressive function described above. The threshold SUV for defining the target is strongly dependent on the mean target SUV of the target, and can be uniquely determined through the proposed iteration process.

Introduction

F-18 fluorodeoxyglucose positron emission tomography (FDG-PET) imaging is now considered the most accurate clinical staging study for non–small-cell lung cancer (NSCLC) and is also important in the staging of multiple other malignancies (1, 2, 3, 4, 5, 6). In addition to its use as a staging tool, PET may also be used as a planning tool for radiotherapy. The benefits of utilizing FDG-PET for gross tumor volume (GTV) definition in three-dimensional conformal radiotherapy or intensity-modulated radiotherapy planning are obvious. NSCLC target volumes have historically been defined based solely on computed tomography (CT) or standard radiographs, and associated with a high degree of inter- and intraobserver variability (7, 8, 9, 10). The prospect of providing a metabolism-based and perhaps more clinically relevant target volume than that historically achieved with either CT or radiographic data alone has prompted many institutions to begin utilizing FDG-PET for treatment planning (11, 12, 13, 14, 15, 16). Unfortunately, there are limited data describing a systematic and uniformly applicable method for accurate FDG-PET-based target volume delineation.

For radiation oncologists, FDG-PET has not been subjected to rigorous validation for GTV definition. Due to the nature of FDG-PET use by nuclear medicine specialists, characterization of a target is focused on a binary decision point (positive vs. negative). Accordingly, minimal literature has addressed the issue of tumor volume definition or edge detection, nor how this may relate to standard CT datasets. In the available literature, GTV definition with FDG-PET has varied, but has typically been based on standardized uptake value (SUV) in some way. Some investigators have chosen to define the PET GTV threshold, or cutoff value, as a percentage of the maximum or peak SUV concentration, whereas others have utilized an absolute SUV value (8, 17). Depending on how that threshold is chosen, the resultant GTV volume may vary significantly. Due to the multiplicity and complexity of patient and tumor variables yielding widely differing maximum intensities, and, thus, equally differing thresholds, the single threshold method may not reliably depict the true volume, particularly for anything other than small solitary nodules.

F-18 fluorodeoxyglucose positron emission tomography imaging output is a depiction of the response of the PET detector to the heterogeneous distribution of radiotracer emissions. Neoplastic tissues with deranged replicative systems and metabolism maintain an elevated basal rate of glucose influx and catabolism to sustain these hyperactive processes. As radiotracer collects and is trapped within all cells of the system, the detector registers the resultant heterogeneous emissions. After complex software manipulation of the signal, a visual representation is reconstructed. In this way, actual or suspected neoplastic tissues are visualized as three-dimensional units (voxels), with higher intensities corresponding to higher concentrations of activity. This activity is directly proportional to the amount of radiotracer present within a particular voxel (represents an aggregate of cells). For instance, 1 cubic centimeter of tissue can contain on the order of 1 billion cells. Each individual cell gives a specific emission equivalent to the amount of radiotracer trapped in that cell. This tiny emission appears as a point source of “pulse”, and produces a sphere of distribution of signals due to the point-spread nature of positron emission. These spread signals are then accumulated by the PET detector for constructing the heterogeneous voxel intensities of a PET image. Although this process can identify a tumor and its probable location, it cannot quantify the boundary of that tumor. This is because we do not know where on the signal the most likely edge is due to the point-spread and the accumulation described above. By assuming uniform emission of tumor cells, the edge of a tumor volume determined could be the function of uptake intensities in the tumor and surrounding normal tissues. Additional complicating factors include, but are not limited to: variable cell density within tumors, variable histology-specific metabolism, variable vascular access, the presence of necrosis, and the background metabolism of different normal tissues.

The question of how to define a volume in FDG-PET can be posed as “What voxel intensity value (commonly with a unit of Bq/cc, μCi/cc, or SUV) should we select above which all activity in the region will contribute accurately to the resultant volume?” As mentioned above, other authors have proposed to characterize this as a universal threshold based on percentages of maximum intensities within a region of interest (ROI) or an absolute SUV value, either of which may not be entirely appropriate (8, 16). Given the paucity of data on FDG-PET-based GTV definition, along with the potential benefits of multimodality targeting, we performed a series of phantom studies to develop a foundation of data with which we might accurately define a GTV with FDG-PET. The goals were to create a systematic method for GTV definition that can be used by radiation oncologists, and remove much of the inter- and intraobserver variation inherent in FDG-PET image-based treatment planning.