Benchmarking of five commercial deformable image registration algorithms for head and neck patients.

Benchmarking is a process in which standardized tests are used to assess system performance. The data produced in the process are important for comparative purposes, particularly when considering the implementation and quality assurance of DIR algorithms. In this work, five commercial DIR algorithms (MIM, Velocity, RayStation, Pinnacle, and Eclipse) were benchmarked using a set of 10 virtual phantoms. The phantoms were previously developed based on CT data collected from real head and neck patients. Each phantom includes a start of treatment CT dataset, an end of treatment CT dataset, and the ground-truth deformation vector field (DVF) which links them together. These virtual phantoms were imported into the commercial systems and registered through a deformable process. The resulting DVFs were compared to the ground-truth DVF to determine the target registration error (TRE) at every voxel within the image set. Real treatment plans were also recalculated on each end of treatment CT dataset and the dose transferred according to both the ground-truth and test DVFs. Dosimetric changes were assessed, and TRE was correlated with changes in the DVH of individual structures. In the first part of the study, results show mean TRE on the order of 0.5 mm to 3 mm for all phan-toms and ROIs. In certain instances, however, misregistrations were encountered which produced mean and max errors up to 6.8 mm and 22 mm, respectively. In the second part of the study, dosimetric error was found to be strongly correlated with TRE in the brainstem, but weakly correlated with TRE in the spinal cord. Several interesting cases were assessed which highlight the interplay between the direction and magnitude of TRE and the dose distribution, including the slope of dosimetric gradients and the distance to critical structures. This information can be used to help clinicians better implement and test their algorithms, and also understand the strengths and weaknesses of a dose adaptive approach.

Near Real-Time Assessment of Anatomic and Dosimetric Variations for Head and Neck Radiation Therapy via Graphics Processing Unit-based Dose Deformation Framework.

PURPOSE: The purpose of this study was to systematically monitor anatomic variations and their dosimetric consequences during intensity modulated radiation therapy (IMRT) for head and neck (H&N) cancer by using a graphics processing unit (GPU)-based deformable image registration (DIR) framework. METHODS AND MATERIALS: Eleven IMRT H&N patients undergoing IMRT with daily megavoltage computed tomography (CT) and weekly kilovoltage CT (kVCT) scans were included in this analysis. Pretreatment kVCTs were automatically registered with their corresponding planning CTs through a GPU-based DIR framework. The deformation of each contoured structure in the H&N region was computed to account for nonrigid change in the patient setup. The Jacobian determinant of the planning target volumes and the surrounding critical structures were used to quantify anatomical volume changes. The actual delivered dose was calculated accounting for the organ deformation. The dose distribution uncertainties due to registration errors were estimated using a landmark-based gamma evaluation. RESULTS: Dramatic interfractional anatomic changes were observed. During the treatment course of 6 to 7 weeks, the parotid gland volumes changed up to 34.7%, and the center-of-mass displacement of the 2 parotid glands varied in the range of 0.9 to 8.8 mm. For the primary treatment volume, the cumulative minimum and mean and equivalent uniform doses assessed by the weekly kVCTs were lower than the planned doses by up to 14.9% (P=.14), 2% (P=.39), and 7.3% (P=.05), respectively. The cumulative mean doses were significantly higher than the planned dose for the left parotid (P=.03) and right parotid glands (P=.006). The computation including DIR and dose accumulation was ultrafast (∼45 seconds) with registration accuracy at the subvoxel level. CONCLUSIONS: A systematic analysis of anatomic variations in the H&N region and their dosimetric consequences is critical in improving treatment efficacy. Nearly real-time assessment of anatomic and dosimetric variations is feasible using the GPU-based DIR framework. Clinical implementation of this technology may enable timely plan adaptation and improved outcome.

A virtual phantom library for the quantification of deformable image registration uncertainties in patients with cancers of the head and neck.

PURPOSE: Deformable image registration (DIR) is being used increasingly in various clinical applications. However, the underlying uncertainties of DIR are not well-understood and a comprehensive methodology has not been developed for assessing a range of interfraction anatomic changes during head and neck cancer radiotherapy. This study describes the development of a library of clinically relevant virtual phantoms for the purpose of aiding clinicians in the QA of DIR software. These phantoms will also be available to the community for the independent study and comparison of other DIR algorithms and processes. METHODS: Each phantom was derived from a pair of kVCT volumetric image sets. The first images were acquired of head and neck cancer patients prior to the start-of-treatment and the second were acquired near the end-of-treatment. A research algorithm was used to autosegment and deform the start-of-treatment (SOT) images according to a biomechanical model. This algorithm allowed the user to adjust the head position, mandible position, and weight loss in the neck region of the SOT images to resemble the end-of-treatment (EOT) images. A human-guided thin-plate splines algorithm was then used to iteratively apply further deformations to the images with the objective of matching the EOT anatomy as closely as possible. The deformations from each algorithm were combined into a single deformation vector field (DVF) and a simulated end-of-treatment (SEOT) image dataset was generated from that DVF. Artificial noise was added to the SEOT images and these images, along with the original SOT images, created a virtual phantom where the underlying "ground-truth" DVF is known. Images from ten patients were deformed in this fashion to create ten clinically relevant virtual phantoms. The virtual phantoms were evaluated to identify unrealistic DVFs using the normalized cross correlation (NCC) and the determinant of the Jacobian matrix. A commercial deformation algorithm was applied to the virtual phantoms to show how they may be used to generate estimates of DIR uncertainty. RESULTS: The NCC showed that the simulated phantom images had greater similarity to the actual EOT images than the images from which they were derived, supporting the clinical relevance of the synthetic deformation maps. Calculation of the Jacobian of the "ground-truth" DVFs resulted in only positive values. As an example, mean error statistics are presented for all phantoms for the brainstem, cord, mandible, left parotid, and right parotid. CONCLUSIONS: It is essential that DIR algorithms be evaluated using a range of possible clinical scenarios for each treatment site. This work introduces a library of virtual phantoms intended to resemble real cases for interfraction head and neck DIR that may be used to estimate and compare the uncertainty of any DIR algorithm.

A comparison of the dosimetric effects of intrafraction motion on step-and-shoot, compensator, and helical tomotherapy-based IMRT.

Intrafraction motion during intensity-modulated radiation therapy can cause differences between the planned and delivered patient dose. The magnitude of these differences is dependent on a number of variables, including the treatment modality. This study was designed to compare the relative susceptibility of plans generated with three different treatment modalities to intrafraction motion. The dosimetric effects of motion were calculated using computational algorithms for seven lung tumor patients. Three delivery techniques - MLC-based step-and-shoot (SNS), beam attenuating compensators, and helical tomotherapy (HT) - were investigated. In total 840 motion-encoded dose-volume histograms (DVHs) were calculated for various combinations of CTV margins and sinusoidal CTV motion including CTV offsets. DVH-based metrics (e.g., D95% and D05%) were used to score plan degradations. For all three modalities, dosimetric degradations were typically smaller than 3% if the CTV displacement was smaller than the CTV margin. For larger displacements, technique and direction-specific sensitivities existed. While the HT plans show similar D95% degradations for motion in the SI and AP directions, SNS and compensator plans showed larger D95% degradations for motion in the SI direction than for motion in the AP direction. When averaged over all motion/margin combinations, compensator plans resulted in 0.9% and 0.6% smaller D95% reductions compared to SNS and HT plans, respectively. These differences were statistically significant. No statistically significant differences in D95% degradations were found between SNS and HT for data averaged over all margin and motion track combinations. For CTV motion that is larger than the CTV margin, the dosimetric impact on the CTV varies with treatment technique and the motion direction. For the cases presented here, the effect of motion on CTV dosimetry was statistically smaller for compensator deliveries than SNS and HT, likely due to the absence of the interplay effect which is present for the more dynamic treatment deliveries. The differences between modalities were, however, small and might not be clinically significant. As expected, margins that envelop the CTV motion provide dosimetric protection against motion for all three modalities.

Dosimetric effects of rotational output variation and x-ray target degradation on helical tomotherapy plans.

In this study, two potential sources of IMRT delivery error have been identified for helical tomotherapy delivery using the HiART system (TomoTherapy, Inc., Madison, WI): Rotational output variation and target degradation. The HiArt system is known to have output variation, typically about +/- 2%, due to the absence of a dose servo system. On the HiArt system, x-ray target replacement is required approximately every 10-12 months due to target degradation. Near the end of target life, the target thins and causes a decrease in the beam energy and a softening of the beam profile at the lateral edges of the beam. The purpose of this study is to evaluate the dosimetric effects of rotational output variation and target degradation by modeling their effects and incorporating them into recalculated treatment plans for three clinical scenarios: Head and neck, partial breast, and prostate. Models were created to emulate both potential sources of error. For output variation, a model was created using a sine function to match the amplitude (+/- 2%), frequency, and phase of the measured rotational output variation data. A second model with a hypothetical variation of +/- 7% was also created to represent the largest variation that could exist without violating the allowable dose window in the delivery system. A measured beam profile near the end of target life was used to create a modified beam profile model for the target degradation. These models were then incorporated into the treatment plan by modifying the leaf opening times in the delivery sinogram. A new beam model was also created to mimic the change in beam energy seen near the end of target life. The plans were then calculated using a research version of the PLANNED ADAPTIVE treatment planning software from TomoTherapy, Inc. Three plans were evaluated in this study: Head and neck, partial breast, and prostate. The D50 of organs at risk, the D95 for planning target volumes (PTVs), and the local dose difference were used to evaluate the changes in the modified treatment plans. Dosimetric effects from rotational variation were found to be low (less than 1%) for a typical variation of +/- 2%. Even using a variation of +/- 7%, DVH values and dose distributions were altered by less than 2% for all scenarios. The dosimetric effects of target degradation were found to be slightly more significant. For a model using data taken just before target failure, dosimetric differences of 2%-4% were observed in the recalculated plans when compared to the original plans. The largest effects (up to 4.5%) were observed for PTVs that were located at deeper depths as seen in the prostate plan. Overall, the recalculated plans show that the dosimetric effects of rotational variation and target degradation are on the order of 1%-4% for helical tomotherapy on the HiART system and do not pose a risk for significant deviations from the original treatment plan.

Megavoltage computed tomography image-based low-dose rate intracavitary brachytherapy planning for cervical carcinoma.

Initial results of megavoltage computed tomography (MVCT) brachytherapy treatment planning are presented, using a commercially available helical tomotherapy treatment unit and standard low dose rate (LDR) brachytherapy applicators used for treatment of cervical carcinoma. The accuracy of MVCT imaging techniques, and dosimetric accuracy of the CT based plans were tested with in-house and commercially-available phantoms. Three dimensional (3D) dose distributions were computed and compared to the two dimensional (2D) dosimetry results. Minimal doses received by the 2 cm3 of bladder and rectum receiving the highest doses (D(B2cc) and D(R2cc), respectively) were computed from dose-volume histograms and compared to the doses computed for the standard ICRU bladder and rectal reference dose points. Phantom test objects in MVCT image sets were localized with sub-millimetric accuracy, and the accuracy of the MVCT-based dose calculation was verified. Fifteen brachytherapy insertions were also analyzed. The ICRU rectal point dose did not differ significantly from D(R2cc) (p=0.749, mean difference was 24 cGy +/- 283 cGy). The ICRU bladder point dose was significantly lower than the D(B2cc) (p=0.024, mean difference was 291 cGy +/- 444 cGy). The median volumes of bladder and rectum receiving at least the corresponding ICRU reference point dose were 6.1 cm(3) and 2.0 cm(3), respectively. Our initial experience in using MVCT imaging for clinical LDR gynecological brachytherapy indicates that the MVCT images are of sufficient quality for use in 3D, MVCT-based dose planning.

Organ and effective doses in pediatric patients undergoing helical multislice computed tomography examination.

As multidetector computed tomography (CT) serves as an increasingly frequent diagnostic modality, radiation risks to patients became a greater concern, especially for children due to their inherently higher radiosensitivity to stochastic radiation damage. Current dose evaluation protocols include the computed tomography dose index (CTDI) or point detector measurements using anthropomorphic phantoms that do not sufficiently provide accurate information of the organ-averaged absorbed dose and corresponding effective dose to pediatric patients. In this study, organ and effective doses to pediatric patients under helical multislice computed tomography (MSCT) examinations were evaluated using an extensive series of anthropomorphic computational phantoms and Monte Carlo radiation transport simulations. Ten pediatric phantoms, five stylized (equation-based) ORNL phantoms (newborn, 1-year, 5-year, 10-year, and 15-year) and five tomographic (voxel-based) UF phantoms (9-month male, 4-year female, 8-year female, 11-year male, and 14-year male) were implemented into MCNPX for simulation, where a source subroutine was written to explicitly simulate the helical motion of the CT x-ray source and the fan beam angle and collimator width. Ionization chamber measurements were performed and used to normalize the Monte Carlo simulation results. On average, for the same tube current setting, a tube potential of 100 kVp resulted in effective doses that were 105% higher than seen at 80 kVp, and 210% higher at 120 kVp regardless of phantom type. Overall, the ORNL phantom series was shown to yield values of effective dose that were reasonably consistent with those of the gender-specific UF phantom series for CT examinations of the head, pelvis, and torso. However, the ORNL phantoms consistently overestimated values of the effective dose as seen in the UF phantom for MSCT scans of the chest, and underestimated values of the effective dose for abdominal CT scans. These discrepancies increased with increasing kVp. Finally, absorbed doses to select radiation sensitive organs such as the gonads, red bone marrow, colon, and thyroid were evaluated and compared between phantom types. Specific anatomical problems identified in the stylized phantoms included excessive pelvic shielding of the ovaries in the female phantoms, enhanced red bone marrow dose to the arms and rib cage for chest exams, an unrealistic and constant torso thickness resulting in excessive x-ray attenuation in the regions of the abdominal organs, and incorrect positioning of the thyroid within the stylized phantom neck resulting in insufficient shielding by clavicles and scapulae for lateral beam angles. To ensure more accurate estimates of organ absorbed dose in multislice CT, it is recommended that voxel-based phantoms, potentially tailored to individual body morphometry, be utilized in any future prospective epidemiological studies of medically exposed children.

Organ and effective doses in infants undergoing upper gastrointestinal (UGI) fluoroscopic examination.

To provide more detailed data on organ and effective doses in digital upper gastrointestinal (UGI) fluoroscopy studies of newborns and infants, the present study was conducted employing the time-sequence videotape-analysis technique used in a companion study of newborn and infant voiding cystourethrograms (VCUG). This technique was originally pioneered [O. H. Suleiman, J. Anderson, B. Jones, G. U. Rao, and M. Rosenstein, Radiology 178, 653-658 (1991)] for adult UGI examinations. Individual video frames were analyzed to include combinations of field size, field center, x-ray projection, image intensifier, and magnification mode. Additionally, the peak tube potential and the mA or mAs values for each segment/subsegment or digital photospot were recorded for both the fluoroscopic and radiographic modes of operation. The data from videotape analysis were then used in conjunction with a patient-scalable newborn tomographic computational phantom to report both organ and effective dose values via Monte Carlo radiation transport. The study includes dose estimates for five simulated UGI examinations representative of patients ranging from three to six months of age. Effective dose values for UGI examinations ranged from 1.17 to 6.47 mSv, with a mean of 3.14 mSv and a large standard deviation of 2.15 mSv. The colon, lungs, stomach, liver, and esophagus absorbed doses in sum were found to constitute between 63 and 75% of the effective dose in these UGI studies. Representing 23-30% of the effective dose, the lungs were found to be the most significant organ in the effective dose calculation. Approximately 80-95% of the effective dose is contributed by the dynamic fluoroscopy segments with larger percentages found in longer studies. The mean effective dose for newborn UGI examinations was not found to be statistically different from that seen in newborn VCUG examinations.

Organ and effective doses in newborns and infants undergoing voiding cystourethrograms (VCUG): a comparison of stylized and tomographic phantoms.

The time-sequence videotape-analysis methodology, developed [Sulieman et al., Radiology 178, 653-658 (1991)] for use in tissue dose estimations in adult fluoroscopy examinations and utilized [Bolch et al., Med. Phys. 30, 667-680 (2003)] for analog fluoroscopy in newborn patients, has been extended to the study of digital fluoroscopic examinations of the urinary bladder in newborn and infant female patients. Individual frames of the fluoroscopic and radiographic video were analyzed with respect to unique combinations of field size, field center, projection, tube potential, and tube current (mA), and integral tube current (mAs), respectively. The dosimetry study was conducted on five female patients of ages ranging from four-days to 66 days. For each patient, three different phantoms were utilized: a stylized computational phantom of the reference newborn (3.5 kg), a tomographic computational phantom of the reference newborn (3.5 kg), and (3) a tomographic computational phantom uniformly rescaled to match patient total-body mass. The latter phantom set circumvented the need for mass-dependent rescaling of recorded technique factors (kVp, mA, mAs, etc.), and thus represented the highest degree of patient specificity in the individual organ dose assessment. Effective dose values for the voiding cystourethrogram examination ranged from 0.6 to 3.2 mSv, with a mean and standard deviation of 1.8+/-0.9 mSv. The ovary and colon equivalent doses contributed in total approximately 65%-80% of the effective dose in these fluoroscopy studies. Percent differences in the effective dose assessed using the two tomographic phantoms (one fixed at 3.5 kg with rescaled technique factors rescaled and one physically rescaled to individual patient masses with no adjustment of recorded technique factors) ranged for -49% to +15%. Percent differences in effective dose found using the 3.5 kg stylized phantom and the 3.5 kg tomographic phantom, both with patient-specific rescaling of technique factors, ranged from -10% to +17%. These differences are due in part to a reduced ovary dose in the tomographic phantom for right posterior oblique (RPO) views when compared to those seen in the stylized phantom.

Organ and effective doses in newborn patients during helical multislice computed tomography examination.

In this study, two computational phantoms of the newborn patient were used to assess individual organ doses and effective doses delivered during head, chest, abdomen, pelvis, and torso examinations using the Siemens SOMATOM Sensation 16 helical multi-slice computed tomography (MSCT) scanner. The stylized phantom used to model the patient anatomy was the revised ORNL newborn phantom by Han et al (2006 Health Phys.90 337). The tomographic phantom used in the study was that developed by Nipper et al (2002 Phys. Med. Biol. 47 3143) as recently revised by Staton et al (2006 Med. Phys. 33 3283). The stylized model was implemented within the MCNP5 radiation transport code, while the tomographic phantom was incorporated within the EGSnrc code. In both codes, the x-ray source was modelled as a fan beam originating from the focal spot at a fan angle of 52 degrees and a focal-spot-to-axis distance of 57 cm. The helical path of the source was explicitly modelled based on variations in collimator setting (12 mm or 24 mm), detector pitch and scan length. Tube potentials of 80, 100 and 120 kVp were considered in this study. Beam profile data were acquired using radiological film measurements on a 16 cm PMMA phantom, which yielded effective beam widths of 14.7 mm and 26.8 mm for collimator settings of 12 mm and 24 mm, respectively. Values of absolute organ absorbed dose were determined via the use of normalization factors defined as the ratio of the CTDI(100) measured in-phantom and that determined by Monte Carlo simulation of the PMMA phantom and ion chamber. Across various technique factors, effective dose differences between the stylized and tomographic phantoms ranged from +2% to +9% for head exams, -4% to -2% for chest exams, +8% to +24% for abdominal exams, -16% to -12% for pelvic exams and -7% to 0% for chest-abdomen-pelvis (CAP) exams. In many cases, however, relatively close agreement in effective dose was accomplished at the expense of compensating errors in individual organ dose. Per cent differences in organ dose between the stylized and tomographic phantoms at 120 kVp and 12 mm collimator setting ranged from -25% (skin) to +164% (muscle) for head exams, -92% (thyroid) to +98% (ovaries) for chest exams, -144% (uterus) to +112% (ovaries) for abdominal exams, -98% (SI wall) to +20% (thymus) for pelvic exams and -60% (extrathoracic airways) to +13% (ovaries) for CAP exams. Better agreement was seen between the two phantom types for organs entirely within the scan field. In these cases, corresponding per cent differences in organ absorbed dose did not vary more than 17%. For all scans, the effective dose was found to range approximately 1-13 mSv across the scan parameters and scan regions. The largest effective dose occurred for CAP scans at 120 kVp.

A tomographic physical phantom of the newborn child with real-time dosimetry. II. Scaling factors for calculation of mean organ dose in pediatric radiography.

Following the recent completion of a tomographic physical newborn dosimetry phantom with incorporated metal-oxide-semiconductor field effect transistor (MOSFET) dosimetry system, it was necessary to derive scaling factors in order to calculate organ doses in the physical phantom given point dose measurements via the MOSFET dosimeters (preceding article in this issue). In this study, we present the initial development of scaling factors using projection radiograph data. These point-to-organ dose scaling factors (SF(POD)) were calculated using a computational phantom created from the same data set as the physical phantom, but which also includes numerous segmented internal organs and tissues. The creation of these scaling factors is discussed, as well as the errors associated when using only point dose measurements to calculate mean organ doses and effective doses in physical phantoms. Scaling factors for various organs ranged from as low as 0.70 to as high as 1.71. Also, the ability to incorporate improvements in the computational phantom into the physical phantom using scaling factors is discussed. An comprehensive set of SF(POD) values is presented in this article for application in pediatric radiography of newborn patients.