3D dose prediction for Gamma Knife radiosurgery using deep learning and data modification

PURPOSE

To develop a machine learning-based, 3D dose prediction methodology for Gamma Knife (GK) radiosurgery. The methodology accounts for cases involving targets of any number, size, and shape.

METHODS

Data from 322 GK treatment plans was modified by isolating and cropping the contoured MRI and clinical dose distributions based on tumor location, then scaling the resulting tumor spaces to a standard size. An accompanying 3D tensor was created for each instance to account for tumor size. The modified dataset for 272 patients was used to train both a generative adversarial network (GAN-GK) and a 3D U-Net model (U-Net-GK). Unmodified data was used to train equivalent baseline models. All models were used to predict the dose distribution of 50 out-of-sample patients. Prediction accuracy was evaluated using gamma, with criteria of 4 %/2mm, 3 %/3mm, 3 %/1mm and 1 %/1mm. Prediction quality was assessed using coverage, selectivity, and conformity indices.

RESULTS

The predictions resulting from GAN-GK and U-Net-GK were similar to their clinical counterparts, with average gamma (4 %/2mm) passing rates of 84.9 ± 15.3 % and 83.1 ± 17.2 %, respectively. In contrast, the gamma passing rate of baseline models were significantly worse than their respective GK-specific models (p < 0.001) at all criterion levels. The quality of GK-specific predictions was also similar to that of clinical plans.

CONCLUSION

Deep learning models can use GK-specific data modification to predict 3D dose distributions for GKRS plans with a large range in size, shape, or number of targets. Standard deep learning models applied to unmodified GK data generated poorer predictions.

Assessment of the accuracy and stability of frameless gamma knife radiosurgery

The aim of this study was to assess the accuracy and stability of frameless gamma knife radiosurgery (GKRS). The accuracies of the radiation isocenter and patient couch movement were evaluated by film dosimetry with a half-year cycle. Radiation isocenter assessment with a diode detector and cone-beam computed tomography (CBCT) image accuracy tests were performed daily with a vendor-provided tool for one and a half years after installation. CBCT image quality was examined twice a month with a phantom. The accuracy of image coregistration using CBCT images was studied using magnetic resonance (MR) and computed tomography (CT) images of another phantom. The overall positional accuracy was measured in whole procedure tests using film dosimetry with an anthropomorphic phantom. The positional errors of the radiation isocenter at the center and at an extreme position were both less than 0.1 mm. The three-dimensional deviation of the CBCT coordinate system was stable for one and a half years (mean 0.04 ± 0.02 mm). Image coregistration revealed a difference of 0.2 ± 0.1 mm between CT and CBCT images and a deviation of 0.4 ± 0.2 mm between MR and CBCT images. The whole procedure test of the positional accuracy of the mask-based irradiation revealed an accuracy of 0.5 ± 0.6 mm. The radiation isocenter accuracy, patient couch movement accuracy, and Gamma Knife Icon CBCT accuracy were all approximately 0.1 mm and were stable for one and a half years. The coordinate system assigned to MR images through coregistration was more accurate than the system defined by fiducial markers. Possible patient motion during irradiation should be considered when evaluating the overall accuracy of frameless GKRS.

Validation of MLC-based linac radiosurgery for trigeminal neuralgia

This study details a validation process for linear accelerator‐based treatment of trigeminal neuralgia using HD‐MLC field collimation. Nine trigeminal neuralgia treatment plans utilizing HD‐MLC were selected for absolute dose measurement at isocenter using a commercial scintillating detector in an anthropomorphic phantom. Four plans were chosen for film dosimetry measurements in each of the three principal planes to assess spatial dose distribution agreement with the treatment planning system. Additionally, trajectory log analysis for each treatment field in the nine cases was performed to assess mechanical positioning accuracy of the MLC system during delivery. Scintillator and film measurements both revealed mean dose agreement at isocenter of better than 3% while FWHM of the 2D dose distribution in each plane showed agreement between plan and measurement within 0.2 mm. Analysis of log files revealed a maximum MLC leaf positioning error of 0.04 mm across 178 treatment fields. In conjunction with a quality‐controlled treatment delivery methodology, an appropriately commissioned treatment planning system can be used for accurate and clinically appropriate design of trigeminal neuralgia treatment plans utilizing HD‐MLC.

Verification of dose profiles generated by the convolution algorithm of the gamma knife® radiosurgery planning system

PURPOSE

A convolution algorithm that takes into account electron-density inhomogeneity was recently introduced to calculate dose distributions for the Gamma Knife (GK) Perfexion™ treatment planning program. The accuracies of the dose distributions computed using the convolution method were assessed using an anthropomorphic phantom and film dosimetry.

METHODS

Absorbed-dose distributions inside a phantom (CIRS Radiosurgery Head Phantom, Model 605) were calculated using the convolution method of the GK treatment-planning software (Leksell Gamma Plan^® version 10.1; LGP) for various combinations of collimator size, location, direction of calculation plane, and number of shots. Computed tomography (CT) images of the phantom and a data set of CT number versus electron density were provided to the LGP. Calculated distributions were exported as digital-image communications in medicine-radiation therapy (DICOM-RT) files. Three types of radiochromic film (GafChromic^® MD-V2-55, MD-V3, and EBT2) were irradiated inside the phantom using GK Perfexion™. Scanned images of the measured films were processed following standard radiochromic film-handling procedures. For a two-dimensional quantitative evaluation, gamma index pass rates (GIPRs) and normalized agreement-test indices (NATIs) were obtained. Image handling and index calculations were performed using a commercial software package (DoseLab Pro version 6.80).

RESULTS

The film-dose calibration data were well fitted with third-order polynomials (R²  ≥ 0.9993). The mean GIPR and NATI of the 93 analyzed films were 99.3 ± 1.1% and 0.8 ± 1.3, respectively, using 3%/1.0 mm criteria. The calculated maximum doses were 4.3 ± 1.7% higher than the measured values for the 4 mm single shots and 1.8 ± 0.7% greater than those for the 8 mm single shots, whereas differences of only 0.3 ± 0.9% were observed for the 16 mm single shots. The accuracy of the calculated distribution was not statistically related to the collimator size, number of shots, or centrality of location (P > 0.05, independent-sample t-test). The plans in the axial planes exhibited poorer agreement with the measured distributions than the plans in the coronal or sagittal planes; however, their GIPR values (≥ 96.9%) were clinically acceptable. The plans for an arbitrary virtual target of volume 1.6 cm³ at an axial plane close to the top of the phantom showed the worst agreement and the greatest fluctuation (GIPR = 96.9 ± 1.2%, NATI = 3.9 ± 1.7).

CONCLUSIONS

The measured accuracies of the dose distributions calculated by the convolution algorithm of the LGP were within the clinically acceptable range (GIPR ≥ 96.9%) for various configurations of collimator size, location, direction of calculation plane, and number of shots. Due to the intrinsic asymmetry in the dose distribution along the z-axis, the treatment plan should also be verified in coronal or sagittal plane.

Tuning of AcurosXB source size setting for small intracranial targets

This study details a method to evaluate the source size selection for small field intracranial stereotactic radiosurgery (SRS) deliveries in Eclipse treatment planning system (TPS) for AcurosXB dose calculation algorithm. Our method uses end‐to‐end dosimetric data to evaluate a total of five source size selections (0.50 mm, 0.75 mm, 1.00 mm, 1.25 mm, and 1.50 mm). The dosimetric leaf gap (DLG) was varied in this analysis (three DLG values were tested for each scenario). We also tested two MLC leaf designs (standard and high‐definition MLC) and two delivery types for intracranial SRS (volumetric modulated arc therapy [VMAT] and dynamic conformal arc [DCA]). Thus, a total of 10 VMAT plans and 10 DCA plans were tested for each machine type (TrueBeam [standard MLC] and Edge [high‐definition MLC]). Each plan was mapped to a solid water phantom and dose was calculated with each iteration of source size and DLG value (15 total dose calculations for each plan). To measure the dose, Gafchromic film was placed in the coronal plane of the solid water phantom at isocenter. The phantom was localized via on‐board CBCT and the plans were delivered at planned gantry, collimator, and couch angles. The planned and measured film dose was compared using Gamma (3.0%, 0.3 mm) criteria. The vendor‐recommended 1.00 mm source size was suitable for TrueBeam planning (both VMAT and DCA planning) and Edge DCA planning. However, for Edge VMAT planning, the 0.50 mm source size yielded the highest passing rates. The difference in dose calculation among the source size variations manifested primarily in two regions of the dose calculation: (1) the shoulder of the high‐dose region, and (2) for small targets (volume ≤ 0.30 cc), in the central portion of the high‐dose region. Selection of a larger than optimal source size can result in increased blurring of the shoulder for all target volume sizes tested, and can result in central axis dose discrepancies in excess of 10% for target volumes sizes ≤ 0.30 cc. Our results indicate a need for evaluation of the source size when AcurosXB is used to model intracranial SRS delivery, and our methods represent a feasible process for many clinics to perform tuning of the AcurosXB source size parameter.

An effective calibration technique for radiochromic films using a single-shot dose distribution in Gamma Knife(®)

PURPOSE

A method of calibrating radiochromic films for Gamma Knife(®) (GK) dosimetry was developed. The applicability and accuracy of the new method were examined.

METHODS

The dose distribution for a sixteen millimeter single-shot from a GK was built using a reference film that was calibrated using the conventional multi-film calibration (MFC) method. Another film, the test film, from a different set of films was irradiated under the same conditions as the reference film. The calibration curve for the second set of films was obtained by assigning the dose distribution of the reference film to the optical density of the test film, point by point. To assess the accuracy of this single-film calibration (SFC) method, differences between gamma index pass rates (GIPRs) were calculated.

RESULTS

The SFC curves were successfully obtained with estimated errors of 1.46%. GIPRs obtained with the SFC method for films irradiated using a single-shot showed differences less than one percentage point when dose difference criterion (ΔD) was 2% and the distance to agreement criterion (Δd) was 1 mm. The GIPRs of the SFC method when the films were irradiated following a virtual target treatment plan were consistent with the GIPRs of the MFC method, with differences of less than 0.2 percentage points for ΔD = 1% and Δd = 1 mm.

CONCLUSION

The accuracy of the SFC method is comparable to that of conventional multi-film calibration method for GK film dosimetry.

Validation and analysis of dose distributions in a new and entirely redesigned cobalt-60 stereotactic radiosurgery units

The objective of this study was to evaluate the reproducibility of dose distributions in stereotactic treatment planning throughout Gamma Knife (GK) stereotactic radiosurgery (SRS) procedures in both GK model C and Perfexion (PFX). An originally-developed phantom and a radiochromic film were used for obtaining actual dose distributions. The phantom, with inserted films, was placed on a Leksell skull frame. Computed tomography (CT) was then acquired with a stereotactic localizer box attached to the frame, dose planning was made using the Leksell GammaPlan treatment planning system, and the phantom was ended up as beam delivery on an equal with clinical radiosurgery process. The reproducibility of the dose plan was provided by distance to agreement (DTA) values between planned and irradiated dose distributions calculated by dedicated film analysis software. The DTA values were determined for the isodose lines at 30%, 50%, 70%, and 90% of the maximum dose. In our study, the reproducibility of dose distributions in GK PFX was lower than in GK model C. As the results common to both units, the mean values of middle dose area (50% isodose) were about half the values of high (90% isodose) and low (30% isodose) dose area. Therefore validation of dose distributions is absolutely essential in commissioning of GK PFX. In addition, when risk organs are close to the target, dose prescription should be normalized for middle isodose line.

MAGAT gel and EBT2 film-based dosimetry for evaluating source plugging-based treatment plan in Gamma Knife stereotactic radiosurgery

This work illustrates a procedure to assess the overall accuracy associated with Gamma Knife treatment planning using plugging. The main role of source plugging or blocking is to create dose falloff in the junction between a target and a critical structure. We report the use of MAGAT gel dosimeter for verification of an experimental treatment plan based on plugging. The polymer gel contained in a head‐sized glass container simulated all major aspects of the treatment process of Gamma Knife radiosurgery. The 3D dose distribution recorded in the gel dosimeter was read using a 1.5T MRI scanner. Scanning protocol was: CPMG pulse sequence with 8 equidistant echoes, TR=7 s, echo step=14 ms, pixel size=0.5 mm x 0.5 mm, and slice thickness of 2 mm. Using a calibration relationship between absorbed dose and spin‐spin relaxation rate (R2), we converted R2 images to dose images. Volumetric dose comparison between treatment planning system (TPS) and gel measurement was accomplished using an in‐house MATLAB‐based program. The isodose overlay of the measured and computed dose distribution on axial planes was in close agreement. Gamma index analysis of 3D data showed more than 94% voxel pass rate for different tolerance criteria of 3%/2 mm, 3%/1 mm and 2%/2 mm. Film dosimetry with GAFCHROMIC EBT 2 film was also performed to compare the results with the calculated TPS dose. Gamma index analysis of film measurement for the same tolerance criteria used for gel measurement evaluation showed more than 95% voxel pass rate. Verification of gamma plan calculated dose on account of shield is not part of acceptance testing of Leksell Gamma Knife (LGK). Through this study we accomplished a volumetric comparison of dose distributions measured with a polymer gel dosimeter and Leksell GammaPlan (LGP) calculations for plans using plugging. We propose gel dosimeter as a quality assurance (QA) tool for verification of plug‐based planning.

PACS number: 87.53.Ly, 87.55.‐x, 87.56.N‐