Deep learning algorithms have recently been developed that utilize patient anatomy and raw imaging information to predict radiation dose, as a means to increase treatment planning efficiency and ...improve radiotherapy plan quality. Current state-of-the-art techniques rely on convolutional neural networks (CNNs) that use pixel-to-pixel loss to update network parameters. However, stereotactic body radiotherapy (SBRT) dose is often heterogeneous, making it difficult to model using pixel-level loss. Generative adversarial networks (GANs) utilize adversarial learning that incorporates image-level loss and is better suited to learn from heterogeneous labels. However, GANs are difficult to train and rely on compromised architectures to facilitate convergence. This study suggests an attention-gated generative adversarial network (DoseGAN) to improve learning, increase model complexity, and reduce network redundancy by focusing on relevant anatomy. DoseGAN was compared to alternative state-of-the-art dose prediction algorithms using heterogeneity index, conformity index, and various dosimetric parameters. All algorithms were trained, validated, and tested using 141 prostate SBRT patients. DoseGAN was able to predict more realistic volumetric dosimetry compared to all other algorithms and achieved statistically significant improvement compared to all alternative algorithms for the V
and V
of the PTV, V
of the rectum, and heterogeneity index.
The goal of this study is to demonstrate the feasibility of a novel fully-convolutional volumetric dose prediction neural network (DoseNet) and test its performance on a cohort of prostate ...stereotactic body radiotherapy (SBRT) patients. DoseNet is suggested as a superior alternative to U-Net and fully connected distance map-based neural networks for non-coplanar SBRT prostate dose prediction. DoseNet utilizes 3D convolutional downsampling with corresponding 3D deconvolutional upsampling to preserve memory while simultaneously increasing the receptive field of the network. DoseNet was implemented on 2 Nvidia 1080 Ti graphics processing units and utilizes a 3 phase learning protocol to help achieve convergence and improve generalization. DoseNet was trained, validated, and tested with 151 patients following Kaggle completion rules. The dosimetric quality of DoseNet was evaluated by comparing the predicted dose distribution with the clinically approved delivered dose distribution in terms of conformity index, heterogeneity index, and various clinically relevant dosimetric parameters. The results indicate that the DoseNet algorithm is a superior alternative to U-Net and fully connected methods for prostate SBRT patients. DoseNet required ~50.1 h to train, and ~0.83 s to make a prediction on a 128 × 128 × 64 voxel image. In conclusion, DoseNet is capable of making accurate volumetric dose predictions for non-coplanar SBRT prostate patients, while simultaneously preserving computational efficiency.
Purpose
Megavoltage computed tomography (MVCT) has been implemented on many radiation therapy treatment machines as a tomographic imaging modality that allows for three‐dimensional visualization and ...localization of patient anatomy. Yet MVCT images exhibit lower contrast and greater noise than its kilovoltage CT (kVCT) counterpart. In this work, we sought to improve these disadvantages of MVCT images through an image‐to‐image‐based machine learning transformation of MVCT and kVCT images. We demonstrated that by learning the style of kVCT images, MVCT images can be converted into high‐quality synthetic kVCT (skVCT) images with higher contrast and lower noise, when compared to the original MVCT.
Methods
Kilovoltage CT and MVCT images of 120 head and neck (H&N) cancer patients treated on an Accuray TomoHD system were retrospectively analyzed in this study. A cycle‐consistent generative adversarial network (CycleGAN) machine learning, a variant of the generative adversarial network (GAN), was used to learn Hounsfield Unit (HU) transformations from MVCT to kVCT images, creating skVCT images. A formal mathematical proof is given describing the interplay between function sensitivity and input noise and how it applies to the error variance of a high‐capacity function trained with noisy input data. Finally, we show how skVCT shares distributional similarity to kVCT for various macro‐structures found in the body.
Results
Signal‐to‐noise ratio (SNR) and contrast‐to‐noise ratio (CNR) were improved in skVCT images relative to the original MVCT images and were consistent with kVCT images. Specifically, skVCT CNR for muscle‐fat, bone‐fat, and bone‐muscle improved to 14.8 ± 0.4, 122.7 ± 22.6, and 107.9 ± 22.4 compared with 1.6 ± 0.3, 7.6 ± 1.9, and 6.0 ± 1.7, respectively, in the original MVCT images and was more consistent with kVCT CNR values of 15.2 ± 0.8, 124.9 ± 27.0, and 109.7 ± 26.5, respectively. Noise was significantly reduced in skVCT images with SNR values improving by roughly an order of magnitude and consistent with kVCT SNR values. Axial slice mean (S‐ME) and mean absolute error (S‐MAE) agreement between kVCT and MVCT/skVCT improved, on average, from −16.0 and 109.1 HU to 8.4 and 76.9 HU, respectively.
Conclusions
A kVCT‐like qualitative aid was generated from input MVCT data through a CycleGAN instance. This qualitative aid, skVCT, was robust toward embedded metallic material, dramatically improves HU alignment from MVCT, and appears perceptually similar to kVCT with SNR and CNR values equivalent to that of kVCT images.
Purpose
This study aims to reduce the delivery time of CyberKnife m6 treatments by allowing for noncoplanar continuous arc delivery. To achieve this, a novel noncoplanar continuous arc delivery ...optimization algorithm was developed for the CyberKnife m6 treatment system (CyberArc‐m6).
Methods and Materials
CyberArc‐m6 uses a five‐step overarching strategy, in which an initial set of beam geometries is determined, the robotic delivery path is calculated, direct aperture optimization is conducted, intermediate MLC configurations are extracted, and the final beam weights are computed for the continuous arc radiation source model. This algorithm was implemented on five prostate and three brain patients, previously planned using a conventional step‐and‐shoot CyberKnife m6 delivery technique. The dosimetric quality of the CyberArc‐m6 plans was assessed using locally confined mutual information (LCMI), conformity index (CI), heterogeneity index (HI), and a variety of common clinical dosimetric objectives.
Results
Using conservative optimization tuning parameters, CyberArc‐m6 plans were able to achieve an average CI difference of 0.036 ± 0.025, an average HI difference of 0.046 ± 0.038, and an average LCMI of 0.920 ± 0.030 compared with the original CyberKnife m6 plans. Including a 5 s per minute image alignment time and a 5‐min setup time, conservative CyberArc‐m6 plans achieved an average treatment delivery speed up of 1.545x ± 0.305x compared with step‐and‐shoot plans.
Conclusions
The CyberArc‐m6 algorithm was able to achieve dosimetrically similar plans compared to their step‐and‐shoot CyberKnife m6 counterparts, while simultaneously reducing treatment delivery times.
Stereotactic body radiotherapy for prostate cancer is rapidly growing in popularity. Stereotactic body radiotherapy plans mimic those of high-dose rate brachytherapy, with tight margins and ...inhomogeneous dose distributions. The impact of interfraction anatomical changes on the dose received by organs at risk under these conditions has not been well documented. To estimate anatomical variation during stereotactic body radiotherapy, 10 patients were identified who received a prostate boost using robotic stereotactic body radiotherapy after completing 25 fractions of pelvic radiotherapy with daily megavoltage computed tomography. Rectal and bladder volumes were delineated on each megavoltage computed tomography, and the stereotactic body radiotherapy boost plan was registered to each megavoltage computed tomography image using a point-based rigid registration with 3 fiducial markers placed in the prostate. The volume of rectum and bladder receiving 75% of the prescription dose (V75%) was measured for each megavoltage computed tomography. The rectal V75% from the daily megavoltage computed tomographies was significantly greater than the planned V75% (median increase of 0.93 cm3, P < .001), whereas the bladder V75% on megavoltage computed tomography was not significantly changed (median decrease of −0.12 cm3, P = .57). Although daily prostate rotation was significantly correlated with bladder V75% (Spearman ρ = .21, P = .023), there was no association between rotation and rectal V75% or between prostate deformation and either rectal or bladder V75%. Planning organ-at-risk volume-based replanning techniques using either a 6-mm isotropic expansion of the plan rectal contour or a 1-cm expansion from the planning target volume in the superior and posterior directions demonstrated significantly improved rectal V75% on daily megavoltage computed tomographies compared to the original stereotactic body radiotherapy plan, without compromising plan quality. Thus, despite tight margins and full translational and rotational corrections provided by robotic stereotactic body radiotherapy, we find that interfraction anatomical variations can lead to a substantial increase in delivered rectal doses during prostate stereotactic body radiotherapy. A planning organ-at-risk volume-based approach to treatment planning may help mitigate the impact of daily organ motion and reduce the risk of rectal toxicity.
High dose rate (HDR) brachytherapy has been established as an excellent monotherapy or after external-beam radiotherapy (EBRT) boost treatment for prostate cancer (PCa). Recently, dosimetric studies ...have demonstrated the potential for achieving similar dosimetry with stereotactic body radiotherapy (SBRT) compared with HDR brachytherapy. Here, we report our technique, PSA nadir, and acute and late toxicity with SBRT as monotherapy and post-EBRT boost for PCa using HDR brachytherapy fractionation.
To date, 38 patients have been treated with SBRT at the University of California-San Francisco with a minimum follow-up of 12 months. Twenty of 38 patients were treated with SBRT monotherapy (9.5 Gy × 4 fractions), and 18 were treated with SBRT boost (9.5 Gy × 2 fractions) post-EBRT and androgen deprivation therapy. PSA nadir to date for 44 HDR brachytherapy boost patients with disease characteristics similar to the SBRT boost cohort was also analyzed as a descriptive comparison.
SBRT was well tolerated. With a median follow-up of 18.3 months (range, 12.6-43.5), 42% and 11% of patients had acute Grade 2 gastrourinary and gastrointestinal toxicity, respectively, with no Grade 3 or higher acute toxicity to date. Two patients experienced late Grade 3 GU toxicity. All patients are without evidence of biochemical or clinical progression to date, and favorably low PSA nadirs have been observed with a current median PSA nadir of 0.35 ng/mL (range, <0.01-2.1) for all patients (0.47 ng/mL, range, 0.2-2.1 for the monotherapy cohort; 0.10 ng/mL, range, 0.01-0.5 for the boost cohort). With a median follow-up of 48.6 months (range, 16.4-87.8), the comparable HDR brachytherapy boost cohort has achieved a median PSA nadir of 0.09 ng/mL (range, 0.0-3.3).
Early results with SBRT monotherapy and post-EBRT boost for PCa demonstrate acceptable PSA response and minimal toxicity. PSA nadir with SBRT boost appears comparable to those achieved with HDR brachytherapy boost.
The treatment planning quality between nonisocentric CyberKnife (CK) and isocentric intensity modulation treatment was studied for hypofractionated prostate body radiotherapy. In particular, the dose ...gradient across the target and the critical structures such as the rectum and bladder was characterized.
In the present study, patients treated with CK underwent repeat planning for nine fixed-field intensity-modulated radiotherapy (IMRT) using identical contour sets and dose-volume constraints. To calculate the dose falloff, the clinical target volume contours were expanded 30 mm anteriorly and posteriorly and 50 mm uniformly in other directions for all patients in the CK and IMRT plans.
We found that all the plans satisfied the dose-volume constraints, with the CK plans showing significantly better conformity than the IMRT plans at a relative greater dose inhomogeneity. The rectal and bladder volumes receiving a low dose were also lower for CK than for IMRT. The average conformity index, the ratio of the prescription isodose volume and clinical target volume, was 1.18 +/- 0.08 for the CK plans vs. 1.44 +/- 0.11 for the IMRT plans. The average homogeneity index, the ratio of the maximal dose and the prescribed dose to the clinical target volume, was 1.45 +/- 0.12 for the CK plans vs. 1.28 +/- 0.06 for the IMRT plans. The average percentage of dose falloff was 2.9% +/- 0.8%/mm for CK and 3.1% +/- 1.0%/mm for IMRT in the anterior direction, 3.8% +/- 1.6%/mm for CK and 3.2% +/- 1.9%/mm for IMRT in the posterior direction, and 3.6% +/- 0.4% for CK and 3.6% +/- 0.4% for IMRT in all directions.
Nonisocentric CK was as capable of producing equivalent fast dose falloff as high-number fixed-field IMRT delivery.
To investigate whether dose fall-off characteristics would be significantly different among intracranial radiosurgery modalities and the influence of these characteristics on fractionation schemes in ...terms of normal tissue sparing.
An analytic model was developed to measure dose fall-off characteristics near the target independent of treatment modalities. Variations in the peripheral dose fall-off characteristics were then examined and compared for intracranial tumors treated with Gamma Knife, Cyberknife, or Novalis LINAC-based system. Equivalent uniform biologic effective dose (EUBED) for the normal brain tissue was calculated. Functional dependence of the normal brain EUBED on varying numbers of fractions (1 to 30) was studied for the three modalities.
The derived model fitted remarkably well for all the cases (R(2) > 0.99). No statistically significant differences in the dose fall-off relationships were found between the three modalities. Based on the extent of variations in the dose fall-off curves, normal brain EUBED was found to decrease with increasing number of fractions for the targets, with alpha/beta ranging from 10 to 20. This decrease was most pronounced for hypofractionated treatments with fewer than 10 fractions. Additionally, EUBED was found to increase slightly with increasing number of fractions for targets with alpha/beta ranging from 2 to 5.
Nearly identical dose fall-off characteristics were found for the Gamma Knife, Cyberknife, and Novalis systems. Based on EUBED calculations, normal brain sparing was found to favor hypofractionated treatments for fast-growing tumors with alpha/beta ranging from 10 to 20 and single fraction treatment for abnormal tissues with low alpha/beta values such as alpha/beta = 2.
To characterize nonrandom intrafraction target motions for spine stereotactic body radiotherapy and to develop a method of correction via image guidance. The dependence of target motions, as well as ...the effectiveness of the correction strategy for lesions of different locations within the spine, was analyzed.
Intrafraction target motions for 64 targets in 64 patients treated with a total of 233 fractions were analyzed. Based on the target location, the cases were divided into three groups, i.e., cervical (n = 20 patients), thoracic (n = 20 patients), or lumbar-sacrum (n = 24 patients) lesions. For each case, time-lag autocorrelation analysis was performed for each degree of freedom of motion that included both translations (x, y, and z shifts) and rotations (roll, yaw, and pitch). A general correction strategy based on periodic interventions was derived to determine the time interval required between two adjacent interventions, to overcome the patient-specific target motions.
Nonrandom target motions were detected for 100% of cases regardless of target locations. Cervical spine targets were found to possess the highest incidence of nonrandom target motion compared with thoracic and lumbar-sacral lesions (p < 0.001). The average time needed to maintain the target motion to within 1 mm of translation or 1 degrees of rotational deviation was 5.5 min, 5.9 min, and 7.1 min for cervical, thoracic, and lumbar-sacrum locations, respectively (at 95% confidence level).
A high incidence of nonrandom intrafraction target motions was found for spine stereotactic body radiotherapy treatments. Periodic interventions at approximately every 5 minutes or less were needed to overcome such motions.