Multileaf Collimator Leaf Positions Research Articles

A novel EPID-based MLC QA method with log files achieving submillimeter accuracy.

The purpose of this study is to develop an electronic portal imaging device-based multi-leaf collimator calibration procedure using log files. Picket fence fields with 2-14mm nominal strip widths were performed and normalized by open field. Normalized pixel intensity profiles along the direction of leaf motion for each leaf pair were taken. Three independent algorithms and an integration method derived from them were developed according to the valley value, valley area, full-width half-maximum (FWHM) of the profile, and the abutment width of the leaf pairs obtained from the log files. Three data processing schemes (Scheme A, Scheme B, and Scheme C) were performed based on different data processing methods. To test the usefulness and robustness of the algorithm, the known leaf position errors along the direction of perpendicular leaf motion via the treatment planning system were introduced in the picket fence field with nominal 5, 8, and 11mm. Algorithm tests were performed every 2weeks over 4 months. According to the log files, about 17.628% and 1.060% of the leaves had position errors beyond ±0.1 and ±0.2mm, respectively. The absolute position errors of the algorithm tests for different data schemes were 0.062±0.067 (Scheme A), 0.041±0.045 (Scheme B), and 0.037±0.043 (Scheme C). The absolute position errors of the algorithms developed by Scheme C were 0.054±0.063 (valley depth method), 0.040±0.038 (valley area method), 0.031±0.031 (FWHM method), and 0.021±0.024 (integrated method). For the efficiency and robustness test of the algorithm, the absolute position errors of the integration method of Scheme C were 0.020±0.024 (5mm), 0.024±0.026 (8mm), and 0.018±0.024 (11mm). Different data processing schemes could affect the accuracy of the developed algorithms. The integration method could integrate the benefits of each algorithm, which improved the level of robustness and accuracy of the algorithm. The integration method can perform multi-leaf collimator (MLC) quality assurance with an accuracy of 0.1mm. This method is simple, effective, robust, quantitative, and can detect a wide range of MLC leaf position errors.

Error detection using a multi-channel hybrid network with a low-resolution detector in patient-specific quality assurance.

This study aimed to develop a hybrid multi-channel network to detect multileaf collimator (MLC) positional errors using dose difference (DD) maps and gamma maps generated from low-resolution detectors in patient-specific quality assurance (QA) for Intensity Modulated Radiation Therapy (IMRT). A total of 68 plans with 358 beams of IMRT were included in this study. The MLC leaf positions of all control points in the original IMRT plans were modified to simulate four types of errors: shift error, opening error, closing error, and random error. These modified plans were imported into the treatment planning system (TPS) to calculate the predicted dose, while the PTW seven29 phantom was utilized to obtain the measured dose distributions. Based on the measured and predicted dose, DD maps and gamma maps, both with and without errors, were generated, resulting in a dataset with 3222 samples. The network's performance was evaluated using various metrics, including accuracy, sensitivity, specificity, precision, F1-score, ROC curves, and normalized confusion matrix. Besides, other baseline methods, such as single-channel hybrid network, ResNet-18, and Swin-Transformer, were also evaluated as a comparison. The experimental results showed that the multi-channel hybrid network outperformed other methods, demonstrating higher average precision, accuracy, sensitivity, specificity, and F1-scores, with values of 0.87, 0.89, 0.85, 0.97, and 0.85, respectively. The multi-channel hybrid network also achieved higher AUC values in the random errors (0.964) and the error-free (0.946) categories. Although the average accuracy of the multi-channel hybrid network was only marginally better than that of ResNet-18 and Swin Transformer, it significantly outperformed them regarding precision in the error-free category. The proposed multi-channel hybrid network exhibits a high level of accuracy in identifying MLC errors using low-resolution detectors. The method offers an effective and reliable solution for promoting quality and safety of IMRT QA.

Improvement of deep learning prediction model in patient-specific QA for VMAT with MLC leaf position map and patient's dose distribution.

Deep learning-based virtual patient-specific quality assurance (QA) is a novel technique that enables patient QA without measurement. However, this method could be improved by further evaluating the optimal data to be used as input. Therefore, a deep learning-based model that uses multileaf collimator (MLC) information per control point and dose distribution in patient's CT as inputs was developed. Overall, 96 volumetric-modulated arc therapy plans generated for prostate cancer treatment were used. We developed a model (Model 1) that can predict measurement-based gamma passing rate (GPR) for a treatment plan using data stored as a map reflecting the MLC leaf position at each control point (MLPM) and data of the dose distribution in patient's CT as inputs. The evaluation of the model was based on the mean absolute error (MAE) and Pearson's correlation coefficient (r) between the measured and predicted GPR. For comparison, we also analyzed models trained with the dose distribution in patient's CT alone (Model 2) and with dose distributions recalculated on a virtual phantom CT (Model 3). At the 2%/2mm criterion, MAE[%] and r for Model 1, Model 2, and Model 3 were 2.32%±0.43% and 0.54±0.03, 2.70%±0.26%, and 0.32±0.08, and 2.96%±0.23% and 0.24±0.22, respectively; at the 3%/3mm criterion, these values were 1.25%±0.05% and 0.36±0.18, 1.57%±0.35% and 0.19±0.20, and 1.39%±0.32% and 0.17±0.22, respectively. This result showed that Model 1 exhibited the lowest MAE and highest r at both criteria of 2%/2mm and 3%3mm. These findings showed that a model that combines the MLPM and dose distribution in patient's CT exhibited a better GPR prediction performance compared with the other two studied models.

Patient-specific quality assurance failure prediction with deep tabular models

Purpose. Patient-specific quality assurance (PSQA) failures in radiotherapy can cause a delay in patient care and increase the workload and stress of staff. We developed a tabular transformer model based directly on the multi-leaf collimator (MLC) leaf positions (without any feature engineering) to predict IMRT PSQA failure in advance. This neural model provides an end-to-end differentiable map from MLC leaf positions to the probability of PSQA plan failure, which could be useful for regularizing gradient-based leaf sequencing optimization algorithms and generating a plan that is more likely to pass PSQA. Method. We retrospectively collected DICOM RT PLAN files of 968 patient plans treated with volumetric arc therapy. We constructed a beam-level tabular dataset with 1873 beams as samples and MLC leaf positions as features. We trained an attention-based neural network FT-Transformer to predict the ArcCheck-based PSQA gamma pass rates. In addition to the regression task, we evaluated the model in the binary classification context predicting the pass or fail of PSQA. The performance was compared to the results of the two leading tree ensemble methods (CatBoost and XGBoost) and a non-learned method based on mean-MLC-gap. Results. The FT-Transformer model achieves 1.44% Mean Absolute Error (MAE) in the regression task of the gamma pass rate prediction and performs on par with XGBoost (1.53 % MAE) and CatBoost (1.40 % MAE). In the binary classification task of PSQA failure prediction, FT-Transformer achieves 0.85 ROC AUC (compared to the mean-MLC-gap complexity metric achieving 0.72 ROC AUC). Moreover, FT-Transformer, CatBoost, and XGBoost all achieve 80% true positive rate while keeping the false positive rate under 20%. Conclusions. We demonstrated that reliable PSQA failure predictors can be successfully developed based solely on MLC leaf positions. FT-Transformer offers an unprecedented benefit of providing an end-to-end differentiable map from MLC leaf positions to the probability of PSQA failure.

Radiotherapy delivery error detection with electronic portal imaging device (EPID)-based in vivo dosimetry.

Radiotherapy delivery error detection with electronic portal imaging device (EPID)-based in vivo dosimetry.

Monte Carlo based continuous aperture optimization for VMAT on conventional and MR-linacs: A proof of concept.

Currently, the commercial treatment planning systems for magnetic-resonance guided linear accelerators (MR-linacs) only support step-and-shoot intensity-modulated radiation therapy (IMRT). However, recent studies have shown the feasibility of delivering arc therapy on MR-linacs, which is expected to improve dose distributions and delivery speed. By accurately accounting for the electron return effect in the presence of a magnetic field, a Monte Carlo (MC) algorithm is ideally suited for the inverse treatment planning of thistechnique. We propose a novel MC-based continuous aperture optimization (MCCAO) algorithm for volumetric modulated arc therapy (VMAT), including applications to VMAT on MR-linacs and trajectory-based VMAT. A unique feature of MCCAO is that the continuous character of gantry rotation and multileaf collimator (MLC) motion is accounted for at every stage of theoptimization. The optimization process uses a multistage simulation of 4D dose distribution. A phase space is scored at the top surface of the MLC and the energy deposition of each particle history is mapped to its position in this phase space. A progressive sampling method is used, where both MLC leaf positions and monitor unit (MU) weights are randomly changed, while respecting the linac mechanical limits. Due to the continuous nature of the leaf motion, such changes affect not only a single control point, but propagate to the adjacent ones as well, and the corresponding dose distribution changes are accounted for. A dose-volume cost function is used, which includes the MC statisticaluncertainty. We applied our optimization technique to various treatment sites, using standard and flattening-filter-free (FFF) 6 MV beam models, with and without a 1.5 T magnetic field. MCCAO generates deliverable plans, whose dose distributions are in good agreement with measurements on ArcCHECK and stereotactic radiosurgery End-To-EndPhantom. We show that the novel MCCAO method generates VMAT plans that meet clinical objectives for both conventional andMR-linacs.

Complex field verification using a large area CMOS MAPS upstream in radiotherapy

A multileaf collimator (MLC) is an integral component in modern radiotherapy machines as it dynamically shapes the photon field used for patient treatment. Currently, the MLC leaves which collimate the treatment field are mechanically calibrated to ±1 mm every 3 months and during pre-treatment calibration are calibrated to the mechanically set leaf positions. Leaf drift can occur between calibration dates and hence exceed the ±1 mm tolerance. Pre-treatment verification, increases LINAC usage time so is seldom performed for each individual patient treatment, but instead for an acceptable sample of patients and/or treatment fractions. Independent real-time treatment verification is therefore desirable. We are developing a large area CMOS MAPS upstream of the patient to monitor MLC leaf positions for real-time treatment verification. CMOS MAPS are radiation hard for photon and electron irradiation, have high readout speeds and low attenuation which makes them an ideal upstream radiation detector for radiotherapy. Previously, we reported on leaf position reconstruction for single leaves using the Lassena, a 12 × 14 cm2, three side buttable MAPS suitable for clinical deployment. Sobel operator based methods were used for edge reconstruction. It was shown that the correspondence between reconstructed and set leaf position was excellent and resolutions ranged between 60.6 ± 8 and 109 ± 12 μm for a single central leaf with leaf extensions ranging from 1 to 35 mm using 0.3 sec of treatment beam time at 400 MU/min. Here, we report on leaf edge reconstruction using updated methods for complex leaf configurations, as occur in clinical use. Results show that leaf positions can be reconstructed with resolutions of 62 ± 6 μm for single leaves and 86 ± 16 μm for adjacent leaves at the isocenter using 0.15 sec at 400 MU/min of treatment beam. These resolutions are significantly better than current calibration standards.

Is a weekly qualitative picket fence test sufficient? A proposed alternate EPID-based weekly MLC QA program.

PurposeWell‐designed routine multileaf collimator (MLC) quality assurance (QA) is important to assure external‐beam radiation treatment delivery accuracy. This study evaluates the clinical necessity of a comprehensive weekly (C‐Weekly) MLC QA program compared to the American Association of Physics in Medicinerecommended weekly picket fence test (PF‐Weekly), based on our seven‐year experience with weekly MLC QA.MethodsThe C‐Weekly MLC QA program used in this study includes 5 tests to analyze: (1) absolute MLC leaf position; (2) interdigitation MLC leaf position; (3) picket fence MLC leaf positions at static gantry angle; (4) minimum leaf‐gap setting; and (5) volumetric‐modulated arc therapy delivery. A total of 20,226 QA images from 16,855 tests (3,371 tests × 5) for 11 linacs at 5 photon clinical sites from May 2014 to June 2021 were analyzed. Failure mode and effects analysis was performed with 5 failure modes related to the 5 tests. For each failure mode, a risk probability number (RPN) was calculated for a C‐Weekly and a PF‐Weekly MLC QA program. The probability of occurrence was evaluated from statistical analyses of the C‐Weekly MLC QA.ResultsThe total number of failures for these 16,855 tests was 143 (0.9%): 39 (27.3%) for absolute MLC leaf position, 13 (9.1%) for interdigitation position, 9 (6.3%) for static gantry picket fence, 2 (1.4%) for minimum leaf‐gap setting, and 80 (55.9%) for VMAT delivery. RPN scores for PF‐Weekly MLC QA ranged from 60 to 192 and from 48 to 96 for C‐Weekly MLC QA.ConclusionRPNs for the 5 failure modes of MLC QA tests were quantitatively determined and analyzed. A comprehensive weekly MLC QA is imperative to lower the RPNs of the 5 failure modes to the desired level (<125); those from the PF‐Weekly MLC QA program were found to be higher (>125). This supports the clinical necessity for comprehensive weekly MLC QA.

Insensitivity of machine log files to MLC leaf backlash and effect of MLC backlash on clinical dynamic MLC motion: An experimental investigation.

PurposeMulti‐leaf‐collimator (MLC) leaf position accuracy is important for accurate dynamic radiotherapy treatment plan delivery. Machine log files have become widely utilized for quality assurance (QA) of such dynamic treatments. The primary aim is to test the sensitivity of machine log files in comparison to electronic portal imaging device (EPID)‐based measurements to MLC position errors caused by leaf backlash. The secondary aim is to investigate the effect of MLC leaf backlash on MLC leaf motion during clinical dynamic plan delivery.MethodsThe sensitivity of machine log files and two EPID‐based measurements were assessed via a controlled experiment, whereby the length of the “T” section of a series of 12 MLC leaf T‐nuts in a Varian Millennium MLC for a Trilogy C‐series type linac was reduced by sandpapering the top of the “T” to introduce backlash. The built‐in machine MLC leaf backlash test as well as measurements for two EPID‐based dynamic MLC positional tests along with log files were recorded pre‐ and post‐T‐nut modification. All methods were investigated for sensitivity to the T‐nut change by assessing the effect on measured MLC leaf positions. A reduced version of the experiment was repeated on a TrueBeam type linac with Millennium MLC.ResultsNo significant differences before and after T‐nut modification were detected in any of the log file data. Both EPID methods demonstrated sensitivity to the introduced change at approximately the expected magnitude with a strong dependence observed with gantry angle. EPID‐based data showed MLC positional error in agreement with the micrometer measured T‐nut length change to 0.07 ± 0.05 mm (1 SD) using the departmental routine QA test. Backlash results were consistent between linac types.ConclusionMachine log files appear insensitive to MLC position errors caused by MLC leaf backlash introduced via the T‐nut. The effect of backlash on clinical MLC motions is heavily gantry angle dependent.

Development and Evaluation of Jaw Position Detection Method in Jaw-tracking Delivery Using an Electronic Portal Imaging Device Cine Mode.

To develop a method for detecting jaw positions during jaw-tracking delivery to ensure an accurate delivery of radiation to patients using an electronic portal imaging device (EPID) in the cine mode on a linear accelerator for radiotherapy. A bidirectional picket fence (BPF) plan was used in a novel application to detect jaw positions during jaw-tracking delivery. In the BPF plan, jaws tracked multileaf collimator (MLC) apertures. The irradiated patterns were acquired by an EPID in the cine mode. The upper- and lower-half leaves in the MLC moved in opposite directions to facilitate detection of jaw positions on EPID images. A picket-fence-like image was created by summing all acquired cine images and evaluated to detect MLC leaf positions. Jaw positions determined on the cine images were compared with those expected from the delivered BPF plan. The absolute differences (average ± 1 standard deviation) were 0.16 ± 0.19 mm for the X1 jaw and 0.11 ± 0.16 mm for the X2 jaw. The maximum error in the MLC leaf positions detected in the picket-fence-like pattern were 0.11 mm. Jaw positions during jaw-tracking delivery were identified using the cine EPID images and could be determined within an accuracy better than 0.5 mm. The BPF plan is also available as a picket fence test and can determine the MLC leaf positions to an accuracy better than 0.5 mm.

Assessment of the Sun Nuclear ArcCHECK to detect errors in 6MV FFF VMAT delivery of brain SABR using ROC analysis.

Institutions use a range of different detector systems for patient‐specific quality assurance (QA) measurements conducted to assure that the dose delivered by a patient’s radiotherapy treatment plan matches the calculated dose distribution. However, the ability of different detectors to detect errors from different sources is often unreported. This study contains a systematic evaluation of Sun Nuclear’s ArcCHECK in terms of the detectability of potential machine‐related treatment errors. The five investigated sources of error were multileaf collimator (MLC) leaf positions, gantry angle, collimator angle, jaw positions, and dose output. The study encompassed the clinical treatment plans of 29 brain cancer patients who received stereotactic ablative radiotherapy (SABR). Six error magnitudes were investigated per source of error. In addition, the Eclipse AAA beam model dosimetric leaf gap (DLG) parameter was varied with four error magnitudes. Error detectability was determined based on the area under the receiver operating characteristic (ROC) curve (AUC). Detectability of DLG errors was good or excellent (AUC >0.8) at an error magnitude of at least ±0.4 mm, while MLC leaf position and gantry angle errors reached good or excellent detectability at error magnitudes of at least 1.0 mm and 0.6°, respectively. Ideal thresholds, that is, gamma passing rates, to maximize sensitivity and specificity ranged from 79.1% to 98.7%. The detectability of collimator angle, jaw position, and dose output errors was poor for all investigated error magnitudes, with an AUC between 0.5 and 0.6. The ArcCHECK device’s ability to detect errors from treatment machine‐related sources was evaluated, and ideal gamma passing rate thresholds were determined for each source of error. The ArcCHECK was able to detect errors in DLG value, MLC leaf positions, and gantry angle. The ArcCHECK was unable to detect the studied errors in collimator angle, jaw positions, and dose output.

Effect of Multileaf Collimator Leaf Position Error Determined by Picket Fence Test on Gamma Index Value in Patient-Specific Quality Assurance of Volumetric-Modulated Arc Therapy Plans.

AimThe correlation between the MLC QA (IBA Dosimetry, Germany) results of the picket fence test created with intentional errors and the patient's quality assurance (QA) evaluation was investigated to assess the impact of multileaf collimator (MLC) positioning error on patient QA.Materials and methodsThe picket fence, including error-free and intentional MLC errors, defined in Bank In, Bank Out, and Bank Both were analyzed using MLC QA. The QA of 15 plans consisting of stereotactic radiosurgery (SRS), stereotactic body radiotherapy (SBRT), and conventionally fractionated volumetric-modulated arc therapy (VMAT) acquired with electronic portal imaging devices (EPID) was evaluated in the presence of error-free and MLC errors. The QA of plans were analyzed with 2%/2 mm and 3%/3 mm criteria.ResultsThe passing rates of the picket fence test were 97%, 92%, 91%, and 87% for error-free and intentional errors. The criterion of 3%/3 mm wasn’t able to detect an MLC error for either SRS/SBRT or conventionally fractionated VMAT. The criterion of 2%/2mm was more sensitive to detect MLC error for the conventionally fractionated VMAT than SRS/SBRT. While only two of SBRT plans had <90%, four of conventionally fractionated VMAT plans had a <90% passing rate.ConclusionWe found that the systematic MLC positioning errors defined with picket fence have a smaller but measurable impact on SRS/SBRT than the VMAT plan for a conventionally fractionated and relatively complex plan such as head and neck and endometrium cases.

An error detection method for real-time EPID-based treatment delivery quality assurance.

To quantify the error detection power of a new treatment delivery error detection method. The method validates monitor unit (MU) resolved beam apertures using real-time EPID images. The on-board EPID imager was used to measure cine-EPID (~10Hz) images for 27 beams from 15 VMAT/SBRT clinical treatment plans and five nonclinical plans. For each frame acquisition, planned apertures were interpolated from the treatment plan multileaf collimator (MLC) positions expected during the frame acquisition interval. Inaccurate deliveries were identified by monitoring in-aperture missed fluence and out-of-aperture excess fluence beyond a specified buffer. Delivery errors were simulated by perturbing the planned MLC positions before comparison with nonperturbed measured apertures. Systematic 1-5mm MLC leaf shifts were used to train a logistic regression model to determine the error detection threshold. Model accuracy was monitored using tenfold cross-validation. The model's error detection ability was tested with other error modes: plan control point (CP) weight perturbations, collimator rotations, random MLC leaf position errors, EPID imager shift, and stuck MLC leaf. The error detection accuracy was evaluated using the Matthews correlation coefficient (MCC) and the false positive rate (FPR). Per-beam error thresholds of >1, >5, and >10% errant frames were tested to label per-beam errors. The model also was tested for its ability to distinguish five cases with highly similar plans and compared with gamma analysis. Delivery errors were detected by monitoring intended per-frame images with a 2mm MLC buffer. Frame-by-frame aperture errors were identified with an optimal threshold of 0.3% of the expected aperture area. The per-frame FPR was 0.02%. The MCC was 1.00 (perfect classification) for detection based on 1% of frames for random CP weight shift, 3mm random MLC shifts, 90° and 180° collimator rotations, and an MLC leaf stuck after 10% of the beam delivery. The MCC for 2°, 4°, and 8° collimator rotation were 0.53, 0.76, and 0.96, respectively, for the 1% of beam delivery threshold. The 3mm EPID shift had poor detection, with a minimum MCC of 0.14. The highly similar plans were reliably detected by the aperture check but were not detectable with gamma analysis. The high error detection sensitivity and low FPR makes the aperture check error detection method well suited to pretreatment and during-treatment beam delivery quality assurance (QA). The aperture check detects subtle beam delivery errors, including those resulting from MLC leaf positioning deviations, CP MU shifts, and stuck MLC leaves. Furthermore, the method can distinguish between highly similar treatment plans. Since the aperture check method monitors for the aperture shapes over a given MU interval, it is also sensitive to errors in MU per CP, without requiring dosimetric calibration of the EPID. The aperture check is one part of a Swiss cheese error detection scheme, which provides redundant error testing of multiple error modes, including nonaperture related errors. The rapid error detection, at 1% of a beam's delivery, make the aperture check a potential candidate for QA of on-line adaptive radiotherapy, or other situations in which pretreatment delivery QA is impractical.

Effect of MU-weighted multi-leaf collimator position error on dose distribution of SBRT radiotherapy in peripheral non-small cell lung cancer.

PurposePosition accuracy of the multi‐leaf collimator (MLC) is essential in stereotactic body radiotherapy (SBRT). This study is aimed to investigate the dosimetric impacts of the MU‐weighted MLC positioning uncertainties of SBRT for patients with early stage peripheral non‐small cell lung cancer (NSCLC).MethodsThree types of MLC position error were simulated: Type 1, random error; Type 2, system shift, in which both MLC banks shifted to the left or right direction; and Type 3, in which both MLC banks moved with same magnitudes in the opposite directions. Two baseline plans were generated: an automatic plan (AP) and a manually optimized plan (MP). Multi‐leaf collimator position errors were introduced to generate simulated plans with the preset MLC leaf position errors, which were then reimported into the Pinnacle system to generate simulated plans, respectively. The dosimetric parameters (CI, nCI, GI, etc.) and gEUD values of PTV and OARs were calculated. Linear regression between MU‐weighted/unweighted MLC position error and gEUD was performed to obtain dose sensitivity.ResultsThe dose sensitivities of the PTVs were −4.93, −38.94, −41.70, −55.55, and 30.33 Gy/mm for random, left shift, right shift, system close, and system open MLC errors, respectively. There were significant differences between the MU‐weighted and the unweighted dose sensitivity, which was −38.94 Gy/mm vs −3.42 Gy/mm (left shift), −41.70 Gy/mm vs −3.56 Gy/mm (right shift), −55.55 Gy/mm vs −4.84 Gy/mm (system close), and 30.33 vs 2.64 Gy/mm (system open). For the system open/close MLC errors, as the PTV volume became larger, the dose sensitivity decreased. APs provided smaller dose sensitivity for the system shift and system close MLC errors compared to the conventional MPs.ConclusionsThere was significant difference in dose sensitivity between MU‐weighted and unweighted MLC position error of SBRT radiotherapy in peripheral NSCLC. MU is suggested to be included in the dosimetric evaluation of the MLC misalignments, since it is much closer to clinical radiotherapy.

Verification of the Machine Delivery Parameters of Treatment Plan via Deep Learning

To develop a generative adversarial network (GAN)-based deep learning approach to estimate the multileaf collimator (MLC) aperture and corresponding monitor units (MUs) from a given three-dimensional (3D) dose distribution. We developed a treatment plan verification framework by implementing a modified pix2pix network to learn a mapping between CT, segment 3D dose and MUs/MLC shapes. The proposed design of adversarial network, which integrates a residual block into pix2pix framework, jointly trains a “U-Net”-like architecture as generator and a convolutional “PatchGAN” classifier as discriminator. 199 patients, including nasopharyngeal, lung and rectum, treated with intensity modulated radiotherapy (IMRT) and volumetric modulated arc therapy (VMAT) techniques were utilized to train the network. The resliced volumetric dose of each segment and CT image datasets were labeled by the corresponding MUs/MLC maps. Additional 47 patients were used to test the prediction accuracy of the proposed deep learning model. The Dice similarity coefficient (DSC) was calculated to evaluate the similarity between the MLC aperture shapes obtained from the treatment planning system (TPS) and the deep learning prediction. The average and standard deviation of the bias between the TPS generated MUs and predicted MUs were calculated to evaluate the MU prediction accuracy. Additionally, the differences between TPS and deep learning-predicted MLC leaf positions were compared. The model performance was first evaluated by five simple fields, including four square fields and one irregular field. The DSC is found to be 0.99, 0.99, 0.99, 0.98 and 0.96, respectively. And the MUs prediction agree with the original plans to within 5%. The average and standard deviation of DSC was 0.94 ± 0.043 for 47 testing patients. The average deviation of predicted MUs from the planned MUs normalized to each beam or arc was within 2% for all the testing patients. The average deviation of the predicted MLC leaf positions was around one pixel for all the testing patients. A novel pix2pix deep learning network was developed to predict MUs/MLC shapes for treatment plan verification. It is the first attempt in using deep learning method to predict fundamental machine delivery parameters and it may provide radiation oncology community a useful tool for improving the efficiency and accuracy of patient specific QA and plan second check process.

Verification of the machine delivery parameters of a treatment plan via deep learning

We developed a generative adversarial network (GAN)-based deep learning approach to estimate the multileaf collimator (MLC) aperture and corresponding monitor units (MUs) from a given 3D dose distribution. The proposed design of the adversarial network, which integrates a residual block into pix2pix framework, jointly trains a ‘U-Net’-like architecture as the generator and a convolutional ‘PatchGAN’ classifier as the discriminator. 199 patients, including nasopharyngeal, lung and rectum, treated with intensity-modulated radiotherapy and volumetric-modulated arc therapy techniques were utilized to train the network. An additional 47 patients were used to test the prediction accuracy of the proposed deep learning model. The Dice similarity coefficient (DSC) was calculated to evaluate the similarity between the MLC aperture shapes obtained from the treatment planning system (TPS) and the deep learning prediction. The average and standard deviation of the bias between the TPS-generated MUs and predicted MUs was calculated to evaluate the MU prediction accuracy. In addition, the differences between TPS and deep learning-predicted MLC leaf positions were compared. The average and standard deviation of DSC was 0.94 ± 0.043 for 47 testing patients. The average deviation of predicted MUs from the planned MUs normalized to each beam or arc was within 2% for all the testing patients. The average deviation of the predicted MLC leaf positions was around one pixel for all the testing patients. Our results demonstrated the feasibility and reliability of the proposed approach. The proposed technique has strong potential to improve the efficiency and accuracy of the patient plan quality assurance process.

Assessing the dependency of the uncertainties in the Elekta Agility MLC calibration procedure on the focal spot position.

The effectiveness of radiotherapy treatments depends on the accuracy of the dose delivery process. The majority of radiotherapy courses are delivered on linear accelerators with a Multi Leaf Collimator (MLC) in 3D conformal Radiation Therapy, Intensity Modulated Radiation Therapy (IMRT) or Volumetric Modulated Arc Therapy (VMAT) modes that require accurate MLC positioning. This study investigates the MLC calibration accuracy, following manufacturer procedures for an Elekta Synergy linac with the Agility head, against the radiation focal spot offset (alignment with the collimator axis of rotation). If the radiation focal spot is not aligned ideally with the collimator axis of rotation then a systematic error can be introduced into the calibration procedure affecting absolute MLC leaf positions. Calibration of diaphrams is equally affected; however they are not investigated here. The results indicate that an estimated 0.15 mm MLC uncertainty in all MLC leaves positions can be introduced due to uncertainty of the radiation focal spot positionof 0.21 mm.

Dynamic fluence map sequencing using piecewise linear leaf position functions

Within the setting of intensity modulated radiation therapy (IMRT) and the fully continuous version of IMRT called volumetric modulated radiation therapy (VMAT), we consider the problem of matching a given fluence map as well as possible in limited time by the use of a linear accelerator (linac) with a multi-leaf collimator (MLC). We introduce two modeling strategies to manage the nonconvexity and the associated local minima of this problem. The first is the use of linear splines to model the MLC leaf positions as functions of time. The second is a progressively controllable smooth model (instead of a step function) of how the leaves block the fluence radiation. We propose a two part solution: an outer loop that optimizes the dose rate pattern over time, and an inner loop that given a dose rate pattern optimizes the leaf trajectories.

Investigating the impact of treatment delivery uncertainties on treatment effectiveness for lung SABR.

To quantify the impact of treatment delivery uncertainties on lung stereotactic ablative body radiotherapy (SABR) plans for step-and-shoot intensity-modulated radiotherapy (ssIMRT) and volumetric modulated arc therapy (VMAT). Baseline ssIMRT and VMAT treatment plans were generated for a cohort of 18 lung SABR patients. Modified plans were generated for each baseline plan by systematically varying gantry and collimator angles between -5 and +5 degrees, as well as multi-leaf collimator (MLC) leaf position errors of magnitude between 1 and 5 mm in both directions (i.e. leaf banks shifted either in the same (Type 1) or opposite (Type 2) directions). Planning target volume (PTV), spinal cord and healthy lung dose-volume histogram (DVH) metrics were compared between the modified and baseline plans. Collimator and gantry angle uncertainties did not significantly impact any of the PTV DVH metrics considered. MLC shifts of 5 mm resulted in average V95% changes of [Formula: see text] (Type 1) and [Formula: see text] (Type 2) and average [Formula: see text] changes of [Formula: see text] (Type 1) and [Formula: see text] (Type 2) for ssIMRT and VMAT plans. Comparatively, MLC shifts of - 2 mm resulted in average [Formula: see text] changes of [Formula: see text] (Type 1) and [Formula: see text] (Type 2) and average [Formula: see text] changes of [Formula: see text] (Type 1) and [Formula: see text] (Type 2) for ssIMRT and VMAT plans. ssIMRT gantry angle uncertainties impacted spinal cord DVH metrics the most, with increases in [Formula: see text] of [Formula: see text] occurring for a 1 degree shift. Type 2 MLC modifications impacted all OAR DVH metrics substantially with differences in spinal cord [Formula: see text] (ssIMRT) and healthy lung [Formula: see text] (VMAT) exceeding [Formula: see text] for 5 mm shifts. Uncertainties in MLC leaf positions affected target and OAR DVH metrics more than collimator or gantry angle uncertainties for lung SABR plans. Less patient-to-patient variation occurred from delivery uncertainties in VMAT than ssIMRT.

Simultaneous delivery time and aperture shape optimization for the volumetric-modulated arc therapy (VMAT) treatment planning problem

In this paper, we propose a novel heuristic algorithm for the volumetric-modulated arc therapy treatment planning problem, optimizing the trade-off between delivery time and treatment quality. We present a new mixed integer programming model in which the multi-leaf collimator leaf positions, gantry speed, and dose rate are determined simultaneously. Our heuristic is based on column generation; the aperture configuration is modeled in the columns and the dose distribution and time restriction in the rows. To reduce the number of voxels and increase the efficiency of the master model, we aggregate similar voxels using a clustering technique. The efficiency of the algorithm and the treatment quality are evaluated on a benchmark clinical prostate cancer case. The computational results show that a high-quality treatment is achievable using a four-thread CPU. Finally, we analyze the effects of the various parameters and two leaf-motion strategies.