Acute hematologic toxicity prediction using dosimetric and radiomics features in patients with cervical cancer: does the treatment regimen matter?

Background: Acute hematologic toxicity (HT) is a prevalent adverse tissue reaction observed in cervical cancer patients undergoing chemoradiotherapy (CRT), which may lead to various negative effects such as compromised therapeutic efficacy and prolonged treatment duration. Accurate prediction of HT occurrence prior to CRT remains challenging. Methods: A discovery dataset comprising 478 continuous cervical cancer patients (140 HT patients) and a validation dataset consisting of 205 patients (52 HT patients) were retrospectively enrolled. Both datasets were categorized into the CRT group and radiotherapy (RT)-alone group based on the treatment regimen, i.e., whether chemotherapy was administered within the focused RT duration. Radiomics features were derived by contouring three regions of interest (ROIs)-bone marrow (BM), femoral head (FH), and clinical target volume (CTV)-on the treatment planning CT images before RT. A comprehensive model combining the radiomics features as well as the demographic, clinical, and dosimetric features was constructed to classify patients exhibiting acute HT symptoms in the CRT group, RT group, and combination group. Furthermore, the time-to-event analysis of the discriminative ROI was performed on all patients with acute HT to understand the HT temporal progression. Results: Among three ROIs, BM exhibited the best performance in classifying acute HT, which was verified across all patient groups in both discovery and validation datasets. Among different patient groups in the discovery dataset, acute HT was more precisely predicted in the CRT group [area under the curve (AUC) = 0.779, 95% CI: 0.657-0.874] than that in the RT-alone (AUC = 0.686, 95% CI: 0.529-0.817) or combination group (AUC = 0.748, 95% CI: 0.655-0.827). The predictive results in the validation dataset similarly coincided with those in the discovery dataset: CRT group (AUC = 0.802, 95% CI: 0.669-0.914), RT-alone group (AUC = 0.737, 95% CI: 0.612-0.862), and combination group (AUC = 0.793, 95% CI: 0.713-0.874). In addition, distinct feature sets were adopted for different patient groups. Moreover, the predicted HT risk of BM was also indicative of the HT temporal progression. Conclusions: HT prediction in cervical patients is dependent on both the treatment regimen and ROI selection, and BM is closely related to the occurrence and progression of HT, especially for CRT patients.

Efficient EPID-based quality assurance of beam time delay for respiratory-gated radiotherapy with validation on Catalyst™ and AlignRT™ systems.

PURPOSE: To propose a straightforward and time-efficient quality assurance (QA) approach of beam time delay for respiratory-gated radiotherapy and validate the proposed method on typical respiratory gating systems, Catalyst™ and AlignRT™. METHODS: The QA apparatus was composed of a motion platform and a Winston-Lutz cube phantom (WL3) embedded with metal balls. The apparatus was first scanned in CT-Sim and two types of QA plans specific for beam on and beam off time delay, respectively, were designed. Static reference images and motion testing images of the WL3 cube were acquired with EPID. By comparing the position differences of the embedded metal balls in the motion and reference images, beam time delays were determined. The proposed approach was validated on three linacs with either Catalyst™ or AlignRT™ respiratory gating systems. To investigate the impact of energy and dose rate on beam time delay, a range of QA plans with Eclipse (V15.7) were devised with varying energy and dose rates. RESULTS: For all energies, the beam on time delays in AlignRT™ V6.3.226, AlignRT™ V7.1.1, and Catalyst™ were 92.13 ± $ \pm $ 5.79 ms, 123.11 ± $ \pm $ 6.44 ms, and 303.44 ± $ \pm $ 4.28 ms, respectively. The beam off time delays in AlignRT™ V6.3.226, AlignRT™ V7.1.1, and Catalyst™ were 121.87 ± $ \pm $ 1.34 ms, 119.33 ± $ \pm $ 0.75 ms, and 97.69 ± $ \pm $ 2.02 ms, respectively. Furthermore, the beam on delays decreased slightly as dose rates increased for all gating systems, whereas the beam off delays remained unaffected. CONCLUSIONS: The validation results demonstrate the proposed QA approach of beam time delay for respiratory-gated radiotherapy was both reproducible and time-efficient to practice for institutions to customize accordingly.

Deep learning-based projection synthesis for low-dose cone-beam computed tomography imaging in image-guided radiotherapy.

Background: The imaging dose of cone-beam computed tomography (CBCT) in image-guided radiotherapy (IGRT) poses adverse effects on patient health. To improve the quality of sparse-view low-dose CBCT images, a projection synthesis convolutional neural network (SynCNN) model is proposed. Methods: Included in this retrospective, single-center study were 223 patients diagnosed with brain tumours from Beijing Cancer Hospital. The proposed SynCNN model estimated two pairs of orthogonally direction-separable spatial kernels to synthesize the missing projection in between the input neighboring sparse-view projections via local convolution operations. The SynCNN model was trained on 150 real patients to learn patterns for inter-view projection synthesis. CBCT data from 30 real patients were used to validate the SynCNN, while data from a phantom and 43 real patients were used to test the SynCNN externally. Sparse-view projection datasets with 1/2, 1/4, and 1/8 of the original sampling rate were simulated, and the corresponding full-view projection datasets were restored using the SynCNN model. The tomographic images were then reconstructed with the Feldkamp-Davis-Kress algorithm. The root-mean-square error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) metrics were measured in both the projection and image domains. Five experts were invited to grade the image quality blindly for 40 randomly selected evaluation groups with a four-level rubric, where a score greater than or equal to 2 was considered acceptable image quality. The running time of the SynCNN model was recorded. The SynCNN model was directly compared with the three other methods on 1/4 sparse-view reconstructions. Results: The phantom and patient studies showed that the missing projections were accurately synthesized. In the image domain, for the phantom study, compared with images reconstructed from sparse-view projections, images with SynCNN synthesis exhibited significantly improved qualities with decreased values in RMSE and increased values in PSNR and SSIM. For the patient study, between the results with and without the SynCNN synthesis, the averaged RMSE decreased by 3.4×10-4, 10.3×10-4, and 21.7×10-4, the averaged PSNR increased by 3.4, 6.6, and 9.4 dB, and the averaged SSIM increased by 5.2×10-2, 18.9×10-2 and 33.9×10-2, for the 1/2, 1/4, and 1/8 sparse-view reconstructions, respectively. In expert subjective evaluation, both the median scores and acceptance rates of the images with SynCNN synthesis were higher than those reconstructed from sparse-view projections. It took the model less than 0.01 s to synthesize an inter-view projection. Compared with the three other methods, the SynCNN model obtained the best scores in terms of the three metrics in both domains. Conclusions: The proposed SynCNN model effectively improves the quality of sparse-view CBCT images at a low time cost. With the SynCNN model, the CBCT imaging dose in IGRT could be reduced potentially.

Towards deep-learning (DL) based fully automated target delineation for rectal cancer neoadjuvant radiotherapy using a divide-and-conquer strategy: a study with multicenter blind and randomized validation.

PURPOSE: Manual clinical target volume (CTV) and gross tumor volume (GTV) delineation for rectal cancer neoadjuvant radiotherapy is pivotal but labor-intensive. This study aims to propose a deep learning (DL)-based workflow towards fully automated clinical target volume (CTV) and gross tumor volume (GTV) delineation for rectal cancer neoadjuvant radiotherapy. MATERIALS & METHODS: We retrospectively included 141 patients with Stage II-III mid-low rectal cancer and randomly grouped them into training (n = 121) and testing (n = 20) cohorts. We adopted a divide-and-conquer strategy to address CTV and GTV segmentation using two separate DL models with DpuUnet as backend-one model for CTV segmentation in the CT domain, and the other for GTV in the MRI domain. The workflow was validated using a three-level multicenter-involved blind and randomized evaluation scheme. Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (95HD) metrics were calculated in Level 1, four-grade expert scoring was performed in Level 2, and head-to-head Turing test in Level 3. RESULTS: For the DL-based CTV contours over the testing cohort, the DSC and 95HD (mean ± SD) were 0.85 ± 0.06 and 7.75 ± 6.42 mm respectively, and 96.4% cases achieved clinical viable scores (≥ 2). The positive rate in the Turing test was 52.3%. For GTV, the DSC and 95HD were 0.87 ± 0.07 and 4.07 ± 1.67 mm respectively, and 100% of the DL-based contours achieved clinical viable scores (≥ 2). The positive rate in the Turing test was 52.0%. CONCLUSION: The proposed DL-based workflow exhibited promising accuracy and excellent clinical viability towards automated CTV and GTV delineation for rectal cancer neoadjuvant radiotherapy.

High-dimensional automated radiation therapy treatment planning via Bayesian optimization.

PURPOSE: Radiation therapy treatment planning can be viewed as an iterative hyperparameter tuning process to balance conflicting clinical goals. In this work, we investigated the performance of modern Bayesian optimization (BO) methods on automated treatment planning problems in high-dimensional settings. METHODS: Twenty locally advanced rectal cancer patients treated with intensity-modulated radiation therapy (IMRT) were retrospectively selected as test cases. The adjustable planning parameters included both dose objectives and their corresponding weights. We implemented an automated treatment planning framework and tested the performance of two BO methods on the treatment planning task: one standard BO method (Gaussian Process with Expected Improvement [GPEI]) and one BO method dedicated to high-dimensional problems (Sparse Axis Aligned Subspace BO [SAAS-BO]). Another derivative-free method (Nelder-Mead simplex search) and the random tuning method were also included as baselines. The four automated methods' plan quality and planning efficiency were compared with the clinical plans regarding target coverage and organs at risk (OAR) sparing. The predictive models in both BO methods were compared to analyze the different search patterns of the two BO methods. RESULTS: For the target structures, the SAAS-BO plans achieved comparable hot spot control ( p = 0.43 $p=0.43$ ) and homogeneity ( p = 0.96 $p=0.96$ ) with the clinical plans, significantly better than the GPEI and Nelder-Mead plans ( p < 0.05 $p < 0.05$ ). Both SAAS-BO and GPEI plans significantly outperformed the clinical plans in conformity and dose spillage ( p < 0.05 $p < 0.05$ ). Compared with the clinical plans, the treatment plans generated by the four automated methods all made reductions in evaluated dosimetric indices for the femoral head and the bladder. The Nelder-Mead plans achieved similar plan quality scores compared with the BO plans, but exhibited poorer control in the target hot spot and dose spillage. The analysis of the underlying predictive models has shown that both BO methods have identified similar sensitive planning parameters. CONCLUSIONS: This work implemented a BO-based hyperparameter tuning framework for automated treatment planning. Both tested BO methods were able to produce high-quality treatment plans and reduce the workload of treatment planners. The model analysis also confirmed the intrinsic low dimensionality of the tested treatment planning problems.

Radiation hematologic toxicity prediction for locally advanced rectal cancer using dosimetric and radiomics features.

BACKGROUND: Hematologic toxicity (HT) is a common adverse tissue reaction during radiotherapy for rectal cancer patients, which may lead to various negative effects such as reduced therapeutic effect, prolonged treatment period and increased treatment cost. Therefore, predicting the occurrence of HT before radiotherapy is necessary but still challenging. PURPOSE: This study proposes a hybrid machine learning model to predict the symptomatic radiation HT in rectal cancer patients using the combined demographic, clinical, dosimetric, and Radiomics features, and ascertains the most effective regions of interest (ROI) in CT images and predictive feature sets. METHODS: A discovery dataset of 240 rectal cancer patients, including 145 patients with HT symptoms and a validation dataset of 96 patients (63 patients with HT) with different dose prescription were retrospectively enrolled. Eight ROIs were contoured on patient CT images to derive Radiomics features, which were then, respectively, combined with the demographic, clinical, and dosimetric features to classify patients with HT symptoms. Moreover, the survival analysis was performed on risky patients with HT in order to understand the HT progression. RESULTS: The classification models in ROIs of bone marrow and femoral head exhibited relatively high accuracies (accuracy = 0.765 and 0.725) in the discovery dataset as well as comparable performances in the validation dataset (accuracy = 0.758 and 0.714). When combining the two ROIs together, the model performance was the best in both discovery and validation datasets (accuracy = 0.843 and 0.802). In the survival analysis test, only the bone marrow ROI achieved statistically significant performance in accessing risky HT (C-index = 0.658, P = 0.03). Most of the discriminative features were Radiomics features, and only gender and the mean dose in Irradvolume was involved in HT. CONCLUSION: The results reflect that the Radiomics features of bone marrow are significantly correlated with HT occurrence and progression in rectal cancer. The proposed Radiomics-based model may help the early detection of radiotherapy induced HT in rectal cancer patients and thus improve the clinical outcome in future.

Retrospective cohort study for thrombocytopenia during concurrent chemoradiotherapy for rectal cancer.

Background: The aim of this article was to establish the clinical prognostic models and identify the predictive radiation dosimetric parameters for thrombocytopenia during concurrent chemoradiotherapy for rectal cancer. Methods: In this retrospective cohort study, patients with rectal adenocarcinoma undergoing concurrent long-term chemoradiotherapy were included. The primary outcome of interest was grade 2 or higher (2+) thrombocytopenia (platelet(PLT) count <75,000/µL). Secondary outcomes included: grade 1 or higher thrombocytopenia (PLT count<100,000/µL) and the PLT count during chemoradiotherapy and its nadir. The risk prediction model was developed by logistic regression to identify clinical predictors of 2+ thrombocytopenia. Univariate linear regression models were used to test correlations between radiation dosimetric parameters and the absolute PLT count at nadirs. Results: This retrospective cohort comprised 238 patients. Fifty-four (22.6%) patients developed thrombocytopenia during concurrent chemoradiotherapy, while 15 (6.3%) patients developed 2+ thrombocytopenia. Four independently associated risk factors, including age, Alb level, PLT count, and chemotherapy regimen, were included in the final model and used to form a 2+ thrombocytopenia probability estimation nomogram. The C-index was 0.87 (95% CI: 0.78-0.96). The calibration plot showed a moderate agreement, and the Brier score was 0.047 (95% CI: 0.025-0.070). The total absolute volume of bone marrow irradiated by 5 Gy, 10 Gy and 15 Gy of radiation (BM-V5ab, BM-V10ab, BM-V15ab), calculated by the volume of bone marrow multiplied by the corresponding Vx, were identified as new predictors. The nadir of PLT was found to be negatively correlated with BM-V5ab (ß = -0.062, P =0.030), BM-V10ab (ß = -0.065, P =0.030) and BM-V15ab (ß = -0.064, P =0.042). Conclusion: The occurrence of 2+ thrombocytopenia during concurrent chemoradiotherapy for rectal cancer can be predicted by the patient's baseline status and chemoradiotherapy regimen, and low dose irradiation of bone marrow can affect the level of platelets during the treatment.

Initial clinical experience of surface guided stereotactic radiation therapy with open-face mask immobilization for improving setup accuracy: a retrospective study.

PURPOSE: To propose a specific surface guided stereotactic radiotherapy (SRT) treatment procedure with open-face mask immobilization and evaluate the initial clinical experience in improving setup accuracy. METHODS AND MATERIALS: The treatment records of 48 SRT patients with head lesions were retrospectively analyzed. For each patient, head immobilization was achieved with a double-shell open-face mask. The anterior shell was left open to expose the forehead, nose, eyes and cheekbones. The exposed facial area was used as region-of-interest for surface tracking by AlignRT (VisionRT Inc, UK). The posterior shell provided a sturdy and personalized headrest. Patient initial setup was guided by 6DoF real-time deltas (RTD) using the reference surface obtained from the skin contour delineated on the planning CT images. The endpoint of initial setup was 1 mm in translational RTD and 1 degree in rotational RTD. CBCT guidance was performed to derive the initial setup errors, and couch shifts for setup correction were applied prior to treatment delivery. CBCT couch shifts, AlignRT RTD values, repositioning rate and setup time were analyzed. RESULTS: The absolute values of median (maximal) CBCT couch shifts were 0.4 (1.3) mm in VRT, 0.1 (2.5) mm in LNG, 0.2 (1.6) mm in LAT, 0.1(1.2) degree in YAW, 0.2 (1.4) degree in PITCH and 0.1(1.3) degree in ROLL. The couch shifts and AlignRT RTD values exhibited highly agreement except in VRT and PITCH (p value < 0.01), of which the differences were as small as negligible. We did not find any case of patient repositioning that was due to out-of-tolerance setup errors, i.e., 3 mm and 2 degree. The surface guided setup time ranged from 52 to 174 s, and the mean and median time was 97.72 s and 94 s respectively. CONCLUSIONS: The proposed surface guided SRT procedure with open-face mask immobilization is a step forward in improving patient comfort and positioning accuracy in the same process. Minimized initial setup errors and repositioning rate had been achieved with reasonably efficiency for routine clinical practice.

Multi-channel respiratory signal detection system for 4D-CT in radiotherapy by measuring the back pressure.

This study proposes a novel respiratory signal detection system for 4D-CT in radiotherapy by measuring back pressure changes at multiple positions on CT couch. The 12-channel pressure sensor is fixed on CT couch to obtain patient's back pressure signal. The 12-channel signal is transmitted to a PC at a sampling rate of 50 Hz after a signal conditioning circuit and an analog-digital converter. The amplitude of pressure changes is characterized to select the optimal channel. This system is validated by comparing with the respiratory signal collected synchronously with a real-time position management (RPM) system on 10 healthy volunteers. The correlation coefficient between the signals is 0.82 ± 0.09 (standard deviation) and the time shift is 0.32 ± 0.15 second. We conclude that the back pressure signal acquired by the proposed system has the potential to replace the clinical RPM system for respiratory signal detection in 4D-CT data acquisition.

[Optimization of Key Techniques for Maintaining Bus Voltage Stability in Variable Speed Turbine of Ventilator].

OBJECTIVE: To explore the optimization scheme of maintaining bus voltage stability during turbo-turbine acceleration and deceleration of ventilator. METHODS: The ideal diode is used to replace the diode in the busbar power supply circuit, and a comparative discharge circuit is added to the busbar. When the busbar voltage is higher than the preset threshold, the comparator can be opened and the energy could be discharged through the power resistor. RESULTS: When the turbine starts and stops rapidly, the optimized scheme can effectively reduce the bus impedance, and the discharge circuit can maintain the bus voltage fluctuation less than 2 V. CONCLUSIONS: The optimization scheme proposed in this study can effectively improve the efficiency and stability of the turbine in the process of acceleration and braking, and provide reference for the design of the stability maintenance circuit of the ventilator turbine bus.

Development and Longitudinal Analysis of Plan-Based Streamlined Quality Assurance on Multiple Positioning Guidance Systems With Single Phantom Setup.

PURPOSE: This study was to propose and validate an efficient and streamlined quality assurance (QA) method with a single phantom setup to check performances of patient positioning guidance systems including six-degree-of-freedom (6DoF) couch, X-ray modalities (kV-kV, MV-MV and CBCT), optical surface imaging system (AlignRT), lasers and optical distance indicator (ODI). METHODS AND MATERIALS: The QA method was based on a pseudo-patient treatment plan using the AlignRT cube phantom. The cube was first randomly set up on the couch, and the initial position offsets were acquired by AlignRT and CBCT. The cube was restored to its reference position by 6DoF couch shift, during which the couch motion accuracy and tracking performances of AlignRT and CBCT were derived. After that, the residual offsets were acquired by kV-kV, MV-MV and AlignRT to derive the isocenter discrepancies. Finally, the laser alignment and ODI values were visually inspected. The QA procedure had been internally approved as a standard weekly QA test, and the results over 50 weeks were longitudinally analyzed for clinical validation. RESULTS: The 6DoF couch motion errors as well as the tracking errors of AlignRT were sub-millimeter and sub-degree, and no deviation over 1 mm or 1 deg was identified. The ROI mode of isocenter (ISO) in AlignRT exhibited more consistent results than the centroid (CEN). While the isocenter discrepancy between CBCT and kV-kV was negligible, the maximal discrepancies between CBCT and MV-MV were 0.4 mm in LNG and 0.3 deg in PITCH. The isocenter discrepancies between CBCT and AlignRT were <0.5 mm in translation and <0.3 deg in rotation. For AlignRT, the isocenter discrepancies between the DICOM and SGRT references were about 0.6 mm in VRT, 0.5 mm in LNG and 0.2 deg in PITCH. As the therapists became familiar with the workflow, the average time to complete the whole procedure was around 23 min. CONCLUSIONS: The streamlined QA exhibits desirable practicality as an efficient multipurpose performance check on positioning guidance systems. The stability, tracking performance and isocenter congruence of the positioning guidance systems have been fully validated for all clinical image guidance RT application, even SRS/SBRT, which requires the strictest tolerance.

Tree-based exploration of the optimization objectives for automatic cervical cancer IMRT treatment planning.

OBJECTIVE: To develop and evaluate a practical automatic treatment planning method for intensity-modulated radiation therapy (IMRT) in cervical cancer cases. METHODS: A novel algorithm named as Optimization Objectives Tree Search Algorithm (OOTSA) was proposed to emulate the planning optimization process and achieve a progressively improving IMRT plan, based on the Eclipse Scripting Application Programming Interface (ESAPI). 30 previously treated cervical cancer cases were selected from the clinical database and comparison was made between the OOTSA-generated plans and clinical treated plans and RapidPlan-based (RP) plans. RESULTS: In clinical evaluation, compared with plan scores of the clinical plans and the RP plans, 22 and 26 of the OOTSA plans were considered as clinically improved in terms of plan quality, respectively. The average conformity index (CI) for the PTV in the OOTSA plans was 0.86 ± 0.01 (mean ± 1 standard deviation), better than those in the RP plans (0.83 ± 0.02) and the clinical plans (0.71 ± 0.11). Compared with the clinical plans, the mean doses of femoral head, rectum, spinal cord and right kidney in the OOTSA plans were reduced by 2.34 ± 2.87 Gy, 1.67 ± 2.10 Gy, 4.12 ± 6.44 Gy and 1.15 ± 2.67 Gy. Compared with the RP plans, the mean doses of femoral head, spinal cord, right kidney and small intestine in the OOTSA plans were reduced by 3.31 ± 1.55 Gy, 4.25 ± 3.69 Gy, 1.54 ± 2.23 Gy and 3.33 ± 1.91 Gy, respectively. In the OOTSA plans, the mean dose of bladder was slightly increased, with 2.33 ± 2.55 Gy (versus clinical plans) and 1.37 ± 1.74 Gy (vs RP plans). The average elapsed time of OOTSA and clinical planning were 59.2 ± 3.47 min and 76.53 ± 5.19 min. CONCLUSION: The plans created by OOTSA have been shown marginally better than the manual plans, especially in preserving OARs. In addition, the time of automatic treatment planning has shown a reduction compared to a manual planning process, and the variation of plan quality was greatly reduced. Although improvement on the algorithm is warranted, this proof-of-concept study has demonstrated that the proposed approach can be a practical solution for automatic planning. ADVANCES IN KNOWLEDGE: The proposed method is novel in the emulation strategy of the physicists' iterative operation during the planning process. Based on the existing optimizers, this method can be a simple yet effective solution for automated IMRT treatment planning.

The structural similarity index for IMRT quality assurance: radiomics-based error classification.

PURPOSE: The implementation of radiomics and machine learning (ML) techniques on analyzing two-dimensional gamma maps has been demonstrated superior to the conventional gamma analysis for error identification in intensity modulated radiotherapy (IMRT) quality assurance (QA). Recently, the Structural SIMilarity (SSIM) sub-index maps were shown to be able to reveal the error types of the dose distributions. In this study, we aimed to apply radiomics analysis on SSIM sub-index maps and develop ML models to classify delivery errors in patient-specific dynamic IMRT QA. METHODS: Twenty-one sliding-window IMRT plans of 180 beams for three treatment sites were involved in this study. Four types of machine-related errors of various magnitudes were simulated for each beam at each control point, including the monitor unit (MU) variations, same-directional and opposite-directional shifts of the multileaf collimators (MLCs) and random mispositioning of the MLCs. In the QA process, a total of 1620 portal dose (PD) images were acquired for the beams with and without errors. The predicted PD images of the original beams were set as references. To quantify the agreement between a measured PD image and the corresponding predicted PD image, four difference maps including three SSIM sub-index maps, and one dose difference-derived map were calculated. Then, radiomic features were extracted from the four difference maps of each measured PD image. We tested four typical classifiers including linear discriminant classifier (LDC), two supporting vector machine (SVM) classifiers, and random forest (RF) for this multiclass classification task. A nested cross-validation scheme was used for model evaluations, where the SVM recursive feature elimination method was applied for feature selection. Finally, the performance of the ML model on identifying the error-free and the erroneous cases was compared to that of the conventional gamma analysis. RESULTS: The statistics of the selected features showed that all of the difference maps and the feature categories made balanced contributions to solve this classification task. Best performance was achieved by the Linear-SVM model with average overall classification accuracy of 0.86. Specifically, the average classification accuracies of the shift, opening, and the random errors were around 0.9. Moreover, ~80% of error-free and MU errors were correctly classified. Using gamma analysis, the 3 mm/3% criterion was found insensitive to errors (sensitivity was only 0.33). Although the sensitivity to errors with the 2 mm/2% criterion increased to 0.79, still 8% worse than that of the ML model. CONCLUSIONS: We proposed an ML-based method for machine-related error identification in patient-specific dynamic IMRT QA, where radiomic analysis on SSIM sub-index maps were used for feature extraction. With extensive validation to select the best features and classifiers, high accuracies in error classification were achieved. Compared with the conventional gamma threshold method, this approach has great potential in error identification for the patient-specific IMRT QA process.

Halcyon clinical performance evaluation: A log file-based study in comparison with a C-arm Linac.

PURPOSE: The aim of this study is to compare the dosimetric and mechanical accuracy of Volumetric Modulation Arc Therapy (VMAT) delivery on the Halcyon, a recent ring-shaped Treatment Delivery System (TDS) featuring fast rotating gantry, with a conventional C-arm Linac. METHODS: The comparison was performed via log file analysis, where mechanical parameters of related components was extracted. 480 and 3951 VMAT log files of clinically delivered fractions from a Halcyon and a TrueBeam Linac were analyzed respectively. The relations between mechanical parameters and errors were extensively explored to further investigate the differences between the two Linacs. The mechanical parameter fluctuations were taken into account for dose recalculations, and the Dose Volume Parameters (DVP) on the PTV were evaluated to quantify such dosimetric variations. RESULTS: The Multi-Leaf Collimator (MLC) leaf mean Root Mean Square (RMS) errors were 0.028 mm and 0.031 mm for Halcyon and TrueBeam respectively. Maximum systematic error on the MLC leaves introduced by the gravity effect were 0.04 mm and 0.01 mm for the Halcyon and TrueBeam respectively. Thanks to the O-ring design, the Halcyon achieved 0.035° in mean RMS error in gantry angle compared with the 0.065° of the TrueBeam. Overall mechanical errors introduced similar levels of dose-volume parameter variations (about 0.1%) on both Linacs. CONCLUSION: The Halcyon TDS can achieve similar mechanical leaf positioning accuracy compared with the TrueBeam TDS with a doubled delivery speed. In terms of dosimetric accuracy, The DVP standard deviations on the studied TB are generally larger than that on the Halcyon.

Radiological tissue equivalence of deformable silicone-based chemical radiation dosimeters (FlexyDos3D).

FlexyDos3D, a silicone-based chemical radiation dosimeter, has great potential to serve as a three-dimensional (3D) deformable dosimetric tool to verify complex dose distributions delivered by modern radiotherapy techniques. To facilitate its clinical application, its radiological tissue needs to be clarified. In this study we investigated its tissue-equivalence in comparison with water and Solid Water (RMI457). We found that its effective and mean atomic numbers were 40% and 20% higher and the total interaction probabilities for kV x-ray photons were larger than those of water respectively. To assess the influence of its over-response to kV photons, its HU value was measured by kV computed tomography (CT) and was found higher than all the soft-tissue substitutes. When applied for dose calculation without correction, this effect led to an 8% overestimation in electron density via HU-value mapping and 0.65% underestimation in target dose. Furthermore, depth dose curves (PDDs) and off-axis ratios (profiles) at various beam conditions as well as the dose distribution of a full-arc VMAT plan in FlexyDos3D and reference materials were simulated by Monte Carlo, where the results showed great agreement. As indicated, FlexyDos3D exhibits excellent radiological water-equivalence for clinical MV x-ray dosimetry, while its nonwater-equivalent effect for low energy x-ray dosimetry requires necessary correction. The key findings of this study provide pertinent reference for further FlexyDos3D characterization research.

Impact of nominal photon energies on normal tissue sparing in knowledge-based radiotherapy treatment planning for rectal cancer patients.

The interactive adjustment of the optimization objectives during the treatment planning process has made it difficult to evaluate the impact of beam quality exclusively in radiotherapy. Without consensus in the published results, the arbitrary selection of photon energies increased the probability of suboptimal plans. This work aims to evaluate the dosimetric impact of various photon energies on the sparing of normal tissues by applying a preconfigured knowledge-based planning (RapidPlan) model to various clinically available photon energies for rectal cancer patients, based on model-generated optimization objectives, which provide a comparison basis with less human interference. A RapidPlan model based on 81 historical VMAT plans for pre-surgical rectal cancer patients using 10MV flattened beam (10X) was used to generate patient-specific objectives for the automated optimization of other 20 patients using 6X, 8X, 10X (reference), 6MV flattening-filter-free (6F) and 10F beams respectively on a TrueBeam accelerator. It was observed that flattened beams produced very comparable target dose coverage yet the conformity index using 6F and 10F were clinically unacceptable (>1.29). Therefore, dose to organs-at-risk (OARs) and normal tissues were only evaluated for flattened beams. RapidPlan-generated objectives for 6X and 8X beams can achieve comparable target dose coverage as that of 10X, yet the dose to normal tissues increased monotonically with decreased energies. Differences were statistically significant except femoral heads. From the radiological perspective of view, higher beam energy is still preferable for deep seated tumors, even if multiple field entries such as VMAT technique can accumulate enough dose to the target using lower energies, as reported in the literature. In conclusion, RapidPlan model configured for flattened beams cannot optimize un-flattened beams before adjusting the target objectives, yet works for flattened beams of other energies. For the investigated 10X, 8X and 6X photons, higher energies provide better normal tissue sparing.

Comparison of Flattening Filter and Flattening Filter-Free Volumetric Modulated Arc Radiotherapy in Patients with Locally Advanced Nasopharyngeal Carcinoma.

BACKGROUND This study aimed to investigate the therapeutic role of flattening filter-free (FFF) mode in volumetric modulated arc therapy (VMAT) compared with flattening filter (FF) mode in patients with locally advanced nasopharyngeal carcinoma (NPC). MATERIAL AND METHODS Ten previously treated patients with NPC underwent treatment re-planning with FFF and FF VMAT. Radiotherapy dose distribution on planning target volume (PTV), organs at risk (OAR), target conformity index (CI), total monitor units (MUs), and therapeutic time were compared. RESULTS Maximum and mean radiotherapy dose in PTV and PGTV (primary lesions of NPC and cervical lymph node metastases) in FFF VMAT planning were significantly increased compared with FF VMAT planning, but PTV and OAR showed no significant differences. The CI value of PTV in FFF VMAT planning was significantly reduced compared with FF planning (P<0.05). No differences were found for the maximum radiotherapy dose in the spinal cord and left and right optic nerve, and the mean radiotherapy dose in the brainstem, left and right parotid gland (P>0.05). The maximum dose in the brainstem in the FFF planning was significantly higher compared with FF planning (P>0.05). The maximum radiotherapy dose in left and right crystalline lens (P<0.05) in FFF planning was significantly reduced compared with FF planning. The total hop count in FFF planning was significantly increased compared with FF planning (P<0.05). CONCLUSIONS Both 6 MV X-ray FFF mode and FF mode in the treatment of patients with NPC showed that FFF VMAT planning provided improved protection for OAR.

An interactive plan and model evolution method for knowledge-based pelvic VMAT planning.

PURPOSE: To test if a RapidPlan DVH estimation model and its training plans can be improved interactively through a closed-loop evolution process. METHODS AND MATERIALS: Eighty-one manual plans (P0 ) that were used to configure an initial rectal RapidPlan model (M0 ) were reoptimized using M0 (closed-loop), yielding 81 P1 plans. The 75 improved P1 (P1+ ) and the remaining 6 P0 were used to configure model M1 . The 81 training plans were reoptimized again using M1 , producing 23 P2 plans that were superior to both their P0 and P1 forms (P2+ ). Hence, the knowledge base of model M2 composed of 6 P0 , 52 P1+ , and 23 P2+ . Models were tested dosimetrically on 30 VMAT validation cases (Pv ) that were not used for training, yielding Pv (M0 ), Pv (M1 ), and Pv (M2 ) respectively. The 30 Pv were also optimized by M2_new as trained by the library of M2 and 30 Pv (M0 ). RESULTS: Based on comparable target dose coverage, the first closed-loop reoptimization significantly (P < 0.01) reduced the 81 training plans' mean dose to femoral head, urinary bladder, and small bowel by 2.65 Gy/15.63%, 2.06 Gy/8.11%, and 1.47 Gy/6.31% respectively, which were further reduced significantly (P < 0.01) in the second closed-loop reoptimization by 0.04 Gy/0.28%, 0.18 Gy/0.77%, 0.22 Gy/1.01% respectively. However, open-loop VMAT validations displayed more complex and intertwined plan quality changes: mean dose to urinary bladder and small bowel decreased monotonically using M1 (by 0.34 Gy/1.47%, 0.25 Gy/1.13%) and M2 (by 0.36 Gy/1.56%, 0.30 Gy/1.36%) than using M0 . However, mean dose to femoral head increased by 0.81 Gy/6.64% (M1 ) and 0.91 Gy/7.46% (M2 ) than using M0 . The overfitting problem was relieved by applying model M2_new . CONCLUSIONS: The RapidPlan model and its constituent plans can improve each other interactively through a closed-loop evolution process. Incorporating new patients into the original training library can improve the RapidPlan model and the upcoming plans interactively.

Fully automated searching for the optimal VMAT jaw settings based on Eclipse Scripting Application Programming Interface (ESAPI) and RapidPlan knowledge-based planning.

PURPOSE: Eclipse treatment planning system has not been able to optimize the jaw positions for Volumetric Modulated Arc Therapy (VMAT). The arbitrary and planner-dependent jaw placements define the maximum field size within which multi-leaf-collimator (MLC) sequences can be optimized to modulate the beam. Considering the mechanical constraints of MLC transitional speed and range, suboptimal X jaw settings may impede the optimization or undermine the deliverability. This work searches optimal VMAT jaw settings automatically based on Eclipse Scripting Application Programming Interface (ESAPI) and RapidPlan knowledge-based planning. METHODS AND MATERIALS: Using an ESAPI script, the X jaws of rectal VMAT plans were initially set to conform the planning-target-volume (PTV), and were gradually extended toward the isocenter (PTV center) in 5-7 mm increments. Using these jaw pairs, 592 plans were automatically created for 10 patients and quantitatively evaluated using a comprehensive scoring function. A published RapidPlan model was evoked by ESAPI to generate patient-specific optimization objectives without manual intervention. All candidate plans were first stored as text files to save storage space, and only the best, worst, and conformal plans were consequently recreated for comparison. RESULTS: Although RapidPlan estimates dose-volume histogram (DVH) based on individual anatomy, the geometry-based expected dose (GED) algorithm does not recognize different jaw settings but uses PTV-conformal jaws as default; hence, identical DVHs were observed regardless of planner-defined jaws. Therefore, ESAPI finalized dose-volume calculation and eliminated the plans with unacceptable hotspots before comparison. The plan quality varied dramatically with different jaw settings. Trade-offs among different organs-at-risk (OARs) were collectively considered by the proposed scoring method, which identified the best and worst plans correctly. The plans using conformal jaws were neither the best nor the worst of all candidates. CONCLUSIONS: VMAT plans using optimal jaw locations can be created automatically using ESAPI and RapidPlan. Conformal jaws are not the optimal choice.

Photon Optimizer (PO) prevails over Progressive Resolution Optimizer (PRO) for VMAT planning with or without knowledge-based solution.

The enhanced dosimetric performance of knowledge-based volumetric modulated arc therapy (VMAT) planning might be jointly contributed by the patient-specific optimization objectives, as estimated by the RapidPlan model, and by the potentially improved Photon Optimizer (PO) algorithm than the previous Progressive Resolution Optimizer (PRO) engine. As PO is mandatory for RapidPlan estimation but optional for conventional manual planning, appreciating the two optimizers may provide practical guidelines for the algorithm selection because knowledge-based planning may not replace the current method completely in a short run. Using a previously validated dose-volume histogram (DVH) estimation model which can produce clinically acceptable plans automatically for rectal cancer patients without interactive manual adjustment, this study reoptimized 30 historically approved plans (referred as clinical plans that were created manually with PRO) with RapidPlan solution (PO plans). Then the PRO algorithm was utilized to optimize the plans again using the same dose-volume constraints as PO plans, where the line objectives were converted as a series of point objectives automatically (PRO plans). On the basis of comparable target dose coverage, the combined applications of new objectives and PO algorithm have significantly reduced the organs-at-risk (OAR) exposure by 23.49-32.72% than the clinical plans. These discrepancies have been largely preserved after substituting PRO for PO, indicating the dosimetric improvements were mostly attributable to the refined objectives. Therefore, Eclipse users of earlier versions may instantly benefit from adopting the model-generated objectives from other RapidPlan-equipped centers, even with PRO algorithm. However, the additional contribution made by the PO relative to PRO accounted for 1.54-3.74%, suggesting PO should be selected with priority whenever available, with or without RapidPlan solution as a purchasable package. Significantly increased monitor units were associated with the model-generated objectives but independent from the optimizers, indicating higher modulation in these plans. As a summary, PO prevails over PRO algorithm for VMAT planning with or without knowledge-based technique.