ABSTRACT
OBJECTIVE: Target volumes for radiotherapy are usually contoured manually, which can be time-consuming and prone to inter- and intra-observer variability. Automatic contouring by convolutional neural networks (CNN) can be fast and consistent but may produce unrealistic contours or miss relevant structures. We evaluate approaches for increasing the quality and assessing the uncertainty of CNN-generated contours of head and neck cancers with PET/CT as input. Approach. Two patient cohorts with head and neck squamous cell carcinoma and baseline 18F-fluorodeoxyglucose positron emission tomography and computed tomography images (FDG-PET/CT) were collected retrospectively from two centers. The union of manual contours of the gross primary tumor and involved nodes was used to train CNN models for generating automatic contours. The impact of image preprocessing, image augmentation, transfer learning and CNN complexity, architecture, and dimension (2D or 3D) on model performance and generalizability across centers was evaluated. A Monte Carlo dropout technique was used to quantify and visualize the uncertainty of the automatic contours. Main results. CNN models provided contours with good overlap with the manually contoured ground truth (median Dice Similarity Coefficient: 0.75 - 0.77), consistent with reported inter-observer variations and previous auto-contouring studies. Image augmentation and model dimension, rather than model complexity, architecture, or advanced image preprocessing, had the largest impact on model performance and cross-center generalizability. Transfer learning on a limited number of patients from a separate center increased model generalizability without decreasing model performance on the original training cohort. High model uncertainty was associated with false positive and false negative voxels as well as low Dice coefficients. Significance. High quality automatic contours can be obtained using deep learning architectures that are not overly complex. Uncertainty estimation of the predicted contours shows potential for highlighting regions of the contour requiring manual revision or flagging segmentations requiring manual inspection and intervention. .
ABSTRACT
BACKGROUND AND PURPOSE: Artificial Intelligence (AI) models in radiation therapy are being developed with increasing pace. Despite this, the radiation therapy community has not widely adopted these models in clinical practice. A cohesive guideline on how to develop, report and clinically validate AI algorithms might help bridge this gap. METHODS AND MATERIALS: A Delphi process with all co-authors was followed to determine which topics should be addressed in this comprehensive guideline. Separate sections of the guideline, including Statements, were written by subgroups of the authors and discussed with the whole group at several meetings. Statements were formulated and scored as highly recommended or recommended. RESULTS: The following topics were found most relevant: Decision making, image analysis, volume segmentation, treatment planning, patient specific quality assurance of treatment delivery, adaptive treatment, outcome prediction, training, validation and testing of AI model parameters, model availability for others to verify, model quality assurance/updates and upgrades, ethics. Key references were given together with an outlook on current hurdles and possibilities to overcome these. 19 Statements were formulated. CONCLUSION: A cohesive guideline has been written which addresses main topics regarding AI in radiation therapy. It will help to guide development, as well as transparent and consistent reporting and validation of new AI tools and facilitate adoption.
Subject(s)
Artificial Intelligence , Delphi Technique , Humans , Radiotherapy Planning, Computer-Assisted/standards , Radiotherapy Planning, Computer-Assisted/methods , Radiation Oncology/standards , Radiotherapy/standards , Radiotherapy/methods , AlgorithmsABSTRACT
[This corrects the article DOI: 10.3389/fmed.2023.1217037.].
ABSTRACT
This retrospective study examined bone flap displacement during radiotherapy in 25 post-operative brain tumour patients. Though never exceeding 2.5 mm, the sheer frequency of displacement highlights the need for future research on larger populations to validate its presence and assess the potential clinical impact on planning tumour volume margins.
ABSTRACT
Objective. Newer cone-beam computed tomography (CBCT) imaging systems offer reconstruction algorithms including metal artifact reduction (MAR) and extended field-of-view (eFoV) techniques to improve image quality. In this study a new CBCT imager, the new Varian HyperSight CBCT, is compared to fan-beam CT and two CBCT imagers installed in a ring-gantry and C-arm linear accelerator, respectively.Approach. The image quality was assessed for HyperSight CBCT which uses new hardware, including a large-size flat panel detector, and improved image reconstruction algorithms. The decrease of metal artifacts was quantified (structural similarity index measure (SSIM) and root-mean-squared error (RMSE)) when applying MAR reconstruction and iterative reconstruction for a dental and spine region using a head-and-neck phantom. The geometry and CT number accuracy of the eFoV reconstruction was evaluated outside the standard field-of-view (sFoV) on a large 3D-printed chest phantom. Phantom size dependency of CT numbers was evaluated on three cylindrical phantoms of increasing diameter. Signal-to-noise and contrast-to-noise were quantified on an abdominal phantom.Main results. In phantoms with streak artifacts, MAR showed comparable results for HyperSight CBCT and CT, with MAR increasing the SSIM (0.97-0.99) and decreasing the RMSE (62-55 HU) compared to iterative reconstruction without MAR. In addition, HyperSight CBCT showed better geometrical accuracy in the eFoV than CT (Jaccard Conformity Index increase of 0.02-0.03). However, the CT number accuracy outside the sFoV was lower than for CT. The maximum CT number variation between different phantom sizes was lower for the HyperSight CBCT imager (â¼100 HU) compared to the two other CBCT imagers (â¼200 HU), but not fully comparable to CT (â¼50 HU).Significance. This study demonstrated the imaging performance of the new HyperSight CBCT imager and the potential of applying this CBCT system in more advanced scenarios by comparing the quality against fan-beam CT.
Subject(s)
Cone-Beam Computed Tomography , Image Processing, Computer-Assisted , Phantoms, Imaging , Cone-Beam Computed Tomography/instrumentation , Cone-Beam Computed Tomography/methods , Image Processing, Computer-Assisted/methods , Humans , Artifacts , Quality ControlABSTRACT
INTRODUCTION: An increase in plan robustness leads to a higher dose to organs-at-risk (OARs), and an increased chance of post-treatment toxicities. In contrast, more conformal plans lead to sparing of healthy surrounding tissue at the expense of a higher sensitivity to anatomical changes, requiring costly adaptations. In this study, we assess the trade-off and impact of treatment plan robustness on the adaptation rate. METHOD: Treatment planning was performed for 40 lung cancer patients, each having a planning 4DCT and up to eight weekly repeated 4DCTs (reCTs). For each patient, plans were made with three different levels of robustness based on setup uncertainty of 3, 6 and 9 mm. These plans were robustly re-evaluated on all reCTs to assess whether the clinical constraints were met. RESULTS: For the 3, 6 and 9 mm robustness levels, adaptation rates of 87.5â¯%, 70.0â¯% and 57.5â¯%, respectively, were observed. A mean absolute normal tissue complication probability (NTCP) gain of 2.9 percentage points (pp) was calculated for pneumonitis gradeâ¯≥â¯2 when transitioning from 9 mm plans to 3 mm plans, 7.6â¯pp for esophagitis gradeâ¯≥â¯2, and 2.5â¯pp for mortality risk 2â¯years post-treatment. CONCLUSION: The lowered risk of post treatment toxicities at lower robustness levels is clinically relevant but comes at the expense of more treatment adaptations, particularly in cases where meeting our clinical goals is not compromised by having a dose that is more conformal to the target. The trade-off between workload and reduced NTCP needs to be individually assessed.
Subject(s)
Lung Neoplasms , Organs at Risk , Proton Therapy , Radiotherapy Planning, Computer-Assisted , Humans , Lung Neoplasms/radiotherapy , Lung Neoplasms/pathology , Proton Therapy/adverse effects , Proton Therapy/methods , Radiotherapy Planning, Computer-Assisted/methods , Organs at Risk/radiation effects , Radiotherapy Dosage , Four-Dimensional Computed Tomography , Male , Female , Aged , Middle AgedABSTRACT
BACKGROUND: Deterioration of neurocognitive function in adult patients with a primary brain tumor is the most concerning side effect of radiotherapy. This study aimed to develop and evaluate normal-tissue complication probability (NTCP) models using clinical and dose-volume measures for 6-month, 1-year, and 2-year Neurocognitive Decline (ND) postradiotherapy. METHODS: A total of 219 patients with a primary brain tumor treated with radical photon and/or proton radiotherapy (RT) between 2019 and 2022 were included. Controlled oral word association test, Hopkins verbal learning test-revised, and trail making test were used to objectively measure ND. A comprehensive set of potential clinical and dose-volume measures on several brain structures were considered for statistical modeling. Clinical, dose-volume and combined models were constructed and internally tested in terms of discrimination (area under the curve, AUC), calibration (mean absolute error, MAE), and net benefit. RESULTS: Fifty percent, 44.5%, and 42.7% of the patients developed ND at 6-month, 1-year, and 2-year time points, respectively. The following predictors were included in the combined model for 6-month ND: age at radiotherapyâ >â 56 years (ORâ =â 5.71), overweight (ORâ =â 0.49), obesity (ORâ =â 0.35), chemotherapy (ORâ =â 2.23), brain V20 Gyâ ≥â 20% (ORâ =â 3.53), brainstem volumeâ ≥â 26 cc (ORâ =â 0.39), and hypothalamus volumeâ ≥â 0.5 cc (ORâ =â 0.4). Decision curve analysis showed that the combined models had the highest net benefits at 6-month (AUCâ =â 0.79, MAEâ =â 0.021), 1-year (AUCâ =â 0.72, MAEâ =â 0.027), and 2-year (AUCâ =â 0.69, MAEâ =â 0.038) time points. CONCLUSIONS: The proposed NTCP models use easy-to-obtain predictors to identify patients at high risk of ND after brain RT. These models can potentially provide a base for RT-related decisions and post-therapy neurocognitive rehabilitation interventions.
Subject(s)
Brain Neoplasms , Cranial Irradiation , Humans , Male , Female , Brain Neoplasms/radiotherapy , Middle Aged , Cranial Irradiation/adverse effects , Adult , Aged , Radiation Injuries/etiology , Radiation Injuries/diagnosis , Follow-Up Studies , Prognosis , Neurocognitive Disorders/etiology , Cognitive Dysfunction/etiology , Risk FactorsABSTRACT
BACKGROUND AND PURPOSE: Studies investigating the application of Artificial Intelligence (AI) in the field of radiotherapy exhibit substantial variations in terms of quality. The goal of this study was to assess the amount of transparency and bias in scoring articles with a specific focus on AI based segmentation and treatment planning, using modified PROBAST and TRIPOD checklists, in order to provide recommendations for future guideline developers and reviewers. MATERIALS AND METHODS: The TRIPOD and PROBAST checklist items were discussed and modified using a Delphi process. After consensus was reached, 2 groups of 3 co-authors scored 2 articles to evaluate usability and further optimize the adapted checklists. Finally, 10 articles were scored by all co-authors. Fleiss' kappa was calculated to assess the reliability of agreement between observers. RESULTS: Three of the 37 TRIPOD items and 5 of the 32 PROBAST items were deemed irrelevant. General terminology in the items (e.g., multivariable prediction model, predictors) was modified to align with AI-specific terms. After the first scoring round, further improvements of the items were formulated, e.g., by preventing the use of sub-questions or subjective words and adding clarifications on how to score an item. Using the final consensus list to score the 10 articles, only 2 out of the 61 items resulted in a statistically significant kappa of 0.4 or more demonstrating substantial agreement. For 41 items no statistically significant kappa was obtained indicating that the level of agreement among multiple observers is due to chance alone. CONCLUSION: Our study showed low reliability scores with the adapted TRIPOD and PROBAST checklists. Although such checklists have shown great value during development and reporting, this raises concerns about the applicability of such checklists to objectively score scientific articles for AI applications. When developing or revising guidelines, it is essential to consider their applicability to score articles without introducing bias.
Subject(s)
Artificial Intelligence , Checklist , Delphi Technique , Radiotherapy Planning, Computer-Assisted , Humans , Radiotherapy Planning, Computer-Assisted/methods , Radiotherapy Planning, Computer-Assisted/standards , Practice Guidelines as Topic , Bias , Reproducibility of Results , Neoplasms/radiotherapyABSTRACT
Background and purpose: Dose calculation on cone-beam computed tomography (CBCT) images has been less accurate than on computed tomography (CT) images due to lower image quality and discrepancies in CT numbers for CBCT. As increasing interest arises in offline and online re-planning, dose calculation accuracy was evaluated for a novel CBCT imager integrated into a ring gantry treatment machine. Materials and methods: The new CBCT system allowed fast image acquisition (5.9 s) by using new hardware, including a large-size flat panel detector, and incorporated image-processing algorithms with iterative reconstruction techniques, leading to accurate CT numbers allowing dose calculation. In this study, CBCT- and CT-based dose calculations were compared based on three anthropomorphic phantoms, after CBCT-to-mass-density calibration was performed. Six plans were created on the CT scans covering various target locations and complexities, followed by CBCT to CT registrations, copying of contours, and re-calculation of the plans on the CBCT scans. Dose-volume histogram metrics for target volumes and organs-at-risk (OARs) were evaluated, and global gamma analyses were performed. Results: Target coverage differences were consistently below 1.2 %, demonstrating the agreement between CT and re-calculated CBCT dose distributions. Differences in Dmean for OARs were below 0.5 Gy for all plans, except for three OARs, which were below 0.8 Gy (<1.1 %). All plans had a 3 %/1mm gamma pass rate > 97 %. Conclusions: This study demonstrated comparable results between dose calculations performed on CBCT and CT acquisitions. The new CBCT system with enhanced image quality and CT number accuracy opens possibilities for off-line and on-line re-planning.
ABSTRACT
PURPOSE: The optimal motion management strategy for patients receiving stereotactic arrhythmia radioablation (STAR) for the treatment of ventricular tachycardia (VT) is not fully known. We developed a framework using a digital phantom to simulate cardiorespiratory motion in combination with different motion management strategies to gain insight into the effect of cardiorespiratory motion on STAR. METHODS AND MATERIALS: The 4-dimensional (4D) extended cardiac-torso (XCAT) phantom was expanded with the 17-segment left ventricular (LV) model, which allowed placement of STAR targets in standardized ventricular regions. Cardiac- and respiratory-binned 4D computed tomography (CT) scans were simulated for free-breathing, reduced free-breathing, respiratory-gating, and breath-hold scenarios. Respiratory motion of the heart was set to population-averaged values of patients with VT: 6, 2, and 1 mm in the superior-inferior, posterior-anterior, and left-right direction, respectively. Cardiac contraction was adjusted by reducing LV ejection fraction to 35%. Target displacement was evaluated for all segments using envelopes encompassing the cardiorespiratory motion. Envelopes incorporating only the diastole plus respiratory motion were created to simulate the scenario where cardiac motion is not fully captured on 4D respiratory CT scans used for radiation therapy planning. RESULTS: The average volume of the 17 segments was 6 cm3 (1-9 cm3). Cardiac contraction-relaxation resulted in maximum segment (centroid) motion of 4, 6, and 3.5 mm in the superior-inferior, posterior-anterior, and left-right direction, respectively. Cardiac contraction-relaxation resulted in a motion envelope increase of 49% (24%-79%) compared with individual segment volumes, whereas envelopes increased by 126% (79%-167%) if respiratory motion also was considered. Envelopes incorporating only the diastole and respiration motion covered on average 68% to 75% of the motion envelope. CONCLUSIONS: The developed LV-segmental XCAT framework showed that free-wall regions display the most cardiorespiratory displacement. Our framework supports the optimization of STAR by evaluating the effect of (cardio)respiratory motion and motion management strategies for patients with VT.
Subject(s)
Heart , Respiration , Humans , Heart/diagnostic imaging , Heart/radiation effects , Heart Ventricles/diagnostic imaging , Heart Ventricles/radiation effects , Motion , Four-Dimensional Computed Tomography , Arrhythmias, Cardiac , Phantoms, ImagingABSTRACT
BACKGROUND AND PURPOSE: Brain tumors are in general treated with a maximal safe resection followed by radiotherapy of remaining tumor including the resection cavity (RC) and chemotherapy. Anatomical changes of the RC during radiotherapy can have impact on the coverage of the target volume. The aim of the current study was to quantify the potential changes of the RC and to identify risk factors for RC changes. MATERIALS AND METHODS: Sixteen patients treated with pencil beam scanning proton therapy between October 2019 and April 2020 were retrospectively analyzed. The RC was delineated on pre-treatment computed tomography (CT) and magnetic resonance imaging, and weekly CT-scans during treatment. Isotropic expansions were applied to the pre-treatment RC (1-5 mm). The percentage of volume of the RC during treatment within the expanded pre-treatment volumes was quantified. Potential risk factors (volume of RC, time interval surgery-radiotherapy and relationship of RC to the ventricles) were evaluated using Spearman's rank correlation coefficient. RESULTS: The average variation in relative RC volume during treatment was 26.1% (SD 34.6%). An expansion of 4 mm was required to cover > 95% of the RC volume in > 90% of patients. There was a significant relationship between the absolute volume of the pre-treatment RC and the volume changes during treatment (Spearman's ρ = - 0.644; p = 0.007). CONCLUSION: RCs are dynamic after surgery. Potentially, an additional margin in brain cancer patients with an RC should be considered, to avoid insufficient target coverage. Future research on local recurrence patterns is recommended.
Subject(s)
Brain Neoplasms , Radiotherapy, Intensity-Modulated , Humans , Retrospective Studies , Combined Modality Therapy , Tomography, X-Ray Computed , Radiotherapy Planning, Computer-Assisted , Brain Neoplasms/diagnostic imaging , Brain Neoplasms/radiotherapy , Brain Neoplasms/surgery , Brain/diagnostic imaging , Brain/surgery , Radiotherapy DosageABSTRACT
PURPOSE: Atlas-based and deep-learning contouring (DLC) are methods for automatic segmentation of organs-at-risk (OARs). The European Particle Therapy Network (EPTN) published a consensus-based atlas for delineation of OARs in neuro-oncology. In this study, geometric and dosimetric evaluation of automatically-segmented neuro-oncological OARs was performed using CT- and MR-models following the EPTN-contouring atlas. METHODS: Image and contouring data from 76 neuro-oncological patients were included. Two atlas-based models (CT-atlas and MR-atlas) and one DLC-model (MR-DLC) were created. Manual contours on registered CT-MR-images were used as ground-truth. Results were analyzed in terms of geometrical (volumetric Dice similarity coefficient (vDSC), surface DSC (sDSC), added path length (APL), and mean slice-wise Hausdorff distance (MSHD)) and dosimetrical accuracy. Distance-to-tumor analysis was performed to analyze to which extent the location of the OAR relative to planning target volume (PTV) has dosimetric impact, using Wilcoxon rank-sum tests. RESULTS: CT-atlas outperformed MR-atlas for 22/26 OARs. MR-DLC outperformed MR-atlas for all OARs. Highest median (95 %CI) vDSC and sDSC were found for the brainstem in MR-DLC: 0.92 (0.88-0.95) and 0.84 (0.77-0.89) respectively, as well as lowest MSHD: 0.27 (0.22-0.39)cm. Median dose differences (ΔD) were within ± 1 Gy for 24/26(92 %) OARs for all three models. Distance-to-tumor showed a significant correlation for ΔDmax,0.03cc-parameters when splitting the data in ≤ 4 cm and > 4 cm OAR-distance (p < 0.001). CONCLUSION: MR-based DLC and CT-based atlas-contouring enable high-quality segmentation. It was shown that a combination of both CT- and MR-autocontouring models results in the best quality.
Subject(s)
Neoplasms , Organs at Risk , Humans , Radiometry , Radiotherapy Planning, Computer-Assisted/methods , Tomography, X-Ray Computed/methodsABSTRACT
Background: Radiomics can provide in-depth characterization of cancers for treatment outcome prediction. Conventional radiomics rely on extraction of image features within a pre-defined image region of interest (ROI) which are typically fed to a classification algorithm for prediction of a clinical endpoint. Deep learning radiomics allows for a simpler workflow where images can be used directly as input to a convolutional neural network (CNN) with or without a pre-defined ROI. Purpose: The purpose of this study was to evaluate (i) conventional radiomics and (ii) deep learning radiomics for predicting overall survival (OS) and disease-free survival (DFS) for patients with head and neck squamous cell carcinoma (HNSCC) using pre-treatment 18F-fluorodeoxuglucose positron emission tomography (FDG PET) and computed tomography (CT) images. Materials and methods: FDG PET/CT images and clinical data of patients with HNSCC treated with radio(chemo)therapy at Oslo University Hospital (OUS; n = 139) and Maastricht University Medical Center (MAASTRO; n = 99) were collected retrospectively. OUS data was used for model training and initial evaluation. MAASTRO data was used for external testing to assess cross-institutional generalizability. Models trained on clinical and/or conventional radiomics features, with or without feature selection, were compared to CNNs trained on PET/CT images without or with the gross tumor volume (GTV) included. Model performance was measured using accuracy, area under the receiver operating characteristic curve (AUC), Matthew's correlation coefficient (MCC), and the F1 score calculated for both classes separately. Results: CNNs trained directly on images achieved the highest performance on external data for both endpoints. Adding both clinical and radiomics features to these image-based models increased performance further. Conventional radiomics including clinical data could achieve competitive performance. However, feature selection on clinical and radiomics data lead to overfitting and poor cross-institutional generalizability. CNNs without tumor and node contours achieved close to on-par performance with CNNs including contours. Conclusion: High performance and cross-institutional generalizability can be achieved by combining clinical data, radiomics features and medical images together with deep learning models. However, deep learning models trained on images without contours can achieve competitive performance and could see potential use as an initial screening tool for high-risk patients.
ABSTRACT
AIM: To identify the optimal STereotactic Arrhythmia Radioablation (STAR) strategy for individual patients, cardiorespiratory motion of the target volume in combination with different treatment methodologies needs to be evaluated. However, an authoritative overview of the amount of cardiorespiratory motion in ventricular tachycardia (VT) patients is missing. METHODS: In this STOPSTORM consortium study, we performed a literature review to gain insight into cardiorespiratory motion of target volumes for STAR. Motion data and target volumes were extracted and summarized. RESULTS: Out of the 232 studies screened, 56 provided data on cardiorespiratory motion, of which 8 provided motion amplitudes in VT patients (n = 94) and 10 described (cardiac/cardiorespiratory) internal target volumes (ITVs) obtained in VT patients (n = 59). Average cardiac motion of target volumes was < 5 mm in all directions, with maximum values of 8.0, 5.2 and 6.5 mm in Superior-Inferior (SI), Left-Right (LR), Anterior-Posterior (AP) direction, respectively. Cardiorespiratory motion of cardiac (sub)structures showed average motion between 5-8 mm in the SI direction, whereas, LR and AP motions were comparable to the cardiac motion of the target volumes. Cardiorespiratory ITVs were on average 120-284% of the gross target volume. Healthy subjects showed average cardiorespiratory motion of 10-17 mm in SI and 2.4-7 mm in the AP direction. CONCLUSION: This review suggests that despite growing numbers of patients being treated, detailed data on cardiorespiratory motion for STAR is still limited. Moreover, data comparison between studies is difficult due to inconsistency in parameters reported. Cardiorespiratory motion is highly patient-specific even under motion-compensation techniques. Therefore, individual motion management strategies during imaging, planning, and treatment for STAR are highly recommended.
ABSTRACT
OBJECTIVE: Several studies have shown that dual-energy CT (DECT) can lead to improved accuracy for proton range estimation. This study investigated the clinical benefit of reduced range uncertainty, enabled by DECT, in robust optimisation for neuro-oncological patients. METHODS: DECT scans for 27 neuro-oncological patients were included. Commercial software was applied to create stopping-power ratio (SPR) maps based on the DECT scan. Two plans were robustly optimised on the SPR map, keeping the beam and plan settings identical to the clinical plan. One plan was robustly optimised and evaluated with a range uncertainty of 3% (as used clinically; denoted 3%-plan); the second plan applied a range uncertainty of 2% (2%-plan). Both plans were clinical acceptable and optimal. The dose-volume histogram parameters were compared between the two plans. Two experienced neuro-radiation oncologists determined the relevant dose difference for each organ-at-risk (OAR). Moreover, the OAR toxicity levels were assessed. RESULTS: For 24 patients, a dose reduction >0.5/1 Gy (relevant dose difference depending on the OAR) was seen in one or more OARs for the 2%-plan; e.g. for brainstem D0.03cc in 10 patients, and hippocampus D40% in 6 patients. Furthermore, 12 patients had a reduction in toxicity level for one or two OARs, showing a clear benefit for the patient. CONCLUSION: Robust optimisation with reduced range uncertainty allows for reduction of OAR toxicity, providing a rationale for clinical implementation. Based on these results, we have clinically introduced DECT-based proton treatment planning for neuro-oncological patients, accompanied with a reduced range uncertainty of 2%. ADVANCES IN KNOWLEDGE: This study shows the clinical benefit of range uncertainty reduction from 3% to 2% in robustly optimised proton plans. A dose reduction to one or more OARs was seen for 89% of the patients, and 44% of the patients had an expected toxicity level decrease.
Subject(s)
Proton Therapy , Protons , Humans , Proton Therapy/methods , Uncertainty , Tomography, X-Ray Computed/methods , Radiotherapy Planning, Computer-Assisted/methodsABSTRACT
BACKGROUND AND PURPOSE: To date, data used in the development of Deep Learning-based automatic contouring (DLC) algorithms have been largely sourced from single geographic populations. This study aimed to evaluate the risk of population-based bias by determining whether the performance of an autocontouring system is impacted by geographic population. MATERIALS AND METHODS: 80 Head Neck CT deidentified scans were collected from four clinics in Europe (n = 2) and Asia (n = 2). A single observer manually delineated 16 organs-at-risk in each. Subsequently, the data was contoured using a DLC solution, and trained using single institution (European) data. Autocontours were compared to manual delineations using quantitative measures. A Kruskal-Wallis test was used to test for any difference between populations. Clinical acceptability of automatic and manual contours to observers from each participating institution was assessed using a blinded subjective evaluation. RESULTS: Seven organs showed a significant difference in volume between groups. Four organs showed statistical differences in quantitative similarity measures. The qualitative test showed greater variation in acceptance of contouring between observers than between data from different origins, with greater acceptance by the South Korean observers. CONCLUSION: Much of the statistical difference in quantitative performance could be explained by the difference in organ volume impacting the contour similarity measures and the small sample size. However, the qualitative assessment suggests that observer perception bias has a greater impact on the apparent clinical acceptability than quantitatively observed differences. This investigation of potential geographic bias should extend to more patients, populations, and anatomical regions in the future.
Subject(s)
Deep Learning , Humans , Tomography, X-Ray Computed , Algorithms , Observer Variation , Europe , Organs at Risk , Radiotherapy Planning, Computer-AssistedABSTRACT
Purpose: Artificial intelligence applications in radiation oncology have been the focus of study in the last decade. The introduction of automated and intelligent solutions for routine clinical tasks, such as treatment planning and quality assurance, has the potential to increase safety and efficiency of radiotherapy. In this work, we present a multi-institutional study across three different institutions internationally on a Bayesian network (BN)-based initial plan review assistive tool that alerts radiotherapy professionals for potential erroneous or suboptimal treatment plans. Methods: Clinical data were collected from the oncology information systems in three institutes in Europe (Maastro clinic - 8753 patients treated between 2012 and 2020) and the United States of America (University of Vermont Medical Center [UVMMC] - 2733 patients, University of Washington [UW] - 6180 patients, treated between 2018 and 2021). We trained the BN model to detect potential errors in radiotherapy treatment plans using different combinations of institutional data and performed single-site and cross-site validation with simulated plans with embedded errors. The simulated errors consisted of three different categories: i) patient setup, ii) treatment planning and iii) prescription. We also compared the strategy of using only diagnostic parameters or all variables as evidence for the BN. We evaluated the model performance utilizing the area under the receiver-operating characteristic curve (AUC). Results: The best network performance was observed when the BN model is trained and validated using the dataset in the same center. In particular, the testing and validation using UVMMC data has achieved an AUC of 0.92 with all parameters used as evidence. In cross-validation studies, we observed that the BN model performed better when it was trained and validated in institutes with similar technology and treatment protocols (for instance, when testing on UVMMC data, the model trained on UW data achieved an AUC of 0.84, compared with an AUC of 0.64 for the model trained on Maastro data). Also, combining training data from larger clinics (UW and Maastro clinic) and using it on smaller clinics (UVMMC) leads to satisfactory performance with an AUC of 0.85. Lastly, we found that in general the BN model performed better when all variables are considered as evidence. Conclusion: We have developed and validated a Bayesian network model to assist initial treatment plan review using multi-institutional data with different technology and clinical practices. The model has shown good performance even when trained on data from clinics with divergent profiles, suggesting that the model is able to adapt to different data distributions.
ABSTRACT
BACKGROUND AND PURPOSE: We aimed to assess if radiation dose escalation to either the whole primary tumour, or to an 18F-FDG-PET defined subvolume within the primary tumour known to be at high risk of local relapse, could improve local control in patients with locally advanced non-small-cell lung cancer. MATERIALS AND METHODS: Patients with inoperable, stage II-III NSCLC were randomised (1:1) to receive dose-escalated radiotherapy to the whole primary tumour or a PET-defined subvolume, in 24 fractions. The primary endpoint was freedom from local failure (FFLF), assessed by central review of CT-imaging. A phase II 'pick-the-winner' design (alpha = 0.05; beta = 0.80) was applied to detect a 15 % increase in FFLF at 1-year. CLINICALTRIALS: gov:NCT01024829. RESULTS: 150 patients were enrolled. 54 patients were randomised to the whole tumour group and 53 to the PET-subvolume group. The trial was closed early due to slow accrual. Median dose/fraction to the boosted volume was 3.30 Gy in the whole tumour group, and 3.50 Gy in the PET-subvolume group. The 1-year FFLF rate was 97 % (95 %CI 91-100) in whole tumour group, and 91 % (95 %CI 82-100) in the PET-subvolume group. Acute grade ≥ 3 adverse events occurred in 23 (43 %) and 20 (38 %) patients, and late grade ≥ 3 in 12 (22 %) and 17 (32 %), respectively. Grade 5 events occurred in 19 (18 %) patients in total, of which before disease progression in 4 (7 %) in the whole tumour group, and 5 (9 %) in the PET-subvolume group. CONCLUSION: Both strategies met the primary objective to improve local control with 1-year rates. However, both strategies led to unexpected high rates of grade 5 toxicity. Dose differentiation, improved patient selection and better sparing of central structures are proposed to improve dose-escalation strategies.
Subject(s)
Carcinoma, Non-Small-Cell Lung , Lung Neoplasms , Humans , Carcinoma, Non-Small-Cell Lung/diagnostic imaging , Carcinoma, Non-Small-Cell Lung/radiotherapy , Carcinoma, Non-Small-Cell Lung/drug therapy , Fluorodeoxyglucose F18 , Lung Neoplasms/diagnostic imaging , Lung Neoplasms/radiotherapy , Lung Neoplasms/drug therapy , Positron-Emission Tomography/methods , Neoplasm Recurrence, Local , Radiotherapy DosageABSTRACT
Purpose: The current knowledge on biological effects associated with proton therapy is limited. Therefore, we investigated the distributions of dose, dose-averaged linear energy transfer (LETd), and the product between dose and LETd (DLETd) for a patient cohort treated with proton therapy. Different treatment planning system features and visualization tools were explored. Methods and Materials: For a cohort of 24 patients with brain tumors, the LETd, DLETd, and dose was calculated for a fixed relative biological effectiveness value and 2 variable models: plan-based and phenomenological. Dose threshold levels of 0, 5, and 20 Gy were imposed for LETd visualization. The relationship between physical dose and LETd and the frequency of LETd hotspots were investigated. Results: The phenomenological relative biological effectiveness model presented consistently higher dose values. For lower dose thresholds, the LETd distribution was steered toward higher values related to low treatment doses. Differences up to 26.0% were found according to the threshold. Maximum LETd values were identified in the brain, periventricular space, and ventricles. An inverse relationship between LETd and dose was observed. Frequency information to the domain of dose and LETd allowed for the identification of clusters, which steer the mean LETd values, and the identification of higher, but sparse, LETd values. Conclusions: Identifying, quantifying, and recording LET distributions in a standardized fashion is necessary, because concern exists over a link between toxicity and LET hotspots. Visualizing DLETd or dose × LETd during treatment planning could allow for clinicians to make informed decisions.