Introduction: Adequate head and neck (HN) organ-at-risk (OAR) delineation is crucial for HN radiotherapy and for investigating the relationships between radiation dose to OARs and radiation-induced side effects. The automatic contouring algorithms that are currently in clinical use, such as atlas-based contouring (ABAS), leave room for improvement. The aim of this study was to use a comprehensive evaluation methodology to investigate the performance of HN OAR auto-contouring when using deep learning contouring (DLC), compared to ABAS. Methods: The DLC neural network was trained on 589 HN cancer patients. DLC was compared to ABAS by providing each method with an independent validation cohort of 104 patients, which had also been manually contoured. For each of the 22 OAR contours-glandular, upper digestive tract and central nervous system (CNS)-related structures-the dice similarity coefficient (DICE), and absolute mean and max dose differences (|Dmean-dose| and |Dmax-dose|) performance measures were obtained. For a subset of 7 OARs, an evaluation of contouring time, inter-observer variation and subjective judgement was performed. Results: DLC resulted in equal or significantly improved quantitative performance measures in 19 out of 22 OARs, compared to the ABAS (DICE/|Dmean dose|/|Dmax dose|: 0.59/4.2/4.1 Gy (ABAS); 0.74/1.1/0.8 Gy (DLC)). The improvements were mainly for the glandular and upper digestive tract OARs. DLC significantly reduced the delineation time for the inexperienced observer. The subjective evaluation showed that DLC contours were more often preferable to the ABAS contours overall, were considered to be more precise, and more often confused with manual contours. Manual contours still outperformed both DLC and ABAS; however, DLC results were within or bordering the inter-observer variability for the manual edited contours in this cohort. Conclusion: The DLC, trained on a large HN cancer patient cohort, outperformed the ABAS for the majority of HN OARs.

a b s t r a c tBackground and purpose: A planning target volume (PTV) in photon treatments aims to ensure that the clinical target volume (CTV) receives adequate dose despite treatment uncertainties. The underlying static dose cloud approximation (the assumption that the dose distribution is invariant to errors) is problematic in intensity modulated proton treatments where range errors should be taken into account as well. The purpose of this work is to introduce a robustness evaluation method that is applicable to photon and proton treatments and is consistent with (historic) PTV-based treatment plan evaluations. Materials and methods: The limitation of the static dose cloud approximation was solved in a multiscenario simulation by explicitly calculating doses for various treatment scenarios that describe possible errors in the treatment course. Setup errors were the same as the CTV-PTV margin and the underlying theory of 3D probability density distributions was extended to 4D to include range errors, maintaining a 90% confidence level. Scenario dose distributions were reduced to voxel-wise minimum and maximum dose distributions; the first to evaluate CTV coverage and the second for hot spots. Acceptance criteria for CTV D98 and D2 were calibrated against PTV-based criteria from historic photon treatment plans. Results: CTV D98 in worst case scenario dose and voxel-wise minimum dose showed a very strong correlation with scenario average D98 (R 2 > 0.99). The voxel-wise minimum dose visualised CTV dose conformity and coverage in 3D in agreement with PTV-based evaluation in photon therapy. Criteria for CTV D98 and D2 of the voxel-wise minimum and maximum dose showed very strong correlations to PTV D98 and D2 (R 2 > 0.99) and on average needed corrections of À0.9% and +2.3%, respectively. Conclusions: A practical approach to robustness evaluation was provided and clinically implemented for PTV-less photon and proton treatment planning, consistent with PTV evaluations but without its static dose cloud approximation. Ó 2019 The Authors. Published by Elsevier B.V. Radiotherapy and Oncology 141 (2019) 267-274 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).The use of margins in photon radiotherapy is a long established and universally adopted method to provide adequate target coverage under the presence of uncertainties. The CTV-PTV margin provides a geometrical buffer zone around the target within which the desired dose is achieved for the majority of treatments; criteria of 95% of the prescription dose in 90% of the patient population has found general appeal [1,2]. The suitability of a geometricallyexpanded buffer zone arises from the (relative) insensitivity of megavoltage photon dose distributions to density changes in the beam path. By and large, the biggest risk to a photon dose distribution is a geometrical miss -a translation of the CTV relative to the beam. Therefore, the static dose cloud approximation (dose distribution is invariant to errors)...

Background Skeletal muscle depletion or sarcopenia is related to multiple adverse clinical outcome. However, frailty questionnaires are currently applied in the daily practice to identify patients who are potentially (un)suitable for treatment but are time consuming and straining for patients and the clinician. Screening for sarcopenia in patients with head and neck cancer (HNC) could be a promising fast biomarker for frailty. Our objective was to quantify sarcopenia with pre‐treatment low skeletal muscle mass from routinely obtained neck computed tomography scans at level of third cervical vertebra in patients diagnosed with HNC and evaluate its association with frailty. Methods A total of 112 HNC patients with Stages III and IV disease were included from a prospective databiobank. The amount of skeletal muscle mass was retrospectively defined using the skeletal muscle index (SMI). Correlation analysis between SMI and continuous frailty data and the observer agreement were analysed with Pearson's r correlation coefficients. Sarcopenia was present when SMI felt below previously published non‐gender specific thresholds (<43.2 cm2/m2). Frailty was evaluated by Geriatrics 8 (G8), Groningen Frailty Indicator, Timed Up and Go test, and Malnutrition Universal Screening Tool. A univariate and multivariate logistic regression analysis was performed for all patients and men separately to obtain odds ratios (ORs) and 95% confidence intervals (95% CIs). Results The cohort included 82 men (73%) and 30 women (27%), with a total mean age of 63 (±9) years. The observer agreement for cross‐sectional measurements was excellent for both intra‐observer variability (r = 0.99, P < 0.001) and inter‐observer variability (r = 0.98, P < 0.001). SMI correlated best with G8 frailty score (r = 0.38, P < 0.001) and did not differ per gender. Sarcopenia was present in 54 (48%) patients, whereof 25 (46%) men and 29 (54%) women. Prevalence of frailty was between 5% and 54% depending on the used screening tool. The multivariate regression analysis for all patients and men separately isolated the G8 questionnaire as the only independent variable associated with sarcopenia (OR 0.76, 95% CI 0.66–0.89, P < 0.001 and OR 0.76, 95% CI 0.66–0.88, P < 0.001, respectively). Conclusions This is the first study that demonstrates that sarcopenia is independently associated with frailty based on the G8 questionnaire in HNC patients. These results suggest that in the future, screening for sarcopenia on routinely obtained neck computed tomography scans may replace time consuming frailty questionnaires and help to select the (un)suitable patients for therapy, which is highly clinically relevant.