To decrease the recurrence rate after complete resection of a brain metastasis, removal of a surgical safety margin is advocated. This is not always feasible when resecting a metastasis in an eloquent location. We aimed to assess the recurrence rate after resection of metastases in an eloquent location followed by postoperative stereotactic radiotherapy to the resection cavity. We retrospectively included patients with 1–3 brain metastases undergoing gross total resection and postoperative stereotactic radiotherapy between 2010 and 2022. Primary endpoint was local recurrence free survival (LRFS). Secondary endpoints were overall survival and distant brain failure free survival. Patients were grouped according to the location of their metastasis into eloquent and non-eloquent. Eloquent localization was considered a surrogate for resection without a surgical safety margin according to our institutional practice. We included 193 patients with 201 resected metastases. Ninety-five metastases (47.3

Introduction: Expression of programmed death-ligand 1 (PD-L1) is the only routinely used tissue biomarker for predicting response to programmed cell death protein 1/ PD-L1 inhibitors. It is to date unclear whether PD-L1 expression is preserved in brain metastases (BMs).Methods: In this single-center, retrospective study, we evaluated PD-L1 expression using the SP263 assay in consecutively resected BMs of lung carcinomas and paired primary tumors, diagnosed from 2000 to 2015, with correlation to clinicopathological and molecular tumor and patient characteristics.Results: PD-L1 tumor proportional score (TPS) could be evaluated on whole tissue slides in 191 BMs and 84 paired primary lung carcinomas. PD-L1 TPS was less than 1% in 113 of 191 (59.2%), 1% to 49% in 34 of 191 (17.8%), and greater than or equal to 50% in 44 of 191 (23.0%) BMs. TPS was concordant between BMs and paired primary lung carcinomas in most cases, with discordance regarding the clinically relevant cutoffs at 1% and 50% in 18 of 84 patients (21.4%). Four of 18 discordant cases had no shared mutations between the primary lung carcinoma and BM. Intratumoral heterogeneity, as assessed using tissue microarray cores, was only significant at the primary site (PWilcoxon signed rank = 0.002) with higher PD-L1 TPS at the infiltration front (mean = 40.4%, interquartile range: 0%- 90%). Neither TPS greater than or equal to 1% nor TPS greater than or equal to 50% nor discordance between the primary lung carcinoma and BMs had prognostic significance regarding overall survival or BM-specific overall survival. Conclusions: PD-L1 expression was mostly concordant between primary lung carcinoma and its BM and between resections of BM and stereotactic biopsies, mirrored by tissue microarray cores. Differences in PD-L1 TPS existed primarily in cases with TPS greater than 10%, for which also human assessment tends to be most error prone.(c) 2022 The Authors. Published by Elsevier Inc. on behalf of the International Association for the Study of Lung Cancer. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). treatment death-ligand with

Background & AimsHigh-dose irradiation is an essential tool to help control the growth of hepatic tumors, but it can cause radiation-induced liver disease (RILD). This life-threatening complication manifests itself months following radiation therapy and is characterized by fibrosis of the pericentral sinusoids. In this study, we aimed to establish a mouse model of RILD to investigate the underlying mechanism of radiation-induced liver fibrosis.MethodsUsing a small animal image-guided radiation therapy platform, an irradiation scheme delivering 50 Gy as a single dose to a focal point in mouse livers was designed. Tissues were analyzed 1 and 6 days, and 6 and 20 weeks post-irradiation. Irradiated livers were assessed by histology, immunohistochemistry, imaging mass cytometry and RNA sequencing. Mitochondrial function was assessed using high-resolution respirometry.ResultsAt 6 and 20 weeks post-irradiation, pericentral fibrosis was visible in highly irradiated areas together with immune cell infiltration and extravasation of red blood cells. RNA sequencing analysis showed gene signatures associated with acute DNA damage, p53 activation, senescence and its associated secretory phenotype and fibrosis. Moreover, gene profiles of mitochondrial damage and an increase in mitochondrial DNA heteroplasmy were detected. Respirometry measurements of hepatocytes in vitro confirmed irradiation-induced mitochondrial dysfunction. Finally, the highly irradiated fibrotic areas showed markers of reactive oxygen species such as decreased glutathione and increased lipid peroxides and a senescence-like phenotype.ConclusionsBased on our mouse model of RILD, we propose that irradiation-induced mitochondrial DNA instability contributes to the development of fibrosis via the generation of excessive reactive oxygen species, p53 pathway activation and a senescence-like phenotype.Lay summaryIrradiation is an efficient cancer therapy, however, its applicability to the liver is limited by life-threatening radiation-induced hepatic fibrosis. We have developed a new mouse model of radiation-induced liver fibrosis, that recapitulates the human disease. Our model highlights the role of mitochondrial DNA instability in the development of irradiation-induced liver fibrosis. This new model and subsequent findings will help increase our understanding of the hepatic reaction to irradiation and to find strategies that protect the liver, enabling the expanded use of radiotherapy to treat hepatic tumors.

Background The management of meningiomas is challenging, and the role of postoperative radiotherapy is not standardized. Methods Radiation oncology experts in Swiss centres were asked to participate in this decision-making analysis on the use of postoperative radiotherapy (RT) for meningiomas. Experts from ten Swiss centres agreed to participate and provided their treatment algorithms. Their input was converted into decision trees based on the objective consensus methodology. The decision trees were used as a basis to identify consensus and discrepancies in clinical routine. Results Several criteria used for decision-making in postoperative RT in meningiomas were identified: histological grading, resection status, recurrence, location of the tumour, zugzwang (therapeutic need to treat and/or severity of symptoms), size, and cell division rate. Postoperative RT is recommended by all experts for WHO grade III tumours as well as for incompletely resected WHO grade II tumours. While most centres do not recommend adjuvant irradiation for WHO grade I meningiomas, some offer this treatment in recurrent situations or routinely for symptomatic tumours in critical locations. The recommendations for postoperative RT for recurrent or incompletely resected WHO grade I and II meningiomas were surprisingly heterogeneous. Conclusions Due to limited evidence on the utility of postoperative RT for meningiomas, treatment strategies vary considerably among clinical experts depending on the clinical setting, even in a small country like Switzerland. Clear majorities were identified for postoperative RT in WHO grade III meningiomas and against RT for hemispheric grade I meningiomas outside critical locations. The limited data and variations in clinical recommendations are in contrast with the high prevalence of meningiomas, especially in elderly individuals.

Background: Brain metastases present a significant complication in lung cancer with an unmet therapeutic need. Methods: In this single-centre, retrospective study, we genotyped a clinico-pathologically well-annotated cohort of consecutively resected brain metastases of lung adenocarcinomas and paired primary tumours, diagnosed from 2000 to 2015, using the Ion Torrent Oncomine Comprehensive Cancer Panel v3. Results: Among 444 consecutive brain metastases, 210 (49%) originated from lung cancer. Analysis was successful in 111 samples, including 54 pairs of brain metastasis and primary tumour. Most driver alterations were preserved in brain metastases. Private alterations exclusive to primary tumours, brain metastases or both sites (intersecting cases) were present in 22%, 26% and 26% of cases, respectively. Seven percent had no shared mutations. KRAS mutations were more frequent in primary tumours metastasised to the brain (32/55, 58%) compared to TCGA (33%, p < 0.005) and own data from routine diagnostics, independent from clinical or pathological characteristics. Fourteen cases showed alterations in the EGFR signalling pathway including additional KRAS alterations that were private to brain metastases. KRAS G12C was detected most frequently (26% of patients) and KRAS G12C and G13C variants were significantly enriched in brain metastases. Synchronous and metachronous cases had a similar mutation profile. Conclusions: Our results suggest an important role of KRAS alterations in the pathobiology of brain metastases from lung adenocarcinomas. This has direct therapeutic implications as inhibitors selectively targeting KRAS G12C are entering the clinics. (C) 2021 The Authors. Published by Elsevier Ltd.

Objective:To assess the factors affecting early local and audiometric outcomes in vestibular schwannoma (VS) patients treated with stereotactic radiosurgery (SRS).Study Design:A retrospective review of medical records.Setting:Tertiary referral center.Patients:Records of all adult patients who underwent SRS between 2010 and 2016 for the treatment of VS were retrospectively reviewed. Patients treated with microsurgery or multi-fractionation schemes, and those who had neurofibromatosis type 2, were excluded.Intervention:SRS, tumor volume/size measurements.Main Outcome Measures:The impact of tumor volume dynamics on the early local and hearing-related outcomes, together with the factors that influence them following SRS, and comparison of different tumor size measurement methods.Results:From 2010 to 2016, 53 patients underwent single fraction SRS of 12 Gy. Median follow-up time was 32 months (range, 6-79). At the last follow-up, only one patient had clinical progression. Age less than or equal to 65 years (p = 0.04; odds ratio [OR]: 0.17; 95% confidence interval [CI]: 0.03-0.93) and baseline pure-tone average (PTA) level less than or equal to 30 dB (p = 0.03; OR: 0.90; 95% CI: 0.84-0.96) were associated with maintenance of serviceable hearing. On multivariate analysis, PTA remained significant (p = 0.01; OR: 0.04; 95% CI: 0.003-0.45). In patients with a loss of serviceable hearing, the mean volume increase tended to be higher than in the patients whose hearing was maintained. The linear measurement method underestimated, and the A x B x C/2 equation overestimated, the radiological progression compared with 3D-volumetric delineations.Conclusion:During the median observation period of almost 3 years, we reported our early outcome results. Tumor volume increase may have an impact on serviceable hearing loss after SRS. Currently there is no widely accepted method for the evaluation of post-SRS response. Linear measurement and the A x B x C/2 equation produce less reliable estimates of radiological progression compared with 3D-volumetric delineations. Accurate volume measurements with 3D delineations should be considered as part of clinical routine for assessing progression and deciding on salvage therapies.

Background Automated brain tumor segmentation methods are computational algorithms that yield tumor delineation from, in this case, multimodal magnetic resonance imaging (MRI). We present an automated segmentation method and its results for resection cavity (RC) in glioblastoma multiforme (GBM) patients using deep learning (DL) technologies. Methods Post-operative, T1w with and without contrast, T2w and fluid attenuated inversion recovery MRI studies of 30 GBM patients were included. Three radiation oncologists manually delineated the RC to obtain a reference segmentation. We developed a DL cavity segmentation method, which utilizes all four MRI sequences and the reference segmentation to learn to perform RC delineations. We evaluated the segmentation method in terms of Dice coefficient (DC) and estimated volume measurements. Results Median DC of the three radiation oncologist were 0.85 (interquartile range [IQR]: 0.08), 0.84 (IQR: 0.07), and 0.86 (IQR: 0.07). The results of the automatic segmentation compared to the three different raters were 0.83 (IQR: 0.14), 0.81 (IQR: 0.12), and 0.81 (IQR: 0.13) which was significantly lower compared to the DC among raters (chi-square = 11.63, p = 0.04). We did not detect a statistically significant difference of the measured RC volumes for the different raters and the automated method (Kruskal-Wallis test: chi-square = 1.46, p = 0.69). The main sources of error were due to signal inhomogeneity and similar intensity patterns between cavity and brain tissues. Conclusions The proposed DL approach yields promising results for automated RC segmentation in this proof of concept study. Compared to human experts, the DC are still subpar.

•We conducted a Swiss survey on patterns of practice during COVID-19 pandemic.•Dedicated IT solutions were implemented in all radiation-oncology departments.•Upfront endocrine treatments for breast and prostate cancer were privileged.•Use of hypofractionation for breast, rectal cancer and palliation was increased.

Deep learning for regression tasks on medical imaging data has shown promising results. However, compared to other approaches, their power is strongly linked to the dataset size. In this study, we evaluate 3D-convolutional neural networks (CNNs) and classical regression methods with hand-crafted features for survival time regression of patients with high-grade brain tumors. The tested CNNs for regression showed promising but unstable results. The best performing deep learning approach reached an accuracy of 51.5% on held-out samples of the training set. All tested deep learning experiments were outperformed by a Support Vector Classifier (SVC) using 30 radiomic features. The investigated features included intensity, shape, location and deep features. The submitted method to the BraTS 2018 survival prediction challenge is an ensemble of SVCs, which reached a cross-validated accuracy of 72.2% on the BraTS 2018 training set, 57.1% on the validation set, and 42.9% on the testing set. The results suggest that more training data is necessary for a stable performance of a CNN model for direct regression from magnetic resonance images, and that non-imaging clinical patient information is crucial along with imaging information.

Second-generation multi leaf collimator (MLC) has been introduced for the CyberKnife Robotic Radiosurgery System (Accuray Inc., Sunnyvale, USA) in 2016. The aim of the present study was to compare the plan quality between InCise 2 MLC and IRIS by analyzing dose parameters for small and large targets such as brain metastases and liver tumors, respectively. Ten brain metastases targets and 10 liver targets were selected. The following parameters were analyzed: coverage, dose distributions in planning target volume (PTV) and organs at risk (OAR), conformity index (CI), Nakamura CI (nCI), homogeneity index (HI), gradient index (GI), V12 Gy within brain, treatment time, number of beams, nodes and monitor units (MU). For brain patient group, CI and nCI were lower with IRIS plans (1.18 versus 1.26; p = 0.048 and 1.19 versus 1.28; p = 0.033, respectively), whereas GI values were higher in InCise 2 MLC plans (0.36 versus 0.33; p < 0.01). PTV mean doses were higher with IRIS plans (1.13 versus 1.16; p = 0.03). For liver tumors, InCise 2 MLC plans reached higher GI values (0.32 versus 0.29; p < 0.01). PTV minimum and mean doses were higher with IRIS plans (0.85 versus 0.93; p < 0.01 and 1.17 versus 1.2; p = 0.046, respectively). OAR’s mean dose for bowel, left-right kidney and healthy liver tissue were lower with MLC plans (0.02 versus 0.04, 0.02 versus 0.05, 0.07 versus 0.10, and 0.20 versus 0.21; p < 0.01, respectively). Shorter delivery time was reached with InCise 2 MLC for both groups (38 versus 56 min, and 42 versus 56 min; p < 0.01, for brain and liver, respectively). Results of this study showed that second generation MLC plans for intracranial and liver targets were dosimetrically feasible. The main advantages of InCise 2 MLC are the improved dose gradient and delivery time.

AIMS Recurrent glioblastoma (GBM) is a disease with poor prognosis. Although several therapeutic approaches such as chemotherapy, irradiation or surgery have been investigated, there is no established standard therapy. A recent survey among Swiss neuro-oncology centres has shown considerable controversy in the treatment recommendations for any specific scenario of recurrent GBM. In view of the cost differences of the available treatment alternatives, the aim of our study was assess the financial impact of different institutional therapeutic strategies for recurrent GBM in Switzerland. METHODS We created a decision analytic model for each of the eight centres participating in the initial study with a centre-specific treatment algorithm to evaluate the average treatment cost per patient. The probability of decision criteria was varied by univariate and probabilistic sensitivity analysis over a wide range to account for the high level of uncertainty. Treatment costs were calculated from the perspective of the Swiss healthcare payer. RESULTS Mean treatment costs per patient calculated on the basis of the institutional treatment algorithms ranged from CHF 13,748 to CHF 22,072 depending on the probability of individual decision criteria. The most influential decision factors for the mean treatment costs were the probability of fit patients and the proportion of patients with resectable tumour recurrences. There was a significant correlation between the complexity of treatment algorithms in a centre and the resulting mean treatment costs. CONCLUSIONS Institutional treatment algorithms can be used to estimate the average treatment costs per patient, which are, however, highly sensitive to probability changes of individual decision criteria. Our study demonstrates a high variability in treatment costs for recurrent GBM among eight Swiss neuro-oncology centres based on individual institutional treatment algorithms.

Background In order to locate an arteriovenous malformation, typically, a digital subtraction angiography (DSA) is carried out. To use the DSA for target definition an accurate image registration between CT and DSA is required. Carrying out a non-invasive, frameless procedure, registration of the 2D-DSA images with the CT is critical. A new software prototype is enabling this frameless procedure. The aim of this work was to evaluate the prototype in terms of targeting accuracy and reliability based on phantom measurements as well as with the aid of patient data. In addition, the user’s ability to recognize registration mismatches and quality was assessed. Methods Targeting accuracy was measured with a simple cubic, as well as with an anthropomorphic head phantom. Clearly defined academic targets within the phantoms were contoured on the CT. These reference structures were compared with the structures generated within the prototype. A similar approach was used with patient data, where the clinically contoured target served as the reference structure. An important error source decreasing the target accuracy comes from registration errors between CT and 2D-DSA. For that reason, the tools in BC provided to the user to check these registrations are very important. In order to check if the user is able to recognize registration errors, a set of different registration errors was introduced to the correctly registered CT and 2D-DSA image data sets of three different patients. Each of six different users rated the whole set of registrations within the prototype. Results The target accuracy of the prototype was found to be below 0.04 cm for the cubic phantom and below 0.05 cm for the anthropomorphic head phantom. The mean target accuracy for the 15 patient cases was found to be below 0.3 cm. In the registration verification part, almost all introduced registration errors above 1° or 0.1 cm were detected by the six users. Nevertheless, in order to quantify and categorize the possibility to detect mismatches in the registration process more data needs to be evaluated. Conclusion Our study shows, that the prototype is a useful tool that has the potential to fill the gap towards a frameless procedure when treating AVMs with the aid of 2D-DSA images in radiosurgery. The target accuracy of the prototype is similar to other systems already established in clinical routine.

Background: Whole-ventricular radiotherapy (WV-RT) followed by a boost to the tumor bed (WV-RT/TB) is recommended for intracranial germ cell tumors (IGCT). As the critical brain areas are mainly in the target volume vicinity, it is unclear if protons indeed substantially spare neurofunctional organs at risk (NOAR). Therefore, a dosimetric comparison study of WV-RT/TB was conducted to assess whether proton or photon radiotherapy achieves better NOAR sparing. Methods: Eleven children with GCT received 24 Gy(RBE) WV-RT and a boost up to 40 Gy(RBE) in 25 fractions of 1.6 Gy(RBE) with pencil beam scanning proton therapy (PBS-PT). Intensity-modulated radiotherapy (IMRT) and volumetric-modulated arc therapy (VMAT) plans were generated for these patients. NOAR were delineated and treatment plans were compared for target volume coverage (TVC), homogeneity index (HI), inhomogeneity coefficient (IC) and (N)OAR sparing. Results: TVC was comparable for all three modalities. Compared to IMRT and VMAT, PBS-PT showed statistically significant optimized IC, as well as dose reduction, among others, in mean and integral dose to the: normal brain (-35.2%, -32.7%; -35.2%, -33.0%, respectively), cerebellum (-53.7%, -33.1%; -53.6%, -32.7%) and right temporal lobe (-14.5%, -31.9%; -14.7%, -29.9%). The Willis' circle was better protected with PBS-PT than IMRT (-7.1%; -7.8%). The left hippocampus sparing was higher with IMRT. Compared to VMAT, the dose to the hippocampi, amygdalae and temporal lobes was significantly decreased in the IMRT plans. Conclusions: Dosimetric comparison of WV-RT/TB in IGCT suggests PBS-PT's advantage over photons in conformality and NOAR sparing, whereas IMRT's superiority over VMAT, thus potentially minimizing long-term sequelae. (C) 2019 The Authors. Published by Elsevier B.V. on behalf of European Society for Radiotherapy and Oncology.

Abstract Background Sarcopenia, the critical depletion of skeletal muscle mass, is an independent prognostic factor in several tumor entities for treatment-related toxicity and survival. In esophageal cancer, there have been conflicting results regarding the value of sarcopenia as prognostic factor, which may be attributed to the heterogeneous patient populations and the retrospective nature of previous studies. The aim of our study was therefore to determine the impact of sarcopenia on prospectively collected specific outcomes in a subgroup of patients treated within the phase III study SAKK 75/08 with trimodality therapy (induction chemotherapy, radiochemotherapy and surgery) for locally advanced esophageal cancer. Methods Sarcopenia was assessed by skeletal muscle index at the 3rd lumbar vertebra (L3) in cross-sectional computed tomography scans before induction chemotherapy, before radiochemotherapy and after neoadjuvant therapy in a subgroup of 61 patients from four centers in Switzerland. Sarcopenia was determined by previously established cut-off values (Martin et al., PMID: 23530101) and correlated with prospectively collected outcomes including treatment-related toxicity, postoperative morbidity, treatment feasibility and survival. Results Using the published cut-off values, the prevalence of sarcopenia increased from 29.5% before treatment to 63.9% during neoadjuvant therapy (p < 0.001). Feasibility of neoadjuvant therapy and surgery was not different in initially sarcopenic and non-sarcopenic patients. We observed in sarcopenic patients significantly increased grade ≥ 3 toxicities during chemoradiation (83.3% vs 52.4%, p = 0.04) and a non-significant trend towards increased postoperative complications (66.7% vs 42.9%, p = 0.16). No difference in survival according to sarcopenia could be observed in this small study population. Conclusions Trimodality therapy in locally advanced esophageal cancer is feasible in selected patients with sarcopenia. Neoadjuvant chemoradiation increased the percentage of sarcopenia. Sarcopenic patients are at higher risk for increased toxicity during neoadjuvant radiochemotherapy and showed a non-significant trend to more postoperative morbidity.

Stereotactic body radiation therapy is an effective and safe treatment modality for bone metastasis which allows clinicians to accurately target lesions to high doses while minimizing dose to organs at risk. The commercially available CyberKnife® Xsight™ Spine Tracking System (Accuray, Inc., Sunnyvale, CA) tracks static skeletal structures and eliminates the need for implanted fiducial markers (FMs). However, the Xsight™ Spine Tracking system is not appropriate for bone metastases outside the spine, which are moving due to respiration and ,typically, FMs have to be implanted close to the lesion. These FMs will be used to track the dynamic target. For targets close to the surface, non-invasive fixation of the FMs to the patient's skin could be an option.

Arteriovenous malformations (AVMs) are rare congenital vascular pathologies. The reported overall annual hemorrhage rate is 3.0%, for unruptured AVMs it is 2.2%, and for ruptured AVMs, 4.5%. The main goal of AVM treatment is to prevent intracerebral hemorrhage. This is achieved by complete nidus eradication. Interventional treatment options include microsurgery, embolization and radiosurgery, as well as multimodal approaches. Radiosurgery is a safe and effective alternative to surgery or embolization, especially for AVMs located in deep or eloquent brain regions, where invasive treatment cannot be performed. With the introduction of the Leksell Gamma Knife, AVMs became one of the most common indications for radiosurgical interventions (nearly 30% of the first 15-year experience). The current review discusses the role of radiosurgery in the treatment of AVMs, with a focus on outcome predictors and a discussion of the relevant literature.

Stereotactic radiosurgery (SRS) is an effective treatment for vestibular schwannoma (VS). Three-dimensional (3D) constructive interference in steady state (CISS) is the preferred magnetic resonance imaging (MRI) sequence for evaluating signal changes in the inner ear endolymph. Previous studies demonstrated a correlation between pretreatment cochlear signal intensity in 3D-CISS and posttherapeutic hearing outcomes. The purpose of our study was to compare 3D-CISS sequences before and after primary SRS of unilateral VSs to evaluate the effect of radiosurgery on the 3D-CISS signal intensities of cochlea and sacculus/utriculus.

Whole ventricular irradiation (WV-RT) followed by a boost to the tumor bed (WV-RT/TB) is recommended for localized intracranial germ cell tumors (GCT). As the eloquent brain areas are mainly in direct vicinity of the target volume, it is unknown if proton therapy indeed substantially spares these organs at risk (OAR). Therefore, a dosimetric comparison study of WV-RT/TB was conducted to assess whether protons or modern photon radiotherapy achieve better critical organ sparing. Eleven children with GCT received 24 Gy(RBE) WV-RT and a boost up to 40 Gy(RBE) in 25 fractions of 1.6 Gy(RBE) with pencil beam scanning proton therapy (PBS-PT). Respective intensity-modulated radiation therapy (IMRT) plans were generated for these patients. Additional critical structures for neurocognition (NCS) have been delineated and plans compared for target volume coverage (TVC) (homogeneity index (HI) and inhomogeneity coefficient (IC)), NCS and OAR sparing. TVC was similar for both modalities. Compared to IMRT, PBS-PT showed statistically significant dose reduction (p<0.05) in: maximum (Dmax), mean (Dmean) and integral dose (ID) of the normal brain (2.6%, 35.4%, 35.7%); Dmean of the Willis’ circle (6.7%), and brainstem (7.4%). Likewise, the volume receiving ≥20 Gy (V20Gy) of the right (24.2%) and left (20.9%) temporal lobes was significantly decreased. No significant difference was observed for the dose-metrics/hippocampus. Dosimetric comparison of WV-RT/TB in GCT demonstrates PBS-PT’s advantage over IMRT in critical organ sparing, while keeping TVC the same. PBS-PT may decrease the likelihood of vascular/neurological sequelae, as well as the risk of radio-induced secondary malignancies.

Uncertainty estimation methods are expected to improve the understanding and quality of computer-assisted methods used in medical applications (e.g., neurosurgical interventions, radiotherapy planning), where automated medical image segmentation is crucial. In supervised machine learning, a common practice to generate ground truth label data is to merge observer annotations. However, as many medical image tasks show a high inter-observer variability resulting from factors such as image quality, different levels of user expertise and domain knowledge, little is known as to how inter-observer variability and commonly used fusion methods affect the estimation of uncertainty of automated image segmentation. In this paper we analyze the effect of common image label fusion techniques on uncertainty estimation, and propose to learn the uncertainty among observers. The results highlight the negative effect of fusion methods applied in deep learning, to obtain reliable estimates of segmentation uncertainty. Additionally, we show that the learned observers' uncertainty can be combined with current standard Monte Carlo dropout Bayesian neural networks to characterize uncertainty of model's parameters.