Purpose Quality assurance (QA) of auto-segmentations is necessary for safe clinical deployment. Deep learning models are optimised for data similar to their training set, but reliability drops with out-of-distribution (OOD) inputs. This study evaluates whether uncertainty quantification (UQ) can detect OOD inputs arising from distribution shift in head and neck organ of interest (OOI) auto-segmentation. Materials and methods Monte Carlo dropout-based uncertainty scores and geometric accuracy of nnU-Net models were evaluated in two experiments. First, as a proof-of-concept, we assessed whether uncertainty scores detect metal artefact (N = 10) from non-artefact cases (N = 10) on a baseline model (mixed training data, N = 620) and a no-metal model (artefact-free training data, N = 620). Second, in a clinically realistic experiment, we split a fine-tuning dataset by uncertainty score to obtain a narrow model (low-uncertainty cases, N = 21) and a broad model (high-uncertainty cases, N = 21). A held-out test set (N = 76) evaluated whether segmentation accuracy differed between uncertainty categories. Results Both experiments showed that uncertainty scores reflect the learned training distribution. In the supervised experiment, the no-metal model demonstrated greater uncertainty score separation between artefact and non-artefact cases (1.1 × 105 vs 0.6 × 105) than the baseline model (1.1 × 105 vs 0.8 × 105). In the unsupervised experiment, the narrow model performed significantly better on certain than uncertain cases (p = 0.04), while no difference was seen for the broad model (p = 0.34), which did include these cases in training. Conclusions Uncertainty quantification can detect distribution shifts in radiotherapy auto-segmentation. This could support automated QA workflows by identifying in-distribution cases and enabling the application of specialised models.
Purpose To evaluate the efficacy and safety of carbon ion radiotherapy (CIRT) in patients with non-metastatic, inoperable pancreatic cancer in a prospective phase 2 setting. Methods In this single-center investigator-initiated phase 2 trial, 25 patients with locally advanced or locally recurrent pancreatic cancer were included. CIRT was performed using an intensity-controlled rasterscanning system for beam application [total dose: 48 Gy (RBE), 12 fractions]. Concurrent chemotherapy with Gemcitabine 300 mg/m2 was administered weekly. The primary endpoint was the 1-year OS rate, defined as time from study inclusion. 1y- and 2y-OS and −PFS were calculated using the Kaplan-Meier method, 95 % confidence intervals are shown. Toxicity was evaluated using CTCAEv5.0. Results Twenty-three patients were included in the final analysis. After the completed 12-month follow-up of the last patient, the 1- and 2-year OS rates were 78 % [CI 63 %; 97 %] and 59 % [CI 41 %; 84 %]. The median OS was 26.5 months. The 1- and 2-year PFS rates were 26 % [CI 13 %; 52 %] and 10 % [CI 3 %; 37 %] with a median PFS of 8.1 months. Study treatment was well tolerated with three treatment-related CTCAE grade 3 toxicities, including one case of abdominal pain, one upper gastrointestinal hemorrhage and one ureteric stricture. No treatment-related CTCAE grade 4–5 toxicities were observed. Conclusion Chemoradiation with carbon ions demonstrated encouraging overall survival and acceptable toxicity in non-metastatic pancreatic cancer patients. A subsequent randomized Phase 3 trial should investigate OS superiority of CIRT compared to conventional chemoradiotherapy or chemotherapy.
Background Locoregional recurrence (LRR) in oropharyngeal squamous cell carcinoma (OPSCC) patients treated with radiotherapy (RT) most often occurs within the high-dose target volume. Anatomical changes during RT may potentially lead to unintended target underdosage, which could contribute to the risk of treatment failure. This study aimed to quantify differences between planned and accumulated dose in the regions of later LRR. Methods Patients treated with curatively intended RT for OPSCC between 2009 and 2017 at Rigshospitalet with biopsy-proven LRR were identified. Recurrence volumes were delineated on follow up imaging and mapped to the planning CT. The point of origin (PO) of each recurrence was determined using the nidus method. Dose accumulation was performed using synthetic CTs based on daily cone beam CT, followed by recalculation and summation of daily dose distributions. Planned and accumulated doses at the PO were compared. Additional analyses evaluated effect of deformable registration methods. Results Forty patients with 54 recurrence volumes were included. The accumulated dose was higher than the planned dose in 57% of recurrences. The mean difference between accumulated and planned dose at the PO was − 0.04 Gy (SD: 0.81; 95% CI: −0.26 to 0.18 Gy; p = 0.72). Neither PO localization nor the dose accumulation process was sensitive to the use of an alternative deformation software. Conclusions Within the evaluated setting, no evidence was found that regions of later LRR were systematically underdosed. Dose differences were small and within clinically acceptable limits. This suggests that anatomical changes during RT were unlikely to cause clinically relevant compromised target dose coverage.
Purpose Sleep-related symptoms in head and neck cancer are poorly characterized and often under-recognized. We examined symptom trajectories, prevalence of moderate/severe symptoms, and associations between severe sleep-related patient-reported symptoms and overall survival (OS) in oropharyngeal cancer (OPC) patients receiving radiotherapy (RT). Methods This retrospective analysis included a longitudinal cohort of 372 patients with OPC treated with curative-intent RT (± chemotherapy) at a tertiary center. Symptoms were assessed at the earliest available pre-RT baseline, RT start, RT end, and 6 months post-RT using the MD Anderson Symptom Inventory–Head and Neck (MDASI-HN). The sleep-related symptom domains were fatigue, sleep disturbance, and drowsiness. OS was evaluated using Kaplan–Meier and multivariable Cox models. Results Severe fatigue, sleep disturbance, and drowsiness were infrequent at baseline but peaked at RT end (26%, 15%, and 18%, respectively) and returned to near-baseline levels by 6 months (6%, 4%, and 2%). Severe longitudinal fatigue (HR 5.05), sleep disturbance (HR 3.86), and drowsiness (HR 8.24) were independently associated with worse OS. Severe symptoms at RT start were also associated with worse OS (fatigue HR 3.58; sleep disturbance HR 5.08; drowsiness HR 7.76; all p < 0.05). Aside from age, severe average daily symptom scores and severe symptoms at RT start were associated with OS after multivariable adjustment. Conclusion Severe sleep-related symptoms peaked during RT. Severe fatigue, sleep disturbance, and drowsiness, both longitudinally and at RT start, were associated with worse OS. These observational associations do not establish causality; symptom assessment may help identify patients who warrant supportive evaluation.
