In this study, intra-fraction motion management using gating and baseline shift corrections on a 1.5 T magnetic resonance linear accelerator was evaluated in 93 prostate cancer patients. Motion tracks were analyzed to assess the frequency, size and duration of baseline shift corrections and derive planning target volume (PTV) margins. Baseline shift corrections were required in 89.2% of patients, with a range of 0–6 corrections per fraction. The resulting PTV margins were 1.2 mm left-right, and 2.3 mm anterior-posterior and cranial-caudal. Because these values were conservatively estimated, a 2 mm isotropic PTV margin was clinically implemented without increasing treatment time.
Background and purpose Four-dimensional computed tomography (4DCT) is essential for motion-resolved imaging in radiotherapy, but stack alignment artifacts caused by irregular breathing may compromise anatomical accuracy. This study established a quantitative framework to evaluate reconstruction differences between clinical reference reconstruction (CRR) and stack artifact-reduced reconstruction (SARR) and identify clinically relevant changes. Materials and methods Retrospective 4DCT datasets from 80 lung cancer patients were analyzed, comprising a rated cohort (n = 30) and an independent control cohort (n = 50). Geometric, intensity-based, and combined metrics, including gamma analysis, were calculated across ten respiratory phases for segmented organs-at-risk. Correlations with reader ratings identified metrics that best reflected clinically relevant reconstruction differences. Results Most similarity and intensity-based measures showed limited association with image quality differences (Pearson correlation coefficient r < 0.5). Gamma-based metrics demonstrated the strongest correlation, with relative gamma area > 0.3 reaching r ≈ 0.59 for phase-based data (95% confidence interval 0.30–0.79). Reconstruction differences were predominantly observed in the liver and lower lung lobes, consistent with respiratory motion, while the spinal cord showed minimal variation. Robust z-score thresholding enabled automated categorization of moderate and substantial deviations. In the control cohort, 12% showed substantial and 14% moderate changes for phase-based reconstructions. In the patient with the largest changes, the liver appeared as a double contour in the CRR and as a single contour in the SARR. Conclusions Gamma-based evaluation provided an objective method to detect and localize reconstruction differences in 4DCT, supporting transparent clinical implementation of advanced reconstruction algorithms in radiotherapy.
Background and Purpose Accurate prostate segmentation in magnetic resonance imaging is critical for prostate cancer radiotherapy, but manual annotation is time-consuming and exhibits high inter-rater variability. This study proposes a Bayesian deep learning framework that improves prostate segmentation under annotation noise and inter-rater variability while providing uncertainty estimation. Materials and methods Using the prostate cancer imaging dataset (140 subjects), simulated prostate annotations modeled boundary ambiguity, annotator subjectivity, and annotation tool-related artifacts across three scenarios: single rater with distinct simulated annotations (S1), majority-vote consensus from three raters (S2), and multi-annotation of all rater masks with majority voting (S3). The framework used Flipout convolutional layers for uncertainty estimation, trained with a combined Kullback-Leibler divergence and Focal Tversky loss. Bayesian learning was implemented first in a standard U-Net and then in nnU-Net to evaluate generalization across backbones. Comparisons included deterministic counterparts, Swin UNETR, and variants with Monte Carlo Dropout, Label Smoothing, and Test-Time Augmentation. Results Bayesian U-Net achieved Dice Similarity Coefficients of 0.730, 0.782, and 0.780 in S1, S2, and S3, respectively, and the lowest Expected Calibration Error in S1 (0.065). Bayesian nnU-Net achieved the highest Dice Similarity Coefficients among nnU-Net-based methods in all scenarios (0.879, 0.903, and 0.900) and the lowest Expected Calibration Error in all scenarios. Predictive uncertainty was consistently higher in incorrectly than in correctly segmented voxels. Conclusions Bayesian learning improves segmentation robustness under multi-rater noisy labels, across both U-Net and nnU-Net while providing voxel-wise uncertainty estimates and may support clinical decision-making and enhance treatment accuracy.
Background Variations in relative biological effectiveness (RBE) and linear energy transfer (LET) in proton therapy may contribute to unexpected side effects in organs of interest (OOIs). This study assessed the feasibility of LET optimization in intensity-modulated proton therapy (IMPT) for sinonasal cancer, aiming to reduce dose-averaged LET (LETd). Materials and methods Robust optimized IMPT reference plans (Ref) were created for 16 patients (clinical target volume (CTV) doses: 64–70 Gy[RBE1.1]) assuming a fixed RBE = 1.1, 2 mm setup and 3.5% range uncertainties. LET optimized plans (LETopt) were generated from Ref plans using identical dose guidance, with additional LETd objectives (LETd < 3 keV/μm at ≥30 Gy[RBE1.1]) to the brainstem, chiasm and optic nerves. Plans were compared in terms of CTV coverage, OOI dose (RBE1.1), mean LETd (LETdmean) and variable RBE-weighted doses (RBEVAR). Results All Ref and LETopt plans met the CTV dose criteria. The median LETdmean across the patients was reduced in the chiasm (3.1 vs 2.3 keV/μm, P = 0.009), right optic nerve (3.1 vs 2.5 keV/μm, P = 0.016), and left optic nerve (2.7 vs 2.7 keV/μm, P = 0.023). RBEVAR doses (D0.03cm3) to OOIs were also significantly decreased, achieving patient-specific reductions from >56 Gy (RBEVAR) to below the 54 Gy (RBEVAR) threshold in the brainstem, chiasm, and optic nerves. Conclusions LET optimization significantly reduced LETdmean and RBEVAR doses to critical OOIs without compromising CTV coverage. However, its effectiveness was patient specific, with notable variability in RBEVAR D0.03cm3 doses across models, underscoring the need for individualized assessment.
