Diffusion kurtosis imaging (DKI) has been shown to be valuable in a wide range of neuroscientific and clinical applications. However, reliable estimation of DKI tensors is often compromised by noise, especially for the kurtosis tensor (KT). Here, we propose a joint denoising and estimating framework that integrates multiple sources of prior information, including nonlocal structural self-similarity (NSS), local spatial smoothness (LSS), physical relevance (PR) of the DKI model, and noise characteristics of magnitude diffusion MRI (dMRI) images for improved estimation of DKI tensors. The local and nonlocal spatial smoothing constraints are complementary to each other, making the proposed framework highly effective in reducing the noise fluctuations on DKI tensors, especially KT. As an additional refinement, we propose to impose a physically relevant constraint within our joint denoising and estimation framework. We further adopt the first-moment noise-corrected fitting model (M ¹ NCM) to remove the noncentral ${\chi }$ -distribution noise bias. The effectiveness of integrating multiple sources of priors into the joint framework is verified by comparing the proposed M ¹ NCM-NSS-LSS-PR method with various versions of M ¹ NCM-based estimators and two state-of-the-art methods. Results show that the proposed method outperformed the compared methods in simulations and in-vivo dMRI datasets of both spatially stationary and nonstationary noise distributions. The in-vivo experiments also show that the proposed M ¹ NCM-NSS-LSS-PR method was robust to the number of diffusion directions.

The purpose of this study was to quantify the effect of neck contour changes and setup errors on spinal cord (SC) doses during the treatment of nasopharyngeal carcinoma (NPC) and to establish a rapid dose estimation method. The setup errors and contour changes in 60 cone-beam computed tomography (CBCT) images of 10 NPC patients were analysed in different regions of the neck (C1–C3, C4–C5 and C6–C7). The actual delivered dose to the SC was calculated using the CBCT images, and univariate simulations were performed using the planning CT to evaluate the dose effects of each factor, and an index ${\mathrm{Dmax}}_{\mathrm{displaced}}$ was introduced to estimate the SC dose. Compared with the planned dose, the mean (maximum) Dmax increases in the C1–C3, C4–C5 and C6–C7 regions of the SC were 2.1% (12.3%), 1.8% (8.2%) and 2.5% (9.2%), respectively. The simulation results showed that the effects of setup error in the C1–C3, C4–C5 and C6–C7 regions were 1.5% (9.7%), 0.9% (8.2%) and 1.3% (6.3%), respectively, and the effects of contour change were 0.4% (1.7%), 0.7% (2.5%) and 1.5% (4.9%), respectively. The linear regression model can be used to estimate the dose effect of contour changes (R2 > 0.975) and setup errors (R2 = 0.989). Setup errors may lead to a significant increase in the SC dose in some patients. This study established a rapid dose estimation method, which is of great significance for the daily dose evaluation and the adaptive re-planning trigger of the SC.

Objective: To develop a deep learning-based super-resolution reconstruction framework that generalizes to ion chamber arrays for treatment planning system (TPS)-consistent patient-specific quality assurance. Methods: A total of 726 clinical intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) plans with field sizes ranging from 3.5 to 24 cm were included. High-resolution TPS dose maps were extracted, and matched low-resolution maps were generated by averaging dose values within detector sensitive volumes. An ATD-Transformer trained with L1 loss was adopted to learn the mapping from low-resolution to high-resolution dose distributions. Model performance was validated on 72 independent cases using TPS-derived data, and model generalizability was further evaluated with real ion chamber array measurements. Picket fence and ROC analyses were performed to assess the model’s sensitivity to clinical delivery errors. Results: The proposed method outperformed all conventional interpolation and deep learning baseline models. It achieved gamma passing rates of 79.39% under the 1%/1 mm criterion and 99.17% under the 3%/2 mm criterion on TPS-derived data. The model also demonstrated excellent generalizability on real measured data, with a stable average gamma passing rate of 98.5% for field sizes larger than 5 cm under the 384 × 384 (0.63 mm/pixel) TPS resolution. Picket fence and ROC analyses verified that the model preserved sensitivity to relevant delivery perturbations while improving TPS-consistent reconstruction accuracy. Conclusion: The proposed framework enables TPS-consistent super-resolution reconstruction of ion chamber array dose distributions. Before this method can be applied in clinical quality assurance, rigorous commissioning and calibration are required.