Background/Objectives: High liver fat content is closely associated with hepatic disease and multiple comorbidities including cancer, diabetes and cardiovascular accidents. Therefore, accurate quantification of hepatic steatosis, especially of borderline cases, is essential for clinical management. Although MRI non-invasively assesses hepatic steatosis, current approaches remain limited by the data variability introduced through use of region-of-interest measurements or classification models that predict discrete fat grades without providing uncertainty estimates. This study proposes a probabilistic approach for hepatic steatosis quantification based on combining a neural network and beta distribution, enabling prediction of hepatic fat percentage with corresponding confidence intervals. Methods: Single in-phase Dixon MRI liver images from a cohort of prepubertal males (n = 84) were used as input to a probabilistic neural network combined with a beta distribution framework to estimate hepatic fat content along with associated confidence intervals. The predicted fat fractions were then compared against reference MRI-derived measurements (ground truth). Results: The methodology achieved a low prediction error and demonstrated good performance for the test set, with predicted values in good agreement with the ground truth measurements. This was reflected by the mean absolute error (MAE = 0.44 percentage points) and the coefficient of determination (R² = 0.98). The empirical standard deviation of the prediction errors on a logarithmic scale was σ = 0.0609. Conclusions: By incorporating uncertainty quantification into hepatic steatosis estimation, this probabilistic framework provides an interpretable measure of variability alongside point estimates. The approach is demonstrated in a specific cohort and requires further validation in broader populations.

High liver fat content is closely associated with hepatic disease and multiple comorbidities including cancer, diabetes and cardiovascular accidents. Therefore, accurate quantification of hepatic steatosis is essential for clinical management. Although MRI provides a non-invasive means for hepatic fat assessment, current approaches remain limited by variability introduced through region-of-interest based methodologies or by classification models that predict discrete fat grades without providing uncertainty estimates. This study proposes a novel approach for hepatic steatosis quantification based on a probabilistic neural network and Beta distribution with single in-phase MRI image inputs to provide estimates of the percentage of hepatic fat with corresponding prediction confidence intervals. The novel model achieved a low prediction error and demonstrated good generalization performance on the test set. By explicitly incorporating uncertainty quantification into the estimates of hepatic steatosis, this probabilistic framework represents a methodological advance that offers an accurate estimate that can be interpreted with greater confidence.

Purpose: Stochastic bowel and rectal gas complicates MRI/CT deformable image registration (DIR) for synthetic CT (sCT) generation, requiring manual corrections. We propose a DIR-free, two-stage deep learning framework to improve intraluminal gas definition in sCT, streamlining MRI-only simulation and enhancing dosimetric reliability. Methods: We developed a two-stage GAN framework for sCT generation. First, a CycleGAN or pix2pix model converted MRIs into segmented map images (SMI); second, a pix2pix conditional GAN generated the final sCT. Ground-truth gas cavities were defined using CNN-based gastrointestinal autosegmentation, manual verification, and a 20% intensity threshold. Trained on a 60-patient dataset, the framework was compared to a single-stage MRI-to-sCT model via Dice-Sørensen coefficient and Mean Absolute Error (MAE). Dosimetric accuracy was assessed by recalculating prostate VMAT plans (36.25 Gy/5 fractions) on synthetic CTs. Results: The gas cavity Dice-Sørensen coefficient for the two-stage method and the single-stage method were 0.63 ± 0.11 and 0.04 ± 0.03 for pix2pix, while for CycleGAN, they were 0.67 ± 0.07 and 0.57 ± 0.11, respectively. The global MAE for the standard single-stage method and the two-stage method for pix2pix were 86±20 HU and 94±20 HU, while for CycleGAN, they were 105±21 HU and 103±22 HU, respectively. Dosimetric analysis demonstrated excellent agreement for the planning target volume (PTV D95%), with mean differences of 0.04 ± 0.08 cGy and −0.89 ± 1.98 cGy for the one-stage and two-stage models, respectively. Both models exhibited negligible deviations across all assessed organ-at-risk (OAR) volumetric metrics, including D0.1cc, D1cc, and D15cc. Conclusions: The two-stage framework significantly enhances sCT intraluminal gas accuracy, improving cavity definition nearly threefold over single-stage models while maintaining tissue fidelity. This overcomes a major MRI-only simulation bottleneck. Future studies will prospectively evaluate its clinical impact on online adaptive radiotherapy.