Inverse‐consistent deformable image registration to propagate biomechanical constraints for online adaptive radiotherapy

GDLS-D: two-stage medical surface reconstruction via geometric diffusion least squares and iterative deformation optimization.

Evaluating synthetic CT from deformable registration in directed adaptive radiotherapy for head and neck cancer

Abstract Background and purposel Anatomical changes over the many weeks of radiotherapy for Head and Neck cancer patients can compromise the original treatment plan, potentially increase the risk of side effects and impact outcomes. It is difficult to decide, based on daily imaging alone, which patients will most benefit from adapted treatment plans. This work presents an automated approach to evidence‐based assessment of treatment plan robustness to anatomy changes. Methods Directed Adaptive Radiotherapy (DART) generates a synthetic CT for calculating dose changes in the treatment planning system (TPS). The process first enhances the CBCT, then uses a double fusion to insert the smaller field‐of‐view CBCT into the larger CT. The double fusion captures both neck tilt and weight loss changes, to accurately capture the position of the head and related critical organs. The deformed CT and contours are sent to TPS for recalculation of the original plan and quantify dose changes. A retrospective study was undertaken using 12 H&N patients who had a replanning CT during their treatment. The quality of the dose distributions calculated on synthetic CT and deformed contours were evaluated, along with the sensitivity and specificity of the synthetic CT as a predictor for adaptation. Results The contour and dose comparison between synthetic CT and clinical replan CT showed good agreement. The average Dice Similarity Coefficient (DSC) was > 80% for all contours, except for the sub‐mandibular (> 75%) and excluding cochlea. The average mean distance‐to‐agreement (MDA) was < 2 mm, and all MDAs were < 3 mm. The average DVH differences were within 1%, and all dose differences were within 3%. The total gamma passing rates were > 95%. Both the sensitivity and specificity of the synthetic CT for predicting the need for adaptation were 92%, based on a tolerance of 2 Gy change for both target coverage and OAR dose constraint limits. Conclusions By quantifying the impact of anatomical changes on dose distributions, the DART process helps determine whether a patient would benefit from adaptive replanning. The DART method improves efficiency by reducing unnecessary imaging and replanning when the treatment remains robust against anatomy changes.

Evaluation of AI‐based segmentation and synthetic CT generation for MRI‐only prostate radiotherapy planning

Abstract Background There is an increasing use of magnetic resonance imaging (MRI) for segmentation of target volumes and organs at risk due to superior soft tissue contrast compared to computed tomography (CT). Conventional workflows require CT‐MRI registration, introducing geometric uncertainty that can propagate to dose delivery. MRI‐only radiotherapy, based on AI‐generated synthetic CT (sCT) and segmentations from T2‐weighted images, offers a single‐modality approach, potentially eliminating registration errors. Purpose Our study aims to evaluate whether an MRI‐only workflow, using commercial AI models for segmentation and sCT generation (MVision), demonstrates performance comparable to expert‐defined segmentations and reference CT‐based calculations within the studied dataset. Methods We analyzed five segmentations on 19 male pelvis patients from the Gold Atlas dataset, providing deformably co‐registered CT and MRI, sCTs, with multi‐observer and consensus segmentations. We investigated segmentation accuracy, image similarity, and evaluated the dosimetric impact by comparing AI‐ to manual segmentations and CTs to sCTs. Wilcoxon signed‐rank tests were used for statistical significance and leave‐one‐out kappa determined segmentation outliers. Results Results show that AI‐segmentations achieved accuracy comparable to interobserver variability across all organs. Median Dice coefficients exceeded 0.9 for the prostate and bladder, and distance metrics closely matched the manual results. The leave‐one‐out kappa showed that AI‐segmentations were indistinguishable from manual observers for all organs. sCTs showed high similarity to the reference CTs, with a median mean absolute error of 34.0 HU and median structural similarity index of 0.93. Dose recalculations showed excellent agreement; median 3D‐gamma pass rate was 99.8% (2%/2 mm), and mean absolute dose difference was 0.036 Gy. All evaluated clinical dose constraints remained fulfilled after recalculation on the sCT. Conclusions Our results confirm that an AI‐segmentation and MRI‐sCT workflow provides geometric, image, and dose accuracy comparable to manual‐ and CT‐based references, supporting their integration into clinical workflows for pelvic radiotherapy.

Eliminating Registration Bias in Synthetic CT Generation using a physics-based simulation framework for pelvic anatomy.

OBJECTIVE Supervised synthetic computed tomography (sCT) generation from cone-beam CT (CBCT) requires spatially registered training pairs, yet perfect registration between separately acquired scans is unattainable. This registration bias propagates into trained models and corrupts intensity-based evaluation, so higher benchmark scores may reward reproduction of registration artifacts over anatomical fidelity. We propose physics-based CBCT simulation for geometrically aligned training pairs by construction, with bias-robust geometric metrics. Approach:A framework simulated pelvic CBCT from fan-beam CT, modeling respiratory motion, X-ray scatter, and noise to yield aligned simulated-CBCT/CT pairs. On a clinical gynecological dataset (deformable registration) and the SynthRAD2023 pelvic dataset (rigid registration), sCT models trained on simulated data were compared against models trained on real pairs, a finetuned variant, and CycleGAN and RegGAN baselines. Evaluation combined intensity metrics (MAE, PSNR, SSIM) with geometric alignment metrics (normalized mutual information, NMI; correlation coefficient, CC) against input CBCT. Downstream segmentation of bladder, rectum and bowel bag was assessed in two modes: an sCT cascade applying a CT-trained model to sCT outputs, and direct segmentation by a model trained on simulated CBCT, plus a physics ablation and five-observer quality assessment. Main results:Simulation-trained models achieved higher geometric alignment than real-trained models (cross-dataset NMI 0.31 vs 0.22) despite lower intensity scores. Intensity metrics correlated inversely with observer ratings under deformable registration, whereas NMI consistently predicted clinical preference (clinical ρ = 0.29, SynthRAD ρ = 0.31). Observers preferred simulation-trained outputs in 87% of cases. In the sCT cascade, simulation-trained models improved segmentation (DSC 0.91/0.86/0.54 vs 0.84/0.77/0.04), while direct simulation-trained CBCT segmentation reached 0.92/0.87/0.83, exceeding a phantom-based baseline on the bowel bag. Significance:Physics-based simulation eliminates registration bias at its source, and downstream segmentation provides a task-based measure of sCT conversion quality that intensity metrics miss. Geometric fidelity, not intensity agreement with biased ground truth, aligns with the spatial-accuracy requirements of adaptive radiotherapy.

Is Deformable Image Registration Ready for Brain Metastasis Reirradiation Dose Accumulation? A Longitudinal MRI Benchmark of Registration Accuracy

Dose accumulation is increasingly important in adaptive radiation therapy and reirradiation, but its clinical validity depends on the performance of deformable image registration (DIR). Reirradiation of brain metastases (BMs) with stereotactic radiosurgery (SRS) provides a controlled but clinically meaningful DIR test case: intra-subject brain deformation is usually limited after rigid alignment, yet recurrent lesions can undergo substantial local shape and volume changes that rigid registration cannot capture and can affect dose accumulation. We benchmarked a wide range of learning-based and optimization-based DIR methods on 87 manually screened longitudinal contrast-enhanced T1-weighted MRI lesion pairs from an institutional BM SRS retreatment cohort. Learning-based methods pretrained on healthy-brain MRI were evaluated zero-shot and after instance-specific optimization (ISO) or tumor-proximity target-specific optimization (TSO). Registration was assessed using lesion overlap (Dice), surface distance metrics (HD95 and sASD), target-volume recovery, and runtime and memory. Pretrained learning-based methods showed variable zero-shot performance, while ISO/TSO improved all tested learning-based families. However, optimization-based methods remained the best-performing approach while maintaining reasonable runtime. These findings suggest that even state-of-the-art DIR methods do not yet provide sufficiently accurate and consistent registration for unmonitored use in brain metastasis reirradiation dose accumulation. Because accurate registration is a prerequisite for deformable dose accumulation, clinical application will require case-level quality control and direct assessment of how registration uncertainty affects downstream dose metrics.

DINE: Distance Is Not Enough - Learning Global Deformation Priors for Robust Soft-Tissue Point Cloud Registration

Non-rigid point cloud registration is central to soft-tissue shape analysis, but large deformations, noise, and outliers make correspondence estimation challenging. Most learning-based methods rely on local objectives such as Chamfer distance, which encourage point-wise proximity but do not constrain the global plausibility of the predicted deformation field. We address this limitation with DINE, a maximum a posteriori framework that augments distance-based registration with a learned statistical prior over displacement vector fields. DINE is applied to two registration backbones, Robust-DefReg and DefTransNet, using a two-stage strategy: a first-stage model is trained with Chamfer distance, its predicted deformation fields are used to estimate a prior, and the model is then refined with a combined distance and negative log-prior objective. We compare a full-field PCA Gaussian prior with a per-vector normalizing-flow prior. Experiments on DeformedTissue and SynBench show lower mean Chamfer distance under deformation and corruption. On DeformedTissue, DINE-PCA reduces Chamfer distance by approximately 27--69\% relative to the corresponding Stage-1 backbone across deformation levels, and improves robustness by up to 66\% for outliers and 83\% for Gaussian noise. On SynBench, improvements are modest at the smallest deformation levels and reach approximately 59--79\% from moderate to severe deformation. These results suggest that global deformation plausibility is an important constraint for reliable soft-tissue point cloud registration. (The code will be published soon.)