Deformable image registration (DIR) is widely used in radiotherapy for dose propagation and accumulation, but uncertainty in the underlying deformation can substantially affect clinically relevant dose estimates. We present a practical probabilistic framework for propagating DIR uncertainty to voxel-wise dose statistics and dose-volume histograms (DVHs). The method models the mapped correspondence at each voxel as a random variable governed by a transparent local certainty map that can be defined by simple safety margins, structure-boundary mismatch, or structure-wise conservative uncertainty values. This yields interpretable quantities such as dose probabilities, expected dose, confidence bounds, and induced DVH envelopes. The framework is designed to remain lightweight and interpretable: it avoids complex biomechanical or ensemble-based uncertainty models and instead emphasizes simple parameterization, computational feasibility, and transparent dose metrics. We further introduce a structure-guided in/out strategy as an optional refinement that restricts mapping probabilities to anatomically plausible target regions. The approach is demonstrated on a prostate radiotherapy case study and used to compare different certainty-map strategies and probability kernels. The experiments show that the certainty-map design has a stronger effect on resulting dose and DVH uncertainty bounds than the specific kernel choice, while the additional benefit of the in/out strategy is case-dependent and modest in the present example. Overall, the proposed framework provides a transparent way to incorporate DIR uncertainty into radiotherapy dose assessment and to study how modelling choices affect propagated dose metrics.

The increasing volume of CT imaging data and the limited availability of radiologists pose challenges for timely and accurate cancer assessment. While lesion diameters are routinely measured in clinical workflows, volumetric analysis remains uncommon due to the time-intensive nature of manual segmentation. Automated segmentation methods enable precise, reproducible, and efficient quantification of tumor burden over time. By leveraging information from prior examinations, longitudinal analysis can further enhance the accuracy and consistency of follow-up segmentations. We propose an approach for longitudinal lesion segmentation through neural instance optimization (NIO).Apre-trained segmentation network is fine-tuned on a patient’s prior scan to capture individual lesion-specific characteristics, which are subsequently leveraged during inference on the follow-up examination. The proposed method is applied on two public datasets. While promising results are achieved on synthetic longitudinal data (median Dice of 0.74/0.83 without/with NIO), no quantitative improvement was achieved on the second real-world longitudinal dataset. Our experiments reveal the potential of the proposed method but also its limited ability to deal with domain shifts between prior and follow-up present in real-world scenarios.

Optimizing patient positioning during CT and MRI examinations is crucial to enhance imaging efficiency and reduce radiation exposure. This study explores the accuracy of bounding box prediction for internal body structures gained from statistical atlases based on four key points in comparison to a regression model. Therefore, we utilized a dataset including 10 828 whole-body MR volumes and extracted corresponding segmentation masks. We evaluated how effectively these boxes encompass the target structures, considering also the size. Our method yields results comparable to the regression model. However, it offers the advantage of rapid adaptability to diverse research questions, making it more flexible for various analytical scenarios.

Segmentation models in medical imaging that are able to segment a wide variety of structures and generalize on different image data is a relevant and recent research topic. With more and more universal segmentation models for both MR and CT being released recently a trend towards generalizing segmentation models for large amounts of structures can be observed. While universal MR segmentation models provide segmentations for a wide variety of structures, other structures are limited to models trained on CT data. One such model is TotalSegmentator which is able to segment up to 117 structures. In our work we present and evaluate a method to leverage models trained on CT data like the TotalSegmentator model for MRI data by training a structure-consistent CycleGAN on unpaired and unregistered data. We demonstrate the feasibility of using domain transfer by leveraging unlabeled and unpaired MR and CT datasets from various scanners and sites, with different sequences and protocols from the public AMOS22 abdomen dataset. This approach translates MR to CT contrast, allowing the synthetic CT image to be used as input for the TotalSegmentator model. Furthermore, we evaluate the segmentation accuracy of our approach on different structure types such as organs, muscles and bones on internal MR and CT datasets and compare them to the recently released TotalSegmentatorMRI and MRSegmentator models.

Purpose:Axial loading, varus and valgus stress lead to meniscal motion towards the joint periphery, defined as meniscal extrusion. Direction and amount of extrusion is unknown as this is a dynamic process within a 3D environment dependent on joint loading as well as individual anatomy. We propose that there is motion in all compartments of the medial and lateral meniscus during valgus and varus stress. Method:MRI scans of 31 healthy subjects in varus or valgus stress positions were acquired with the help of a pneumatic loading device. Semiautomatic segmentation of the menisci, the femur and the tibia (with corresponding cartilages) was carried out. An individual 3D model of the joint was generated. The meniscal movement was calculated within a tibia-based coordinate system and broken down into total and partial meniscal movement (anterior/posterior horn, intermediate part). Results:Under valgus load the medial meniscus (MM) showed average movement of 1.5 (±0.5) mm in lateral-posterior direction with most lateral motion of 1.4 (±0.7) mm in the intermediate part. The lateral meniscus averaged 1.6 (±1.0) mm in lateral-anterior motion, exhibiting maximal lateral motion in the anterior horn (AH) 0.7 (±0.8) mm and posterior horn 0.6 (±0.6) mm. In response to the varus load, average MM motion was 0.9 (±0.5) mm in medial-anterior direction with the largest medial movement in the AH 0.9 (±1.1) mm. The lateral meniscus moved in average 1.6 (±0.8) mm into lateral-posterior direction with the intermediate part showing the largest medial motion of 0.6 (±0.4) mm. Conclusion:In a healthy population, the menisci extrude up to 1.5 mm during varus and valgus loading. The anterior and posterior horn show greater dynamic extrusion than the intermediate part. However, an in vivo dynamic intrusion mechanism of meniscus when discharged (medial 1.45 mm, lateral 1.56 mm) could be demonstrated. Quantification and reconstruction of this phenomenon might be of crucial importance during meniscal root or meniscal transplantation surgery. Level of Evidence:Level II, descriptive laboratory study.

The knowledge of anatomical structures' locations prior to radiological imaging allows for the improvement of MRI and CT examination efficiency, as well as a reduction in the required radiation dose for CT imaging. The objective of this work is to develop a deep learning model for the prediction of 2D projections of bone, organ, muscle and vessel masks from a 2D depth image of a patient, which resembles a depth image taken by time-of-flight cameras. To achieve this goal, CT and MRI images are segmented in 3D and projected down to a 2D depth map of the patient surface and 2D projections of corresponding anatomical labels. The models are trained using the TotalSegmentator CT and German National Cohort MRI data and additionally evaluated using the ACRIN-FLT-Breast data. The model achieves an average symmetric surface distance (mean +/- standard deviation) of 6.97 +/- 12.52, 4.55 +/- 4.91 and 6.39 +/- 7.54 mm, respectively.

BACKGROUND:Deformable dose accumulation (DDA) uncertainty models can inform treatment decisions by communicating the dosimetric impact of deformable image registration (DIR) errors over multiple fractions. Currently there is limited guidance on how end-users can validate such models in the clinic. PURPOSE:We propose an end-user validation sequence for DDA uncertainty modelling tools, akin to an acceptance test, using existing patient data and a clinical treatment planning system (TPS) as the evaluation platform. METHODS:The proposed test sequence begins with a single "fixed" image (planning CT with associated contours and treatment plan) and a "moving" image (e.g., fractional synthetic CT with calculated dose) and uses the TPS to simulate multiple fractional DIRs as input to a DDA uncertainty model. Outputs of the uncertainty tool-including volumetric images of DIR spatial uncertainties, and associated uncertainties on propagated dose-are imported back to the TPS and a series of visual and semi-quantitative (point-based and DVH-based) cross-checks are carried out. Emphasis is placed on the use of standard dose and distance measurement tools, in conjunction with vendor-provided equations, to evaluate correctness of the uncertainty tool outputs at contour boundaries and in bulk tissue, considering variable DIR quality for both targets and organs at risk. RESULTS:The test sequence is demonstrated for a non-clinical uncertainty tool using a clinical bladder case. Agreement within 2 voxels (for spatial uncertainties) and up to 5% of the prescribed dose (for dose uncertainties) is shown to be achievable for regions of plausible deformation and stable dose gradient (e.g., < 5%/voxel). CONCLUSIONS:As the use of DDA in adaptive treatment becomes more commonplace, use of DDA uncertainty tools will become increasingly important to inform adaptive treatment decisions. This work represents an important effort to formalize an end-user validation process using standardly available clinical tools.

Accurate alignment of preoperative planning data with intraoperative conditions is critical for effective surgical navigation. In neurosurgery, this alignment is complicated by brainshift, which occurs when the brain shifts within the skull after the dura mater is opened and cerebrospinal fluid (CSF) is drained. We propose a novel method for intraoperative alignment of MRI-based planning data with photographs taken during awake brain surgery, using a landmark-driven 2D/3D registration technique. Landmarks are interactively selected on both the intraoperative photographs and 3D renderings of the brain's structure and vasculature, derived from T1-weighted MR images with and without contrast enhancement. Registration is performed by minimizing the average distance between corresponding landmark pairs in the two modalities. The precision of this method is evaluated by comparing the positions of physically placed markers on the brain surface through multiple photographs taken from different angles. The landmark projections are reproducible with an average standard deviation of 0.925mm.

Automated patient positioning is a crucial step in streamlining MRI workflows and enhancing patient throughput. RGB-D camera-based systems offer a promising approach to automate this process by leveraging depth information to estimate internal organ positions. This paper investigates the feasibility of a learning-based framework to infer approximate internal organ positions from the body surface. Our approach utilizes a large-scale dataset of MRI scans to train a deep learning model capable of accurately predicting organ positions and shapes from depth images alone. We demonstrate the effectiveness of our method in localization of multiple internal organs, including bones and soft tissues. Our findings suggest that RGB-D camera-based systems integrated into MRI workflows have the potential to streamline scanning procedures and improve patient experience by enabling accurate and automated patient positioning.

The alignment of tissue between histopathological whole-slide-images (WSI) is crucial for research and clinical applications. Advances in computing, deep learning, and availability of large WSI datasets have revolutionised WSI analysis. Therefore, the current state-of-the-art in WSI registration is unclear. To address this, we conducted the ACROBAT challenge, based on the largest WSI registration dataset to date, including 4,212 WSIs from 1,152 breast cancer patients. The challenge objective was to align WSIs of tissue that was stained with routine diagnostic immunohistochemistry to its H&E-stained counterpart. We compare the performance of eight WSI registration algorithms, including an investigation of the impact of different WSI properties and clinical covariates. We find that conceptually distinct WSI registration methods can lead to highly accurate registration performances and identify covariates that impact performances across methods. These results provide a comparison of the performance of current WSI registration methods and guide researchers in selecting and developing methods.

Previous work on methods for cross domain generalization in medical imaging found a simple but very effective method called "global intensity non-linear" (GIN) augmentation. Our goal in this study is to use the GIN approach to train a model as powerful as TotalSegmentator for MRI data, despite having neither sufficient amounts of MRI data nor ground truth organ contours. Instead, we employ the GIN augmentation approach to show qualitatively and quantitatively that this is indeed feasible for a diverse set of anatomical structures including abdominal and thoracic organs as well as bones. The models are trained on the TotalSegmentator and AMOS22 datasets. For evaluation we apply them to whole body MRI scans from the German National Cohort (NAKO) study with a set of in-house reference masks. With GIN augmentation the mean Dice score of the model increases from 0.18 to 0.52 on Dixon water images, when using TotalSegmentator data for training. The improvements can be further split into 0.47 to 0.66 for abdominal organs, 0.55 to 0.79 for thoracic organs and 0.00 to 0.40 for bones.

3D visualisation and modelling of anatomical structures of the human body play a significant role in diagnosis, computer-aided surgery, surgical planning, and patient follow-up. However, 2D X-ray images are often used in clinical routine. We propose and validate a method for reconstructing 3D shapes from 2D X-ray scans. This method comprises automatic segmentation and labelling, automated construction of 3D statistical shape models (SSM), and automatic fitting of the SSM to standard 2D X-ray images. This workflow is applied to finger bone shape reconstruction and validated for each finger bone using a set of five synthetic reference configurations and 34 CT/X-ray data pairs. We reached submillimetre accuracy for 91.59% of the synthetic data, while 79.65% of the clinical cases show surface errors below 2 mm. Thus, applying the proposed method can add valuable 3D information where 3D imaging is not indicated. Moreover, 3D imaging can be avoided if the 2D-3D reconstruction accuracy is sufficient.

Purpose: Analyzing the anatomy of the aorta and left ventricular outflow tract (LVOT) is crucial for risk assessment and planning of transcatheter aortic valve implantation (TAVI). A comprehensive analysis of the aortic root and LVOT requires the extraction of the patient-individual anatomy via segmentation. Deep learning has shown good performance on various segmentation tasks. If this is formulated as a supervised problem, large amounts of annotated data are required for training. Therefore, minimizing the annotation complexity is desirable. Approach: We propose two-dimensional (2D) cross-sectional annotation and point cloud-based surface reconstruction to train a fully automatic 3D segmentation network for the aortic root and the LVOT. Our sparse annotation scheme enables easy and fast training data generation for tubular structures such as the aortic root. From the segmentation results, we derive clinically relevant parameters for TAVI planning. Results: The proposed 2D cross-sectional annotation results in high inter-observer agreement [Dice similarity coefficient (DSC): 0.94]. The segmentation model achieves a DSC of 0.90 and an average surface distance of 0.96 mm. Our approach achieves an aortic annulus maximum diameter difference between prediction and annotation of 0.45 mm (inter-observer variance: 0.25 mm). Conclusions: The presented approach facilitates reproducible annotations. The annotations allow for training accurate segmentation models of the aortic root and LVOT. The segmentation results facilitate reproducible and quantifiable measurements for TAVI planning.

Various uncertainty estimation methods have been proposed for deep learning-based image segmentation models. An uncertainty measure is treated useful if it can be used to accurately predict segmentation quality. Therefore, structure-wise uncertainty measures are frequently correlated with measures like the Dice score. However, it is known that the Dice score highly depends on the size of the structure of interest. It is less well-known that popular structure-wise uncertainty measures also correlate with structure size. Therefore, the structure size acts as confounding variable when trying to quantify the performance of such uncertainty measures via correlation. We investigate this for the popular uncertainty measures structure-wise epistemic uncertainty, mean pairwise Dice and volume variation coefficient based on test-time-augmentation, Monte Carlo Dropout and model ensembles. We propose to use a partial correlation coefficient to address structure size as confounding variable and arrive at lower correlation estimates which better reflect the true relationship between segmentation quality and structure-wise uncertainty.

Purpose To help radiologists examine the growing number of computed tomography (CT) scans, automatic anomaly detection is an ongoing focus of medical imaging research. Radiologists must analyze a CT scan by searching for any deviation from normal healthy anatomy. We propose an approach to detecting abnormalities in axial 2D CT slice images of the brain. Although much research has been done on detecting abnormalities in magnetic resonance images of the brain, there is little work on CT scans, where abnormalities are more difficult to detect due to the low image contrast that must be represented by the model used. Approach We use a generative adversarial network (GAN) to learn normal brain anatomy in the first step and compare two approaches to image reconstruction: training an encoder in the second step and using iterative optimization during inference. Then, we analyze the differences from the original scan to detect and localize anomalies in the brain. Results Our approach can reconstruct healthy anatomy with good image contrast for brain CT scans. We obtain median Dice scores of 0.71 on our hemorrhage test data and 0.43 on our test set with additional tumor images from publicly available data sources. We also compare our models to a state-of-the-art autoencoder and a diffusion model and obtain qualitatively more accurate reconstructions. Conclusions Without defining anomalies during training, a GAN-based network was used to learn healthy anatomy for brain CT scans. Notably, our approach is not limited to the localization of hemorrhages and tumors and could thus be used to detect structural anatomical changes and other lesions. (c) The Authors. Published by SPIE under a Creative Commons Attribution 4.0 International License.Distribution or reproduction of this work in whole or in part requires full attribution of the originalpublication, including its DOI. [DOI:10.1117/1.JMI.11.4.044508]

Intuitive visualization of relevant changes between radiological image pairs in the form of change maps has the potential to not only increase efficiency in diagnostic reading, but also to decrease the number of missed abnormalities. Classically, change maps are created from difference images after an image registration step, which requires a careful balance in order to neither generate artifacts nor disguise relevant changes.We propose jointly learning registration and change map in order to address these limitations. As a proof of concept, the method was tested on NLST lung CT images and synthetically generated data, and shows comparable results to the conventional approach. In a reader study, the use of change maps resulted in a 23% reduction in reading time while maintaining similar recall.

Follow-up assessment of lesions for cancer patients is an important part of radiologists’ work. Image registration is a key technology to facilitate this task, as it allows for the automatic establishment of correspondences between previous findings and current observations. However, as the number of examinations increases, more registrations must be computed to allow full correspondence assessment between longitudinal studies. We address the challenge of increased computational time and complexity by identifying and eliminating redundant registration procedures and composing deformations from previously performed registrations, thereby significantly reducing the number of registrations required. We evaluate our proposed methods on a dataset consisting of oncological thoracic follow-up CT scans from 260 patients. By grouping series within a study and identifying reference series, we can reduce the total number of registrations required for a patient by an average factor of 27.5 while maintaining comparable registration quality. Additionally composing deformations further reduces the number of registrations by a factor of 1.86, resulting in an overall average reduction factor of 51.4. Since the number of registrations is directly related to the time required to process the input data, the information is available more quickly and subsequent examinations can be performed sooner. For a single subject, this results in an exemplary reduction of total computation time from 37.4 to 1.3 min.

Analyzing the anatomy of the aorta and left ventricular outflow tract (LVOT) is crucial for risk assessment and planning of transcatheter aortic valve implantation (TAVI). We propose 2D cross-sectional annotation and point cloud-based surface reconstruction to train a fully automatic 3D segmentation network for the aortic root and the LVOT. Our sparse annotation scheme enables easy and fast training data generation for tubular structures like the aortic root. Based on this annotation concept, we trained a 3D segmentation model that achieves a Dice similarity coefficient (DSC) of 0.9 and an average surface distance (ASD) of 0.96 mm. In addition, we show that our fully automatic segmentation approach facilitates reproducible and quantifiable measurements for TAVI planning. Our approach achieves an aortic annulus maximum diameter difference between prediction and annotation of 0.45 mm (inter-observer variance: 0.25 mm).