Non-rigid Registration Research Articles

Use of a Template-Matching Clavicle Fracture Atlas for Surrogate Dating of Humeral and Femoral Fractures in Young Infants: A Six-Reader Study.

Background: Fracture dating is important in suspected infant abuse. Birth-related clavicle fractures are common and may provide surrogates to aid infant long bone fracture dating. Objective: To assess the impact of a template-matching clavicle fracture timeline atlas on radiologists' performance in dating birth-related clavicle, humerus, and femur fractures in young infants. Methods: This retrospective study included infants with age ≤90 days who underwent a radiograph of a birth-related clavicle fracture from April 2021 to July 2024, or of a birth-related humerus or femur fracture from December 2011 to July 2024. All eligible radiographs of each fracture were identified, representing distinct observations for purposes of analysis. Patient age in days at the time of radiograph acquisition served as the reference standard for fracture ages. A non-rigid image registration technique was applied to a nonoverlapping pre-assembled database of birth-related clavicle fracture radiographs, to create a fracture dating atlas. Six readers (three trainees, three pediatric radiologists) independently reviewed radiographs in separate sessions without and with the atlas to estimate fracture ages. Interreader agreement was assessed using intraclass correlation coefficients (ICCs). Fracture aging performance was assessed using mean absolute errors (MAEs). Results: The analysis included 145 infants (87 male, 58 female) with 269 fracture radiographs (104 clavicle, 128 humerus, 37 femur). Mean fracture age was 26±19, 22±14, and 20±13 days for clavicle, humerus, and femur fractures, respectively. Interreader agreement for estimating fracture ages improved from moderate (ICC=0.69) without, to excellent (ICC=0.91) with, the atlas. MAE in fracture dating was significantly lower (P<.05) with than without the atlas for all six readers for clavicle fractures (range, 4.8-5.5 vs 5.8-10.1 days), all six readers for humerus fractures (6.0-12.1 vs 3.0-3.8 days), and five of six readers for femur fractures (7.4-17.2 vs 3.3-4.8 days). MAE without and with the atlas was 8.8 versus 4.3 days, respectively, across trainee readers and 8.4 versus 4.0 days, respectively, across attending readers. Conclusion: The fracture dating atlas yielded significant improvements in radiologists' performance for dating infant clavicle, humerus, and femur fractures. Clinical Impact: Clavicle fracture healing patterns can serve as surrogates for dating long bone fractures commonly encountered in infant abuse.

Assessment and mitigation of geometric distortions in Cartesian MR images at 15.2T for preclinical radiation research.

Ultra-high field (UHF) magnetic resonance (MR) systems are advancing in preclinical imaging offering the potential to enhance radiation research. However, system-dependent factors, such as magnetic field inhomogeneities ( ) and gradient non-linearity (GNL), induce geometric distortions compromising the sub-millimeter accuracy required for radiationresearch. This study tackles system-dependent distortions in 15.2T MR images by prospective shimming strategies optimization and comparing two imaging methods for voxel displacement correction. The methods were evaluated on a 3D-printed grid phantom and validated on in vivo mouse brain MR images. Additionally, a phantom-based displacement map was tested for GNL correction in mouse brainimages. Phantom MR and CT images were acquired with 200 resolution. In vivo mouse brain MR and CT images had 140 and 200 resolutions, respectively. Three shimming strategies were established to assess displacements ( ) in phantom MR images. was calculated using the acquired static field maps in three volumes of interest (VOIs) via Python script. A one-step distortion correction (1SDC) method, which simultaneously corrects and GNL distortions via non-rigid registration with CT, and a two-step distortion correction (2SDC) method, which corrects separately in two consecutive steps and GNL displacements, were assessed on phantom and in vivo images. For in vivo 2SDC validation, a phantom displacement map generated by MR to CT non-rigid registration was applied to correct GNL on the mouse brain. Total displacements ( ) were quantified in phantom VOIs and the in vivo skull region by measuring landmarks'positions. The in the phantom increased with distance from the VOI center and magnet isocenter. Shimming scenario-2 showed the lowest maximum displacement (0.26mm) for the largest VOI but required a longer acquisition time. Distortion correction methods were necessary for large VOIs (13-25mm, along the z-axis) in the phantom where 0.2mm. The 2SDC method outperformed 1SDC by achieving a 0.2mm accuracy in 100%, 92.1%, and 59.3% of the landmarks from the smallest to the largest VOI. Phantom dice scores confirmed the improvement in geometric precision after each correction step. In vivo results showed that 1SDC correction overcorrected MR images, increasing voxel displacements. The 2SDC exceeded the 1SDC, reducing by 85%, in accordance with the dice score analysis (0.97 2SDC vs. 0.84 1SDC). At 15.2T, in vivo MR images of even small regions (e.g., mouse brain) require geometric distortion correction for radiation research. The 2SDC method outperformed the 1SDC, emphasizing the need for separate and GNL corrections. Moreover, a phantom-based displacement map shows promise for in vivo GNL correction.

DefTransNet: a transformer-based method for non-rigid point cloud registration in the simulation of soft tissue deformation

Abstract Soft-tissue surgeries, such as tumor resections, are complicated by tissue deformations that can obscure the accurate location and shape of tissues. By representing tissue surfaces as point clouds and applying non-rigid point cloud registration (PCR) methods, surgeons can better understand tissue deformations before, during, and after surgery. Existing non-rigid PCR methods, such as feature-based approaches, struggle with robustness against challenges like noise, outliers, partial data, and large deformations, making accurate point correspondence difficult. Although learning-based PCR methods, particularly Transformer-based approaches, have recently shown promise due to their attention mechanisms for capturing interactions, their robustness remains limited in challenging scenarios. In this paper, we present DefTransNet, a novel end-to-end Transformer-based architecture for non-rigid PCR. DefTransNet is designed to address the key challenges of deformable registration—including large deformations, outliers, noise, and partial data—by inputting source and target point clouds and outputting displacement vector fields. The proposed method incorporates a learnable transformation matrix to enhance robustness to affine transformations, integrates global and local geometric information, and captures long-range dependencies among points using Transformers. We validate our approach on four datasets: ModelNet, SynBench, 4DMatch, and DeformedTissue, using both synthetic and real-world data to demonstrate the generalization of our proposed method. Experimental results demonstrate that DefTransNet outperforms current state-of-the-art registration networks across various challenging conditions. Our code and data are publicly available.&amp;#xD;

Artificial intelligence empowers functional preservation and safety guarantee in laparoscopic colorectal surgery

Transabdominal and transanal endoscopic approaches have become mainstream in colorectal surgery. With the substantial improvement in survival outcomes for colorectal cancer patients, a growing number of colorectal surgeons are increasingly focusing on enhancing postoperative quality of life, prioritizing functional preservation, especially the intraoperative preservation of pelvic autonomic nerves. Recently, with the gradual deepening of artificial intelligence (AI) applications in the medical field, colorectal surgeons have begun exploring its implementation in colorectal surgery. Current achievements primarily involve the identification and protection of nerves and organs. However, most AI applications remain at preclinical exploration stages, limiting their clinical application. Furthermore, AI faces challenges in recognizing blood vessels with significant deformation and movement. Thus, the precise real-time navigation and protection of blood vessels during surgery have yet to be achieved. Therefore, future developments in this field should focus on resolving issues such as non-rigid registration, real-time calibration etc., thereby deepening the application of AI in functional preservation and surgical safety assurance during laparoscopic colorectal surgery.

Refining cardiac segmentation from MRI volumes with CT labels for fine anatomy of the ascending aorta.

Magnetic resonance imaging (MRI) is time-consuming, posing challenges in capturing clear images of moving organs, such as cardiac structures, including complex structures such as the Valsalva sinus. This study evaluates a computed tomography (CT)-guided refinement approach for cardiac segmentation from MRI volumes, focused on preserving the detailed shape of the Valsalva sinus. Owing to the low spatial contrast around the Valsalva sinus in MRI, labels from separate computed tomography (CT) volumes are used to refine the segmentation. Deep learning techniques are employed to obtain initial segmentation from MRI volumes, followed by the detection of the ascending aorta's proximal point. This detected proximal point is then used to select the most similar label from CT volumes of other patients. Non-rigid registration is further applied to refine the segmentation. Experiments conducted on 20 MRI volumes with labels from 20 CT volumes exhibited a slight decrease in quantitative segmentation accuracy. The CT-guided method demonstrated the precision (0.908), recall (0.746), and Dice score (0.804) for the ascending aorta compared with those obtained by nnU-Net alone (0.903, 0.770, and 0.816, respectively). Although some outputs showed bulge-like structures near the Valsalva sinus, an improvement in quantitative segmentation accuracy could not be validated.

Comparative evaluation of image registration techniques in functional ultrasound imaging

Abstract Functional ultrasound imaging (fUSI) is an emerging hemodynamic-based neuroimaging technique that combines high spatiotemporal resolution and sensitivity with extensive brain coverage, enabling a wide range of applications in preclinical brain research. Based on power Doppler imaging, fUSI measures changes in cerebral blood volume by detecting the back-scattered echoes from red blood cells moving within its field of view. Despite the significant contribution of fUSI technology to neuroscience research, its full potential is partly constrained by the challenge of accurately co-registering power Doppler vascular maps acquired across different sessions and/or animals to a single reference: an approach that could widen the scope of experimental paradigms and enable the utilization of advanced data analysis tools. This study aims to address this critical limitation by comparing eight image registration techniques to align 2D sagittal whole-brain fUSI datasets acquired from 82 anesthetized mice. The results showed a significant improvement in the alignment of fUSI images across all techniques. However, the non-rigid registration methods demonstrated either similar or superior performance in similarity metrics compared to rigid approaches, with the non-rigid version of Elastix and Imregdeform emerging as the top-performing techniques. Further analysis revealed that both methods maintained comparable high levels of geometric integrity, as evidenced by similar mean Jacobian determinants (close to 1) and low folding rates. In summary, our study offers the first comparative analysis of image registration techniques specifically tailored for 2D fUSI mouse brain datasets, paving the groundwork for enhanced utilization of fUSI in future research applications.

An Elastic Fine-Tuning Dual Recurrent Framework for Non-Rigid Point Cloud Registration

Non-rigid transformation is based on rigid transformation by adding distortions to form a more complex but more consistent common scene. Many advanced non-rigid alignment models are implemented using supervised learning; however, the large number of labels required for the training process makes their application difficult. Here, an elastic fine-tuning dual recurrent computation for unsupervised non-rigid registration is proposed. At first, we transform a non-rigid transformation into a series of combinations of rigid transformations using an outer recurrent computational network. Then, the inner loop layer computes elastic-controlled rigid incremental transformations by controlling the threshold to obtain a finely coherent rigid transformation. Finally, we design and implement loss functions that constrain deformations and keep transformations as rigid as possible. Extensive experiments validate that the proposed method achieves state-of-the-art performance with 0.01219 earth mover’s distances (EMDs) and 0.0153 root mean square error (RMSE) in non-rigid and rigid scenes, respectively.

Nonrigid temporal registration of multiphase CT pulmonary angiography using low-kV and low contrast: a feasibility study with dual-source CT.

Nonrigid temporal registration of multiphase CT pulmonary angiography using low-kV and low contrast: a feasibility study with dual-source CT.

Longitudinal analysis of coal workers' pneumoconiosis using enhanced resolution-computed tomography images: unveiling patterns in lung structure, function, and clinical correlations.

Pneumoconiosis, caused by prolonged exposure to mineral dust, leads to progressive structural and functional lung alterations. Quantitative computed tomography (qCT) has emerged as a critical tool for assessing these changes, yet there is limited research on the longitudinal patterns in pneumoconiosis patients. This study examined a cohort of 31 former coal workers with pneumoconiosis over a 1-year period. Inspiratory qCT images were enhanced using a deep learning-based super-resolution model and then processed to extract lung functional and airway structural metrics. A non-rigid image registration process was performed with baseline images as fixed and follow-up images as moving. Registration-derived metrics, including anisotropic deformation index (ADI), slab rod index (SRI), and Jacobian (J), were extracted to quantify regional deformation longitudinally. Pulmonary function tests, including forced expiratory volume in one second (FEV1) and forced vital capacity (FVC), were recorded at both time points to assess functional decline. The study identified significant airway changes in angles, diameters, and geometry, with a decrease in normal lung tissue in the right upper lobe. Blood vessel volumes declined, indicating vascular remodeling. Registration metrics revealed regional heterogeneity, with higher ADI and SRI values and localized volume loss (J) in the lower lobes. FEV1/FVC progression correlated positively with tracheal angle, emphysema, and consolidation but negatively with normal lung tissue, semi-consolidation, and fibrosis. ADI, SRI, and J were associated with structural deformation, airway remodeling, and parenchymal loss, linking these changes to lung function decline. qCT imaging and registration metrics effectively monitor structural and functional lung changes in pneumoconiosis. Registering baseline and follow-up inspiration images offers additionally valuable insights into disease progression.

Advanced Quantification Pipeline Reveals New Spatial and Temporal Tumor Characteristics in Preclinical Multiple Myeloma

Background.Radiological imaging plays an indispensable role in both preclinical and clinical studies of multiple myeloma (MM). However, manual quantification in longitudinal small animal PET/CT is limited by annotator bias, signal artifacts from urinary/fecal excretion, and voxel misalignment due to non-rigid registration. To address these challenges and improve characterization of tumor biology, we developed a semi-automated PET/CT quantification pipeline targeting defined regions of interest (ROIs) within the bone marrow-rich mouse skeleton, achieving sub-organ spatial resolution, including in anatomically complex sites such as the pelvis. We applied this MM-specific preclinical pipeline to analyze tumor distribution in a longitudinal molecular PET study using an immunocompetent mouse model of skeletally disseminated MM. An Attention U-Net was trained to segment the thoracolumbar spine, pelvis and pelvic joints, sacrum, and femurs from 2D CT slices. A custom algorithm masked spillover signal from physiological excretion, and a PCA-based projection was used to map tumor distribution along the skeletal axis. Quantification metrics included mean and maximum standardized uptake values (SUVmean, SUVmax) from PET and Hounsfield Units (HU) from CT to assess tumor burden, spatiotemporal tumor distribution, and bone involvement.Results.Tumor burden localized preferentially to skeletal regions near joints. Using precise CT-based alignment (DICE = 0.966 ± 0.005), we detected early disease progression and aggressive phenotypes. A marked increase in tumor uptake was observed by day 18 post-implantation, with significant SUVmean increases in the spine (p = 0.012), left/right femurs (p = 0.007/0.006), pelvis and pelvic joints (p = 0.018), and sacrum (p = 0.02). Notably, sex-based differences were identified: female mice showed greater bone loss near the hip joint at later stages, with significant HUmean reductions at days 25 (p = 0.008) and 32 (p = 0.002).Conclusions.This pipeline enables reproducible, anatomically precise quantification of region-specific trends in MM progression, including joint-specific lesion tropism and sex-based differences, from longitudinal PET/CT scans. By mitigating common challenges such as excretion artifacts and inconsistent mouse positioning, our approach overcomes limitations of manual analysis and enhances evaluation of tumor biology and treatment response in preclinical models of bone-involved cancers.

Nonrigid Multimodal Registration Based on Fuzzy Inference System for Retinal Image Registration.

Retinal imaging employs various modalities, each providing distinct perspectives on ocular structures. However, the integration of information from these modalities, which often have differing resolutions, requires effective image registration techniques. Existing retinal image registration methods typically rely on rigid or affine transformations, which may not adequately address the complexities of multimodal retinal images. This study introduces a nonrigid fuzzy image registration approach designed to align optical coherence tomography (OCT) images with fundus images. The method employs a fuzzy inference system (FIS) that uses vessel locations as key features for registration. The FIS applies specific rules to map points from the source image to the reference image, facilitating accurate alignment. The proposed method achieved a mean absolute registration error of 44.57 ± 39.38 µm in the superior-inferior orientation and 11.46 ± 10.06 µm in the nasal-temporal orientation. These results underscore the method's precision in aligning multimodal retinal images. The nonrigid fuzzy image registration approach demonstrates robust and versatile performance in integrating multimodal retinal imaging data. Despite its straightforward implementation, the method effectively addresses the challenges of multimodal retinal image registration, providing a reliable tool for advanced ocular imaging analysis.

Brain multi modality image inpainting via deep learning based edge region generative adversarial network.

A brain tumor (BT) is considered one of the most crucial and deadly diseases in the world, as it affects the central nervous system and its main functions. Headaches, nausea, and balance problems are caused by tumors pressing on nearby brain tissue and affecting its function. The existing techniques are challenging to analyze diseased brain images since abnormal brain tissues lead to distorted or biased results during image processing, like tissue segmentation and non-rigid registration. To overcome these issues, proposed a DS-GAN model for inpainting brain MRI images. Initially, the input MRI images are segmented using a Gated shape convolution neural network (GS-CNN). In the first GAN, grayscale pixel intensities and the remaining image edges are utilized to create edge generators or edge reconstruction Generative Adversarial Networks (EGAN), which are capable of creating false edges in areas that are missing. The results of the experimental results demonstrated that the Jaccard Index (JI) was 0.82, while the Dice Index (DI) was 0.86. The proposed DS-GAN in terms of L1 loss, PSNR, SSIM, and MSE obtained was 2.18, 0.972, 32.04, and 26.42. As compared to existing techniques, the proposed DS-GAN model achieves an overall accuracy of 99.18%.

Guided ultrasound acquisition for nonrigid image registration using reinforcement learning.

We propose a guided registration method for spatially aligning a fixed preoperative image and untracked ultrasound image slices. We exploit the unique interactive and spatially heterogeneous nature of this application to develop a registration algorithm that interactively suggests and acquires ultrasound images at optimised locations (with respect to registration performance). Our framework is based on two trainable functions: (1) a deep hyper-network-based registration function, which is generalisable over varying location and deformation, and adaptable at test-time; (2) a reinforcement learning function for producing test-time estimates of image acquisition locations and adapted deformation regularisation (the latter is required due to varying acquisition locations). We evaluate our proposed method with real preoperative patient data, and simulated intraoperative data with variable field-of-view. In addition to simulation of intraoperative data, we simulate global alignment based on previous work for efficient training, and investigate probe-level guidance towards an improved deformable registration. The evaluation in a simulated environment shows statistically significant improvements in overall registration performance across a variety of metrics for our proposed method, compared to registration without acquisition guidance or adaptable deformation regularisation, and to commonly used classical iterative methods and learning-based registration. For the first time, efficacy of proactive image acquisition is demonstrated in a simulated surgical interventional registration, in contrast to most existing work addressing registration post-data-acquisition, one of the reasons we argue may have led to previously under-constrained nonrigid registration in such applications. Code: https://github.com/s-sd/rl_guided_registration.

A robust approach for analyzing and mapping hierarchical brain connectome towards laminar-specific neural networks

Abstract Probing neuronal activity and functional connectivity at cortical layer and sub-cortical nucleus level provides opportunities for mapping local and remote neural circuits and resting-state networks (RSN) critical for understanding cognition and behaviors. However, conventional resting-state fMRI (rs-fMRI) has been applied predominantly at relatively low spatial resolution and macroscopic level, unable to obtain laminar-specific information and neural circuits across the cortex at mesoscopic level. In addition, it is lack of sophisticated processing pipeline to deal with small laminar structures in rodent brains. To fill this gap, we conducted a high-resolution rs-fMRI study of mouse brain at ultra-high field and developed an fMRI preprocessing pipeline that features in random matrix theory-based principal component analysis to remove thermal noise, non-rigid image registration strategy to improve head motion estimation, one-time image voxel shift correction to minimize multi-interpolation-induced spatial blur, and improve subject-level alignment to facilitate group analysis. By applying this pipeline to the high-resolution mouse rs-fMRI with atlas-based connectivity analysis, we achieved high-quality hierarchical connectomes covering from large brain regions to cortical layers, and between white matter bundle fibers and cortices in mice. We demonstrate the hierarchical connectomes connecting to three representative brain regions: somatosensory areas, hippocampal regions, and lateral forebrain white matter bundles, showing previously undetected networks. The distinct laminar-specific networks evidence that the spontaneous neuronal activity is not uniform across the cortical layers in the resting brain, consistent with the layer-specific neuronal projection patterns that were observed in AAV viral tracer projections. Additionally, we also observed extended functional connections in areas with sparse viral tracer projections. The feasibility of achieving laminar-specific connectomes with distinct RSNs provides opportunities to study neural circuits and brain functions at multiple scales, though achieving high fidelity and specificity in mapping laminar-specific connectomes may require even higher spatial resolution.

Abstract 150: Integrating spatial metabolomics and spatial transcriptomics on the same cancer tissue sections to detect gene-metabolite correlations

Abstract The combination of spatial metabolomics and spatial transcriptomics can reveal powerful connections between gene expression and metabolism in heterogeneous tissues. However, section-to-section variability can convolute data integration. We present a novel approach, termed Desium, which integrates spatial metabolomics by Desorption Electrospray Ionization Mass Spectrometry Imaging (DESI-MSI) and Visium Spatial Transcriptomics (VST) on the same tissue section. This work demonstrates that DESI-MSI does not significantly impact RNA quality, enabling robust analysis of gene-metabolite relationships in human breast and lung cancer tissues. The Desium workflow involves sequentially performing DESI-MSI at 100 μm spatial resolution, hematoxylin and eosin (H&amp;E) staining, and VST CytAssist analyses on the same tissue section. We applied this workflow to six human cancer tissues (3 breast, 3 lung) and analyzed a serial section of two samples (1 breast, 1 lung) with standard VST CytAssist as a control. H&amp;E and VST data were automatically aligned, while DESI-MSI and H&amp;E data were coregistered manually using nonrigid registration. The MSI data was then converted into VST spot coordinates using a Gaussian granularity matching algorithm. Pearson correlation scores were calculated for each tissue to link gene expression and metabolite abundances. Comparing sequencing results between Desium and the standard VST protocol using UMI counts/gene, we found a correlation coefficient of 1 for both lung and breast samples, indicating minimal impact of DESI-MSI on gene expression measurements. Across all samples, we identified 15, 111 transcript-metabolite correlations from 1, 136 transcripts and 439 metabolites, with correlation scores above 0.5 or below -0.5. Notably, we observed a 0.82 correlation between glycerophosphorylethanolamine and XBP1 in breast cancer and a 0.77 correlation between phosphatidylglycerol PG 22:6_22:5 and MSLN in lung cancer. Our study demonstrates that Desium enables unambiguous spatial correlations between metabolites and RNA transcripts, uncovering novel relationships and unique patterns of intratumor heterogeneity not evident through histologic analysis. These findings suggest that Desium could serve as a powerful tool to explore the gene expression-metabolic phenotype relationships within the complex tumor microenvironment. Citation Format: Trevor M. Godfrey, Yasmin Shanneik, Wanqiu Zhang, Thao Tran, Nico Verbeeck, Nathan H. Patterson, Faith E. Jackobs, Maheswhari Ramineni, Chandandeep Nagi, Livia S. Eberlin. Integrating spatial metabolomics and spatial transcriptomics on the same cancer tissue sections to detect gene-metabolite correlations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 150.

Multi-contrast computed tomography atlas of healthy pancreas with dense displacement sampling registration.

Diverse population demographics can lead to substantial variation in the human anatomy. Therefore, standard anatomical atlases are needed for interpreting organ-specific analyses. Among abdominal organs, the pancreas exhibits notable variability in volumetric morphology, shape, and appearance, complicating the generalization of population-wide features. Understanding the common features of a healthy pancreas is crucial for identifying biomarkers and diagnosing pancreatic diseases. We propose a high-resolution CT atlas framework optimized for the healthy pancreas. We introduce a deep-learning-based preprocessing technique to extract abdominal ROIs and leverage a hierarchical registration pipeline to align pancreatic anatomy across populations. Briefly, DEEDS affine and non-rigid registration techniques are employed to transfer patient abdominal volumes to a fixed high-resolution atlas template. To generate and evaluate the pancreas atlas, multi-phase contrast CT scans of 443 subjects (aged 15 to 50 years, with no reported history of pancreatic disease) were processed. The two-stage DEEDS affine and non-rigid registration outperforms other state-of-the-art tools, achieving the highest scores for pancreas label transfer across all phases (non-contrast: 0.497, arterial: 0.505, portal venous: 0.494, delayed: 0.497). External evaluation with 100 portal venous scans and 13 labeled abdominal organs shows a mean Dice score of 0.504. The low variance between the pancreases of registered subjects and the obtained pancreas atlas further illustrates the generalizability of the proposed method. We introduce a high-resolution pancreas atlas framework to generalize healthy biomarkers across populations with multi-contrast abdominal CT. The atlases and the associated pancreas organ labels are publicly available through the Human Biomolecular Atlas Program (HuBMAP).

Wavelet-Guided Multi-Scale ConvNeXt for Unsupervised Medical Image Registration

Medical image registration is essential in clinical practices such as surgical navigation and image-guided diagnosis. The Transformer architecture of TransMorph demonstrates better accuracy in non-rigid registration tasks. However, its weaker spatial locality priors necessitate large-scale training datasets and a heavy number of parameters, which conflict with the limited annotated data and real-time demands of clinical workflows. Moreover, traditional downsampling and upsampling always degrade high-frequency anatomical features such as tissue boundaries or small lesions. We proposed WaveMorph, a wavelet-guided multi-scale ConvNeXt method for unsupervised medical image registration. A novel multi-scale wavelet feature fusion downsampling module is proposed by integrating the ConvNeXt architecture with Haar wavelet lossless decomposition to extract and fuse features from eight frequency sub-images using multi-scale convolution kernels. Additionally, a lightweight dynamic upsampling module is introduced in the decoder to reconstruct fine-grained anatomical structures. WaveMorph integrates the inductive bias of CNNs with the advantages of Transformers, effectively mitigating topological distortions caused by spatial information loss while supporting real-time inference. In both atlas-to-patient (IXI) and inter-patient (OASIS) registration tasks, WaveMorph demonstrates state-of-the-art performance, achieving Dice scores of 0.779 ± 0.015 and 0.824 ± 0.021, respectively, and real-time inference (0.072 s/image), validating the effectiveness of our model in medical image registration.

Evaluation of Image Quality of Temporal Maximum Intensity Projection and Average Intensity Projection of Adaptive 4D-Spiral CT Scans: A Phantom Study.

Adaptive four-dimensional (4D) spiral computed tomography (CT) scans facilitate the acquisition of volume perfusion data for organs or long-range vessels; however, optimizing image quality and reducing noise while minimizing radiation doses remains challenging. Thus, image-processing techniques such as temporal maximum intensity projection (MIP) and average intensity projection (AIP) are crucial in this context. This ex vivo study aimed to compare the image noise, spatial resolution, and measurements of temporal MIP and AIP images generated from low radiation dose 4D CT scans data with those of conventional CT images using phantoms. Three phantoms were scanned with equivalent radiation doses using single helical and adaptive 10-phase 4D spiral scans using a third-generation dual-source CT scanner. Temporal MIP and AIP images of 4D CT scans were generated by summing varying numbers of phases, incorporating automatic motion correction with non-rigid registration and noise reduction algorithm. The CT values and image noise of the temporal MIP and AIP images were compared to conventional CT images. The task transfer function (TTF) was calculated using static phantoms. Vessel diameters of the phantoms for each image dataset were evaluated using motion phantoms. Temporal AIP images showed comparable CT values with those of the reference image. In contrast, the CT values of the temporal MIP images were significantly higher than those of the reference images (p<0.01). The image noise of temporal AIP images with six or more phases was equal to or lower than that of the reference images. In contrast, temporal MIP images exhibited consistently high noise levels regardless of the number of summed phases. The TTF of temporal AIP images was comparable to that of the reference CT images. However, the TTF of temporal MIP images gradually decreased as the number of summed phases increased. No significant differences were observed in vessel diameter measurements among the three groups or with varying numbers of summed phases (p>0.05). In conclusion, temporal MIP and AIP images generated from low radiation dose 4D CT scans could effectively reduce noise while preserving measurement reliability in the motion phantom, achieving performance comparable to conventional CT images.

Research on automatic spraying technology for 3D shoe uppers

Purpose This study aims to address the challenges of boundary detection and path discontinuities in the 3D shoe upper gluing process by proposing a novel method for robotic trajectory planning. Design/methodology/approach The method begins with aligning sole and upper point clouds using non-rigid registration. The overlapping upper region is segmented via the K-nearest neighbor algorithm, and boundary points are extracted. Offset path is calculated based on local geometry to generate the robot’s glue path, ensuring smooth spraying by minimizing pose changes between adjacent points. Findings Experiments conducted on six shoe types demonstrate that the proposed method achieves a boundary recognition accuracy within 2 mm. Real-world glue spraying experiments confirm its effectiveness and practical feasibility for automated processes. Originality/value This study introduces a robust method for recognizing shoe upper boundaries and generating continuous robotic spray paths, addressing key limitations in current approaches and advancing automation in footwear manufacturing.

Constructing nasal prosthesis morphological data based on a nonrigid registration algorithm.

Constructing nasal prosthesis morphological data based on a nonrigid registration algorithm.