Deformable image registration estimates voxel-wise correspondences between images through spatial transformations, and plays a key role in medical imaging. While deep learning methods have significantly reduced runtime, efficiently handling large deformations remains a challenging task. Convolutional networks aggregate local features but lack direct modeling of voxel correspondences, promoting recent works to explore explicit feature matching. Among them, voxel-to-region matching is more efficient for direct correspondence modeling by computing local correlation features within neighbourhoods, while region-to-region matching incurs higher redundancy due to excessive correlation pairs across large regions. However, the inherent locality of voxel-to-region matching hinders the capture of long-range correspondences required for large deformations. To address this, we propose a Recurrent Correlation-based framework that dynamically relocates the matching region toward more promising positions. At each step, local matching is performed with low cost, and the estimated offset guides the next search region, supporting efficient convergence toward large deformations. In addition, we uses a lightweight recurrent update module with memory capacity and decouples motion-related and texture features to suppress semantic redundancy. We conduct extensive experiments on brain MRI and abdominal CT datasets under two settings: with and without affine pre-registration. Results show our method exhibits a strong accuracy-computation trade-off, surpassing or matching the state-of-the-art performance. For example, it achieves comparable performance on the non-affine OASIS dataset, while using only 9.5% of the FLOPs and running 96% faster than RDP, a representative high-performing method.

Real-time guidewire tracking and position estimation are crucial for intraoperative navigation during percutaneous coronary intervention (PCI). Compared with other interventional procedures, PCI employs smaller-sized guidewires that are extremely difficult to distinguish from surrounding anatomical structures due to their slenderness and complex anatomical backgrounds. Moreover, guidewires typically exhibit low signal-to-noise ratios (SNR) in fluoroscopic video sequences. Additionally, complex motion artifacts caused by patient breathing and cardiac movements further complicate real-time guidewire localization. To address these challenges, we propose a novel end-to-end framework for guidewire segmentation and localization. Given that background information in PCI images predominantly occupies low-frequency components, we designed a multi-scale feature aggregation module based on low-frequency suppression to enhance the network's capability for extracting slender structures such as guidewires. Furthermore, to handle discontinuities caused by low SNR and structural occlusions in preliminary segmentation results, we incorporated a lightweight reconstruction network along with a generative adversarial network (GAN) to repair broken regions. The two networks are trained jointly, and their outputs are ultimately fused to achieve accurate guidewire segmentation. Extensive experiments conducted on multi-center private datasets demonstrate the superior performance of our approach, with F1 score of 76.45% and a localization precision of 0.18 mm. With a compact model size of only 12.64 M parameters and a real-time processing speed of 46 FPS, our method offers a highly promising solution for future clinical applications.

Susceptibility-induced distortions in diffusion MRI (dMRI) data significantly affect the study of human brain fiber pathways. Most correction methods using B0 images in the opposite phase encodings (PEs) struggle to correct the distortion in certain regions due to the low intensity distribution in B0 images. Recently, some methods use complementary information from Fiber Orientation Distribution (FOD) images to further correct the residual distortion. However, these methods follow a two-step correction pipeline and require dMRI data acquisition with both PEs to estimate FOD images, limiting their applicability. To accommodate different acquisition protocols, we propose a novel distortion correction framework MAC. Specifically, by mapping the B0 image onto the intensity distribution of the FOD image, our proposed MAC extracts complementary features from multiple modalities and progressively estimates the distortion field through cross-modal attention embedding. Extensive experiments in multiple datasets demonstrated the superiority and robustness of our method in all modal combinations.

Encoder-Decoder architectures are widely used in deep learning-based Deformable Image Registration (DIR), where the encoder extracts multi-scale features and the decoder predicts deformation fields by recovering spatial locations. However, current methods lack specialized extraction of features (that are useful for registration) and predict deformation jointly and homogeneously in all three directions. In this paper, we propose a novel expert-guided DIR network with Mixture of Experts (MoE) mechanism applied in both encoder and decoder, named SHMoAReg. Specifically, we incorporate Mixture of Attention heads (MoA) into encoder layers, while Spatial Heterogeneous Mixture of Experts (SHMoE) into the decoder layers. The MoA enhances the specialization of feature extraction by dynamically selecting the optimal combination of attention heads for each image token. Meanwhile, the SHMoE predicts deformation fields heterogeneously in three directions for each voxel using experts with varying kernel sizes. Extensive experiments conducted on two publicly available datasets show consistent improvements over various methods, with a notable increase from 60.58% to 65.58% in Dice score for the abdominal CT dataset. To the best of our knowledge, we are the first to introduce MoE mechanism into DIR tasks.

Optical coherence tomography angiography (OCTA) plays a crucial role in quantifying and analyzing retinal vascular diseases. However, the limited field of view (FOV) inherent in most commercial OCTA imaging systems poses a significant challenge for clinicians, restricting the possibility to analyze larger retinal regions of high resolution. Automatic stitching of OCTA scans in adjacent regions may provide a promising solution to extend the region of interest. However, commonly-used stitching algorithms face difficulties in achieving effective alignment due to noise, artifacts and dense vasculature present in OCTA images. To address these challenges, we propose a novel retinal OCTA image stitching network, named -Net, which integrates multi-scale representation learning and dynamic location guidance. In the first stage, an image registration network with a progressive multi-resolution feature fusion is proposed to derive deep semantic information effectively. Additionally, we introduce a dynamic guidance strategy to locate the foveal avascular zone (FAZ) and constrain registration errors in overlapping vascular regions. In the second stage, an image fusion network based on multiple mask constraints and adjacent image aggregation (AIA) strategies is developed to further eliminate the artifacts in the overlapping areas of stitched images, thereby achieving precise vessel alignment. To validate the effectiveness of our method, we conduct a series of experiments on two delicately constructed datasets, i.e., OPTOVUE-OCTA and SVision-OCTA. Experimental results demonstrate that our method outperforms other image stitching methods and effectively generates high-quality wide-field OCTA images, achieving a structural similarity index (SSIM) score of 0.8264 and 0.8014 on the two datasets, respectively.

Susceptibility induced distortion is a major artifact that affects the diffusion MRI (dMRI) data analysis. In the Human Connectome Project (HCP), the state-of-the-art method adopted to correct this kind of distortion is to exploit the displacement field from the B0 image in the reversed phase encoding images. However, both the traditional and learning-based approaches have limitations in achieving high correction accuracy in certain brain regions, such as brainstem. By utilizing the fiber orientation distribution (FOD) computed from the dMRI, we propose a novel deep learning framework named DistoRtion Correction Net (DrC-Net), which consists of the U-Net to capture the latent information from the 4D FOD images and the spatial transformer network to propagate the displacement field and back propagate the losses between the deformed FOD images. The experiments are performed on two datasets acquired with different phase encoding (PE) directions including the HCP and the Human Connectome Low Vision (HCLV) dataset. Compared to two traditional methods topup and FODReg and two deep learning methods S-Net and flow-net, the proposed method achieves significant improvements in terms of the mean squared difference (MSD) of fractional anisotropy (FA) images and minimum angular difference between two PEs in white matter and also brainstem regions. In the meantime, the proposed DrC-Net takes only several seconds to predict a displacement field, which is much faster than the FODReg method.

Optical Coherence Tomography Angiography (OCTA) is a novel, non-invasive imaging modality of retinal capillaries at micron resolution. Recent studies have correlated macular OCTA vascular measures with retinal disease severity and supported their use as a diagnostic tool. However, these measurements mostly rely on a few summary statistics in retinal layers or regions of interest in the two-dimensional (2D) en face projection images. To enable 3D and localized comparisons of retinal vasculature between longitudinal scans and across populations, we develop a novel approach for mapping retinal vessel density from OCTA images. We first obtain a high-quality 3D representation of OCTA-based vessel networks via curvelet-based denoising and optimally oriented flux (OOF). Then, an effective 3D retinal vessel density mapping method is proposed. In this framework, a vessel density image (VDI) is constructed by diffusing the vessel mask derived from OOF-based analysis to the entire image volume. Subsequently, we utilize a non-linear, 3D OCT image registration method to provide localized comparisons of retinal vasculature across subjects. In our experimental results, we demonstrate an application of our method for longitudinal qualitative analysis of two pathological subjects with edema during the course of clinical care. Additionally, we quantitatively validate our method on synthetic data with simulated capillary dropout, a dataset obtained from a normal control (NC) population divided into two age groups and a dataset obtained from patients with diabetic retinopathy (DR). Our results show that we can successfully detect localized vascular changes caused by simulated capillary loss, normal aging, and DR pathology even in presence of edema. These results demonstrate the potential of the proposed framework in localized detection of microvascular changes and monitoring retinal disease progression.

Diabetic retinopathy (DR) is a significant microvascular complication of diabetes mellitus and a leading cause of vision impairment in working age adults. Optical coherence tomography (OCT) is a routinely used clinical tool to observe retinal structural and thickness alterations in DR. Pathological changes that alter the normal anatomy of the retina, such as intraretinal edema, pose great challenges for conventional layer-based analysis of OCT images. We present an alternative approach for the automated analysis of OCT volumes in DR research based on nonlinear registration. In this paper, we first obtain an anatomically consistent volume of interest (VOI) in different OCT images via carefully designed masking and affine registration. After that, efficient B-spline transformations are computed using stochastic gradient descent optimization. Using the OCT volumes of normal controls, for which layer-based segmentation works well, we demonstrate the accuracy of our registration-based analysis in aligning layer boundaries. By nonlinearly registering the OCT volumes of DR subjects to an atlas constructed from normal controls and measuring the Jacobian determinant of the deformation, we can simultaneously visualize tissue contraction and expansion due to DR pathology. Tensor-based morphometry (TBM) can also be performed for quantitative analysis of local structural changes. In our experimental results, we apply our method to a dataset of 105 subjects and demonstrate that volumetric OCT registration and TBM analysis can successfully detect local retinal structural alterations due to DR.

3D optical coherence tomography angiography (OCT-A) is a novel and non-invasive imaging modality for analyzing retinal diseases. The studies of microvasculature in 2D en face projection images have been widely implemented, but comprehensive 3D analysis of OCT-A images with rich depth-resolved microvascular information is rarely considered. In this paper, we propose a robust, effective, and automatic 3D shape modeling framework to provide a high-quality 3D vessel representation and to preserve valuable 3D geometric and topological information for vessel analysis. Effective vessel enhancement and extraction steps by means of curvelet denoising and optimally oriented flux (OOF) filtering are first designed to produce 3D microvascular networks. Afterwards, a novel 3D data representation of OCT-A microvasculature is reconstructed via advanced mesh reconstruction techniques. Based on the 3D surfaces, shape analysis is established to extract novel shape-based microvascular area distortion via the Laplace-Beltrami eigen-projection. The extracted feature is integrated into a graph-cut segmentation system to categorize large vessels and small capillaries for more precise shape analysis. The proposed framework is validated on a dedicated repeated scan dataset including 260 volume images and shows high repeatability. Statistical analysis using the surface area biomarker is performed on small capillaries to avoid the effect of tailing artifact from large vessels. It shows significant differences ( ) between DR stages on 100 subjects in a OCTA-DR dataset. The proposed shape modeling and analysis framework opens the possibility for further investigating OCT-A microvasculature in a new perspective.

Stochastic gradient descent (SGD) is commonly used to solve (parametric) image registration problems. In the case of badly scaled problems, SGD, however, only exhibits sublinear convergence properties. In this paper, we propose an efficient preconditioner estimation method to improve the convergence rate of SGD. Based on the observed distribution of voxel displacements in the registration, we estimate the diagonal entries of a preconditioning matrix, thus rescaling the optimization cost function. The preconditioner is efficient to compute and employ and can be used for mono-modal as well as multi-modal cost functions, in combination with different transformation models, such as the rigid, the affine, and the B-spline model. Experiments on different clinical datasets show that the proposed method, indeed, improves the convergence rate compared with SGD with speedups around 2~5 in all tested settings while retaining the same level of registration accuracy.

Fast automatic image registration is an important prerequisite for image-guided clinical procedures. However, due to the large number of voxels in an image and the complexity of registration algorithms, this process is often very slow. Stochastic gradient descent is a powerful method to iteratively solve the registration problem, but relies for convergence on a proper selection of the optimization step size. This selection is difficult to perform manually, since it depends on the input data, similarity measure and transformation model. The Adaptive Stochastic Gradient Descent (ASGD) method is an automatic approach, but it comes at a high computational cost. In this paper, we propose a new computationally efficient method (fast ASGD) to automatically determine the step size for gradient descent methods, by considering the observed distribution of the voxel displacements between iterations. A relation between the step size and the expectation and variance of the observed distribution is derived. While ASGD has quadratic complexity with respect to the transformation parameters, fast ASGD only has linear complexity. Extensive validation has been performed on different datasets with different modalities, inter/intra subjects, different similarity measures and transformation models. For all experiments, we obtained similar accuracy as ASGD. Moreover, the estimation time of fast ASGD is reduced to a very small value, from 40 s to less than 1 s when the number of parameters is 105, almost 40 times faster. Depending on the registration settings, the total registration time is reduced by a factor of 2.5-7 × for the experiments in this paper.