Generalizing Spatial Transformers to Projective Geometry with Applications to 2D/3D Registration

PRSCS-Net: Progressive 3D/2D rigid Registration network with the guidance of Single-view Cycle Synthesis

The 3D/2D registration for 3D pre-operative images (computed tomography, CT) and 2D intra-operative images (X-ray) plays an important role in image-guided spine surgeries. Conventional iterative-based approaches suffer from time-consuming processes. Existing learning-based approaches require high computational costs and face poor performance on large misalignment because of projection-induced losses or ill-posed reconstruction. In this paper, we propose a Progressive 3D/2D rigid Registration network with the guidance of Single-view Cycle Synthesis, named PRSCS-Net. Specifically, we first introduce the differentiable backward/forward projection operator into the single-view cycle synthesis network, which reconstructs corresponding 3D geometry features from two 2D intra-operative view images (one from the input, and the other from the synthesis). In this way, the problem of limited views during reconstruction can be solved. Subsequently, we employ a self-reconstruction path to extract latent representation from pre-operative 3D CT images. The following pose estimation process will be performed in the 3D geometry feature space, which can solve the dimensional gap, greatly reduce the computational complexity, and ensure that the features extracted from pre-operative and intra-operative images are as relevant as possible to pose estimation. Furthermore, to enhance the ability of our model for handling large misalignment, we develop a progressive registration path, including two sub-registration networks, aiming to estimate the pose parameters via two-step warping volume features. Finally, our proposed method has been evaluated on a public dataset CTSpine1k and an in-house dataset C-ArmLSpine for 3D/2D registration. Results demonstrate that PRSCS-Net achieves state-of-the-art registration performance in terms of registration accuracy, robustness, and generalizability compared with existing methods. Thus, PRSCS-Net has potential for clinical spinal disease surgical planning and surgical navigation systems.

Deformable motion compensation in interventional cone-beam CT with a context-aware learned autofocus metric

PURPOSE

Interventional Cone-Beam CT (CBCT) offers 3D visualization of soft-tissue and vascular anatomy, enabling 3D guidance of abdominal interventions. However, its long acquisition time makes CBCT susceptible to patient motion. Image-based autofocus offers a suitable platform for compensation of deformable motion in CBCT, but it relies on handcrafted motion metrics based on first-order image properties and that lack awareness of the underlying anatomy. This work proposes a data-driven approach to motion quantification via a learned, context-aware, deformable metric,VI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ , that quantifies the amount of motion degradation as well as the realism of the structural anatomical content in the image.

METHODS

The proposedVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ was modeled as a deep convolutional neural network (CNN) trained to recreate a reference-based structural similarity metric-visual information fidelity (VIF). The deep CNN acted on motion-corrupted images, providing an estimation of the spatial VIF map that would be obtained against a motion-free reference, capturing motion distortion, and anatomic plausibility. The deep CNN featured a multi-branch architecture with a high-resolution branch for estimation of voxel-wise VIF on a small volume of interest. A second contextual, low-resolution branch provided features associated to anatomical context for disentanglement of motion effects and anatomical appearance. The deep CNN was trained on paired motion-free and motion-corrupted data obtained with a high-fidelity forward projection model for a protocol involving 120 kV and 9.90 mGy. The performance ofVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ was evaluated via metrics of correlation with ground truth VIF ${\bm{VIF}}$ and with the underlying deformable motion field in simulated data with deformable motion fields with amplitude ranging from 5 to 20 mm and frequency from 2.4 up to 4 cycles/scan. Robustness to variation in tissue contrast and noise levels was assessed in simulation studies with varying beam energy (90-120 kV) and dose (1.19-39.59 mGy). Further validation was obtained on experimental studies with a deformable phantom. Final validation was obtained via integration ofVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ on an autofocus compensation framework, applied to motion compensation on experimental datasets and evaluated via metric of spatial resolution on soft-tissue boundaries and sharpness of contrast-enhanced vascularity.

RESULTS

The magnitude and spatial map ofVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ showed consistent and high correlation levels with the ground truth in both simulation and real data, yielding average normalized cross correlation (NCC) values of 0.95 and 0.88, respectively. Similarly,VI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ achieved good correlation values with the underlying motion field, with average NCC of 0.90. In experimental phantom studies,VI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ properly reflects the change in motion amplitudes and frequencies: voxel-wise averaging of the localVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ across the full reconstructed volume yielded an average value of 0.69 for the case with mild motion (2 mm, 12 cycles/scan) and 0.29 for the case with severe motion (12 mm, 6 cycles/scan). Autofocus motion compensation usingVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ resulted in noticeable mitigation of motion artifacts and improved spatial resolution of soft tissue and high-contrast structures, resulting in reduction of edge spread function width of 8.78% and 9.20%, respectively. Motion compensation also increased the conspicuity of contrast-enhanced vascularity, reflected in an increase of 9.64% in vessel sharpness.

CONCLUSION

The proposedVI F D L ${\bm{VI}}{{\bm{F}}}_{DL}$ , featuring a novel context-aware architecture, demonstrated its capacity as a reference-free surrogate of structural similarity to quantify motion-induced degradation of image quality and anatomical plausibility of image content. The validation studies showed robust performance across motion patterns, x-ray techniques, and anatomical instances. The proposed anatomy- and context-aware metric poses a powerful alternative to conventional motion estimation metrics, and a step forward for application of deep autofocus motion compensation for guidance in clinical interventional procedures.

Scan-Free and Fully Automatic Tracking of Native Knee Anatomy from Dynamic Stereo-Radiography with Statistical Shape and Intensity Models

Kinematic tracking of native anatomy from stereo-radiography provides a quantitative basis for evaluating human movement. Conventional tracking procedures require significant manual effort and call for acquisition and annotation of subject-specific volumetric medical images. The current work introduces a framework for fully automatic tracking of native knee anatomy from dynamic stereo-radiography which forgoes reliance on volumetric scans. The method consists of three computational steps. First, captured radiographs are annotated with segmentation maps and anatomic landmarks using a convolutional neural network. Next, a non-convex polynomial optimization problem formulated from annotated landmarks is solved to acquire preliminary anatomy and pose estimates. Finally, a global optimization routine is performed for concurrent refinement of anatomy and pose. An objective function is maximized which quantifies similarities between masked radiographs and digitally reconstructed radiographs produced from statistical shape and intensity models. The proposed framework was evaluated against manually tracked trials comprising dynamic activities, and additional frames capturing a static knee phantom. Experiments revealed anatomic surface errors routinely below 1.0 mm in both evaluation cohorts. Median absolute errors of individual bone pose estimates were below 1.0∘ or mm for 15 out of 18 degrees of freedom in both evaluation cohorts. Results indicate that accurate pose estimation of native anatomy from stereo-radiography may be performed with significantly reduced manual effort, and without reliance on volumetric scans.

A Fully Differentiable Framework for 2D/3D Registration and the Projective Spatial Transformers

Image-based 2D/3D registration is a critical technique for fluoroscopic guided surgical interventions. Conventional intensity-based 2D/3D registration approa- ches suffer from a limited capture range due to the presence of local minima in hand-crafted image similarity functions. In this work, we aim to extend the 2D/3D registration capture range with a fully differentiable deep network framework that learns to approximate a convex-shape similarity function. The network uses a novel Projective Spatial Transformer (ProST) module that has unique differentiability with respect to 3D pose parameters, and is trained using an innovative double backward gradient-driven loss function. We compare the most popular learning-based pose regression methods in the literature and use the well-established CMAES intensity-based registration as a benchmark. We report registration pose error, target registration error (TRE) and success rate (SR) with a threshold of 10mm for mean TRE. For the pelvis anatomy, the median TRE of ProST followed by CMAES is 4.4mm with a SR of 65.6% in simulation, and 2.2mm with a SR of 73.2% in real data. The CMAES SRs without using ProST registration are 28.5% and 36.0% in simulation and real data, respectively. Our results suggest that the proposed ProST network learns a practical similarity function, which vastly extends the capture range of conventional intensity-based 2D/3D registration. We believe that the unique differentiable property of ProST has the potential to benefit related 3D medical imaging research applications. The source code is available at https://github.com/gaocong13/Projective-Spatial-Transformers.

Fully automatic tracking of native glenohumeral kinematics from stereo-radiography

The current work introduces a system for fully automatic tracking of native glenohumeral kinematics in stereo-radiography sequences. The proposed method first applies convolutional neural networks to obtain segmentation and semantic key point predictions in biplanar radiograph frames. Preliminary bone pose estimates are computed by solving a non-convex optimization problem with semidefinite relaxations to register digitized bone landmarks to semantic key points. Initial poses are then refined by registering computed tomography-based digitally reconstructed radiographs to captured scenes, which are masked by segmentation maps to isolate the shoulder joint. A particular neural net architecture which exploits subject-specific geometry is also introduced to improve segmentation predictions and increase robustness of subsequent pose estimates. The method is evaluated by comparing predicted glenohumeral kinematics to manually tracked values from 17 trials capturing 4 dynamic activities. Median orientation differences between predicted and ground truth poses were 1.7∘ and 8.6∘ for the scapula and humerus, respectively. Joint-level kinematics differences were less than 2∘ in 65%, 13%, and 63% of frames for XYZ orientation DoFs based on Euler angle decompositions. Automation of kinematic tracking can increase scalability of tracking workflows in research, clinical, or surgical applications.

2D/3D Non-Rigid Image Registration via Two Orthogonal X-ray Projection Images for Lung Tumor Tracking

Two-dimensional (2D)/three-dimensional (3D) registration is critical in clinical applications. However, existing methods suffer from long alignment times and high doses. In this paper, a non-rigid 2D/3D registration method based on deep learning with orthogonal angle projections is proposed. The application can quickly achieve alignment using only two orthogonal angle projections. We tested the method with lungs (with and without tumors) and phantom data. The results show that the Dice and normalized cross-correlations are greater than 0.97 and 0.92, respectively, and the registration time is less than 1.2 seconds. In addition, the proposed model showed the ability to track lung tumors, highlighting the clinical potential of the proposed method.

Mixed Reality Interfaces for Achieving Desired Views with Robotic X-ray Systems

Robotic X-ray C-arm imaging systems can precisely achieve any position and orientation relative to the patient. Informing the system, however, what pose exactly corresponds to a desired view is challenging. Currently these systems are operated by the surgeon using joysticks, but this interaction paradigm is not necessarily effective because users may be unable to efficiently actuate more than a single axis of the system simultaneously. Moreover, novel robotic imaging systems, such as the Brainlab Loop-X, allow for independent source and detector movements, adding even more complexity. To address this challenge, we consider complementary interfaces for the surgeon to command robotic X-ray systems effectively. Specifically, we consider three interaction paradigms: (1) the use of a pointer to specify the principal ray of the desired view relative to the anatomy, (2) the same pointer, but combined with a mixed reality environment to synchronously render digitally reconstructed radiographs from the tool's pose, and (3) the same mixed reality environment but with a virtual X-ray source instead of the pointer. Initial human-in-the-loop evaluation with an attending trauma surgeon indicates that mixed reality interfaces for robotic X-ray system control are promising and may contribute to substantially reducing the number of X-ray images acquired solely during "fluoro hunting" for the desired view or standard plane.

CAR-Net: A Deep Learning-Based Deformation Model for 3D/2D Coronary Artery Registration

Percutaneous coronary intervention is widely applied for the treatment of coronary artery disease under the guidance of X-ray coronary angiography (XCA) image. However, the projective nature of XCA causes the loss of 3D structural information, which hinders the intervention. This issue can be addressed by the deformable 3D/2D coronary artery registration technique, which fuses the pre-operative computed tomography angiography volume with the intra-operative XCA image. In this study, we propose a deep learning-based neural network for this task. The registration is conducted in a segment-by-segment manner. For each vessel segment pair, the centerlines that preserve topological information are decomposed into an origin tensor and a spherical coordinate shape tensor as network input through independent branches. Features of different modalities are fused and processed for predicting angular deflections, which is a special type of deformation field implying motion and length preservation constraints for vessel segments. The proposed method achieves an average error of 1.13 mm on the clinical dataset, which shows the potential to be applied in clinical practice.

Deep learning-based 2D/3D registration of an atlas to biplanar X-ray images

Purpose

The registration of a 3D atlas image to 2D radiographs enables 3D pre-operative planning without the need to acquire costly and high-dose CT-scans. Recently, many deep-learning-based 2D/3D registration methods have been proposed which tackle the problem as a reconstruction by regressing the 3D image immediately from the radiographs, rather than registering an atlas image. Consequently, they are less constrained against unfeasible reconstructions and have no possibility to warp auxiliary data. Finally, they are, by construction, limited to orthogonal projections.

Methods

We propose a novel end-to-end trainable 2D/3D registration network that regresses a dense deformation field that warps an atlas image such that the forward projection of the warped atlas matches the input 2D radiographs. We effectively take the projection matrix into account in the regression problem by integrating a projective and inverse projective spatial transform layer into the network.

Results

Comprehensive experiments conducted on simulated DRRs from patient CT images demonstrate the efficacy of the network. Our network yields an average Dice score of 0.94 and an average symmetric surface distance of 0.84 mm on our test dataset. It has experimentally been determined that projection geometries with 80\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{\circ }$$\end{document}∘ to 100\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{\circ }$$\end{document}∘ projection angle difference result in the highest accuracy.

Conclusion

Our network is able to accurately reconstruct patient-specific CT-images from a pair of near-orthogonal calibrated radiographs by regressing a deformation field that warps an atlas image or any other auxiliary data. Our method is not constrained to orthogonal projections, increasing its applicability in medical practices. It remains a future task to extend the network for uncalibrated radiographs.

The Impact of Machine Learning on 2D/3D Registration for Image-Guided Interventions: A Systematic Review and Perspective

Image-based navigation is widely considered the next frontier of minimally invasive surgery. It is believed that image-based navigation will increase the access to reproducible, safe, and high-precision surgery as it may then be performed at acceptable costs and effort. This is because image-based techniques avoid the need of specialized equipment and seamlessly integrate with contemporary workflows. Furthermore, it is expected that image-based navigation techniques will play a major role in enabling mixed reality environments, as well as autonomous and robot-assisted workflows. A critical component of image guidance is 2D/3D registration, a technique to estimate the spatial relationships between 3D structures, e.g., preoperative volumetric imagery or models of surgical instruments, and 2D images thereof, such as intraoperative X-ray fluoroscopy or endoscopy. While image-based 2D/3D registration is a mature technique, its transition from the bench to the bedside has been restrained by well-known challenges, including brittleness with respect to optimization objective, hyperparameter selection, and initialization, difficulties in dealing with inconsistencies or multiple objects, and limited single-view performance. One reason these challenges persist today is that analytical solutions are likely inadequate considering the complexity, variability, and high-dimensionality of generic 2D/3D registration problems. The recent advent of machine learning-based approaches to imaging problems that, rather than specifying the desired functional mapping, approximate it using highly expressive parametric models holds promise for solving some of the notorious challenges in 2D/3D registration. In this manuscript, we review the impact of machine learning on 2D/3D registration to systematically summarize the recent advances made by introduction of this novel technology. Grounded in these insights, we then offer our perspective on the most pressing needs, significant open problems, and possible next steps.