Reference-free learning-based similarity metric for motion compensation in cone-beam CT

Purpose. Patient motion artifacts present a prevalent challenge to image quality in interventional cone-beam CT (CBCT). We propose a novel reference-free similarity metric (DL-VIF) that leverages the capability of deep convolutional neural networks (CNN) to learn features associated with motion artifacts within realistic anatomical features. DL-VIF aims to address shortcomings of conventional metrics of motion-induced image quality degradation that favor characteristics associated with motion-free images, such as sharpness or piecewise constancy, but lack any awareness of the underlying anatomy, potentially promoting images depicting unrealistic image content. DL-VIF was integrated in an autofocus motion compensation framework to test its performance for motion estimation in interventional CBCT.Methods. DL-VIF is a reference-free surrogate for the previously reported visual image fidelity (VIF) metric, computed against a motion-free reference, generated using a CNN trained using simulated motion-corrupted and motion-free CBCT data. Relatively shallow (2-ResBlock) and deep (3-Resblock) CNN architectures were trained and tested to assess sensitivity to motion artifacts and generalizability to unseen anatomy and motion patterns. DL-VIF was integrated into an autofocus framework for rigid motion compensation in head/brain CBCT and assessed in simulation and cadaver studies in comparison to a conventional gradient entropy metric.Results. The 2-ResBlock architecture better reflected motion severity and extrapolated to unseen data, whereas 3-ResBlock was found more susceptible to overfitting, limiting its generalizability to unseen scenarios. DL-VIF outperformed gradient entropy in simulation studies yielding average multi-resolution structural similarity index (SSIM) improvement over uncompensated image of 0.068 and 0.034, respectively, referenced to motion-free images. DL-VIF was also more robust in motion compensation, evidenced by reduced variance in SSIM for various motion patterns (σ_DL-VIF = 0.008 versusσ_{gradient entropy} = 0.019). Similarly, in cadaver studies, DL-VIF demonstrated superior motion compensation compared to gradient entropy (an average SSIM improvement of 0.043 (5%) versus little improvement and even degradation in SSIM, respectively) and visually improved image quality even in severely motion-corrupted images.Conclusion: The studies demonstrated the feasibility of building reference-free similarity metrics for quantification of motion-induced image quality degradation and distortion of anatomical structures in CBCT. DL-VIF provides a reliable surrogate for motion severity, penalizes unrealistic distortions, and presents a valuable new objective function for autofocus motion compensation in CBCT.

Accelerated 3D image reconstruction with a morphological pyramid and noise-power convergence criterion

Model-based iterative reconstruction (MBIR) for cone-beam CT (CBCT) offers better noise-resolution tradeoff and image quality than analytical methods for acquisition protocols with low x-ray dose or limited data, but with increased computational burden that poses a drawback to routine application in clinical scenarios. This work develops a comprehensive framework for acceleration of MBIR in the form of penalized weighted least squares optimized with ordered subsets separable quadratic surrogates. The optimization was scheduled on a set of stages forming a morphological pyramid varying in voxel size. Transition between stages was controlled with a convergence criterion based on the deviation between the mid-band noise power spectrum (NPS) measured on a homogeneous region of the evolving reconstruction and that expected for the converged image, computed with an analytical model that used projection data and the reconstruction parameters. A stochastic backprojector was developed by introducing a random perturbation to the sampling position of each voxel for each ray in the reconstruction within a voxel-based backprojector, breaking the deterministic pattern of sampling artifacts when combined with an unmatched Siddon forward projector. This fast, forward and backprojector pair were included into a multi-resolution reconstruction strategy to provide support for objects partially outside of the field of view. Acceleration from ordered subsets was combined with momentum accumulation stabilized with an adaptive technique that automatically resets the accumulated momentum when it diverges noticeably from the current iteration update. The framework was evaluated with CBCT data of a realistic abdomen phantom acquired on an imaging x-ray bench and with clinical CBCT data from an angiography robotic C-arm (Artis Zeego, Siemens Healthineers, Forchheim, Germany) acquired during a liver embolization procedure. Image fidelity was assessed with the structural similarity index (SSIM) computed with a converged reconstruction. The accelerated framework provided accurate reconstructions in 60 s (SSIM = 0.97) and as little as 27 s (SSIM = 0.94) for soft-tissue evaluation. The use of simple forward and backprojectors resulted in 9.3× acceleration. Accumulation of momentum provided extra ∼3.5× acceleration with stable convergence for 6-30 subsets. The NPS-driven morphological pyramid resulted in initial faster convergence, achieving similar SSIM with 1.5× lower runtime than the single-stage optimization. Acceleration of MBIR to provide reconstruction time compatible with clinical applications is feasible via architectures that integrate algorithmic acceleration with approaches to provide stable convergence, and optimization schedules that maximize convergence speed.

Deformable motion compensation for interventional cone-beam CT

Image-guided therapies in the abdomen and pelvis are often hindered by motion artifacts in cone-beam CT (CBCT) arising from complex, non-periodic, deformable organ motion during long scan times (5-30 s). We propose a deformable image-based motion compensation method to address these challenges and improve CBCT guidance. Motion compensation is achieved by selecting a set of small regions of interest in the uncompensated image to minimize a cost function consisting of an autofocus objective and spatiotemporal regularization penalties. Motion trajectories are estimated using an iterative optimization algorithm (CMA-ES) and used to interpolate a 4D spatiotemporal motion vector field. The motion-compensated image is reconstructed using a modified filtered backprojection approach. Being image-based, the method does not require additional input besides the raw CBCT projection data and system geometry that are used for image reconstruction. Experimental studies investigated: (1) various autofocus objective functions, analyzed using a digital phantom with a range of sinusoidal motion magnitude (4, 8, 12, 16, 20 mm); (2) spatiotemporal regularization, studied using a CT dataset from The Cancer Imaging Archive with deformable sinusoidal motion of variable magnitude (10, 15, 20, 25 mm); and (3) performance in complex anatomy, evaluated in cadavers undergoing simple and complex motion imaged on a CBCT-capable mobile C-arm system (Cios Spin 3D, Siemens Healthineers, Forchheim, Germany). Gradient entropy was found to be the best autofocus objective for soft-tissue CBCT, increasing structural similarity (SSIM) by 42%-92% over the range of motion magnitudes investigated. The optimal temporal regularization strength was found to vary widely (0.5-5 mm^-2) over the range of motion magnitudes investigated, whereas optimal spatial regularization strength was relatively constant (0.1). In cadaver studies, deformable motion compensation was shown to improve local SSIM by ∼17% for simple motion and ∼21% for complex motion and provided strong visual improvement of motion artifacts (reduction of blurring and streaks and improved visibility of soft-tissue edges). The studies demonstrate the robustness of deformable motion compensation to a range of motion magnitudes, frequencies, and other factors (e.g. truncation and scatter).

Convergence criterion for MBIR based on the local noise-power spectrum: Theory and implementation in a framework for accelerated 3D image reconstruction with a morphological pyramid

Model-based iterative reconstruction (MBIR) offers improved noise-resolution tradeoffs and artifact reduction in cone-beam CT compared to analytical reconstruction, but carries increased computational burden. An important consideration in minimizing computation time is reliable selection of the stopping criterion to perform the minimum number of iterations required to obtain the desired image quality. Most MBIR methods rely on a fixed number of iterations or relative metrics on image or cost-function evolution, and it would be desirable to use metrics that are more representative of the underlying image properties. A second front for reduction of computation time is the use of acceleration techniques (e.g. subsets or momentum). However, most of these techniques do not strictly guarantee convergence of the resulting MBIR method. A data-dependent analytical model of noise-power spectrum (NPS) for penalized weighted least squares (PWLS) reconstruction is proposed as an absolute metric of image properties for the fully converged volume. Distance to convergence is estimated as the root mean squared error (RMSE) between the estimated NPS and an NPS measured on a uniform region of interest (ROI) in the evolving volume. Iterations are stopped when the RMSE falls below a threshold directly related with the properties of the target image. Further acceleration was achieved by combining the spectral stopping criterion with a morphological pyramid (mPyr) in which the minimization of the PWLS cost-function is divided in a cascade of stages. The algorithm parameters (voxel size in this work) change between stages to achieve faster evolution in early stages, and a final stage with the target parameters to guarantee convergence. Transition between stages is governed by the spectral stopping criterion. The approach was evaluated on simulated CBCT data of a realistic digital abdomen phantom. Accuracy of the NPS model and evolution with time of the distance from the measured NPS was assessed in two ROIs. Performance of the spectrally-driven mPyr architecture was compared to a conventional, single stage, PWLS, and to two mPyr designs running a fixed number of iterations. The spectrally-driven mPyr achieved faster convergence, with 40% lower RMSE than the single stage PWLS, and between 10% and 20% RMSE reduction compared to other mPyr designs. The proposed spectral stopping criterion proved to be a suitable choice for a stopping rule, and, in particular, to govern mPyr stage transition.

Correction of patient motion in cone-beam CT using 3D-2D registration

Cone-beam CT (CBCT) is increasingly common in guidance of interventional procedures, but can be subject to artifacts arising from patient motion during fairly long (~5-60 s) scan times. We present a fiducial-free method to mitigate motion artifacts using 3D-2D image registration that simultaneously corrects residual errors in the intrinsic and extrinsic parameters of geometric calibration. The 3D-2D registration process registers each projection to a prior 3D image by maximizing gradient orientation using the covariance matrix adaptation-evolution strategy optimizer. The resulting rigid transforms are applied to the system projection matrices, and a 3D image is reconstructed via model-based iterative reconstruction. Phantom experiments were conducted using a Zeego robotic C-arm to image a head phantom undergoing 5-15 cm translations and 5-15° rotations. To further test the algorithm, clinical images were acquired with a CBCT head scanner in which long scan times were susceptible to significant patient motion. CBCT images were reconstructed using a penalized likelihood objective function. For phantom studies the structural similarity (SSIM) between motion-free and motion-corrected images was  >0.995, with significant improvement (p  <  0.001) compared to the SSIM values of uncorrected images. Additionally, motion-corrected images exhibited a point-spread function with full-width at half maximum comparable to that of the motion-free reference image. Qualitative comparison of the motion-corrupted and motion-corrected clinical images demonstrated a significant improvement in image quality after motion correction. This indicates that the 3D-2D registration method could provide a useful approach to motion artifact correction under assumptions of local rigidity, as in the head, pelvis, and extremities. The method is highly parallelizable, and the automatic correction of residual geometric calibration errors provides added benefit that could be valuable in routine use.

Task-Driven Orbit Design and Implementation on a Robotic C-Arm System for Cone-Beam CT

PURPOSE

This work applies task-driven optimization to the design of non-circular orbits that maximize imaging performance for a particular imaging task. First implementation of task-driven imaging on a clinical robotic C-arm system is demonstrated, and a framework for orbit calculation is described and evaluated.

METHODS

We implemented a task-driven imaging framework to optimize orbit parameters that maximize detectability index d'. This framework utilizes a specified Fourier domain task function and an analytical model for system spatial resolution and noise. Two experiments were conducted to test the framework. First, a simple task was considered consisting of frequencies lying entirely on the f_z-axis (e.g., discrimination of structures oriented parallel to the central axial plane), and a "circle + arc" orbit was incorporated into the framework as a means to improve sampling of these frequencies, and thereby increase task-based detectability. The orbit was implemented on a robotic C-arm (Artis Zeego, Siemens Healthcare). A second task considered visualization of a cochlear implant simulated within a head phantom, with spatial frequency response emphasizing high-frequency content in the (f_y , f_z ) plane of the cochlea. An optimal orbit was computed using the task-driven framework, and the resulting image was compared to that for a circular orbit.

RESULTS

For the f_z -axis task, the circle + arc orbit was shown to increase d' by a factor of 1.20, with an improvement of 0.71 mm in a 3D edge-spread measurement for edges located far from the central plane and a decrease in streak artifacts compared to a circular orbit. For the cochlear implant task, the resulting orbit favored complementary views of high tilt angles in a 360° orbit, and d' was increased by a factor of 1.83.

CONCLUSIONS

This work shows that a prospective definition of imaging task can be used to optimize source-detector orbit and improve imaging performance. The method was implemented for execution of non-circular, task-driven orbits on a clinical robotic C-arm system. The framework is sufficiently general to include both acquisition parameters (e.g., orbit, kV, and mA selection) and reconstruction parameters (e.g., a spatially varying regularizer).

Self-calibration of cone-beam CT geometry using 3D-2D image registration

Robotic C-arms are capable of complex orbits that can increase field of view, reduce artifacts, improve image quality, and/or reduce dose; however, it can be challenging to obtain accurate, reproducible geometric calibration required for image reconstruction for such complex orbits. This work presents a method for geometric calibration for an arbitrary source-detector orbit by registering 2D projection data to a previously acquired 3D image. It also yields a method by which calibration of simple circular orbits can be improved. The registration uses a normalized gradient information similarity metric and the covariance matrix adaptation-evolution strategy optimizer for robustness against local minima and changes in image content. The resulting transformation provides a 'self-calibration' of system geometry. The algorithm was tested in phantom studies using both a cone-beam CT (CBCT) test-bench and a robotic C-arm (Artis Zeego, Siemens Healthcare) for circular and non-circular orbits. Self-calibration performance was evaluated in terms of the full-width at half-maximum (FWHM) of the point spread function in CBCT reconstructions, the reprojection error (RPE) of steel ball bearings placed on each phantom, and the overall quality and presence of artifacts in CBCT images. In all cases, self-calibration improved the FWHM-e.g. on the CBCT bench, FWHM  =  0.86 mm for conventional calibration compared to 0.65 mm for self-calibration (p  <  0.001). Similar improvements were measured in RPE-e.g. on the robotic C-arm, RPE  =  0.73 mm for conventional calibration compared to 0.55 mm for self-calibration (p  <  0.001). Visible improvement was evident in CBCT reconstructions using self-calibration, particularly about high-contrast, high-frequency objects (e.g. temporal bone air cells and a surgical needle). The results indicate that self-calibration can improve even upon systems with presumably accurate geometric calibration and is applicable to situations where conventional calibration is not feasible, such as complex non-circular CBCT orbits and systems with irreproducible source-detector trajectory.

Self-Calibration of Cone-Beam CT Geometry Using 3D-2D Image Registration: Development and Application to Task-Based Imaging with a Robotic C-Arm

PURPOSE

Robotic C-arm systems are capable of general noncircular orbits whose trajectories can be driven by the particular imaging task. However obtaining accurate calibrations for reconstruction in such geometries can be a challenging problem. This work proposes a method to perform a unique geometric calibration of an arbitrary C-arm orbit by registering 2D projections to a previously acquired 3D image to determine the transformation parameters representing the system geometry.

METHODS

Experiments involved a cone-beam CT (CBCT) bench system, a robotic C-arm, and three phantoms. A robust 3D-2D registration process was used to compute the 9 degree of freedom (DOF) transformation between each projection and an existing 3D image by maximizing normalized gradient information with a digitally reconstructed radiograph (DRR) of the 3D volume. The quality of the resulting "self-calibration" was evaluated in terms of the agreement with an established calibration method using a BB phantom as well as image quality in the resulting CBCT reconstruction.

RESULTS

The self-calibration yielded CBCT images without significant difference in spatial resolution from the standard ("true") calibration methods (p-value >0.05 for all three phantoms), and the differences between CBCT images reconstructed using the "self" and "true" calibration methods were on the order of 10^-3 mm^-1. Maximum error in magnification was 3.2%, and back-projection ray placement was within 0.5 mm.

CONCLUSION

The proposed geometric "self" calibration provides a means for 3D imaging on general non-circular orbits in CBCT systems for which a geometric calibration is either not available or not reproducible. The method forms the basis of advanced "task-based" 3D imaging methods now in development for robotic C-arms.

Task-driven imaging in cone-beam computed tomography

PURPOSE

Conventional workflow in interventional imaging often ignores a wealth of prior information of the patient anatomy and the imaging task. This work introduces a task-driven imaging framework that utilizes such information to prospectively design acquisition and reconstruction techniques for cone-beam CT (CBCT) in a manner that maximizes task-based performance in subsequent imaging procedures.

METHODS

The framework is employed in jointly optimizing tube current modulation, orbital tilt, and reconstruction parameters in filtered backprojection reconstruction for interventional imaging. Theoretical predictors of noise and resolution relates acquisition and reconstruction parameters to task-based detectability. Given a patient-specific prior image and specification of the imaging task, an optimization algorithm prospectively identifies the combination of imaging parameters that maximizes task-based detectability. Initial investigations were performed for a variety of imaging tasks in an elliptical phantom and an anthropomorphic head phantom.

RESULTS

Optimization of tube current modulation and view-dependent reconstruction kernel was shown to have greatest benefits for a directional task (e.g., identification of device or tissue orientation). The task-driven approach yielded techniques in which the dose and sharp kernels were concentrated in views contributing the most to the signal power associated with the imaging task. For example, detectability of a line pair detection task was improved by at least three fold compared to conventional approaches. For radially symmetric tasks, the task-driven strategy yielded results similar to a minimum variance strategy in the absence of kernel modulation. Optimization of the orbital tilt successfully avoided highly attenuating structures that can confound the imaging task by introducing noise correlations masquerading at spatial frequencies of interest.

CONCLUSIONS

This work demonstrated the potential of a task-driven imaging framework to improve image quality and reduce dose beyond that achievable with conventional imaging approaches.