Weighted simultaneous algebraic reconstruction technique for tomosynthesis imaging of objects with high-attenuation features

PURPOSE

This paper introduces a nonlinear weighting scheme into the backprojection operation within the simultaneous algebraic reconstruction technique (SART). It is designed for tomosynthesis imaging of objects with high-attenuation features in order to reduce limited angle artifacts.

METHODS

The algorithm estimates which projections potentially produce artifacts in a voxel. The contribution of those projections into the updating term is reduced. In order to identify those projections automatically, a four-dimensional backprojected space representation is used. Weighting coefficients are calculated based on a dissimilarity measure, evaluated in this space. For each combination of an angular view direction and a voxel position an individual weighting coefficient for the updating term is calculated.

RESULTS

The feasibility of the proposed approach is shown based on reconstructions of the following real three-dimensional tomosynthesis datasets: a mammography quality phantom, an apple with metal needles, a dried finger bone in water, and a human hand. Datasets have been acquired with a Siemens Mammomat Inspiration tomosynthesis device and reconstructed using SART with and without suggested weighting. Out-of-focus artifacts are described using line profiles and measured using standard deviation (STD) in the plane and below the plane which contains artifact-causing features. Artifacts distribution in axial direction is measured using an artifact spread function (ASF). The volumes reconstructed with the weighting scheme demonstrate the reduction of out-of-focus artifacts, lower STD (meaning reduction of artifacts), and narrower ASF compared to nonweighted SART reconstruction. It is achieved successfully for different kinds of structures: point-like structures such as phantom features, long structures such as metal needles, and fine structures such as trabecular bone structures.

CONCLUSIONS

Results indicate the feasibility of the proposed algorithm to reduce typical tomosynthesis artifacts produced by high-attenuation features. The proposed algorithm assigns weighting coefficients automatically and no segmentation or tissue-classification steps are required. The algorithm can be included into various iterative reconstruction algorithms with an additive updating strategy. It can also be extended to computed tomography case with the complete set of angular data.

A fully 3D approach for metal artifact reduction in computed tomography

PURPOSE

In computed tomography imaging metal objects in the region of interest introduce inconsistencies during data acquisition. Reconstructing these data leads to an image in spatial domain including star-shaped or stripe-like artifacts. In order to enhance the quality of the resulting image the influence of the metal objects can be reduced. Here, a metal artifact reduction (MAR) approach is proposed that is based on a recomputation of the inconsistent projection data using a fully three-dimensional Fourier-based interpolation. The success of the projection space restoration depends sensitively on a sensible continuation of neighboring structures into the recomputed area. Fortunately, structural information of the entire data is inherently included in the Fourier space of the data. This can be used for a reasonable recomputation of the inconsistent projection data.

METHODS

The key step of the proposed MAR strategy is the recomputation of the inconsistent projection data based on an interpolation using nonequispaced fast Fourier transforms (NFFT). The NFFT interpolation can be applied in arbitrary dimension. The approach overcomes the problem of adequate neighborhood definitions on irregular grids, since this is inherently given through the usage of higher dimensional Fourier transforms. Here, applications up to the third interpolation dimension are presented and validated. Furthermore, prior knowledge may be included by an appropriate damping of the transform during the interpolation step. This MAR method is applicable on each angular view of a detector row, on two-dimensional projection data as well as on three-dimensional projection data, e.g., a set of sequential acquisitions at different spatial positions, projection data of a spiral acquisition, or cone-beam projection data.

RESULTS

Results of the novel MAR scheme based on one-, two-, and three-dimensional NFFT interpolations are presented. All results are compared in projection data space and spatial domain with the well-known one-dimensional linear interpolation strategy.

CONCLUSIONS

In conclusion, it is recommended to include as much spatial information into the recomputation step as possible. This is realized by increasing the dimension of the NFFT. The resulting image quality can be enhanced considerably.

A generic framework for modeling brain deformation as a constrained parametric optimization problem to aid non-diffeomorphic image registration in brain tumor imaging

OBJECTIVES

In the present paper a novel computational framework for modeling tumor induced brain deformation as a biophysical prior for non-rigid image registration is described. More precisely, we aim at providing a generic building block for non-rigid image registration that can be used to resolve inherent irregularities in non-diffeomorphic registration problems that naturally arise in serial and cross-population brain tumor imaging studies due to the presence (or progression) of pathology.

METHODS

The model for the description of brain cancer dynamics on a tissue level is based on an initial boundary value problem (IBVP). The IBVP follows the accepted assumption that the progression of primary brain tumors on a tissue level is governed by proliferation and migration of cancerous cells into surrounding healthy tissue. The model of tumor induced brain deformation is phrased as a parametric, constrained optimization problem. As a basis of comparison and to demonstrate generalizability additional soft constraints (penalties) are considered. A back-tracking line search is implemented in conjunction with a limited memory Broyden-Fletcher-Goldfarb-Shanno (LBFGS) method in order to handle the numerically delicate log-barrier strategy for confining volume change.

RESULTS

Numerical experiments are performed to test the flexible control of the computed deformation patterns in terms of varying model parameters. The results are qualitatively and quantitatively related to patterns in patient individual magnetic resonance imaging data.

CONCLUSIONS

Numerical experiments demonstrate the flexible control of the computed deformation patterns. This in turn strongly suggests that the model can be adapted to patient individual imaging patterns of brain tumors. Qualitative and quantitative comparison of the computed cancer profiles to patterns in medical imaging data of an exemplary patient demonstrates plausibility. The designed optimization problem is based on computational tools widely used in non-rigid image registration, which in turn makes the model generally applicable for integration into non-rigid image registration algorithms.

Magnetic particle imaging: visualization of instruments for cardiovascular intervention

PURPOSE

To evaluate the feasibility of different approaches of instrument visualization for cardiovascular interventions guided by using magnetic particle imaging (MPI).

MATERIALS AND METHODS

Two balloon (percutaneous transluminal angioplasty) catheters were used. The balloon was filled either with diluted superparamagnetic iron oxide (SPIO) ferucarbotran (25 mmol of iron per liter) or with sodium chloride. Both catheters were inserted into a vessel phantom that was filled oppositional to the balloon content with sodium chloride or diluted SPIO (25 mmol of iron per liter). In addition, the administration of a 1.4-mL bolus of pure SPIO (500 mmol of iron per liter) followed by 5 mL of sodium chloride through a SPIO-labeled balloon catheter into the sodium chloride-filled vessel phantom was recorded. Images were recorded by using a preclinical MPI demonstrator. All images were acquired by using a field of view of 3.6 × 3.6 × 2.0 cm.

RESULTS

By using MPI, both balloon catheters could be visualized with high temporal (21.54 msec per image) and sufficient spatial (≤ 3 mm) resolution without any motion artifacts. The movement through the field of view, the inflation and deflation of the balloon, and the application of the SPIO bolus were visualized at a rate of 46 three-dimensional data sets per second.

CONCLUSION

Visualization of SPIO-labeled instruments for cardiovascular intervention at high temporal resolution as well as monitoring the application of a SPIO-based tracer by using labeled instruments is feasible. Further work is necessary to evaluate different labeling approaches for diagnostic catheters and guidewires and to demonstrate their navigation in the vascular system after administration of contrast material.

SUPPLEMENTAL MATERIAL

http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.12120424/-/DC1.

Biophysical modeling of brain tumor progression: from unconditionally stable explicit time integration to an inverse problem with parabolic PDE constraints for model calibration

PURPOSE

A novel unconditionally stable, explicit numerical method is introduced to the field of modeling brain cancer progression on a tissue level together with an inverse problem (IP) based on optimal control theory that allows for automated model calibration with respect to observations in clinical imaging data.

METHODS

Biophysical models of cancer progression on a tissue level are in general based on the assumption that the spatiotemporal spread of cancerous cells is determined by cell division and net migration. These processes are typically described in terms of a parabolic partial differential equation (PDE). In the present work a parallelized implementation of an unconditionally stable, explicit Euler (EE(⋆)) time integration method for the solution of this PDE is detailed. The key idea of the discussed EE(⋆) method is to relax the strong stability requirement on the spectral radius of the coefficient matrix by introducing a subdivision regime for a given outer time step. The performance is related to common implicit numerical methods. To quantify the numerical error, a simplified model that has a closed form solution is considered. To allow for a systematic, phenomenological validation a novel approach for automated model calibration on the basis of observations in medical imaging data is developed. The resulting IP is based on optimal control theory and manifests as a large scale, PDE constrained optimization problem.

RESULTS

The numerical error of the EE(⋆) method is at the order of standard implicit numerical methods. The computing times are well below those obtained for implicit methods and by that demonstrate efficiency. Qualitative and quantitative analysis in 12 patients demonstrates that the obtained results are in strong agreement with observations in medical imaging data. Rating simulation success in terms of the mean overlap between model predictions and manual expert segmentations yields a success rate of 75% (9 out of 12 patients).

CONCLUSIONS

The discussed EE(⋆) method provides desirable features for image-based model calibration or hybrid image registration algorithms in which the model serves as a biophysical prior. This is due to (i) ease of implementation, (ii) low memory requirements, (iii) efficiency, (iv) a straightforward interface for parameter updates, and (v) the fact that the method is inherently matrix-free. The explicit time integration method is confirmed via experiments for automated model calibration. Qualitative and quantitative analysis demonstrates that the proposed framework allows for recovering observations in medical imaging data and by that phenomenological model validity.

Magnetic particle imaging: introduction to imaging and hardware realization

Magnetic Particle Imaging (MPI) is a recently invented tomographic imaging method that quantitatively measures the spatial distribution of a tracer based on magnetic nanoparticles. The new modality promises a high sensitivity and high spatial as well as temporal resolution. There is a high potential of MPI to improve interventional and image-guided surgical procedures because, today, established medical imaging modalities typically excel in only one or two of these important imaging properties. MPI makes use of the non-linear magnetization characteristics of the magnetic nanoparticles. For this purpose, two magnetic fields are created and superimposed, a static selection field and an oscillatory drive field. If superparamagnetic iron-oxide nanoparticles (SPIOs) are subjected to the oscillatory magnetic field, the particles will react with a non-linear magnetization response, which can be measured with an appropriate pick-up coil arrangement. Due to the non-linearity of the particle magnetization, the received signal consists of the fundamental excitation frequency as well as of harmonics. After separation of the fundamental signal, the nanoparticle concentration can be reconstructed quantitatively based on the harmonics. The spatial coding is realized with the static selection field that produces a field-free point, which is moved through the field of view by the drive fields. This article focuses on the frequency-based image reconstruction approach and the corresponding imaging devices while alternative concepts like x-space MPI and field-free line imaging are described as well. The status quo in hardware realization is summarized in an overview of MPI scanners.

Toward cardiovascular interventions guided by magnetic particle imaging: first instrument characterization

Magnetic particle imaging has emerged as a new technique for the visualization and quantification of superparamagnetic iron oxide nanoparticles. It seems to be a very promising application for cardiovascular interventional radiology. A prerequisite for interventions is the artifact-free visualization of the required instruments and implants. Various commercially available catheters, guide wires, and a catheter experimentally coated with superparamagnetic iron oxide nanoparticles were tested regarding their signal characteristics using magnetic particle spectroscopy to evaluate their performance in magnetic particle imaging. The results indicate that signal-generating and non-signal-generating instruments can be distinguished. Furthermore, coating or loading non-signal-generating instruments with superparamagnetic iron oxide nanoparticles seems to be a promising approach, but optimized nanoparticles need yet to be developed.

Multi-object active shape model construction for abdomen segmentation: preliminary results

The automatic segmentation of abdominal organs is a pre-requisite for many medical applications. Successful methods typically rely on prior knowledge about the to be segmented anatomy as it is for instance provided by means of active shape models (ASMs). Contrary to most previous ASM based methods, this work does not focus on individual organs. Instead, a more holistic approach that aims at exploiting inter-organ relationships to eventually segment a complex of organs is proposed. Accordingly, a flexible framework for automatic construction of multi-object ASMs is introduced, employed for coupled shape modeling, and used for co-segmentation of liver and spleen based on a new coupled shape/separate pose approach. Our first results indicate feasible segmentation accuracies, whereas pose decoupling leads to substantially better segmentation results and performs in average also slightly better than the standard single-object ASM approach.

A computational multiscale model of glioblastoma growth: regulation of cell migration and proliferation via microRNA-451, LKB1 and AMPK

A new computational multiscale model of glioblastoma growth is introduced. This model combines an agent-based model for representing processes on the cellular level with a molecular interaction network for each cell on the subcellular scale. The network is based on recently published work on the interaction of microRNA-451, LKB1 and AMPK in the regulation of glioblastoma cell migration and proliferation. We translated this network into a mathematical description by the use of 17 ordinary differential equations. In our model, we furthermore establish a link from the molecular interaction network of a single cell to cellular actions (e.g. chemotactic movement) on the microscopic level. First results demonstrate that the computational model reproduces a tumor cell development comparable to that observed in in vitro experiments.