Enhanced detection of threat materials by dark-field x-ray imaging combined with deep neural networks.

X-ray imaging has been boosted by the introduction of phase-based methods. Detail visibility is enhanced in phase contrast images, and dark-field images are sensitive to inhomogeneities on a length scale below the system's spatial resolution. Here we show that dark-field creates a texture which is characteristic of the imaged material, and that its combination with conventional attenuation leads to an improved discrimination of threat materials. We show that remaining ambiguities can be resolved by exploiting the different energy dependence of the dark-field and attenuation signals. Furthermore, we demonstrate that the dark-field texture is well-suited for identification through machine learning approaches through two proof-of-concept studies. In both cases, application of the same approaches to datasets from which the dark-field images were removed led to a clear degradation in performance. While the small scale of these studies means further research is required, results indicate potential for a combined use of dark-field and deep neural networks in security applications and beyond.

Enhanced diffuse optical tomographic reconstruction using concurrent ultrasound information.

Multimodal imaging is an active branch of research as it has the potential to improve common medical imaging techniques. Diffuse optical tomography (DOT) is an example of a low resolution, functional imaging modality that typically has very low resolution due to the ill-posedness of its underlying inverse problem. Combining the functional information of DOT with a high resolution structural imaging modality has been studied widely. In particular, the combination of DOT with ultrasound (US) could serve as a useful tool for clinicians for the formulation of accurate diagnosis of breast lesions. In this paper, we propose a novel method for US-guided DOT reconstruction using a portable time-domain measurement system. B-mode US imaging is used to retrieve morphological information on the probed tissues by means of a semi-automatical segmentation procedure based on active contour fitting. A two-dimensional to three-dimensional extrapolation procedure, based on the concept of distance transform, is then applied to generate a three-dimensional edge-weighting prior for the regularization of DOT. The reconstruction procedure has been tested on experimental data obtained on specifically designed dual-modality silicon phantoms. Results show a substantial quantification improvement upon the application of the implemented technique. This article is part of the theme issue 'Synergistic tomographic image reconstruction: part 2'.

Networks for Nonlinear Diffusion Problems in Imaging.

A multitude of imaging and vision tasks have seen recently a major transformation by deep learning methods and in particular by the application of convolutional neural networks. These methods achieve impressive results, even for applications where it is not apparent that convolutions are suited to capture the underlying physics. In this work, we develop a network architecture based on nonlinear diffusion processes, named DiffNet. By design, we obtain a nonlinear network architecture that is well suited for diffusion-related problems in imaging. Furthermore, the performed updates are explicit, by which we obtain better interpretability and generalisability compared to classical convolutional neural network architectures. The performance of DiffNet is tested on the inverse problem of nonlinear diffusion with the Perona-Malik filter on the STL-10 image dataset. We obtain competitive results to the established U-Net architecture, with a fraction of parameters and necessary training data.

Novel time-resolved camera based on compressed sensing.

Time-resolved cameras with high temporal resolution (down to ps) enable a huge set of novel applications ranging from biomedicine and environmental science to material and device characterization. In this work, we propose, and experimentally validate, a novel detection scheme for time-resolved imaging based on a compressed sampling approach. The proposed scheme unifies into a single element all the required operations, i.e. space modulation, space integration and time-resolved detection, paving the way to dramatic cost reduction, performance improvement and ease of use.

Heterodyne frequency-domain multispectral diffuse optical tomography of breast cancer in the parallel-plane transmission geometry.

PURPOSE: The authors introduce a state-of-the-art all-optical clinical diffuse optical tomography (DOT) imaging instrument which collects spatially dense, multispectral, frequency-domain breast data in the parallel-plate geometry. METHODS: The instrument utilizes a CCD-based heterodyne detection scheme that permits massively parallel detection of diffuse photon density wave amplitude and phase for a large number of source-detector pairs (10(6)). The stand-alone clinical DOT instrument thus offers high spatial resolution with reduced crosstalk between absorption and scattering. Other novel features include a fringe profilometry system for breast boundary segmentation, real-time data normalization, and a patient bed design which permits both axial and sagittal breast measurements. RESULTS: The authors validated the instrument using tissue simulating phantoms with two different chromophore-containing targets and one scattering target. The authors also demonstrated the instrument in a case study breast cancer patient; the reconstructed 3D image of endogenous chromophores and scattering gave tumor localization in agreement with MRI. CONCLUSIONS: Imaging with a novel parallel-plate DOT breast imager that employs highly parallel, high-resolution CCD detection in the frequency-domain was demonstrated.

Rapid processing of PET list-mode data for efficient uncertainty estimation and data analysis.

In this technical note we propose a rapid and scalable software solution for the processing of PET list-mode data, which allows the efficient integration of list mode data processing into the workflow of image reconstruction and analysis. All processing is performed on the graphics processing unit (GPU), making use of streamed and concurrent kernel execution together with data transfers between disk and CPU memory as well as CPU and GPU memory. This approach leads to fast generation of multiple bootstrap realisations, and when combined with fast image reconstruction and analysis, it enables assessment of uncertainties of any image statistic and of any component of the image generation process (e.g. random correction, image processing) within reasonable time frames (e.g. within five minutes per realisation). This is of particular value when handling complex chains of image generation and processing. The software outputs the following: (1) estimate of expected random event data for noise reduction; (2) dynamic prompt and random sinograms of span-1 and span-11 and (3) variance estimates based on multiple bootstrap realisations of (1) and (2) assuming reasonable count levels for acceptable accuracy. In addition, the software produces statistics and visualisations for immediate quality control and crude motion detection, such as: (1) count rate curves; (2) centre of mass plots of the radiodistribution for motion detection; (3) video of dynamic projection views for fast visual list-mode skimming and inspection; (4) full normalisation factor sinograms. To demonstrate the software, we present an example of the above processing for fast uncertainty estimation of regional SUVR (standard uptake value ratio) calculation for a single PET scan of (18)F-florbetapir using the Siemens Biograph mMR scanner.

Fluorescence diffuse optical tomography using the split Bregman method.

PURPOSE: Standard image reconstruction methods for fluorescence Diffuse Optical Tomography (fDOT) generally make use of L2-regularization. A better choice is to replace the L2 by a total variation functional that effectively removes noise while preserving edges. Among the wide range of approaches available, the recently appeared Split Bregman method has been shown to be optimal and efficient. Furthermore, additional constraints can be easily included. We propose the use of the Split Bregman method to solve the image reconstruction problem for fDOT with a nonnegativity constraint that imposes the reconstructed concentration of fluorophore to be positive. METHODS: The proposed method is tested with simulated and experimental data, and results are compared with those yielded by an equivalent unconstrained optimization approach based on Gauss-Newton (GN) method, in which the negative part of the solution is projected to zero after each iteration. In addition, the method dependence on the parameters that weigh data fidelity and nonnegativity constraints is analyzed. RESULTS: Split Bregman yielded a reduction of the solution error norm and a better full width at tenth maximum for simulated data, and higher signal-to-noise ratio for experimental data. It is also shown that it led to an optimum solution independently of the data fidelity parameter, as long as the number of iterations is properly selected, and that there is a linear relation between the number of iterations and the inverse of the data fidelity parameter. CONCLUSIONS: Split Bregman allows the addition of a nonnegativity constraint leading to improve image quality.

Accelerated boundary element method for diffuse optical imaging.

The boundary element method (BEM) is a useful tool in diffuse optical imaging (DOI) when modelling large optical regions whose parameters are piecewise constant, but are computationally expensive. We present here an acceleration technique, the single-level fast multipole method, for a highly lossy medium. The enhanced practicability of the BEM in DOI is demonstrated through test examples on single-layer problems, where order of magnitude reduction factors on solution time are achieved and on a realistic three-layer model of the neonatal head. Our experimental results agree very closely with theoretical predictions of computational complexity.

3D level set reconstruction of model and experimental data in Diffuse Optical Tomography.

The level set technique is an implicit shape-based image reconstruction method that allows the recovery of the location, size and shape of objects of distinct contrast with well-defined boundaries embedded in a medium of homogeneous or moderately varying background parameters. In the case of diffuse optical tomography, level sets can be employed to simultaneously recover inclusions that differ in their absorption or scattering parameters from the background medium. This paper applies the level set method to the three-dimensional reconstruction of objects from simulated model data and from experimental frequency-domain data of light transmission obtained from a cylindrical phantom with tissue-like parameters. The shape and contrast of two inclusions, differing in absorption and diffusion parameters from the background, respectively, are reconstructed simultaneously. We compare the performance of level set recons uction with results from an image-based method using a Gauss-Newton iterative approach, and show that the level set technique can improve the detection and localisation of small, high-contrast targets.

A combined reconstruction-classification method for diffuse optical tomography.

We present a combined classification and reconstruction algorithm for diffuse optical tomography (DOT). DOT is a nonlinear ill-posed inverse problem. Therefore, some regularization is needed. We present a mixture of Gaussians prior, which regularizes the DOT reconstruction step. During each iteration, the parameters of a mixture model are estimated. These associate each reconstructed pixel with one of several classes based on the current estimate of the optical parameters. This classification is exploited to form a new prior distribution to regularize the reconstruction step and update the optical parameters. The algorithm can be described as an iteration between an optimization scheme with zeroth-order variable mean and variance Tikhonov regularization and an expectation-maximization scheme for estimation of the model parameters. We describe the algorithm in a general Bayesian framework. Results from simulated test cases and phantom measurements show that the algorithm enhances the contrast of the reconstructed images with good spatial accuracy. The probabilistic classifications of each image contain only a few misclassified pixels.

Estimating chromophore distributions from multiwavelength photoacoustic images.

Biomedical photoacoustic tomography (PAT) can provide qualitative images of biomedical soft tissue with high spatial resolution. However, whether it is possible to give accurate quantitative estimates of the spatially varying concentrations of the sources of photoacoustic contrast-endogenous or exogenous chromophores-remains an open question. Even if the chromophores' absorption spectra are known, the problem is nonlinear and ill-posed. We describe a framework for obtaining such quantitative estimates. When the optical scattering distribution is known, adjoint and gradient-based optimization techniques can be used to recover the concentration distributions of the individual chromophores that contribute to the overall tissue absorption. When the scattering distribution is unknown, prior knowledge of the wavelength dependence of the scattering is shown to be sufficient to overcome the absorption-scattering nonuniqueness and allow both distributions of chromophore concentrations and scattering to be recovered from multiwavelength photoacoustic images.

Anisotropic wave propagation and apparent conductivity estimation in a fast electrophysiological model: application to XMR interventional imaging.

Cardiac arrhythmias are increasingly being treated using ablation procedures. Development of fast electrophysiological models and estimation of parameters related to conduction pathologies can aid in the investigation of better treatment strategies during Radio-frequency ablations. We present a fast electrophysiological model incorporating anisotropy of the cardiac tissue. A global-local estimation procedure is also outlined to estimate a hidden parameter (apparent electrical conductivity) present in the model. The proposed model is tested on synthetic and real data derived using XMR imaging. We demonstrate a qualitative match between the estimated conductivity parameter and possible pathology locations. This approach opens up possibilities to directly integrate modelling in the intervention room.

k-space propagation models for acoustically heterogeneous media: application to biomedical photoacoustics.

Biomedical applications of photoacoustics, in particular photoacoustic tomography, require efficient models of photoacoustic propagation that can incorporate realistic properties of soft tissue, such as acoustic inhomogeneities both for purposes of simulation and for use in model-based image reconstruction methods. k-space methods are well suited to modeling high-frequency acoustics applications as they require fewer mesh points per wavelength than conventional finite element and finite difference models, and larger time steps can be taken without a loss of stability or accuracy. They are also straightforward to encode numerically, making them appealing as a general tool. The rationale behind k-space methods and the k-space approach to the numerical modeling of photoacoustic waves in fluids are covered in this paper. Three existing k-space models are applied to photoacoustics and demonstrated with examples: an exact model for homogeneous media, a second-order model that can take into account heterogeneous media, and a first-order model that can incorporate absorbing boundary conditions.

Three dimensional optical imaging of blood volume and oxygenation in the neonatal brain.

Optical methods provide a means of monitoring cerebral oxygenation in newborn infants at risk of brain injury. A 32-channel optical imaging system has been developed with the aim of reconstructing three-dimensional images of regional blood volume and oxygenation. Full image data sets were acquired from 14 out of 24 infants studied; successful images have been reconstructed in 8 of these infants. Regional variations in cerebral blood volume and tissue oxygen saturation are present in healthy preterm infants. In an infant with a large unilateral intraventricular haemorrhage, a corresponding region of low oxygen saturation was detected. These results suggest that optical tomography may provide an appropriate technique for investigating regional cerebral haemodynamics and oxygenation at the cotside.

Reconstructing absorption and diffusion shape profiles in optical tomography by a level set technique.

A shape reconstruction algorithm for optical tomography is introduced that uses a level-set formulation for the shapes. Evolution laws based on gradient directions for a cost functional are derived for two different level-set functions, one describing the absorption and one the diffusion parameter, as well as for the parameter values inside these shapes. Numerical experiments are presented in 2D that show that the new method is able to simultaneously recover shapes and contrast values of absorbing and scattering objects embedded in a moderately heterogeneous background medium from simulated noisy data.

Three-dimensional whole-head optical tomography of passive motor evoked responses in the neonate.

Optical tomography has been used to reconstruct three-dimensional images of the entire neonatal head during motor evoked responses. Data were successfully acquired during passive movement of each arm on four out of six infants examined, from which eight sets of bilateral images of hemodynamic parameters were reconstructed. Six out of the eight images showed the largest change in total hemoglobin in the region of the contralateral motor cortex. The mean distance between the peak response in the image and the estimated position of the contralateral motor cortex was 10.8 mm. These results suggest that optical tomography may provide an appropriate technique for non-invasive cot-side imaging of brain function.

Approximation errors and model reduction in optical tomography.

Model reduction is often required in optical diffusion tomography (ODT), typically due to limited available computation time or computer memory. In practice, this often means that we are bound to use sparse meshes in the model for the forward problem. Conversely, if we are given more and more accurate measurements, we have to employ increasingly accurate forward problem solvers in order to exploit the information in the measurements. In this paper we apply the approximation error theory to ODT. We show that if the approximation errors are estimated and employed, it is possible to use mesh densities that would be unacceptable with a conventional measurement model.

Recent advances in diffuse optical imaging.

We review the current state-of-the-art of diffuse optical imaging, which is an emerging technique for functional imaging of biological tissue. It involves generating images using measurements of visible or near-infrared light scattered across large (greater than several centimetres) thicknesses of tissue. We discuss recent advances in experimental methods and instrumentation, and examine new theoretical techniques applied to modelling and image reconstruction. We review recent work on in vivo applications including imaging the breast and brain, and examine future challenges.

Diffusion tensor magnetic resonance image regularization.

As multi-dimensional complex data become more common, new regularization schemes tailored to those data are needed. In this paper we present a scheme for regularising diffusion tensor magnetic resonance (DT-MR) data, and more generally multi-dimensional data defined by a direction map and one or several magnitude maps. The scheme is divided in two steps. First, a variational method is proposed to restore direction fields with preservation of discontinuities. Its theoretical aspects are presented, as well as its application to the direction field that defines the main orientation of the diffusion tensors. The second step makes use of an anisotropic diffusion process to regularize the magnitude maps. The main idea is that for a range of data it is possible to use the restored direction as a prior to drive the regularization process in a way that preserves discontinuities and respects the local coherence of the magnitude map. We show that anisotropic diffusion is a convenient framework to implement that idea, and define a regularization process for the magnitude maps from our DT-MR data. Both steps are illustated on synthetic and real diffusion tensor magnetic resonance data.

A method for generating patient-specific finite element meshes for head modelling.

Finite element modelling of fields within the body, whether electrical or optical, requires knowledge of the geometry of the object being examined. It can be clinically impractical to obtain accurate surface information for individual patients, although a limited set of measurements such as the locations of sensors attached to the body, can be acquired more readily. In this paper, we describe how a generic surface taken from an adult head is warped to fit points measured on a neonatal head surface to provide a new, individual surface from which a finite element mesh was generated. Simulations show that data generated from this mesh and from the original neonatal head surface are similar to within experimental errors. However, data generated from a mesh of the best fit sphere were significantly different from data generated from the original neonatal head surface.