Deep Gauss-Newton for phase retrieval.

We propose the deep Gauss-Newton (DGN) algorithm. The DGN allows one to take into account the knowledge of the forward model in a deep neural network by unrolling a Gauss-Newton optimization method. No regularization or step size needs to be chosen; they are learned through convolutional neural networks. The proposed algorithm does not require an initial reconstruction and is able to retrieve simultaneously the phase and absorption from a single-distance diffraction pattern. The DGN method was applied to both simulated and experimental data and permitted large improvements of the reconstruction error and of the resolution compared with a state-of-the-art iterative method and another neural-network-based reconstruction algorithm.

Nonlinear primal-dual algorithm for the phase and absorption retrieval from a single phase contrast image.

We propose a nonlinear primal-dual algorithm for the retrieval of phase shift and absorption from a single x ray in-line phase contrast, or Fresnel diffraction, image. The algorithm permits us to regularize phase and absorption separately. We demonstrate that taking into account the nonlinearity in the reconstruction improves reconstruction compared with linear methods. We also demonstrate that choosing different regularizers for absorption and phase can improve the reconstructions. The use of the total variation and its generalization in a primal-dual approach allows us to exploit the sparsity of the investigated sample. On both simulated and real datasets, the proposed nonlinear primal-dual hybrid gradient (NL-PDHG) method yields reconstructions with considerably fewer artifacts and improved the normalized mean squared error compared with its linearized version.

Mixed scale dense convolutional networks for x-ray phase contrast imaging.

X-ray in-line phase contrast imaging relies on the measurement of Fresnel diffraction intensity patterns due to the phase shift and the attenuation induced by the object. The recovery of phase and attenuation from one or several diffraction patterns is a nonlinear ill-posed inverse problem. In this work, we propose supervised learning approaches using mixed scale dense (MS-D) convolutional neural networks to simultaneously retrieve the phase and the attenuation from x-ray phase contrast images. This network architecture uses dilated convolutions to capture features at different image scales and densely connects all feature maps. The long range information in images becomes quickly available, and greater receptive field size can be obtained without losing resolution. This network architecture seems to account for the effect of the Fresnel operator very efficiently. We train the networks using simulated data of objects consisting of either homogeneous components, characterized by a fixed ratio of the induced refractive phase shifts and attenuation, or heterogeneous components, consisting of various materials. We also train the networks in the image domain by applying a simple initial reconstruction using the adjoint of the Fréchet derivative. We compare the results obtained with the MS-D network to reconstructions using U-Net, another popular network architecture, as well as to reconstructions using the contrast transfer function method, a direct phase and attenuation retrieval method based on linearization of the direct problem. The networks are evaluated using simulated noisy data as well as images acquired at NanoMAX (MAX IV, Lund, Sweden). In all cases, large improvements of the reconstruction errors are obtained on simulated data compared to the linearized method. Moreover, on experimental data, the networks improve the reconstruction quantitatively, improving the low-frequency behavior and the resolution.

Impact of the training loss in deep learning-based CT reconstruction of bone microarchitecture.

PURPOSE: Computed tomography (CT) is a technique of choice to image bone structure at different scales. Methods to enhance the quality of degraded reconstructions obtained from low-dose CT data have shown impressive results recently, especially in the realm of supervised deep learning. As the choice of the loss function affects the reconstruction quality, it is necessary to focus on the way neural networks evaluate the correspondence between predicted and target images during the training stage. This is even more true in the case of bone microarchitecture imaging at high spatial resolution where both the quantitative analysis of bone mineral density (BMD) and bone microstructure is essential for assessing diseases such as osteoporosis. Our aim is thus to evaluate the quality of reconstruction on key metrics for diagnosis depending on the loss function that has been used for training the neural network. METHODS: We compare and analyze volumes that are reconstructed with neural networks trained with pixelwise, structural, and adversarial loss functions or with a combination of them. We perform realistic simulations of various low-dose acquisitions of bone microarchitecture. Our comparative study is performed with metrics that have an interest regarding the diagnosis of bone diseases. We therefore focus on bone-specific metrics such as bone volume and the total volume (BV and TV), resolution, connectivity assessed with the Euler number, and quantitative analysis of BMD to evaluate the quality of reconstruction obtained with networks trained with the different loss functions. RESULTS: We find that using L 1 $L_1$ norm as the pixelwise loss is the best choice compared to L 2 $L_2$ or no pixelwise loss since it improves resolution without deteriorating other metrics. Visual Geometry Group (VGG) perceptual loss, especially when combined with an adversarial loss, allows to better retrieve topological and morphological parameters of bone microarchitecture compared to Structural SIMilarity (SSIM) index. This however leads to a decreased resolution performance. The adversarial loss enhances the reconstruction performance in terms of BMD distribution accuracy. CONCLUSIONS: In order to retrieve the quantitative and structural characteristics of bone microarchitecture that are essential for postreconstruction diagnosis, our results suggest to use L 1 $L_1$ norm as part of the loss function. Then, trade-offs should be made depending on the application: VGG perceptual loss improves accuracy in terms of connectivity at the cost of a deteriorated resolution, and adversarial losses help better retrieve BMD distribution while significantly increasing the training time.

Comparison of five one-step reconstruction algorithms for spectral CT.

Over the last decade, dual-energy CT scanners have gone from prototypes to clinically available machines, and spectral photon counting CT scanners are following. They require a specific reconstruction process, consisting of two steps: material decomposition and tomographic reconstruction. Image-based methods perform reconstruction, then decomposition, while projection-based methods perform decomposition first, and then reconstruction. As an alternative, 'one-step inversion' methods have been proposed, which perform decomposition and reconstruction simultaneously. Unfortunately, one-step methods are typically slower than their two-step counterparts, and in most CT applications, reconstruction time is critical. This paper therefore proposes to compare the convergence speeds of five one-step algorithms. We adapted all these algorithms to solve the same problem: spectral photon-counting CT reconstruction from five energy bins, using a three materials decomposition basis and spatial regularization. The paper compares a Bayesian method which uses non-linear conjugate gradient for minimization (Cai et al 2013 Med. Phys. 40 111916-31), three methods based on quadratic surrogates (Long and Fessler 2014 IEEE Trans. Med. Imaging 33 1614-26, Weidinger et al 2016 Int. J. Biomed. Imaging 2016 1-15, Mechlem et al 2018 IEEE Trans. Med. Imaging 37 68-80), and a primal-dual method based on MOCCA, a modified Chambolle-Pock algorithm (Barber et al 2016 Phys. Med. Biol. 61 3784). Some of these methods have been accelerated by using µ-preconditioning, i.e. by performing all internal computations not with the actual materials the object is made of, but with carefully chosen linear combinations of those. In this paper, we also evaluated the impact of three different µ-preconditioners on convergence speed. Our experiments on simulated data revealed vast differences in the number of iterations required to reach a common image quality objective: Mechlem et al (2018 IEEE Trans. Med. Imaging 37 68-80) needed ten iterations, Cai et al (2013 Med. Phys. 40 111916-31), Long and Fessler (2014 IEEE Trans. Med. Imaging 33 1614-26) and Weidinger et al (2016 Int. J. Biomed. Imaging 2016 1-15) several hundreds, and Barber et al (2016 Phys. Med. Biol. 61 3784) several thousands. We also sum up other practical aspects, like memory footprint and the need to tune extra parameters.

Regularization of nonlinear decomposition of spectral x-ray projection images.

PURPOSE: Exploiting the x-ray measurements obtained in different energy bins, spectral computed tomography (CT) has the ability to recover the 3-D description of a patient in a material basis. This may be achieved solving two subproblems, namely the material decomposition and the tomographic reconstruction problems. In this work, we address the material decomposition of spectral x-ray projection images, which is a nonlinear ill-posed problem. METHODS: Our main contribution is to introduce a material-dependent spatial regularization in the projection domain. The decomposition problem is solved iteratively using a Gauss-Newton algorithm that can benefit from fast linear solvers. A Matlab implementation is available online. The proposed regularized weighted least squares Gauss-Newton algorithm (RWLS-GN) is validated on numerical simulations of a thorax phantom made of up to five materials (soft tissue, bone, lung, adipose tissue, and gadolinium), which is scanned with a 120 kV source and imaged by a 4-bin photon counting detector. To evaluate the method performance of our algorithm, different scenarios are created by varying the number of incident photons, the concentration of the marker and the configuration of the phantom. The RWLS-GN method is compared to the reference maximum likelihood Nelder-Mead algorithm (ML-NM). The convergence of the proposed method and its dependence on the regularization parameter are also studied. RESULTS: We show that material decomposition is feasible with the proposed method and that it converges in few iterations. Material decomposition with ML-NM was very sensitive to noise, leading to decomposed images highly affected by noise, and artifacts even for the best case scenario. The proposed method was less sensitive to noise and improved contrast-to-noise ratio of the gadolinium image. Results were superior to those provided by ML-NM in terms of image quality and decomposition was 70 times faster. For the assessed experiments, material decomposition was possible with the proposed method when the number of incident photons was equal or larger than 105 and when the marker concentration was equal or larger than 0.03 g·cm-3 . CONCLUSIONS: The proposed method efficiently solves the nonlinear decomposition problem for spectral CT, which opens up new possibilities such as material-specific regularization in the projection domain and a parallelization framework, in which projections are solved in parallel.

Nonlinear approaches for the single-distance phase retrieval problem involving regularizations with sparsity constraints.

The phase retrieval process is a nonlinear ill-posed problem. The Fresnel diffraction patterns obtained with hard x-ray synchrotron beam can be used to retrieve the phase contrast. In this work, we present a convergence comparison of several nonlinear approaches for the phase retrieval problem involving regularizations with sparsity constraints. The phase solution is assumed to have a sparse representation with respect to an orthonormal wavelets basis. One approach uses alternatively a solution of the nonlinear problem based on the Fréchet derivative and a solution of the linear problem in wavelet coordinates with an iterative thresholding. A second method is the one proposed by Ramlau and Teschke which generalizes to a nonlinear problem the classical thresholding algorithm. The algorithms were tested on a 3D Shepp-Logan phantom corrupted by white Gaussian noise. The best simulation results are obtained by the first method for the various noise levels and initializations investigated. The reconstruction errors are significantly decreased with respect to the ones given by the classical linear phase retrieval approaches.

Clay dispersion and aspect ratios in polymer-clay nanocomposites.

It has clearly been shown in the literature that the properties achieved by polymer clay nanocomposites are often related to their structures and to the states of dispersion of the silicate platelets in the polymer matrices. Unfortunately, up to date most techniques used in a standard procedure do not allow a correct interpretation of polymer-clay nanocomposite structure and dispersion. In a recent work, we proposed an image analysis procedure (I.A.P.) based on TEM/OM observations to characterize the clay dispersion in polymer clay nanocomposites. The I.A.P. allows a very fine description of the nanocomposites microstructure. Nevertheless this analysis method shows some limits like the representativity of the sample analyzed volume. The purpose of this work is to discuss about the accuracy of the parameters extracted from the I.A.P. We propose SAXS experimental developments for evaluating the thickness distribution of the clay tactoids. The good agreement between the results of the two techniques confirms the validity of the I.A.P. methodology. Moreover, other experiments were performed in order to understand the abnormally low platelet lengths and aspect ratios determined from TEM micrographs. Wet-STEM observations revealed that clay platelets were not broken during the extrusion process. And, low platelet lengths and aspect ratios were shown to originate from the preparation of the ultramicrotomed sections and from TEM projection effects induced by the clay platelet wavy shape.

Characterization of the drastic increase in molecular mobility of a deformed amorphous polymer.

Original experiments of dynamic mechanical analysis and small angle x-ray scattering on a deformed amorphous polymer below its glass transition temperature are reported. The mechanical treatment reveals high mobility zones induced by shearing and leads to a drastic increase in the molecular mobility of the system. These domains are evidenced by small angle x-ray scattering measurements, and their geometrical characteristics are independent of the applied deformation. An experimental procedure is proposed to determine an apparent activation energy associated with high mobility domains. The energy values obtained for intermediate modes rise from the beta to the alpha relaxation ones.