Abstract-Deep convolutional neural networks have been successfully applied to many image processing problems in recent works. Popular network architectures often add additional operations and connections to the standard architecture to enable training deeper networks. To achieve accurate results in practice, a large number of trainable parameters is often required. Here, we introduce a network architecture based on using dilated convolutions to capture features at different image scales, and densely connecting all feature maps with each other. The resulting architecture is able to achieve accurate results with relatively few parameters and consists of a single set of operations, making it easier to implement, train, and apply in practice, and automatically adapts to different problems. We compare results of the proposed network architecture with popular existing architectures for several segmentation problems, showing that the proposed architecture is able to achieve accurate results with fewer parameters, with a reduced risk of overfitting the training data.

No abstract

In many applications of tomography, the acquired data are limited in one or more ways due to unavoidable experimental constraints. In such cases, popular direct reconstruction algorithms tend to produce inaccurate images, and more accurate iterative algorithms often have prohibitively high computational costs. Using machine learning to improve the image quality of direct algorithms is a recently proposed alternative, for which promising results have been shown. However, previous attempts have focused on using encoder–decoder networks, which have several disadvantages when applied to large tomographic images, preventing wide application in practice. Here, we propose the use of the Mixed-Scale Dense convolutional neural network architecture, which was specifically designed to avoid these disadvantages, to improve tomographic reconstruction from limited data. Results are shown for various types of data limitations and object types, for both simulated data and large-scale real-world experimental data. The results are compared with popular tomographic reconstruction algorithms and machine learning algorithms, showing that Mixed-Scale Dense networks are able to significantly improve reconstruction quality even with severely limited data, and produce more accurate results than existing algorithms.

Recovering a high-quality image from noisy indirect measurements is an important problem with many applications. For such inverse problems, supervised deep convolutional neural network (CNN)-based denoising methods have shown strong results, but the success of these supervised methods critically depends on the availability of a high-quality training dataset of similar measurements. For image denoising, methods are available that enable training without a separate training dataset by assuming that the noise in two different pixels is uncorrelated. However, this assumption does not hold for inverse problems, resulting in artifacts in the denoised images produced by existing methods. Here, we propose Noise2Inverse, a deep CNN-based denoising method for linear image reconstruction algorithms that does not require any additional clean or noisy data. Training a CNN-based denoiser is enabled by exploiting the noise model to compute multiple statistically independent reconstructions. We develop a theoretical framework which shows that such training indeed obtains a denoising CNN, assuming the measured noise is element-wise independent, and zero-mean. On simulated CT datasets, Noise2Inverse demonstrates an improvement in peak signal-to-noise ratio and structural similarity index compared to state-of-the-art image denoising methods, and conventional reconstruction methods, such as Total-Variation Minimization. We also demonstrate that the method is able to significantly reduce noise in challenging real-world experimental datasets.

Image reconstruction from a small number of projections is a challenging problem in tomography. Advanced algorithms that incorporate prior knowledge can sometimes produce accurate reconstructions, but they typically require long computation times. Furthermore, the required prior knowledge can be very specific, limiting the type of images that can be reconstructed. Here, we present a reconstruction method that automatically learns prior knowledge using an artificial neural network. We show that this method can be viewed as a combination of filtered backprojection steps, and, therefore, has a relatively low computational cost. Results for two different cases show that the new method is able to use the learned information to produce high quality reconstructions in a short time, even when presented with a small number of projections.

The integration of two Python toolboxes used for processing tomographic data, TomoPy and the ASTRA toolbox, is presented.

We present an efficient and robust method based on Monte Carlo simulations for predicting crystal structures at finite temperature. We apply this method, which is surprisingly easy to implement, to a variety of systems, demonstrating its effectiveness for hard, attractive, and anisotropic interactions, binary mixtures, semi-long-range soft interactions, and truly long-range interactions where the truly long-range interactions are treated using Ewald sums. In the case of binary hard-sphere mixtures, star polymers, and binary Lennard-Jones mixtures, the crystal structures predicted by this algorithm are consistent with literature, providing confidence in the method. Finally, we predict new crystal structures for hard asymmetric dumbbell particles, bowl-like particles and hard oblate cylinders and present the phase diagram for the oblate cylinders based on full free energy calculations.

There is a widening gap between the fast advancement of computational methods for tomographic reconstruction and their successful implementation in production software at various synchrotron facilities. This is due in part to the lack of readily available instrument datasets and phantoms representative of real materials for validation and comparison of new numerical methods. Recent advancements in detector technology have made sub-second and multi-energy tomographic data collection possible (Gibbs et al 2015 Sci. Rep. 5 11824), but have also increased the demand to develop new reconstruction methods able to handle in situ