Open Access

Show Abstract ▼

Abstract: Recently, several convolutional neural networks have been proposed not only for 2D images, but also for 3D and 4D volume segmentation. Nevertheless, due to the large data size of the latter, acquiring a sufficient amount of training annotations is much more strenuous than in 2D images. For 4D time-series tomograms, this is usually handled by segmenting the constituent tomograms independently through time with 3D convolutional neural networks. Inter-volume information is therefore not utilized, potentially leading to temporal incoherence. In this paper, we attempt to resolve this by proposing two hidden Markov model variants that refine 4D segmentation labels made by 3D convolutional neural networks working on each time point. Our models utilize not only inter-volume information, but also the prediction confidence generated by the 3D segmentation convolutional neural networks themselves. To the best of our knowledge, this is the first attempt to refine 4D segmentations made by 3D convolutional neural networks using hidden Markov models. During experiments we test our models, qualitatively, quantitatively and behaviourally, using prespecified segmentations. We demonstrate in the domain of time series tomograms which are typically undersampled to allow more frequent capture; a particularly challenging problem. Finally, our dataset and code is publicly available.

Show Abstract ▼

Abstract: The challenge of processing big data effectively and efficiently is crucial for many synchrotron facilities which can collect up to several petabytes of data annually. At Diamond Light Source, the tomographic data is reconstructed with Python-based software Savu which utilises Message Passing Interface protocols to efficiently reconstruct parallel beam geometry data. When projection data is undersampled and/or noisy, regularised iterative reconstruction methods can provide a better reconstruction quality than direct methods. The iterative methods, however, require significantly more computational resources than direct methods and their usability is impeded by the choice of additional parameters. Notably, the use of 2D regularised iterative methods for reconstruction of 3D objects results in inconsistent (saw-shaped) features in a perpendicular to slicing orientation. Due to large data sizes, fully 3D regularised model-based iterative reconstruction is problematic or impossible in practice due to high memory requirements and long processing times. In this work, we demonstrate a practical solution which enables an approximated full 3D regularised iterative reconstruction running in parallel on a computing cluster. This modification delivers an equivalent to exact 3D reconstruction quality of images with a high computational efficiency.

Open Access

Show Abstract ▼

Abstract: Multi-scale structural assessment of biological soft tissue is challenging but essential to gain insight into structure–function relationships of tissue/organ. Using the human placenta as an example, this study brings together sophisticated sample preparation protocols, advanced imaging and robust, validated machine-learning segmentation techniques to provide the first massively multi-scale and multi-domain information that enables detailed morphological and functional analyses of both maternal and fetal placental domains. Finally, we quantify the scale-dependent error in morphological metrics of heterogeneous placental tissue, estimating the minimal tissue scale needed in extracting meaningful biological data. The developed protocol is beneficial for high-throughput investigation of structure–function relationships in both normal and diseased placentas, allowing us to optimize therapeutic approaches for pathological pregnancies. In addition, the methodology presented is applicable in the characterization of tissue architecture and physiological behaviours of other complex organs with similarity to the placenta, where an exchange barrier possesses circulating vascular and avascular fluid spaces.

Open Access

Show Abstract ▼

Abstract: Over recent years, many approaches have been proposed for the denoising or semantic segmentation of X-ray computed tomography (CT) scans. In most cases, high-quality CT reconstructions are used; however, such reconstructions are not always available. When the X-ray exposure time has to be limited, undersampled tomograms (in terms of their component projections) are attained. This low number of projections offers low-quality reconstructions that are difficult to segment. Here, we consider CT time-series (i.e. 4D data), where the limited time for capturing fast-occurring temporal events results in the time-series tomograms being necessarily undersampled. Fortunately, in these collections, it is common practice to obtain representative highly sampled tomograms before or after the time-critical portion of the experiment. In this paper, we propose an end-to-end network that can learn to denoise and segment the time-series’ undersampled CTs, by training with the earlier highly sampled representative CTs. Our single network can offer two desired outputs while only training once, with the denoised output improving the accuracy of the final segmentation. Our method is able to outperform state-of-the-art methods in the task of semantic segmentation and offer comparable results in regard to denoising. Additionally, we propose a knowledge transfer scheme using synthetic tomograms. This not only allows accurate segmentation and denoising using less real-world data, but also increases segmentation accuracy. Finally, we make our datasets, as well as the code, publicly available.

Open Access

Show Abstract ▼

Abstract: X-ray computed tomography and, specifically, time-resolved volumetric tomography data collections (4D datasets) routinely produce terabytes of data, which need to be effectively processed after capture. This is often complicated due to the high rate of data collection required to capture at sufficient time-resolution events of interest in a time-series, compelling the researchers to perform data collection with a low number of projections for each tomogram in order to achieve the desired `frame rate'. It is common practice to collect a representative tomogram with many projections, after or before the time-critical portion of the experiment without detrimentally affecting the time-series to aid the analysis process. For this paper these highly sampled data are used to aid feature detection in the rapidly collected tomograms by assisting with the upsampling of their projections, which is equivalent to upscaling the θ-axis of the sinograms. In this paper, a super-resolution approach is proposed based on deep learning (termed an upscaling Deep Neural Network, or UDNN) that aims to upscale the sinogram space of individual tomograms in a 4D dataset of a sample. This is done using learned behaviour from a dataset containing a high number of projections, taken of the same sample and occurring at the beginning or the end of the data collection. The prior provided by the highly sampled tomogram allows the application of an upscaling process with better accuracy than existing interpolation techniques. This upscaling process subsequently permits an increase in the quality of the tomogram's reconstruction, especially in situations that require capture of only a limited number of projections, as is the case in high-frequency time-series capture. The increase in quality can prove very helpful for researchers, as downstream it enables easier segmentation of the tomograms in areas of interest, for example. The method itself comprises a convolutional neural network which through training learns an end-to-end mapping between sinograms with a low and a high number of projections. Since datasets can differ greatly between experiments, this approach specifically develops a lightweight network that can easily and quickly be retrained for different types of samples. As part of the evaluation of our technique, results with different hyperparameter settings are presented, and the method has been tested on both synthetic and real-world data. In addition, accompanying real-world experimental datasets have been released in the form of two 80 GB tomograms depicting a metallic pin that undergoes corruption from a droplet of salt water. Also a new engineering-based phantom dataset has been produced and released, inspired by the experimental datasets.