Comparison of Two-Dimensional- and Three-Dimensional-Based U-Net Architectures for Brain Tissue Classification in One-Dimensional Brain CT

Brain tissue segmentation plays a crucial role in feature extraction, volumetric quantification, and morphometric analysis of brain scans. For the assessment of brain structure and integrity, CT is a non-invasive, cheaper, faster, and more widely available modality than MRI. However, the clinical application of CT is mostly limited to the visual assessment of brain integrity and exclusion of copathologies. We have previously developed two-dimensional (2D) deep learning-based segmentation networks that successfully classified brain tissue in head CT. Recently, deep learning-based MRI segmentation models successfully use patch-based three-dimensional (3D) segmentation networks. In this study, we aimed to develop patch-based 3D segmentation networks for CT brain tissue classification. Furthermore, we aimed to compare the performance of 2D- and 3D-based segmentation networks to perform brain tissue classification in anisotropic CT scans. For this purpose, we developed 2D and 3D U-Net-based deep learning models that were trained and validated on MR-derived segmentations from scans of 744 participants of the Gothenburg H70 Cohort with both CT and T1-weighted MRI scans acquired timely close to each other. Segmentation performance of both 2D and 3D models was evaluated on 234 unseen datasets using measures of distance, spatial similarity, and tissue volume. Single-task slice-wise processed 2D U-Nets performed better than multitask patch-based 3D U-Nets in CT brain tissue classification. These findings provide support to the use of 2D U-Nets to segment brain tissue in one-dimensional (1D) CT. This could increase the application of CT to detect brain abnormalities in clinical settings.

Brain Morphometry and Chronic Inflammation in Bangladeshi Children Growing up in Extreme Poverty

Over 300 million children grow up in environments of extreme poverty, and the biological and psychosocial hazards endemic to these environments often expose these children to infection, disease, and consequent inflammatory responses. Chronic inflammation in early childhood has been associated with diminished cognitive outcomes and despite this established relationship, the mechanisms explaining how inflammation affects brain development are not well known. Importantly, chronic inflammation is very common in areas of extreme poverty, raising the possibility that it may serve as a mechanism explaining the known relationship between low socioeconomic status (SES) and atypical brain development. To examine these potential pathways, seventy-nine children growing up in an extremely poor, urban area of Bangladesh underwent structural MRI scanning at six years of age. Structural brain images were submitted to Mindboggle software, a Docker-compliant and high-reproducibility tool for tissue segmentation and regional estimations of volume, surface area, cortical thickness, sulcal depth, and mean curvature. Concentration of C-reactive protein was assayed at eight time points between infancy and five years of age and the frequency with which children had elevated concentrations of inflammatory marker served as the measure of chronic inflammation. SES was measured with years of maternal education and income-to-needs. Chronic inflammation predicted total brain volume, total white matter volume, average sulcal depth, and bilateral putamen volumes. Chronic inflammation also mediated the link between maternal education and bilateral putamen volumes. These findings suggest that chronic inflammation is associated with brain morphometry globally and in the putamen, and further suggests that inflammation may be a potential mechanism linking SES to brain development.

Learning Cortical Parcellations Using Graph Neural Networks

Deep learning has been applied to magnetic resonance imaging (MRI) for a variety of purposes, ranging from the acceleration of image acquisition and image denoising to tissue segmentation and disease diagnosis. Convolutional neural networks have been particularly useful for analyzing MRI data due to the regularly sampled spatial and temporal nature of the data. However, advances in the field of brain imaging have led to network- and surface-based analyses that are often better represented in the graph domain. In this analysis, we propose a general purpose cortical segmentation method that, given resting-state connectivity features readily computed during conventional MRI pre-processing and a set of corresponding training labels, can generate cortical parcellations for new MRI data. We applied recent advances in the field of graph neural networks to the problem of cortical surface segmentation, using resting-state connectivity to learn discrete maps of the human neocortex. We found that graph neural networks accurately learn low-dimensional representations of functional brain connectivity that can be naturally extended to map the cortices of new datasets. After optimizing over algorithm type, network architecture, and training features, our approach yielded mean classification accuracies of 79.91% relative to a previously published parcellation. We describe how some hyperparameter choices including training and testing data duration, network architecture, and algorithm choice affect model performance.

The future of artificial intelligence in digital pathology - results of a survey across stakeholder groups

Artificial intelligence (AI) provides a powerful tool to extract information from digitized histopathology whole slide images. In the last five years, academic and commercial actors have developed new technical solutions for a diverse set of tasks, including tissue segmentation, cell detection, mutation prediction, prognostication and prediction of treatment response. In the light of limited overall resources, it is presently unclear for researchers, practitioners and policymakers which of these topics are stable enough for clinical use in the near future and which topics are still experimental, but worth investing time and effort into. To identify potentially promising applications of AI in pathology, we performed an anonymous online survey of 75 computational pathology domain experts from academia and industry. Participants enrolled in 2021 were queried about their subjective opinion on promising and appealing sub-fields of computational pathology with a focus on solid tumors. The results of this survey indicate that the prediction of treatment response directly from routine pathology slides is regarded as the most promising future application. This item was ranked highest in the overall analysis and in sub-groups by age and professional background. Furthermore, prediction of genetic alterations, gene expression and survival directly from routine pathology images scored consistently high across subgroups. Together, these data demonstrate a possible direction for the development of computational pathology systems in clinical, academic and industrial research in the near future.

Analysis of the paraspinal muscle morphology of the lumbar spine using a convolutional neural network (CNN)

Purpose This single-center study aimed to develop a convolutional neural network to segment multiple consecutive axial magnetic resonance imaging (MRI) slices of the lumbar spinal muscles of patients with lower back pain and automatically classify fatty muscle degeneration. Methods We developed a fully connected deep convolutional neural network (CNN) with a pre-trained U-Net model trained on a dataset of 3,650 axial T2-weighted MRI images from 100 patients with lower back pain. We included all qualities of MRI; the exclusion criteria were fractures, tumors, infection, or spine implants. The training was performed using k-fold cross-validation (k = 10), and performance was evaluated using the dice similarity coefficient (DSC) and cross-sectional area error (CSA error). For clinical correlation, we used a simplified Goutallier classification (SGC) system with three classes. Results The mean DSC was high for overall muscle (0.91) and muscle tissue segmentation (0.83) but showed deficiencies in fatty tissue segmentation (0.51). The CSA error was small for the overall muscle area of 8.42%, and fatty tissue segmentation showed a high mean CSA error of 40.74%. The SGC classification was correctly predicted in 75% of the patients. Conclusion Our fully connected CNN segmented overall muscle and muscle tissue with high precision and recall, as well as good DSC values. The mean predicted SGC values of all available patient axial slices showed promising results. With an overall Error of 25%, further development is needed for clinical implementation. Larger datasets and training of other model architectures are required to segment fatty tissue more accurately.

Automated Segmentation of Trigeminal Nerve and Cerebrovasculature in MR-Angiography Images by Deep Learning

Trigeminal neuralgia caused by paroxysmal and severe pain in the distribution of the trigeminal nerve is a rare chronic pain disorder. It is generally accepted that compression of the trigeminal root entry zone by vascular structures is the major cause of primary trigeminal neuralgia, and vascular decompression is the prior choice in neurosurgical treatment. Therefore, accurate preoperative modeling/segmentation/visualization of trigeminal nerve and its surrounding cerebrovascular is important to surgical planning. In this paper, we propose an automated method to segment trigeminal nerve and its surrounding cerebrovascular in the root entry zone, and to further reconstruct and visual these anatomical structures in three-dimensional (3D) Magnetic Resonance Angiography (MRA). The proposed method contains a two-stage neural network. Firstly, a preliminary confidence map of different anatomical structures is produced by a coarse segmentation stage. Secondly, a refinement segmentation stage is proposed to refine and optimize the coarse segmentation map. To model the spatial and morphological relationship between trigeminal nerve and cerebrovascular structures, the proposed network detects the trigeminal nerve, cerebrovasculature, and brainstem simultaneously. The method has been evaluated on a dataset including 50 MRA volumes, and the experimental results show the state-of-the-art performance of the proposed method with an average Dice similarity coefficient, Hausdorff distance, and average surface distance error of 0.8645, 0.2414, and 0.4296 on multi-tissue segmentation, respectively.

Tissue Segmentation from Whole-Slide Images Using Lightweight Neural Networks

Pathology slides of malignancies are segmented using lightweight convolutional neural networks (CNNs) that may be deployed on mobile devices. This is made possible by preprocessing candidate images to make CNN analysis tractable and also to exclude regions unlikely to be diagnostically relevant. In a training phase, labeled whole-slide histopathology images are first downsampled and decomposed into square tiles. Tiles corresponding to diseased regions are analyzed to determine boundary values of a visual criterion, image entropy. A lightweight CNN is then trained to distinguish tiles of diseased and non-diseased tissue, and if more than one disease type is present, to discriminate among these as well. A segmentation is generated by downsampling and tiling a candidate image, and retaining only those tiles with values of the visual criterion falling within the previously established extrema. The sifted tiles, which now exclude much of the non-diseased image content, are efficiently and accurately classified by the trained CNN. Tiles classified as diseased tissue ¾ or in the case of multiple possible subtypes, as the dominant subtype in the tile set ¾ are combined, either as a simple union or at a pixel level, to produce a segmentation mask or map. This approach was applied successfully to two very different datasets of large whole-slide images, one (PAIP2020) involving multiple subtypes of colorectal cancer and the other (CAMELYON16) single-type breast-cancer metastases. Scored using standard similarity metrics, the segmentations exhibited notably high recall, even when tiles were large relative to tumor features.