The success of deep convolutional neural networks (NNs) on image classification and recognition tasks has led to new applications in very diversified contexts, including the field of medical imaging. In this paper, we investigate and propose NN architectures for automated multiclass segmentation of anatomical organs in chest radiographs (CXRs), namely for lungs, clavicles, and heart. We address several open challenges including model overfitting, reducing number of parameters, and handling of severely imbalanced data in CXR by fusing recent concepts in convolutional networks and adapting them to the segmentation problem task in CXR. We demonstrate that our architecture combining delayed subsampling, exponential linear units, highly restrictive regularization, and a large number of high-resolution low-level abstract features outperforms state-of-the-art methods on all considered organs, as well as the human observer on lungs and heart. The models use a multiclass configuration with three target classes and are trained and tested on the publicly available Japanese Society of Radiological Technology database, consisting of 247 X-ray images the ground-truth masks for which are available in the segmentation in CXR database. Our best performing model, trained with the loss function based on the Dice coefficient, reached mean Jaccard overlap scores of 95% for lungs, 86.8% for clavicles, and 88.2% for heart. This architecture outperformed the human observer results for lungs and heart.
SummaryIdentifying distinct anatomical structures within the brain and developing genetic tools to target them are fundamental steps for understanding brain function. We hypothesize that enhancer expression patterns can be used to automatically identify functional units such as neuropils and fiber tracts. We used two recent, genome-scale Drosophila GAL4 libraries and associated confocal image datasets to segment large brain regions into smaller subvolumes. Our results (available at https://strawlab.org/braincode) support this hypothesis because regions with well-known anatomy, namely the antennal lobes and central complex, were automatically segmented into familiar compartments. The basis for the structural assignment is clustering of voxels based on patterns of enhancer expression. These initial clusters are agglomerated to make hierarchical predictions of structure. We applied the algorithm to central brain regions receiving input from the optic lobes. Based on the automated segmentation and manual validation, we can identify and provide promising driver lines for 11 previously identified and 14 novel types of visual projection neurons and their associated optic glomeruli. The same strategy can be used in other brain regions and likely other species, including vertebrates.
Figure 1: We render high-quality implicit surfaces on regular grids, e.g., distance fields or medical CT scans, in real-time without pre-computing additional per-voxel information. Gradients with C 1 continuity, second-order derivatives, and surface curvature are computed exactly for each output pixel using tri-cubic filtering. Applications include surface interrogation and visualizing levelset computations by color mapping curvature measures (center), and ridge and valley lines (left and right). AbstractThis paper presents a real-time rendering pipeline for implicit surfaces defined by a regular volumetric grid of samples. We use a ray-casting approach on current graphics hardware to perform a direct rendering of the isosurface. A two-level hierarchical representation of the regular grid is employed to allow object-order and image-order empty space skipping and circumvent memory limitations of graphics hardware. Adaptive sampling and iterative refinement lead to high-quality ray/surface intersections. All shading operations are deferred to image space, making their computational effort independent of the size of the input data. A continuous third-order reconstruction filter allows on-the-fly evaluation of smooth normals and extrinsic curvatures at any point on the surface without interpolating data computed at grid points. With these local shape descriptors, it is possible to perform advanced shading using high-quality lighting and non-photorealistic effects in real-time.
Surgical approaches tailored to an individual patient's anatomy and pathology have become standard in neurosurgery. Precise preoperative planning of these procedures, however, is necessary to achieve an optimal therapeutic effect. Therefore, multiple radiological imaging modalities are used prior to surgery to delineate the patient's anatomy, neurological function, and metabolic processes. Developing a three-dimensional perception of the surgical approach, however, is traditionally still done by mentally fusing multiple modalities. Concurrent 3D visualization of these datasets can, therefore, improve the planning process significantly. In this paper we introduce an application for planning of individual neurosurgical approaches with high-quality interactive multimodal volume rendering. The application consists of three main modules which allow to (1) plan the optimal skin incision and opening of the skull tailored to the underlying pathology; (2) visualize superficial brain anatomy, function and metabolism; and (3) plan the patient-specific approach for surgery of deep-seated lesions. The visualization is based on direct multi-volume raycasting on graphics hardware, where multiple volumes from different modalities can be displayed concurrently at interactive frame rates. Graphics memory limitations are avoided by performing raycasting on bricked volumes. For preprocessing tasks such as registration or segmentation, the visualization modules are integrated into a larger framework, thus supporting the entire workflow of preoperative planning.
down to single cells and molecules). The review is in line with major European initiatives, such as COMULIS (CA17121), a COST Action to promote and foster Correlated Multimodal Imaging in Life Sciences.
Segmentation in 3D scans is playing an increasingly important role in current clinical practice supporting diagnosis, tissue quantification, or treatment planning. The current 3D approaches based on Convolutional Neural Networks (CNN) usually suffer from at least three main issues caused predominantly by implementation constraints -first, they require resizing the volume to the lower-resolutional reference dimensions, second, the capacity of such approaches is very limited due to memory restrictions, and third, all slices of volumes have to be available at any given training or testing time. We address these problems by a U-Net-like [1] architecture consisting of bidirectional Convolutional Long Short-Term Memory (C-LSTM) [2] and convolutional, pooling, upsampling and concatenation layers enclosed into timedistributed wrappers. Our network can either process the full volumes in a sequential manner, or segment slabs of slices on demand. We demonstrate performance of our architecture on vertebrae and liver segmentation tasks in 3D CT scans.
Fig. 1: Neural projections in the brain of the fruit fly visualized using the BrainGazer system. Abstract-Neurobiology investigates how anatomical and physiological relationships in the nervous system mediate behavior. Molecular genetic techniques, applied to species such as the common fruit fly Drosophila melanogaster, have proven to be an important tool in this research. Large databases of transgenic specimens are being built and need to be analyzed to establish models of neural information processing. In this paper we present an approach for the exploration and analysis of neural circuits based on such a database. We have designed and implemented BrainGazer, a system which integrates visualization techniques for volume data acquired through confocal microscopy as well as annotated anatomical structures with an intuitive approach for accessing the available information. We focus on the ability to visually query the data based on semantic as well as spatial relationships. Additionally, we present visualization techniques for the concurrent depiction of neurobiological volume data and geometric objects which aim to reduce visual clutter. The described system is the result of an ongoing interdisciplinary collaboration between neurobiologists and visualization researchers.
Summary. Visualization and quantitative analysis of vessel data is an important preprocessing step in diagnosis of vascular diseases, monitoring, surgery planning, blood flow simulation, education and training of surgeons. This paper surveys several geometrie methods to solve basie visualization and quantification problems like centerline computation, boundary detection, projection teehniques, and geometrie model generation.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
334 Leonard St
Brooklyn, NY 11211
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.