Methods for the analysis of digital-image texture are reviewed. The functions of MaZda, a computer program for quantitative texture analysis developed within the framework of the European COST ...(Cooperation in the Field of Scientific and Technical Research) B11 program, are introduced. Examples of texture analysis in magnetic resonance images are discussed.
In this paper, we propose a method for reducing thermal noise in diffusion-weighted magnetic resonance images (DWI MRI) of the brain using a convolutional neural network (CNN) trained on realistic, ...synthetic MR data. Two reference methods are considered: a) averaging of repeated scans, a widespread method used in clinics to improve signal-to-noise ratio of MR images and b) the blockwise Non-Local Means (NLM) filter, one of the post-processing methods frequently used in DWI denoising. To obtain training data for transfer learning, the effects of echo-planar imaging (EPI) – Nyquist ghosting and ramp sampling – are modelled in a data-driven fashion. These effects are introduced to the digital phantom of brain anatomy (BrainWeb). Real noise maps are obtained from the MRI scanner with a brain-DWI-designed protocol and later combined with simulated, noise-free EPI images. The Point Spread Function is measured in a DW image of an AJR-approved geometrical phantom. Inter-scan patient movement is captured from a brain scan of a healthy volunteer using image registration. The denoising methods are applied to the simulated EPI brain images and in real EPI DWI of the brain. The quality of denoised images is evaluated at several signal-to-noise ratios. The characteristics of noise residuals are studied thoroughly. A diffusion phantom is used to investigate the influence of denoising on ADC measurements. The method is also evaluated on a GRAPPA dataset. We show that our method outperforms NLM and image averaging and allows for a significant reduction in scan time by lowering the number of repeated scans. We also analyse the trained CNN denoisers and point out the challenges accompanying this denoising method.
Magnetic resonance imaging (MRI) modalities have achieved an increasingly important role in the clinical work-up of chronic kidney diseases (CKD). This comprises among others assessment of ...hemodynamic parameters by arterial spin labeling (ASL) or dynamic contrast-enhanced (DCE-) MRI. Especially in the latter, images or volumes of the kidney are acquired over time for up to several minutes. Therefore, they are hampered by motion, e.g., by pulsation, peristaltic, or breathing motion. This motion can hinder subsequent image analysis to estimate hemodynamic parameters like renal blood flow or glomerular filtration rate (GFR). To overcome motion artifacts in time-resolved renal MRI, a wide range of strategies have been proposed. Renal image registration approaches could be grouped into (1) image acquisition techniques, (2) post-processing methods, or (3) a combination of image acquisition and post-processing approaches. Despite decades of progress, the translation in clinical practice is still missing. The aim of the present article is to discuss the existing literature on renal image registration techniques and show today’s limitations of the proposed techniques that hinder clinical translation. This paper includes transformation, criterion function, and search types as traditional components and emerging registration technologies based on deep learning. The current trend points towards faster registrations and more accurate results. However, a standardized evaluation of image registration in renal MRI is still missing.
Due to inherent physical and hardware limitations, 3D MR images are often acquired in the form of orthogonal thick slices, resulting in highly anisotropic voxels. This causes the partial volume ...effect, which introduces blurring of image details, appearance of staircase artifacts and significantly decreases the diagnostic value of images. To restore high resolution isotropic volumes, we propose to use a convolutional neural network (CNN) driven by patches taken from three orthogonal thick-slice images. To assess the validity and efficiency of this postprocessing approach, we used 1x1x1 mm3-voxel brain images of different modalities, available via the well known BrainWeb database. They served as a high resolution reference and were numerically preprocessed to create input images of different slice thickness and anatomical orientation, for CNN training, validation and testing. The visual quality of reconstructed images was indeed superior, compared to images obtained by fusion of interpolated thick-slice images, or to images reconstructed with the CNN using a single input MR scan. The significant increase of objectively computed figures of merit, e.g. the Structural Similarity Image Metric, was also noticed. Keeping in mind that any single value of such quality metrics represents a number of psychophysical effects, we applied the CNN trained on brain images for superresolution reconstruction of synthetic and acquired blood vessel tree images. We then used the restored superresolution volumes for estimation of vessel radii. It was demonstrated that vessel radius values derived from superresolution images of simulated vessel trees are significantly more accurate than those obtained from a standard fusion of interpolated thick-slice orthogonal scans. Superiority of the CNN-based superresolution images was also observed for scanner-acquired MR scans according to the evaluated parameters. These three experiments show the efficiency of CNN-based image reconstruction for qualitative and quantitative improvement of its diagnostic quality, as well as illustrates the practical usefulness of transfer learning - networks trained on example images of one kind can be used to restore superresolution images of physically different objects.
In this paper we present a numerical framework for validating methods of quantitative analysis of non-invasive MR angiography imaging protocols such as Time-of-Flight (ToF) and Phase Contrast ...Angiography (PCA). Our study is motivated by the need to reliably and objectively verify blood flow and geometry measurements derived from image data. Both factors are important predictors in diagnosing of carotid artery stenosis. Credibility of the tested image processing methods is verified by comparing their results against reference models designed using integrated flow and MRA imaging simulator.
Vascularity diseases comprise one of the major health problems worldwide. Their diagnosis, therapy planning, treatment, and etiology understanding require personalized geometrical modeling of the ...vessel trees and measurement of their parameters, e.g. local lumen radius, volume of calcification or centerline course. Arteries'and veins' wall surface models are essential for visualization and for blood flow numerical simulation - in support ofbiomedical research and education. Medical imaging is the main technique that makes those measurements possible in a least invasive way, with 3D magnetic resonance (MR) and computed tomography (CT) modalities being the most popular. This tutorial is focused on methods of automated processing of 3D images for objective quantification of blood vessel lumen. One of the aims of the search for such methods is to release medical experts from the tedious, time-consuming (and overall subjective) annotation of the enormous number of voxels in 3D space. Other expectations include better repeatability, objectivity and accuracy/precision of the obtained diagnostic data and shorter time of their extraction from images. Accurate segmentation of the vascular objects in the image is a challenging task - the images contain noise, artefacts, the vessels feature high shape complexity and are closely surrounded by other tissues and organs of similar appearance. The vessel branches' diameter take values in a wide range - from tens of millimeters to tens of micrometers - while the resolution of images is limited and can not be improved without increasing the noise level or time of examination. The noise introduces uncertainty to segmentation results, expanded acquisition time may result in significant artefacts caused by movement of organs and patient body. A widely adopted solution - acquiring images composed of anisotropic voxels (small in-plane elements extending deep into a few times thicker slice) - is a compromise which does not help in resolving details of 3D shapes. Superresolution preprocessing might be helpful in reducing the segmentation errors in such cases. There are two general approaches to vessel segmentation - via 2D cross-sections perpendicular to the vessel centerline and by direct 3D volumetric segmentation. The first one is usually preceded by image "vesselness" filtering to define the centerline, the second may use long-lasting iterative computations, e.g. those of the level-set algorithm. There is an open-source and free software available for 3D vascular image segmentation, e.g. SimVascular or ITK-Snap, as well as a few free-access databases of annotated vasculature images. Reference to those resources and examples of their use will be provided. Centerlinebased methods and code, developed at TUL Institute of Electronics in collaboration with researchers from Jena University and Medical University of Łódź will also be illustrated. Various schemes of 2D cross-sections segmentation and quantification will be compared, with application to computer-simulated 3D images, MR images of 3D-printed vascularity models, human coronary arteries visualized in X-ray CT volumes, and 3D MR images of human brain blood vessels.
Highlights • Introduction of a framework for quantitative and observer-independent validation of vessel segmentation in the MRA images • Demonstration of a practical application of a custom-designed ...magnetic resonance angiography simulator as a reliable tool for quantitative validation of vessel segmentation algorithms that provides objective and reproducible results. • Development of a realistic digital phantom of an intracranial arterial tree based on a real Time-of-Flight data set. • Performing thorough validation of a level-set vessel segmentation algorithm implemented in the Vascular Modeling Toolkit (VMTK), an open-source software library. • A comparison study of three vessel segmentation algorithms: 1) the VMTK's level-set method, 2) the Chan-Vese formulation of level-set method and 3) the multi-scale Hessian-based vessel enhancement funciton.
Abstract A method is proposed for quantitative description of blood-vessel trees, which can be used for tree classification and/or physical parameters indirect monitoring. The method is based on ...texture analysis of 3D images of the trees. Several types of trees were defined, with distinct tree parameters (number of terminal branches, blood viscosity, input and output flow). A number of trees were computer-simulated for each type. 3D image was computed for each tree and its texture features were calculated. Best discriminating features were found and applied to 1-NN nearest neighbor classifier. It was demonstrated that (i) tree images can be correctly classified for realistic signal-to-noise ratio, (ii) some texture features are monotonously related to tree parameters, (iii) 2D texture analysis is not sufficient to represent the trees in the discussed sense. Moreover, applicability of texture model to quantitative description of vascularity images was also supported by unsupervised exploratory analysis. Eventually, the experimental confirmation was done, with the use of confocal microscopy images of rat brain vasculature. Several classes of brain tissue were clearly distinguished based on 3D texture numerical parameters, including control and different kinds of tumours – treated with NG2 proteoglycan to promote angiogenesis-dependent growth of the abnormal tissue. The method, applied to magnetic resonance imaging e.g. real neovasculature or retinal images can be used to support noninvasive medical diagnosis of vascular system diseases.
OBJECTIVES:To (1) determine whether magnetic resonance (MR) image interpolation at the pixel or k-space level can improve the results of texture-based pattern classification, and (2) compare the ...effects of image interpolation on texture features of different categories, with regard to their ability to distinguish between different patterns.
MATERIALS AND METHODS:We obtained T2-weighted, multislice multiecho MR images of 2 sets of each 3 polystyrene spheres and agar gel (PSAG) phantoms with different nodular patterns (sphere diameterPSAG-1, 0.8-1.25 mm; PSAG-2, 1.25-2.0 mm; PSAG-3, 2.0-3.15 mm), using a 3.0 Tesla scanner equipped with a dedicated microimaging gradient insert. Image datasets, which consisted of 20 consecutive axial slices each, were obtained with a constant field of view (30 × 30 mm), but with variations of matrix size (MTX)16 × 16; 32 × 32; 64 × 64; 128 × 128; and 256 × 256. Original images were interpolated to higher matrix sizes (up to 256 × 256) by means of linear and cubic B-spline (pixel level) as well as zero-fill (k-space level) interpolation. For both original and interpolated image datasets, texture features derived from the co-occurrence (COC) and run-length matrix (RUN), absolute gradient (GRA), autoregressive model, and wavelet transform (WAV) were calculated independently. Based on the 3 best texture features of each category, as determined by calculation of Fisher coefficients using images from the first set of PSAG phantoms (training dataset), k-means clustering was performed to separate PSAG-1, PSAG-2, and PSAG-3 images belonging to the second set of phantoms (test dataset). This was done independently for all original and interpolated image datasets. Rates of misclassified data vectors were used as primary outcome measures.
RESULTS:For images based on a very low original resolution (MTX = 16 × 16), misclassification rates remained high, despite the use of interpolation. For higher resolution images (MTX = 32 × 32 and 64 × 64), interpolation enhanced the ability of texture features, in all categories except WAV, to discriminate between the 3 phantoms. This positive effect was particularly pronounced for COC and RUN features, and to a lesser degree, also GRA features. No consistent improvements, and even some negative effects, were observed for WAV features, after interpolation. Although there was no clear superiority of any single interpolation techniques at very low resolution (MTX = 16 × 16), zero-fill interpolation outperformed the two pixel interpolation techniques, for images based on higher original resolutions (MTX = 32 × 32 and 64 × 64). We observed the most considerable improvements after interpolation by a factor of 2 or 4.
CONCLUSIONS:MR image interpolation has the potential to improve the results of pattern classification, based on COC, RUN, and GRA features. Unless spatial resolution is very poor, zero-filling is the interpolation technique of choice, with a recommended maximum interpolation factor of 4.