Two approaches to Hessian-based estimation of tubular blood-vessel radius from 3D raster images are compared. In the proposed approach, binary skeleton is found for each tubular vessel-tree branch by ...thresholding the Hessian-derived vesselness image. Coordinates of the binary skeleton are approximated with smooth 3D spline functions. Their derivatives with respect to arc length give local tangent vectors, and thus planes normal to the vessel centerline. A proposed image intensity profile model is then least-squares fitted to the vessel cross-section by those planes, at each skeleton point. The circular vessel local radius is one of the model parameters. In the reference method, the vessel centerline direction is defined by the local Hessian eigenvector corresponding to the smallest eigenvalue. The radius is estimated using a square root of the vessel cross-section area (as obtained by an adaptive thresholding), divided by π. The impact of Frangi Hessian filter parameters and scale selection on the methods' performance is examined. Higher accuracy, precision and robustness to image noise and artifacts is demonstrated for the proposed method. Example of the method suitability for modeling of brain vasculature magnetic resonance images is also presented in this paper.
It has been demonstrated recently by the authors that texture analysis of 3D images of vascular trees can be used to describe the trees quantitatively (see References section). Computer-simulated ...trees and their raster images were used in that study. This paper presents experimental confirmation of the findings using supervised and unsupervised classification methods.
A technique is proposed for modeling the surface of normal cerebral vasculature based on three-dimensional magnetic resonance images. The Frangi multiscale image filtering is the starting point, ...followed by thresholding and skeletonization. The skeleton of tubular branches is approximated by a smooth function in 3D, allowing accurate estimation of tangent vector to the vessel centerline and planes normal to it. Vessel radius is then computed by least-squares fitting of the image intensity model to vessel cross-sections by normal planes. In effect, each tubular branch of the vessel tree is represented by centerline-radius description. The usage of Frangi filtering results in tubular branch discontinuities in places where the vessels do not feature the assumed elongated shape, e.g. at bifurcations and intensity artefacts. This paper proposes algorithms for modeling the vessel tree surface discontinuities. The resulting integrated surface of the macroscale (of diameter comparable or larger than the voxel side) vessels model is waterproof. This is important for future usage of the model for blood flow simulation. A network of mesoscale vessels (of diameter smaller than the voxel side) is synthesized at the branch terminations of the macroscale surface model, using constrained numerical optimization. This is a step toward modeling the whole brain vasculature.
T2-weighted magnetic resonance images (T2W MRI) of prostate cancer are usually acquired with a large slice thickness compared to in-plane voxel dimensions and to the minimal significant malignant ...prostate tumour size. This causes a negative partial volume effect, decreasing the precision of tumour volumetry and complicating 3D texture analysis of the images. At the same time, three orthogonal, anisotropic acquisitions with overlapping fields of view are often acquired to allow insight into the prostate from different anatomical planes. It is desirable to reconstruct an isotropic prostate T2W image, using the 3 orthogonal volumes computationally, instead of directly acquiring a high-resolution MR image, which typically requires elongated scanning time, with higher cost, less patient comfort and lower signal-to-noise ratio. In our previous work, we followed the above rationale applying a Markov-Random-Field(MRF)-based combination of 3 orthogonal T2W images of the prostate. Our initial results were, however, biased by the quality of input orthogonal images. These were first preprocessed using spline interpolation to yield the same voxel dimensions and later registered. In this paper, we apply a dictionary learning approach to interpolation in order to increase the resolution of a coronal T2W MRI image. We compose a low-resolution dictionary from the original axial image, calculate its sparse representation by Orthogonal Matching Pursuit and finally derive the high-resolution dictionary to improve the original coronal image. We assess the improvement in visual image quality as satisfying and propose further studies.
Traditionally, analysis of Dynamic Contrast-Enhanced Magnetic Resonance Images (DCE MRI) requires pharmacokinetic modelling to derive quantitative physiological parameters of the tissue. Modelling, ...however, is a complex task and many competing models of contrast agent kinetics and tissue structure were proposed. Alternatively, raw DCE data could be analysed to find correlation with pathology in the tissue or other desired effects, for example by clustering. In this paper, we propose a new method for DCE MRI timeseries clustering. We model the data space as a Conditional Random Field (CRF) and optimize the objective function in order to find cluster labels for all timeseries. The method is unsupervised and fully automatic. We also propose a strategy to speed up the clustering process using Support Vector Machines. We demonstrate the utility of our method on two distinct problems: prostate cancer localization and healthy kidney compartment segmentation.
This paper describes a pilot study aimed at evaluating the usefulness of texture parameters computed from blood pharmacokinetic maps as biomarkers of cancers. The maps are extracted from DCE MRI ...images which, in turn, visualize blood perfusion in tissue by means of a contrast medium. Each voxel of a DCE image is characterised by a curve in time domain (enhancement signal). The shape of such curves is characteristic to different tissues and is quantified by fitting a proposed parameterized signal model. Different-modality MR images were acquired for a small number of patients, prior to diagnosed cancer surgical extraction. Six images were collected for each patient, four of them were distributions of blood pharmacokinetic model parameters and two - ADC and VIBE MR images. The extracted carcinoma tissue for each patient was investigated by pathologists. This histological grade provided labels to images for supervised data exploration. Both unsupervised (clustering) and supervised (LDA) experiments demonstrated the image texture can be a promising biomarker to reflect correlations between tumour tissue appearance at structural and functional MRI and the corresponding histological characteristics.
In this paper, we present a new unsupervised prostate cancer (PCa) localization algorithm for the peripheral zone (PZ), utilizing well-established rules used in clinical PCa diagnosis from mpMRI ...data. We perform clustering on ADC and DWI images accompanied by T2W examination of clustered regions and then combined with DCE findings. For each of the 10 analysed patients, we obtain a likelihood map showing suspicious areas. We evaluate our method by comparison against radiological MR tumor segmentations and delineations in histopathological whole-mount sections automatically registered to the MR, using voxel-wise ROC analysis. The resulting mean AUC values for our algorithm were 0.81 and 0.67 with radiological and histopathological ground truth, respectively, while the mean AUC for the radiological segmentation with the histopathological segmentation as the ground truth was 0.60. We conclude that the proposed approach can localize PZ PCa with good accuracy and could be used as an aid for radiologists.
High resolution (HR) volume reconstruction is a collection of post-processing algorithms applied to enhance out-of-plane or sub-voxel image quality. In this work, we use HR reconstruction to combine ...three orthogonal magnetic resonance (MR) acquisitions of the prostate, i.e. axial, sagittal and coronal. The three orthogonal MR volumes are first resampled, registered and intensity-corrected. Then we reconstruct the HR volumes using maximum a posteriori (MAP) approach with Markov-Random-Field-based (MRF) regularization. Our preliminary results are promising and show the usefulness of the method to reconstruct HR prostate MR volumes. Future work includes the use of this method for calculation of 3D features in the context of in vivo detection of PCa.
An algorithm is developed for automated modeling of tubular blood vessel segments, based on their noisy 3D raster image. The approach is based on continuous-function approximation of binary skeleton ...lines extracted from thresholded multiscale vesselness images. The continuous centerline functions allow robust computation of tangent vectors, to define normal planes and 3D image cross-sections on those planes. A vessel intensity profile model is next least-squares fitted to the image cross-section along straight lines segments - anchored at centerline and extended toward vessel walls, at a number of directions covering the full angle. Vessel parameters, such as local radius for circular vessels, distances between the centerline and edges for non-circular shapes or intensity profile corresponding to blood velocity distribution, are estimated through the model fitting. Subvoxel accuracy vessel representation, robustness to noise and image inhomogeneity are of primary concern. The algorithm is applied to 3D synthetic and real-life magnetic resonance images. It is demonstrated that the proposed method facilitates automated extraction of geometric vessel-tree models from images and outperforms the well-known Hessian vector approach in terms of accurate estimation of the centerline local direction in noisy images.
Magnetic resonance imaging (MRI) is currently widely used in medical image diagnosis. However, MR scanners are extensively used in climes and thus are rarely accessible for experimentation. In ...consequence, the number of images available for image processing methods evaluation is too low and there appears a need for a method to generate synthetic images. In their previous works, the authors studied various methods for blood vessels segmentation and tracking. Effectiveness of the designed algorithms requires objective verification which implies repetition of experiments for large number of images and comparing the results with some ground truth models. Therefore, this study aims at designing a computer system which implements numerical routines for generation of synthetic MRA images. In particular, in this paper we study the performance of various configurations of assembled computer grid and analyze their potential in angiographic image synthesis.