Ordered subset reconstruction for x-ray CT

Statistical methods for image reconstruction such as the maximum likelihood expectation maximization are more robust and flexible than analytical inversion methods and allow for accurate modelling of the counting statistics and photon transport during acquisition of projection data. Statistical reconstruction is prohibitively slow when applied to clinical x-ray CT due to the large data sets and the high number of iterations required for reconstructing high-resolution images. Recently, however, powerful methods for accelerating statistical reconstruction have been proposed which, instead of accessing all projections simultaneously for updating an image estimate, are based on accessing a subset of projections at the time during iterative reconstruction. In this paper we study images generated by the convex algorithm accelerated by the use of ordered subsets (the OS convex algorithm (OSC)) for data sets with sizes, noise levels and spatial resolution representative of x-ray CT imaging. It is only in the case of extremely high acceleration factors (higher than 50, corresponding to fewer than 20 projections per subset), that areas with incorrect grey values appear in the reconstructed images, and that image noise increases compared with the standard convex algorithm. These image degradations can be adequately corrected for by running the final iteration of OSC with a reduced number of subsets. Even by applying such a relatively slow final iteration, OSC produces almost an equal resolution and lesion contrast as the standard convex algorithm, but more than two orders of magnitude faster.     

Off-line determination of the optimal number of iterations of the robust anisotropic diffusion filter applied to denoising of brain MR images

Although anisotropic diffusion filters have been used extensively and with great success in medical image denoising, one limitation of this iterative approach, when used on fully automatic medical image processing schemes, is that the quality of the resulting denoised image is highly dependent on the number of iterations of the algorithm. Using many iterations may excessively blur the edges of the anatomical structures, while a few may not be enough to remove the undesirable noise. In this work, a mathematical model is proposed to automatically determine the number of iterations of the robust anisotropic diffusion filter applied to the problem of denoising three common human brain magnetic resonance (MR) images (T1-weighted, T2-weighted and proton density). The model is determined off-line by means of the maximization of the mean structural similarity index, which is used in this work as metric for quantitative assessment of the resulting processed images obtained after each iteration of the algorithm. After determining the model parameters, the optimal number of iterations of the algorithm is easily determined without requiring any extra computation time. The proposed method was tested on 3D synthetic and clinical human brain MR images and the results of qualitative and quantitative evaluation have shown its effectiveness.     

Generalized Gibbs priors based positron emission tomography reconstruction

Bayesian methods have been widely applied to the ill-posed problem of image reconstruction. Typically the prior information of the objective image is needed to produce reasonable reconstructions. In this paper, we propose a novel generalized Gibbs prior (GG-Prior), which exploits the basic affinity structure information in an image. The motivation for using the GG-Prior is that it has been shown to be effective noise suppression, while also maintaining sharp edges without oscillations. This feature makes it particularly attractive for the reconstruction of positron emission tomography (PET) where the aim is to identify the shape of objects from the background by sharp edges. We show that the standard paraboloidal surrogate coordinate ascent (PSCA) algorithm can be modified to incorporate the GG-Prior using a local linearized scheme in each iteration process. The proposed GG-Prior MAP reconstruction algorithm based on PSCA has been tested on simulated, real phantom data. Comparison studies with conventional filtered backprojection (FBP) method and Huber prior clearly demonstrate that the proposed GG-Prior performs better in lowering the noise, preserving the image edge and in higher signal noise ratio (SNR) condition.     

Accelerated median root prior reconstruction for pinhole single-photon emission tomography (SPET)

Pinhole collimation can be used to improve spatial resolution in SPET. However, the resolution improvement is achieved at the cost of reduced sensitivity, which leads to projection images with poor statistics. Images reconstructed from these projections using the maximum likelihood expectation maximization (ML-EM) algorithms, which have been used to reduce the artefacts generated by the filtered backprojection (FBP) based reconstruction, suffer from noise/bias trade-off: noise contaminates the images at high iteration numbers, whereas early abortion of the algorithm produces images that are excessively smooth and biased towards the initial estimate of the algorithm. To limit the noise accumulation we propose the use of the pinhole median root prior (PH-MRP) reconstruction algorithm. MRP is a Bayesian reconstruction method that has already been used in PET imaging and shown to possess good noise reduction and edge preservation properties. In this study the PH-MRP algorithm was accelerated with the ordered subsets (OS) procedure and compared to the FBP, OS-EM and conventional Bayesian reconstruction methods in terms of noise reduction, quantitative accuracy, edge preservation and visual quality. The results showed that the accelerated PH-MRP algorithm was very robust. It provided visually pleasing images with lower noise level than the FBP or OS-EM and with smaller bias and sharper edges than the conventional Bayesian methods.     

采用MRF二次混合多阶先验的PET图像的贝叶斯重建

在正电子发射成像（肿）中，很多方法被用来抑制重建图像中的噪声效果，其中。在所有方法中，贝叶斯重建或者最大化后验估计的方法被证明具有在重建图像质量方面相对于其他方法的优越性。基于贝叶斯重建，本研究提出了一种应用于贝叶斯重建中新的综合了二次一阶先验和二次二阶先验的马尔可夫随机场混合多阶先验。基于不同阶数的二次平滑先验的自身的不同性质，该新先验的设计的目的是实现自适应的发挥这些算子的作用。该混合先验能够保持其先验能量函数凸性，从而保证整体目标贝叶斯后验能量函数的凹性。模拟实验和实验结果的比较证明了对于PET重建，该先验在抑制背景噪声和保持边缘方面均具有很好的表现。     

Non-iterative methods incorporating a priori source distribution and data information for suppression of image noise and artefacts in 3D SPECT

Non-iterative methods have been developed for image reconstruction in 3D SPECT with uniform attenuation and distance-dependent spatial resolution. It was observed that these methods can, in general, be susceptible to data noise and other errors, yielding conspicuous image artefacts. In this work, we developed and evaluated a regularized inverse-filtering approach for effective suppression of noise and artefacts in 3D SPECT images without significantly compromising image resolution. The proposed approach allows the incorporation of a priori random image field and data information and can thus robustly control the degree of suppression of noise and artefacts in 3D SPECT images. Using computer simulations, we evaluated and compared quantitatively images reconstructed from data sets of various noise levels by the use of the proposed methods and the existing non-iterative methods. These numerical results clearly demonstrated that the proposed regularized inverse-filtering approach can effectively suppress image noise and artefacts that plague the existing non-iterative methods, thus yielding quantitatively more accurate 3D SPECT images. The proposed regularized inverse-filtering approach can also be generalized to other imaging modalities.     

Noise and signal decoupling in maximum-likelihood reconstructions and Metz filters for PET brain images

Images reconstructed with the maximum-likelihood-by-expectation-maximization (ML) algorithm have lower noise in some regions, particularly low count areas, compared with images reconstructed with filtered backprojection (FBP). The use of statistically correct noise model coupled with the positivity constraint in the ML algorithm provides this noise improvement, but whether this model confers a general advantage for ML over FBP with no noise model and any reconstruction filter, is unclear. We have studied the quantitative impact of the correct noise model in the ML algorithm applied to simulated and real PET fluoro-deoxyglucose (FDG) brain images, given a simplified but accurate reconstruction model with spatially invariant resolution. For FBP reconstruction, several Metz filters were chosen and images with different resolution were obtained depending on the order (1-400) of the Metz filters. Comparisons were made based on the mean Fourier spectra of the projection amplitudes, the noise-power spectra, and the mean region-of-interest signal and noise behaviour in the images. For images with resolution recovery beyond the intrinsic detector resolution, the noise increased significantly for FBP compared with ML. This indicates that in the process of signal recovery using ML, the noise is decoupled from the signal. Such noise decoupling is not possible for FBP. However, for image resolution equivalent to or less than the intrinsic detector resolution, FBP with Metz filters of various orders can achieve a performance similar to ML. The significance of the noise decoupling advantage in ML is dependent on the reconstructed image resolution required for specific imaging tasks.     

3.5D dynamic PET image reconstruction incorporating kinetics-based clusters

Standard 3D dynamic positron emission tomographic (PET) imaging consists of independent image reconstructions of individual frames followed by application of appropriate kinetic model to the time activity curves at the voxel or region-of-interest (ROI). The emerging field of 4D PET reconstruction, by contrast, seeks to move beyond this scheme and incorporate information from multiple frames within the image reconstruction task. Here we propose a novel reconstruction framework aiming to enhance quantitative accuracy of parametric images via introduction of priors based on voxel kinetics, as generated via clustering of preliminary reconstructed dynamic images to define clustered neighborhoods of voxels with similar kinetics. This is then followed by straightforward maximum a posteriori (MAP) 3D PET reconstruction as applied to individual frames; and as such the method is labeled '3.5D' image reconstruction. The use of cluster-based priors has the advantage of further enhancing quantitative performance in dynamic PET imaging, because: (a) there are typically more voxels in clusters than in conventional local neighborhoods, and (b) neighboring voxels with distinct kinetics are less likely to be clustered together. Using realistic simulated (11)C-raclopride dynamic PET data, the quantitative performance of the proposed method was investigated. Parametric distribution-volume (DV) and DV ratio (DVR) images were estimated from dynamic image reconstructions using (a) maximum-likelihood expectation maximization (MLEM), and MAP reconstructions using (b) the quadratic prior (QP-MAP), (c) the Green prior (GP-MAP) and (d, e) two proposed cluster-based priors (CP-U-MAP and CP-W-MAP), followed by graphical modeling, and were qualitatively and quantitatively compared for 11 ROIs. Overall, the proposed dynamic PET reconstruction methodology resulted in substantial visual as well as quantitative accuracy improvements (in terms of noise versus bias performance) for parametric DV and DVR images. The method was also tested on a 90 min (11)C-raclopride patient study performed on the high-resolution research tomography. The proposed method was shown to outperform the conventional method in visual as well as quantitative accuracy improvements (in terms of noise versus regional DVR value performance).     

Image filtering for improved dose resolution in CT polymer gel dosimetry

X-ray computed tomography (CT) has been established as a feasible method of performing dosimetry using polyacrylamide gels (PAGs). A small density change occurs in PAG upon irradiation that provides contrast in PAG CT images. However, low dose resolution limits the clinical usefulness of the technique. This work investigates the potential of using image filtering techniques on PAG CT images in order to reduce image noise and improve dose resolution. CT image noise for the scanner and protocol used for the gel images is analyzed and found to be Gaussian distributed and independent of the contrast level in the images. As a result, several filters for reducing spatially invariant noise are investigated: mean, median, midpoint, adaptive mean, alpha-trimmed mean, sigma mean, and a relatively new filter called SUSAN (smallest univalue segment assimilating nucleus). All filters are applied, using 3x3, 5x5, and 7x7 pixel masks, to a CT image of a PAG irradiated with a stereotactic radiosurgery dose distribution. The dose resolution within 95% confidence (D(delta)95%) is calculated and compared for each filtered image, as well the unfiltered image. In addition, the ability of the filters to maintain the spatial integrity of the dose distribution is evaluated and compared. Results clearly indicate that the filters are not equal in their ability to improve D(delta)95% or in their effect on the spatial integrity of the dose distribution. In general, increasing mask size improves D(delta)95% but simultaneously degrades spatial dose information. The mean filter provides the greatest improvement in D(delta)95%, but also the greatest loss of spatial dose information. The SUSAN, mean adaptive, and alpha-trimmed mean filters all provide comparable, but slightly poorer dose resolution. In addition, the SUSAN and adaptive filters both excel at maintaining the spatial distribution of dose and overall are the best performing filters for this application. The midpoint filter, normally useful for Gaussian noise, is poor all-round, dramatically distorting the dose distribution for masks greater than 3x3. The median filter, a common edge preserving noise reduction filter, performs moderately well, but artificially increases high dose gradients. The sigma filter preserves the spatial distribution of dose very well but is least effective at improving dose resolution. In summary, dose resolution can be significantly improved in CT PAG dosimetry through postprocessing of CT images using spatial noise reduction filters. However, such filters are not equal in their ability to improve dose resolution or to maintain the spatial integrity of the dose distribution and an appropriate filter must be chosen depending on clinical demands of the application.     

Diffuse optical tomography with a priori anatomical information

Diffuse optical tomography (DOT) poses a typical ill-posed inverse problem with a limited number of measurements and inherently low spatial resolution. In this paper, we propose a hierarchical Bayesian approach to improve spatial resolution and quantitative accuracy by using a priori information provided by a secondary high resolution anatomical imaging modality, such as magnetic resonance (MR) or x-ray. In such a dual imaging approach, while the correlation between optical and anatomical images may be high, it is not perfect. For example, a tumour may be present in the optical image, but may not be discernable in the anatomical image. The proposed hierarchical Bayesian approach allows incorporation of partial a priori knowledge about the noise and unknown optical image models, thereby capturing the function-anatomy correlation effectively. We present a computationally efficient iterative algorithm to simultaneously estimate the optical image and the unknown a priori model parameters. Extensive numerical simulations demonstrate that the proposed method avoids undesirable bias towards anatomical prior information and leads to significantly improved spatial resolution and quantitative accuracy.     

Half-scan fan-beam computed tomography with improved noise and resolution properties

Half-scan strategy can be used for reducing scanning time and radiation dose delivered to the patient in fan-beam computed tomography (CT). In helical CT, the data weighting/interpolation functions are often devised based upon half-scan configurations. The half-scan fan-beam filtered backprojection (FFBP) algorithm is generally used for image reconstruction from half-scan data. It can, however, be susceptible to sample aliasing and data noise for configurations with short focal lengths and/or large fan-angles, leading to nonuniform resolution and noise properties in reconstructed images. Uniform resolution and noise properties are generally desired because they may lead to an increased utility of reconstructed images in estimation and/or detection/classification tasks. In this work, we propose an algorithm for reconstruction of images with uniform noise and resolution properties in half-scan CT. In an attempt to evaluate the image-noise properties, we derive analytic expressions for image variances obtained by use of the half-scan algorithms. We also perform numerical studies to assess quantitatively the resolution and noise properties of the algorithms. The results in these studies confirm that the proposed algorithm yields images with more uniform spatial resolution and with lower and more uniform noise levels than does the half-scan FFBP algorithm. Empirical results obtained in noise studies also verify the validity of the derived expressions for image variances. The proposed algorithm would be particularly useful for image reconstruction from data acquired by use of configurations with short focal lengths and large field of measurement, which may be encountered in compact micro-CT and radiation therapeutic CT applications. The analytic results of the image-noise properties can be used for image-quality assessment in detection/classification tasks by use of model-observers.     

Evaluation of the ordered subset convex algorithm for cone-beam CT

Statistical methods for image reconstruction such as maximum likelihood expectation maximization (ML-EM) are more robust and flexible than analytical inversion methods and allow for accurate modelling of the photon transport and noise. Statistical reconstruction is prohibitively slow when applied to clinical x-ray cone-beam CT due to the large data sets and the high number of iterations required for reconstructing high resolution images. One way to reduce the reconstruction time is to use ordered subsets of projections during the iterations, which has been successfully applied to fan-beam x-ray CT. In this paper, we quantitatively analyse the use of ordered subsets in concert with the convex algorithm for cone-beam x-ray CT reconstruction, for the case of circular acquisition orbits. We focus on the reconstructed image accuracy of a 3D head phantom. Acceleration factors larger than 300 were obtained with errors smaller than 1%, with the preservation of signal-to-noise ratio. Pushing the acceleration factor towards 600 by using an increasing number of subsets increases the reconstruction error up to 5% and significantly increases noise. The results indicate that the use of ordered subsets can be extremely useful for cone-beam x-ray CT.     

Convergence of the maximum likelihood reconstruction algorithm for emission computed tomography

Convergence properties of the maximum likelihood estimator (MLE) for emission computed tomographic (ECT) image reconstruction are evaluated as a function of Poisson noise, precision of the assumed system resolution model and iteration number up to 10,000 iterations. In the ECT reconstruction problem, the photon-emitting source distribution is to be estimated from measurements of projections of the emitted photon flux. The MLE algorithm seeks a source distribution which will maximise the maximum likelihood function relating the estimated and the measured projections. A Monte Carlo model of the system transfer function of a single photon emission computed tomographic (SPECT) system allowed realistic projection data to be simulated from a known source distribution. Poisson noise was added to the Monte Carlo simulations. By using projection data from a known source distribution generated through a known system transfer function, we were able to simultaneously evaluate the convergence of both the projection estimations as well as the source distribution estimations. As predicted by theory, the estimates of the projections did continue to improve (or remain the same) for all combinations of Poisson noise (up to 10% RMS) and system resolution (+/- 10% of true value) tested. Convergence of source distribution estimates to the true value was found for up to 10,000 iterations only for low noise (0.1% RMS) with the correct resolution function. For all other combinations, there was some optimum iteration (between 30 and 400) after which the source estimate was degraded even though the estimate of the projections was improved.     

Multi-Resolution Level Sets with Shape Priors: A Validation Report for 2D Segmentation of Prostate Gland in T2W MR Images

The level set approach to segmentation of medical images has received considerable attention in recent years. Evolving an initial contour to converge to anatomical boundaries of an organ or tumor is a very appealing method, especially when it is based on a well-defined mathematical foundation. However, one drawback of such evolving method is its high computation time. It is desirable to design and implement algorithms that are not only accurate and robust but also fast in execution. Bresson et al. have proposed a variational model using both boundary and region information as well as shape priors. The latter can be a significant factor in medical image analysis. In this work, we combine the variational model of level set with a multi-resolution approach to accelerate the processing. The question is whether a multi-resolution context can make the segmentation faster without affecting the accuracy. As well, we investigate the question whether a premature convergence, which happens in a much shorter time, would reduce accuracy. We examine multiple semiautomated configurations to segment the prostate gland in T2W MR images. Comprehensive experimentation is conducted using a data set of a 100 patients (1,235 images) to verify the effectiveness of the multi-resolution level set with shape priors. The results show that the convergence speed can be increased by a factor of ≈?2.5 without affecting the segmentation accuracy. Furthermore, a premature convergence approach drastically increases the segmentation speed by a factor of ≈?17.9.     

Statistical and heuristic image noise extraction (SHINE): a new method for processing Poisson noise in scintigraphic images

Poisson noise is one of the factors degrading scintigraphic images, especially at low count level, due to the statistical nature of photon detection. We have developed an original procedure, named statistical and heuristic image noise extraction (SHINE), to reduce the Poisson noise contained in the scintigraphic images, preserving the resolution, the contrast and the texture. The SHINE procedure consists in dividing the image into 4 x 4 blocks and performing a correspondence analysis on these blocks. Each block is then reconstructed using its own significant factors which are selected using an original statistical variance test. The SHINE procedure has been validated using a line numerical phantom and a hot spots and cold spots real phantom. The reference images are the noise-free simulated images for the numerical phantom and an extremely high counts image for the real phantom. The SHINE procedure has then been applied to the Jaszczak phantom and clinical data including planar bone scintigraphy, planar Sestamibi scintigraphy and Tl-201 myocardial SPECT. The SHINE procedure reduces the mean normalized error between the noisy images and the corresponding reference images. This reduction is constant and does not change with the count level. The SNR in a SHINE processed image is close to that of the corresponding raw image with twice the number of counts. The visual results with the Jaszczak phantom SPECT have shown that SHINE preserves the contrast and the resolution of the slices well. Clinical examples have shown no visual difference between the SHINE images and the corresponding raw images obtained with twice the acquisition duration. SHINE is an entirely automatic procedure which enables halving the acquisition time or the injected dose in scintigraphic acquisitions. It can be applied to all scintigraphic images, including PET data, and to all low-count photon images.     

Noise reduction methods for hybrid subtraction

In digital subtraction angiography, hybrid subtraction provides selective vessel images free of soft-tissue motion artifacts but with a lower signal-to-noise ratio (SNR) than temporal subtraction images. An image processing method called measurement-dependent filtering has been developed to enhance the SNR of hybrid images without losing resolution or selectivity. Linear combinations of four images consisting of a pre- and postcontrast dual-energy measurement pair form both the hybrid image and a lower noise but less selective vessel image. The noise-reduced image is derived by combining the low-frequency components of the hybrid image with the high-frequency components of the lower noise image in a variety of ways. The results of the filtering method, when tested on both phantom and clinical data, display images with about the same degree of conspicuity as the hybrid image and a SNR approaching that of the temporal image.     

结合各向异性扩散滤波Thin Plate先验正电子发射断层图像重建的算法

背景：在正电子发射断层成像中,MAP重建方法通过引入先验分布约束,可以明显提高重建图像的质量,但不合适的先验分布项可能会造成重建图像过度平滑或出现阶梯状边缘伪影。 目的：针对基于传统局部先验信息的MAP方法易于导致重建图像过平滑或产生阶梯状边缘伪影的问题,提出了一种结合各向异性扩散滤波的、基于Thin Plate先验的改进MAP重建算法。 方法：重建算法由两步组成：基于双向扩散系数的PDE各向异性扩散滤波和基于Thin Plate先验的MAP估计。重建图像通过这两步交替迭代得到。文中采用归一化均方根误差和信噪比定量评价重建图像质量。 结果与结论：结合了基于双向扩散系数的PDE各向异性扩散滤波,并将Thin Plate二次二阶先验模型引入到MAP重建算法中,所获得的重建结果图像在抑制噪声、边缘保持方面取得了良好的效果,SNR、RMSE以及视觉评价等方面均有较大程度的改善。     

Statistical image reconstruction from limited projection data with intensity priors

The radiation dose generated from x-ray computed tomography (CT) scans and its responsibility for increasing the risk of malignancy became a major concern in the medical imaging community. Accordingly, investigating possible approaches for image reconstruction from low-dose imaging protocols, which minimize the patient radiation exposure without affecting image quality, has become an issue of interest. Statistical reconstruction (SR) methods are known to achieve a superior image quality compared with conventional analytical methods. Effective physical noise modeling and possibilities of incorporating priors in the image reconstruction problem are the main advantages of the SR methods. Nevertheless, the high computation cost limits its wide use in clinical scanners. This paper presents a framework for SR in x-ray CT when the angular sampling rate of the projection data is low. The proposed framework is based on the fact that, in many CT imaging applications, some physical and anatomical structures and the corresponding attenuation information of the scanned object can be a priori known. Therefore, the x-ray attenuation distribution in some regions of the object can be expected prior to the reconstruction. Under this assumption, the proposed method is developed by incorporating this prior information into the image reconstruction objective function to suppress streak artifacts. We limit the prior information to only a set of intensity values that represent the average intensity of the normal and expected homogeneous regions within the scanned object. This prior information can be easily computed in several x-ray CT applications. Considering the theory of compressed sensing, the objective function is formulated using the ?(1) norm distance between the reconstructed image and the available intensity priors. Experimental comparative studies applied to simulated data and real data are used to evaluate the proposed method. The comparison indicates a significant improvement in image quality when the proposed method is used.     

An anisotropic diffusion method for denoising dynamic susceptibility contrast-enhanced magnetic resonance images.

The purpose of this study was to present an application of a novel denoising technique for improving the accuracy of cerebral blood flow (CBF) images generated from dynamic susceptibility contrast-enhanced magnetic resonance imaging (DSC-MRI). The method presented in this study was based on anisotropic diffusion (AD). The usefulness of this method was firstly investigated using computer simulations. We applied this method to patient data acquired using a 1.5 T MR system. After a bolus injection of Gd-DTPA, we obtained 40-50 dynamic images with a 1.32-2.08 s time resolution in 4-6 slices. The dynamic images were processed using the AD method, and then the CBF images were generated using pixel-by-pixel deconvolution analysis. For comparison, the CBF images were also generated with or without processing the dynamic images using a median or Gaussian filter. In simulation studies, the standard deviation of the CBF values obtained after processing by the AD method was smaller than that of the CBF values obtained without any processing, while the mean value agreed well with the true CBF value. Although the median and Gaussian filters also reduced image noise, the mean CBF values were considerably underestimated compared with the true values. Clinical studies also suggested that the AD method was capable of reducing the image noise while preserving the quantitative accuracy of CBF images. In conclusion, the AD method appears useful for denoising DSC-MRI, which will make the CBF images generated from DSC-MRI more reliable.     

Removal of bone in CT angiography by multiscale matched mask bone elimination

For clear visualization of vessels in CT angiography (CTA) images of the head and neck using maximum intensity projection (MIP) or volume rendering (VR) bone has to be removed. In the past we presented a fully automatic method to mask the bone [matched mask bone elimination (MMBE)] for this purpose. A drawback is that vessels adjacent to bone may be partly masked as well. We propose a modification, multiscale MMBE, which reduces this problem by using images at two scales: a higher resolution than usual for image processing and a lower resolution to which the processed images are transformed for use in the diagnostic process. A higher in-plane resolution is obtained by the use of a sharper reconstruction kernel. The out-of-plane resolution is improved by deconvolution or by scanning with narrower collimation. The quality of the mask that is used to remove bone is improved by using images at both scales. After masking, the desired resolution for the normal clinical use of the images is obtained by blurring with Gaussian kernels of appropriate widths. Both methods (multiscale and original) were compared in a phantom study and with clinical CTA data sets. With the multiscale approach the width of the strip of soft tissue adjacent to the bone that is masked can be reduced from 1.0 to 0.2 mm without reducing the quality of the bone removal. The clinical examples show that vessels adjacent to bone are less affected and therefore better visible. Images processed with multiscale MMBE have a slightly higher noise level or slightly reduced resolution compared with images processed by the original method and the reconstruction and processing time is also somewhat increased. Nevertheless, multiscale MMBE offers a way to remove bone automatically from CT angiography images without affecting the integrity of the blood vessels. The overall image quality of MIP or VR images is substantially improved relative to images processed with the original MMBE method.