Split-and-merge segmentation of magnetic resonance medical images: performance evaluation and extension to three dimensions

Intensity- or edge-based methods of segmentation are often insufficiently robust to be applied to images containing complex anatomical objects, such as those seen in high-resolution magnetic resonance imaging systems. Split-and-merge techniques attempt to overcome these difficulties by using homogeneity measures. Simple modifications to the basic 2D split-and-merge method, based on the principles of simulated annealing and controlled boundary elimination, are developed and discussed. Simulated annealing reduced the number of regions by 22% with a further reduction of 21% achieved through boundary elimination. Smoother regional boundaries are also produced. These methods are extended to true 3D and quantitatively compared with their 2D counterparts. The main advantage of 3D methods is that they produce segmented volumes by directly preserving the connectivity between slices, whereas in 2D, segments have to be grouped together in a post-split-and-merge process. Finally, the properties of the 3D approach are demonstrated by the automatic quantitation of brain ventricle volume, producing estimates to within 7% of validated manual methods.     

Edge detection in brain SPET perfusion imaging: a comparison of several methods

Four methods of brain edge detection on brain SPET perfusion (99Tcm-hexamethylpropylene amine oxime) images were compared: ellipse adaptation, simple thresholding (four threshold values), a low threshold (40%) followed by 1, 2 or 3 pixel erosion, and the Deriche 3D adaptive cut-off frequency method (four filter widths: alpha = 1, 2, 3 or 4). The SPET data of six patients were reconstructed to obtain 10 axial slices, each 10 mm thick, covering the whole brain. On the 60 axial slices, the methods were compared based on automaticity, computation time and accuracy of edge detection compared with morphological edges drawn manually on the patients' 3D co-registered magnetic resonance imaging (MRI) scans. The proportion of pixels inside the contour defined by the MRI scan but outside the SPET edge (p(i)), and the proportion of pixels inside the contour defined by the SPET image but outside the MRI contour (pe), were calculated. The thresholding methods provided interesting results, particularly the application of a low threshold value (40%), followed by a 2 pixel erosion, which required a computation time of 12 s (p(i) = 5.7 +/- 2.2%; pe = 2.7 +/- 0.9%). Because of adjustments to each slice of the ellipse axis, the processing time of this method was about 3 min (p(i) = 1.5 +/- 1.4%; pe = 11.3 +/- 3.4%). The Deriche 3D filter was time-consuming (6 min for 10 slices on a NXT workstation, SMV International). With this method, the best edge fitting was found with a filter width of 3 and 4 (p(i) = 9.6 +/- 11.1%; pe = 14.1 +/- 23.2%; alpha = 3). Three-dimensional filtering methods must be refined to reduce the computation time and to improve brain edge fitting accuracy when compared with the eroded thresholding method.     

An image-processing system for qualitative and quantitative volumetric analysis of brain images

In this work, we developed, implemented, and validated an image-processing system for qualitative and quantitative volumetric analysis of brain images. This system allows the visualization and quantitation of global and regional brain volumes. Global volumes were obtained via an automated adaptive Bayesian segmentation technique that labels the brain into white matter, gray matter, and cerebrospinal fluid. Absolute volumetric errors for these compartments ranged between 1 and 3% as indicated by phantom studies. Quantitation of regional brain volumes was performed through normalization and tessellation of segmented brain images into the Talairach space with a 3D elastic warping model. Retest reliability of regional volumes measured in Talairach space indicated errors of < 1.5% for the frontal, parietal, temporal, and occipital brain regions. Additional regional analysis was performed with an automated hybrid method combining a region-of-interest approach and voxel-based analysis, named Regional Analysis of Volumes Examined in Normalized Space (RAVENS). RAVENS analysis for several subcortical structures showed good agreement with operator-defined volumes. This system has sufficient accuracy for longitudinal imaging data and is currently being used in the analysis of neuroimaging data of the Baltimore Longitudinal Study of Aging.     

Volume estimation from multiplanar 2D ultrasound images using a remote electromagnetic position and orientation sensor

A system is described for calculating volume from a sequence of multiplanar 2D ultrasound images. Ultrasound images are captured using a video digitising card (Hauppauge Win/TV card) installed in a personal computer, and regions of interest transformed into 3D space using position and orientation data obtained from an electromagnetic device (Polhemus, Fastrak). The accuracy of the system was assessed by scanning 10 water filled balloons (13-141 mL), 10 kidneys (147-200 mL) and 16 fetal livers (8-37 mL) in water using an Acuson 128XP/10 (5 MHz curvilinear probe). Volume was calculated using the ellipsoid, planimetry, tetrahedral and ray tracing methods and compared with the actual volume measured by weighing (balloons) and water displacement (kidneys and livers). The mean percentage error for the ray tracing method was 0.9 +/- 2.4%, 2.7 +/- 2.3%, 6.6 +/- 5.4% for balloons, kidneys and livers, respectively. So far the system has been used clinically to scan fetal livers and lungs, neonate brain ventricles and adult prostate glands.     

Watershed-based segmentation of 3D MR data for volume quantization

The aim of this work is the development of a semiautomatic segmentation technique for efficient and accurate volume quantization of Magnetic Resonance (MR) data. The proposed technique uses a 3D variant of Vincent and Soilles immersion-based watershed algorithm that is applied to the gradient magnitude of the MR data and that produces small volume primitives. The known drawback of the watershed algorithm, oversegmentation, is strongly reduced by a priori application of a 3D adaptive anisotropic diffusion filter to the MR data. Furthermore, oversegmentation is a posteriori reduced by properly merging small volume primitives that have similar gray level distributions. The outcome of the proceeding image processing steps is presented to the user for manual segmentation. Through selection of volume primitives, the user quickly segments of first slice, which contains the object of interest. Afterwards, the subsequent slices are automatically segmented by extrapolation. Segmentation results are contingently manually corrected. The proposed segmentation technique is tested on phantom objects, where segmentation errors less than 2% are observed. In addition, the technique is demonstrated on 3D MR data of the mouse head from which the cerebellum is extracted. Volumes of the mouse cerebellum and the mouse brains in toto are calculated.     

Three-dimensional reconstruction of the alpha D and beta C-hemocyanins of Helix pomatia from frozen-hydrated specimens

The three-dimensional (3D) reconstructions of the di-decameric forms of alpha D and beta C-hemocyanins of the Roman snail Helix pomatia and of the decameric half molecules of alpha D-hemocyanin were carried out on frozen-hydrated specimens observed in the electron microscope by using the random conical tilt series method. The three 3D volumes were examined by computing solid-body surface representations and slices through the volume and by eroding the structure progressively through raising of the threshold. The di-decameric molecule of alpha D and beta C-hemocyanins, reconstructed from side views, are very similar and are composed of a cylindrical wall, comprising ten oblique wall units, and of two collar complexes located at both ends of the cylinder, comprising each five arches and an annular collar made up of five collar units. Erosion of the structure reveals that the wall looks like a segment of a five-stranded right-handed helix and that each oblique wall unit resembles a figure 8 inclined to the right. The decameric half molecule of alpha D-hemocyanin, reconstructed from end-on views, resembles the whole molecule, except that the collar is thinner and appears composed of five independent collar complex units. It is suggested that the difference in structural appearance of the collar complex between the whole and the half alpha D-hemocyanin may be due to the missing cone artifact, induced by the angular limitations imposed by the goniometer of the electron microscope. The comparison between the alpha D-hemocyanin and the beta C-di-decameric hemocyanin at high thresholds suggests that in the beta C-hemocyanin the oblique wall units of each half molecule may be linked by two connections, whereas in alpha D-hemocyanin there may be only one. This difference in the number of connections may be responsible for the lower stability of the alpha D molecule at high salt concentration.     

Volumetric spectroscopic imaging with spiral-based k-space trajectories

Spiral-based k-space trajectories were applied in a spectroscopic imaging sequence with time-varying readout gradients to collect volumetric chemical shift information. In addition to spectroscopic imaging of low signal-to-noise ratio (SNR) brain metabolites, the spiral trajectories were used to rapidly collect reference signals from the high SNR water signal to automatically phase the spectra and to aid the reconstruction of metabolite maps. Spectral-spatial pulses were used for excitation and water suppression. The pulses were designed to achieve stable phase profiles in the presence of up to 20% variation in the radiofrequency field. A gridding algorithm was used to resample the data onto a rectilinear grid before fast Fourier transforms. This method was demonstrated by in vivo imaging of brain metabolites at 1.5 T with 10 slices of 18 x 18 pixels each. Nominal voxel size was 1.1 cc, spectral bandwidth was 400 Hz, scan time was 18 min for the metabolite scan and 3 min for the reference scan.     

Depth-Averaged Model of Open-Channel Flows over an Arbitrary 3D Surface and Its Applications to Analysis of Water Surface Profile

A new set of depth-averaged equations is introduced to study the flow over an arbitrary three-dimensional (3D) surface. These equations are derived based on a generalized curvilinear coordinate system attached to the 3D bed surface, therefore it allows us to include the effect of centrifugal force due to the bottom curvature. These general equations make it possible to analyze flows over complex terrain without the limitation of mild slope assumption used in conventional depth-averaged models. This new model is then applied to calculate the water surface profiles of (1) flow over a cylindrical surface; (2) flow over a circular surface; and (3) flow with an air-core vortex at a vertical intake. A simple hydraulic experiment is conducted in the laboratory to observe the water surface profile of flow over a circular surface. The results obtained from the model are in good agreement with experimental measurements and calculation by an empirical formula. Consequently, it demonstrates the applicability of the model in cases of flow over a highly curved bottom.     

Three-dimensional coronary MR angiography using zonal echo planar imaging

Using an adapted two-dimensional spatially selective RF excitation scheme, a novel yet practical three-dimensional (3D) zonal echo-planar imaging technique for MR coronary angiography has been developed. The robustness of the technique compared with the two-dimensional (2D) segmented fast low angle shot (FLASH) method was evaluated using the right coronary artery images of 16 asymptomatic volunteers with a 0.5-T mobile scanner. Each 3D acquisition required multiple breath-holds, and real-time navigator echoes were used to ensure consistent breath-holding. Advantages of the technique include an improved signal-to-noise ratio, clearer depiction of tortuous coronary vessels due to decreased partial volume effects, and reduced motion blurring by the use of a short echo-planar readout.     

A digital model of trabecular bone

A 3D microCT dataset of bovine bone was used to create a digital 3D model simulating trabecular bone. The model serves a dual purpose: It allows for standard quantitative histomorphometric analysis and it approximates the reality e.g. of high resolution CT in vivo datasets of trabecular bone. Thus the model can potentially be used as a reference to develop 2D and 3D structural analysis algorithms applicable in vivo while it simultaneously allows verification of the results of these algorithms by standard histomorphometry. The model can be used as a standard to evaluate the impact of image processing techniques and of restrictions of imaging systems on the quantitative analysis of structural parameters describing a trabecular network. The model can be used for a comparison of 2D and 3D structural analysis methods and for an analysis of decreasing spatial resolution. The effects of segmentation and filtration can be studied separately and grayscale analysis is possible. As examples standard 2D histomorphometry and the analysis of topological parameters like node number and trabecular network length were applied to the model. The influence of spatial resolution was investigated by decreasing the spatial resolution of the digital model. The bone surface area determined by 3D surface triangulation was only 4% smaller than the surface area determined from the traditional 2D bone histomorphometric parameter bone surface/tissue volume (BS/TV) when 2D results were averaged over all slices of the 3D volume. However, BS/TV showed large (10%) variations among slices within the volume. Both histomorphometric and topological parameters were heavily influenced by spatial resolution and image segmentation. Our initial experience with the digital model indicates a need to investigate bone microstructure based on volume data or to average the 2D results of many slices.     

Morphologic image processing operators: reduction of partial volume effects for improved 3D visualization of CT data

AIM: The quality of segmentation and three-dimensional reconstruction of anatomical structures in tomographic slices is often impaired by disturbances due to partial volume effects (PVE). The potential for artefact reduction by use of the morphological image processing operators (MO) erosion and dilation is investigated. DESIGN: The CT examinations of 31 patients with pathological alterations in lung or brain were segmented using automatic region growing and the MO were applied in a different number of iterations. The processed regions were 3D-reconstructed (shaded surface display, MIP, volume rendering) and the occurrence of PVE-related artefacts using the signal-to-background ratio (SBR) prior to and after MO application was compared. RESULTS: For all patients under review, the artefacts caused by PVE were significantly reduced by erosion (lung: mean SBRpre = 1.67, SBRpost = 4.83; brain: SBRpre = 1.06, SBRpost = 1.29) even with only a small number of iterations. Region dilation was applied to integrate further structures (e.g. at tumor borders) into a configurable neighbourhood for segmentation and quantitative analysis. CONCLUSIONS: The MO represent an efficient approach for the reduction of PVE artefacts in 3D-CT reconstructions and allow optimized visualization of individual objects.     

Azoxymethane-induced colon tumors and aberrant crypt foci in mice of different genetic susceptibility

INTRODUCTION: Virtual endoscopy is a new 3D technique which permits to depict the inner surface of anatomic cavities. We report our experience in the study of the nasal cavity and paranasal sinuses. MATERIAL AND METHODS: CT examinations of the maxillofacial region were obtained using 1.5 or 3.0 mm slices, 1.5 or 3.0 mm table feed, 120 kV, 140 mA, 2 s scan time, standard and high resolution algorithms for bony structures and a field of view of 14-16 cm. The images were transferred on a workstation and processed with the Navigator virtual endoscopy software (General Electric). A threshold value ranging -300 to -550 UH was applied. RESULTS: We report a series of virtual images of the nasal cavity which includes rhinopharynx vault, torius tubarius, choanae, turbinates, tubal orifice and osteomeatal complex. Moreover, images of frontal sinus and tear duct, of polyps and fracture of maxillary sinus are also presented. 3D virtual endoscopy provides a clear visualization of the anatomic structures of the nasal cavity and sinuses. The images are similar to those of conventional endoscopy. Moreover virtual endoscopy visualizes the paranasal sinuses, which are not accessible at conventional endoscopy. The main limitations of this new technique are the arbitrary choice of the threshold value and the homogenization of different tissue densities, which reduces the contrast resolution. CONCLUSIONS: Virtual endoscopy can presently be considered a complementary technique of the standard axial and coronal CT examination. It provides an effective demonstration of the anatomy of these structures and shows areas which are difficult to visualize with conventional endoscopy. This technique could be of help in didactical activity; its clinical application has to be verified.     

Intensity modulation with electrons: calculations, measurements and clinical applications

Intensity modulation of electron beams is one step towards truly conformal therapy. This can be realized with the MM50 racetrack microtron that utilizes a scanning beam technique. By adjusting the scan pattern it is possible to obtain arbitrary fluence distributions. Since the monitor chambers in the treatment head are segmented in both x- and y-directions it is possible to verify the fluence distribution to the patient at any time during the treatment. Intensity modulated electron beams have been measured with film and a plane parallel chamber and compared with calculations. The calculations were based on a pencil beam method. An intensity distribution at the multileaf collimator (MLC) level was calculated by superposition of measured pencil beams over scan patterns. By convolving this distribution with a Gaussian pencil beam, which has propagated from the MLC to the isocentre, a fluence distribution at isocentre level was obtained. The agreement between calculations and measurements was within 2% in dose or 1 mm in distance in the penumbra zones. A standard set of intensity modulated electron beams has been developed. These beams have been implemented in a treatment planning system and are used for manual optimization. A clinical example (prostate) of such an application is presented and compared with a standard irradiation technique.     

Computer-assisted three-dimensional reconstruction of the corticospinal system as a reference for CT and MRI

We present a three-dimensional (3D) anatomical computer-graphics model of the corticospinal system acquired from equidistant serial anatomical slices of six intracranially-fixed human brains. This model is part of a neuroanatomical reference system (NeuRef) which enables 3D visualization of the brain and shows the relationship of its components such as anatomical structures, functional fibre tracts and arteries. Sections through the models can be matched with corresponding CT or MR images. This allows the probable localisation of corticospinal fibres on CT or MRI.     

Unilateral spatial neglect in the late stage of Alzheimer''s disease

This paper presents a method designed to register preoperative computed tomography (CT) images to vertebral surface points acquired intraoperatively from ultrasound (US) images or via a tracked probe. It also presents a comparison of the registration accuracy achievable with surface points acquired from the entire posterior surface of the vertebra to the accuracy achievable with points acquired only from the spinous process and central laminar regions. Using a marker-based method as a reference, this work shows that submillimetric registration accuracy can be obtained even when a small portion of the posterior vertebral surface is used for registration. It also shows that when selected surface patches are used, CT slice thickness is not a critical parameter in the registration process. Furthermore, the paper includes qualitative results of registering vertebral surface points in US images to multiple CT slices. The method has been tested with US points and physical points on a plastic spine phantom and with simulated data on a patient CT scan.     

Dynamic Barite Sag in Oil Based Drilling Fluids

Abstract

A rheological technique is described which utilises flat parallel plate geometry. The technique is sensitive enough to pick up the effect of syneresis and resulting movement of denser materials from the surface layers into the bulk of the drilling mud samples over time. Examples of down flow curves are given which demonstrate that dynamic sag while in flow is represented by a dip in this curve depending flow on particular samples, when expressed as log/log plots of viscosity vsshear rate. The up curves illustrate the degree of syneresis and static sag in the samples. Log/log plots of viscosity vs shear rate for both up and down flow curves exhibit one or two crossover points. The lower crossover points can be plotted on a 3D graph for easy visualisation of the sagging tendencies of various oil mud samples.     

基于扫描线的带钢低对比度缺陷图像分割技术

在带钢表面质量视觉检测过程中,一些低对比度缺陷图像应用梯度算子进行分割时,往往找不到图像中的缺陷,从而出现漏检的情况.为了改善带钢图像中低对比度缺陷的分割效果,提出了一种基于扫描线的阈值图像分割算法,该算法可以有效实现平整花、点划伤等低对比度缺陷的分割,同时也能较好地实现对其他缺陷的分割.     

Validation of an MRI/PET landmark registration method using 3D simulated PET images and point simulations

Three-dimensional (3D) simulated PET images generated from MRI were used to validate a multimodality registration technique based on the identification of internal anatomical landmarks. In addition, point-based simulations were compared with registration datasets acquired over 3 yr of routine use of the technique. Registration errors were found to range from 1.0 mm at the brain centre to 2.8 mm in each dimension at the brain surface.