Unsupervised segmentation and quantification of anatomical knee features: data from the Osteoarthritis Initiative

This paper presents a fully automated method for segmenting articular knee cartilage and bone from in vivo 3-D dual echo steady state images. The magnetic resonance imaging (MRI) datasets were obtained from the Osteoarthritis Initiative (OAI) pilot study and include longitudinal images from controls and subjects with knee osteoarthritis (OA) scanned twice at each visit (baseline, 24 month). Initially, human experts segmented six MRI series. Five of the six resultant sets served as reference atlases for a multiatlas segmentation algorithm. The methodology created precise knee segmentations that were used to extract articular cartilage volume, surface area, and thickness as well as subchondral bone plate curvature. Comparison to manual segmentation showed Dice similarity coefficient (DSC) of 0.88 and 0.84 for the femoral and tibial cartilage. In OA subjects, thickness measurements showed test-retest precision ranging from 0.014 mm (0.6%) at the femur to 0.038 mm (1.6%) at the femoral trochlea. In the same population, the curvature test-retest precision ranged from 0.0005 mm(-1) (3.6%) at the femur to 0.0026 mm(-1) (11.7%) at the medial tibia. Thickness longitudinal changes showed OA Pearson correlation coefficient of 0.94 for the femur. In conclusion, the fully automated segmentation methodology produces reproducible cartilage volume, thickness, and shape measurements valuable for the study of OA progression.     

A technique for regional analysis of femorotibial cartilage thickness based on quantitative magnetic resonance imaging

The objective of this work was to develop a methodology for measuring cartilage thickness in anatomically based subregions in the tibial and in the central weight-bearing femoral cartilage from magnetic resonance (MR) images. The tibial plateau was divided into a central area of the total subchondral bone area (tAB), and anterior, posterior, internal, and external subregions surrounding it. In the weight-bearing femoral condyles, central, internal, and external subregions were determined. The Euclidean distance between the tAB and cartilage surface was used for determining cartilage thickness. The reproducibility of the method was evaluated on test-retest data sets of 12 participants (six healthy, six with osteoarthritis). The subregion size was varied systematically to study the influence on the reproducibility. The size of the subregions was highly consistent under conditions of repositioning (standard deviation 0.0%-0.3%). The precision errors for regional mean cartilage thickness measurements ranged from 19 mum (1.5%) to 84 mum (4.7%). The computation of regional cartilage thickness values from segmented MR images is shown to be highly reproducible and robust under conditions of joint repositioning. In longitudinal studies, this technique may substantially enhance the ability of quantitative MRI to monitor structural changes in osteoarthritis at narrow time intervals.     

Monitoring osteoarthritis in the rat model using optical coherence tomography

OBJECTIVE: A need exists for an animal model to assess therapeutics for osteoarthritis (OA) without sacrificing the animal. Our goal is to assess the progression of experimentally induced osteoarthritis in the rat knee joint by monitoring articular cartilage thickness, surface abnormalities, and collagen organization using a new technology known as optical coherence tomography (OCT). DESIGN: OA was generated in Wistar Hanover rats via injection of sodium iodoacetate into the left articular joint of the knee while normal saline was injected as a control in the contralateral right knee. Rats were sacrificed at 1-, 2-, 3-, 4-, and 8-week intervals and the knee joints were subsequently harvested and imaged using normal and polarization sensitive OCT (PS-OCT). Treated knees were compared to normal counterparts in the contralateral leg. Following imaging, knees underwent both routine histological processing and picrosirus staining for organized collagen. RESULTS: OCT images indicate that injection of sodium iodoacetate resulted in a progressive decrease in cartilage thickness and loss of the bone-cartilage interface which correlated with histology. In addition, PS-OCT was able to detect collagen disorganization, an early indicator of OA. CONCLUSIONS: The use of OCT in combination with the induction of OA in rats is a promising new animal model for assessing articular changes with the goal of monitoring therapeutics longitudinally. Future work will extend the model to in vivo assessments.     

Limits on the accuracy of 3-D thickness measurement in magnetic resonance images--effects of voxel anisotropy

Measuring the thickness of sheet-like thin anatomical structures, such as articular cartilage and brain cortex, in three-dimensional (3-D) magnetic resonance (MR) images is an important diagnostic procedure. This paper investigates the fundamental limits on the accuracy of thickness determination in MR images. We defined thickness here as the distance between the two sides of boundaries measured at the subvoxel resolution, which are the zero-crossings of the second directional derivatives combined with Gaussian blurring along the normal directions of the sheet surface. Based on MR imaging and computer postprocessing parameters, characteristics for the accuracy of thickness determination were derived by a theoretical simulation. We especially focused on the effects of voxel anisotropy in MR imaging with variable orientation of sheet-like structure. Improved and stable accuracy features were observed when the standard deviation of Gaussian blurring combined with thickness determination processes was around square root of 2/2 times as large as the pixel size. The relation between voxel anisotropy in MR imaging and the range of sheet normal orientation within which acceptable accuracy is attainable was also clarified, based on the dependences of voxel anisotropy and the sheet normal orientation obtained by numerical simulations. Finally, in vitro experiments were conducted using an acrylic plate phantom and a resected femoral head to validate the results of theoretical simulation. The simulated thickness was demonstrated to be well-correlated with the actual in vitro thickness.     

Rapid Dynamic Image Registration of the Beating Heart for Diagnosis and Surgical Navigation

Dynamic cardiac magnetic resonance imaging (MR) and computed tomography (CT) provide cardiologists and cardiac surgeons with high-quality 4-D images for diagnosis and therapy, yet the effective use of these high-quality anatomical models remains a challenge. Ultrasound (US) is a flexible imaging tool, but the US images produced are often difficult to interpret unless they are placed within their proper 3-D anatomical context. The ability to correlate real-time 3-D US volumes (RT3D US) with dynamic MR/CT images would offer a significant contribution to improve the quality of cardiac procedures. In this paper, we present a rapid two-step method for registering RT3D US to high-quality dynamic 3-D MR/CT images of the beating heart. This technique overcomes some major limitations of image registration (such as the correct registration result not necessarily occurring at the maximum of the mutual information (MI) metric) using the MI metric. We demonstrate the effectiveness of our method in a dynamic heart phantom (DHP) study and a human subject study. The achieved mean target registration error of CT+US images in the phantom study is 2.59 mm. Validation using human MR/US volumes shows a target registration error of 1.76 mm. We anticipate that this technique will substantially improve the quality of cardiac diagnosis and therapies.      

Web-Based Multilayer Viewing Interface for Knee Cartilage

Many adults suffer from osteoarthritis (OA) with the majority of people over 65 showing radiographic evidence of the disease. To carry out effective diagnosis and treatment, it is necessary to understand the progression of cartilage loss and study the effectiveness of therapeutic interventions. Hence, it is important to have accurate, fast diagnosis of the disease. In this paper, we describe a Web-based user interface that enables the direct viewing of 2-D and 3-D image data from the visceral and tissue levels of the biological continuum (i.e., the continuum comprising systems, viscera, tissue, cells, proteins, and genes)–-while preserving geometric integrity. This is achieved despite the fact that the data are from different modalities (i.e., magnetic resonance (MR) and light microscopy). The user interface was tested using image data acquired from a study of articular cartilage thickness in the porcine knee. The interface allows the clinician to view both MR and light microscopy images in an integrated manner—with the information linked geometrically.      

Automatic construction of 3-D statistical deformation models of the brain using nonrigid registration

In this paper, we show how the concept of statistical deformation models (SDMs) can be used for the construction of average models of the anatomy and their variability. SDMs are built by performing a statistical analysis of the deformations required to map anatomical features in one subject into the corresponding features in another subject. The concept of SDMs is similar to statistical shape models (SSMs) which capture statistical information about shapes across a population, but offers several advantages over SSMs. First, SDMs can be constructed directly from images such as three-dimensional (3-D) magnetic resonance (MR) or computer tomography volumes without the need for segmentation which is usually a prerequisite for the construction of SSMs. Instead, a nonrigid registration algorithm based on free-form deformations and normalized mutual information is used to compute the deformations required to establish dense correspondences between the reference subject and the subjects in the population class under investigation. Second, SDMs allow the construction of an atlas of the average anatomy as well as its variability across a population of subjects. Finally, SDMs take the 3-D nature of the underlying anatomy into account by analysing dense 3-D deformation fields rather than only information about the surface shape of anatomical structures. We show results for the construction of anatomical models of the brain from the MR images of 25 different subjects. The correspondences obtained by the nonrigid registration are evaluated using anatomical landmark locations and show an average error of 1.40 mm at these anatomical landmark positions. We also demonstrate that SDMs can be constructed so as to minimize the bias toward the chosen reference subject.     

Volumetric texture segmentation by discriminant feature selection and multiresolution classification

In this paper, a multiresolution volumetric texture segmentation (M-VTS) algorithm is presented. The method extracts textural measurements from the Fourier domain of the data via subband filtering using an orientation pyramid (Wilson and Spann, 1988). A novel Bhattacharyya space, based on the Bhattacharyya distance, is proposed for selecting the most discriminant measurements and producing a compact feature space. An oct tree is built of the multivariate features space and a chosen level at a lower spatial resolution is first classified. The classified voxel labels are then projected to lower levels of the tree where a boundary refinement procedure is performed with a three-dimensional (3-D) equivalent of butterfly filters. The algorithm was tested with 3-D artificial data and three magnetic resonance imaging sets of human knees with encouraging results. The regions segmented from the knees correspond to anatomical structures that can be used as a starting point for other measurements such as cartilage extraction.     

Semantic image retrieval in magnetic resonance brain volumes

Practitioners in the area of neurology often need to retrieve multimodal magnetic resonance (MR) images of the brain to study disease progression and to correlate observations across multiple subjects. In this paper, a novel technique for retrieving 2-D MR images (slices) in 3-D brain volumes is proposed. Given a 2-D MR query slice, the technique identifies the 3-D volume among multiple subjects in the database, associates the query slice with a specific region of the brain, and retrieves the matching slice within this region in the identified volumes. The proposed technique is capable of retrieving an image in multimodal and noisy scenarios. In this study, support vector machines (SVM) are used for identifying 3-D MR volume and for performing semantic classification of the human brain into various semantic regions. In order to achieve reliable image retrieval performance in the presence of misalignments, an image registration-based retrieval framework is developed. The proposed retrieval technique is tested on various modalities. The test results reveal superior robustness performance with respect to accuracy, speed, and multimodality.     

Volume registration using needle paths and point landmarks for evaluation of interventional MRI treatments

We created a method for three-dimensional (3-D) registration of medical images (e.g., magnetic resonance imaging (MRI) or computed tomography) to images of physical tissue sections or to other medical images and evaluated its accuracy. Our method proved valuable for evaluation of animal model experiments on interventional-MRI guided thermal ablation and on a new localized drug delivery system. The method computes an optimum set of rigid body registration parameters by minimization of the Euclidean distances between automatically chosen correspondence points, along manually selected fiducial needle paths, and optional point landmarks, using the iterative closest point algorithm. For numerically simulated experiments, using two needle paths over a range of needle orientations, mean voxel displacement errors depended mostly on needle localization error when the angle between needles was at least 20 degrees. For parameters typical of our in vivo experiments, the mean voxel displacement error was < 0.35 mm. In addition, we determined that the distance objective function was a useful diagnostic for predicting registration quality. To evaluate the registration quality of physical specimens, we computed the misregistration for a needle not considered during the optimization procedure. We registered an ex vivo sheep brain MR volume with another MR volume and tissue section photographs, using various combinations of needle and point landmarks. Mean registration error was always < or = 0.54 mm for MR-to-MR registrations and < or = 0.52 mm for MR to tissue section registrations. We also applied the method to correlate MR volumes of radio-frequency induced thermal ablation lesions with actual tissue destruction. In this case, in vivo rabbit thigh volumes were registered to photographs of ex vivo tissue sections using two needle paths. Mean registration errors were between 0.7 and 1.36 mm over all rabbits, the largest error less than two MR voxel widths. We conclude that our method provides sufficient spatial correspondence to facilitate comparison of 3-D image data with data from gross pathology tissue sections and histology.     

Medical image compression by using three-dimensional wavelet transformation

This paper proposes a three-dimensional (3-D) medical image compression method for computed tomography (CT) and magnetic resonance (MR) that uses a separable nonuniform 3-D wavelet transform. The separable wavelet transform employs one filter bank within two-dimensional (2-D) slices and then a second filter bank on the slice direction. CT and MR image sets normally have different resolutions within a slice and between slices. The pixel distances within a slice are normally less than 1 mm and the distance between slices can vary from 1 mm to 10 mm. To find the best filter bank in the slice direction, the authors use the various filter banks in the slice direction and compare the compression results. The results from the 12 selected MR and CT image sets at various slice thickness show that the Haar transform in the slice direction gives the optimum performance for most image sets, except for a CT image set which has 1 mm slice distance. Compared with 2-D wavelet compression, compression ratios of the 3-D method are about 70% higher for CT and 35% higher for MR image sets at a peak signal to noise ratio (PSNR) of 50 dB, In general, the smaller the slice distance, the better the 3-D compression performance.     

Registration and tracking to integrate X-ray and MR images in an XMR facility

We describe a registration and tracking technique to integrate cardiac X-ray images and cardiac magnetic resonance (MR) images acquired from a combined X-ray and MR interventional suite (XMR). Optical tracking is used to determine the transformation matrices relating MR image coordinates and X-ray image coordinates. Calibration of X-ray projection geometry and tracking of the X-ray C-arm and table enable three-dimensional (3-D) reconstruction of vessel centerlines and catheters from bi-plane X-ray views. We can, therefore, combine single X-ray projection images with registered projection MR images from a volume acquisition, and we can also display 3-D reconstructions of catheters within a 3-D or multi-slice MR volume. Registration errors were assessed using phantom experiments. Errors in the combined projection images (two-dimensional target registration error--TRE) were found to be 2.4 to 4.2 mm, and the errors in the integrated volume representation (3-D TRE) were found to be 4.6 to 5.1 mm. These errors are clinically acceptable for alignment of images of the great vessels and the chambers of the heart. Results are shown for two patients. The first involves overlay of a catheter used for invasive pressure measurements on an MR volume that provides anatomical context. The second involves overlay of invasive electrode catheters (including a basket catheter) on a tagged MR volume in order to relate electrophysiology to myocardial motion in a patient with an arrhythmia. Visual assessment of these results suggests the errors were of a similar magnitude to those obtained in the phantom measurements.     

Four-chamber heart modeling and automatic segmentation for 3-D cardiac CT volumes using marginal space learning and steerable features

We propose an automatic four-chamber heart segmentation system for the quantitative functional analysis of the heart from cardiac computed tomography (CT) volumes. Two topics are discussed: heart modeling and automatic model fitting to an unseen volume. Heart modeling is a nontrivial task since the heart is a complex nonrigid organ. The model must be anatomically accurate, allow manual editing, and provide sufficient information to guide automatic detection and segmentation. Unlike previous work, we explicitly represent important landmarks (such as the valves and the ventricular septum cusps) among the control points of the model. The control points can be detected reliably to guide the automatic model fitting process. Using this model, we develop an efficient and robust approach for automatic heart chamber segmentation in 3-D CT volumes. We formulate the segmentation as a two-step learning problem: anatomical structure localization and boundary delineation. In both steps, we exploit the recent advances in learning discriminative models. A novel algorithm, marginal space learning (MSL), is introduced to solve the 9-D similarity transformation search problem for localizing the heart chambers. After determining the pose of the heart chambers, we estimate the 3-D shape through learning-based boundary delineation. The proposed method has been extensively tested on the largest dataset (with 323 volumes from 137 patients) ever reported in the literature. To the best of our knowledge, our system is the fastest with a speed of 4.0 s per volume (on a dual-core 3.2-GHz processor) for the automatic segmentation of all four chambers.      

The boundary shift integral: an accurate and robust measure ofcerebral volume changes from registered repeat MRI

We propose the boundary shift integral (BSI) as a measure of cerebral volume changes derived from registered repeat three-dimensional (3-D) magnetic resonance (MR) [3D MR] scans. The BSI determines the total volume through which the boundaries of a given cerebral structure have moved and, hence, the volume change, directly from voxel intensities. We found brain and ventricular BSI's correlated tightly (r=1.000 and r=0.999) with simulated volumes of change. Applied to 21 control scan pairs and 11 scan pairs from Alzheimer's disease (AD) patients (mean interval 386 days) the BSI yielded mean brain volume loss of 1.8 cc (controls) and 34.7 cc (AD); the control group was tightly bunched (SD=3.8 cc) and there was wide group separation, the group means differing by 8.7 control group standard deviations (SDs). A measure based on the same segmentation used by the BSI yielded similar group means, but wide spread in the control group (SD=13.4 cc) and group overlap, the group means differing by 2.8 control group SDs. The BSI yielded mean ventricular volume losses of 0.4 cc (controls) and 10.1 cc (AD). Good linear correlation (r=0.997) was obtained between the ventricular BSI and the difference in their segmented volumes. We conclude the BSI is an accurate and robust measure of regional and global cerebral volume changes     

Cancellation of image crosstalk in time-sequential displays ofstereoscopic video

Stereoscopic visualization systems based on liquid crystal shutter (LCS) eyewear and cathode-ray tube (CRT) displays provide today the best overall quality of three-dimensional (3-D) images and therefore have a dominant position in commercial as well as professional markets. Due to the CRT and LCS characteristics, however, such systems suffer from perceptual crosstalk ("shadows") at object boundaries that can reduce, and at times inhibit, the ability to perceive depth. In this paper, we propose a method to reduce such crosstalk. We present a simple model for intensity leak, we assess model parameters for a time-sequential LCS/CRT system and we propose a computationally efficient algorithm to eliminate the crosstalk. Since the full crosstalk elimination implies an unacceptable image degradation (reduction of contrast), we study the tradeoff between crosstalk elimination and image contrast. We describe experiments on synthetic and natural stereoscopic images and we discuss informal subjective viewing of processed images. Overall, the viewer response has been very positive; 3-D perception of many objects became either much easier or even effortless. Since the proposed algorithm can be easily implemented in real time (only linear scaling and table look-up are needed), we believe that it can be successfully used today in various stereoscopic applications suffering from image crosstalk. This is particularly true for PC-based 3-D viewing where the algorithm can be executed by the CPU or by an advanced graphics board.     

Volume delineation by fusion of fuzzy sets obtained frommultiplanar tomographic images

Techniques of three-dimensional (3-D) volume delineation from tomographic medical imaging are usually based on 2-D contour definition. For a given structure, several different contours can be obtained depending on the segmentation method used or the user's choice. The goal of this work is to develop a new method that reduces the inaccuracies generally observed. A minimum volume that is certain to be included in the volume concerned (membership degree mu = 1), and a maximum volume outside which no part of the volume is expected to be found (membership degree mu = 0), are defined semi-automatically. The intermediate fuzziness region (0 < mu < 1) is processed using the theory of possibility. The resulting fuzzy volume is obtained after data fusion from multiplanar slices. The influence of the contrast-to-noise ratio was tested on simulated images. The influence of slice thickness as well as the accuracy of the method were studied on phantoms. The absolute volume error was less than 2% for phantom volumes of 2-8 cm3, whereas the values obtained with conventional methods were much larger than the actual volumes. Clinical experiments were conducted, and the fuzzy logic method gave a volume lower than that obtained with the conventional method. Our fuzzy logic method allows volumes to be determined with better accuracy and reproducibility.     

Left ventricular deformation recovery from cine MRI using an incompressible model

This paper presents a method for 3-D deformation recovery of the left ventricular (LV) wall from anatomical cine magnetic resonance imaging (MRI). The method is based on a deformable model that is incompressible, a desired property since the myocardium has been shown to be nearly incompressible. The LV wall needs to be segmented in an initial frame after which the method automatically determines the deformation everywhere in the LV wall throughout the cardiac cycle. Two studies were conducted to validate the method. In the first study, the deformation recovered from a 3-D anatomical cine MRI of a healthy volunteer was compared against the manual segmentation of the LV wall and against the corresponding 3-D tagged cine MRI. The average volume agreement between the model and the manual segmentation had a false positive rate of 3%, false negative rate of 3%, and true positive rate of 93%. The average distance between the model and manually determined intersections of perpendicular tag planes was 1.6 mm (1.1 pixel). Another set of 3-D anatomical and tagged MRI scans was taken of the same volunteer four months later. The method was applied to the second set and the recovered deformation was very similar to the one obtained from the first set. In the second study, the method was applied to 3-D anatomical cine MRI scans of three patients with ventricular dyssynchrony and three age-matched healthy volunteers. The LV wall deformations recovered for the three normals agreed well and the recovered strains were similar to those reported by other researchers for normal subjects. Strains and displacements of the three patients were clearly smaller than those of the three normals indicating reduced cardiac function. The deformation recovered for the three normals and the three patients was validated against manual segmentation and corresponding tag cine MRI scans and the agreement was similar to that of the first validation study.     

Robust white matter lesion segmentation in FLAIR MRI

This paper discusses a white matter lesion (WML) segmentation scheme for fluid attenuation inversion recovery (FLAIR) MRI. The method computes the volume of lesions with subvoxel precision by accounting for the partial volume averaging (PVA) artifact. As WMLs are related to stroke and carotid disease, accurate volume measurements are most important. Manual volume computation is laborious, subjective, time consuming, and error prone. Automated methods are a nice alternative since they quantify WML volumes in an objective, efficient, and reliable manner. PVA is initially modeled with a localized edge strength measure since PVA resides in the boundaries between tissues. This map is computed in 3-D and is transformed to a global representation to increase robustness to noise. Significant edges correspond to PVA voxels, which are used to find the PVA fraction α (amount of each tissue present in mixture voxels). Results on simulated and real FLAIR images show high WML segmentation performance compared to ground truth (98.9% and 83% overlap, respectively), which outperforms other methods. Lesion load studies are included that automatically analyze WML volumes for each brain hemisphere separately. This technique does not require any distributional assumptions/parameters or training samples and is applied on a single MR modality, which is a major advantage compared to the traditional methods.     

Bayesian reconstruction of functional images using anatomical information as priors

Proposes a Bayesian method whereby maximum a posteriori (MAP) estimates of functional (PET and SPECT) images may be reconstructed with the aid of prior information derived from registered anatomical MR images of the same slice. The prior information consists of significant anatomical boundaries that are likely to correspond to discontinuities in an otherwise spatially smooth radionuclide distribution. The authors' algorithm, like others proposed recently, seeks smooth solutions with occasional discontinuities; the contribution here is the inclusion of a coupling term that influences the creation of discontinuities in the vicinity of the significant anatomical boundaries. Simulations on anatomically derived mathematical phantoms are presented. Although computationally intense in its current implication, the reconstructions are improved (ROI-RMS error) relative to filtered backprojection and EM-ML reconstructions. The simulations show that the inclusion of position-dependent anatomical prior Information leads to further improvement relative to Bayesian reconstructions without the anatomical prior. The algorithm exhibits a certain degree of robustness with respect to errors in the location of anatomical boundaries.     

A new validation method for X-ray mammogram registration algorithms using a projection model of breast X-ray compression

Establishing spatial correspondence between features visible in X-ray mammograms obtained at different times has great potential to aid assessment and quantitation of change in the breast indicative of malignancy. The literature contains numerous nonrigid registration algorithms developed for this purpose, but existing approaches are flawed by the assumption of inappropriate 2-D transformation models and quantitative estimation of registration accuracy is limited. In this paper, we describe a novel validation method which simulates plausible mammographic compressions of the breast using a magnetic resonance imaging (MRI) derived finite element model. By projecting the resulting known 3-D displacements into 2-D and generating pseudo-mammograms from these same compressed magnetic resonance (MR) volumes, we can generate convincing images with known 2-D displacements with which to validate a registration algorithm. We illustrate this approach by computing the accuracy for two conventional nonrigid 2-D registration algorithms applied to mammographic test images generated from three patient MR datasets. We show that the accuracy of these algorithms is close to the best achievable using a 2-D one-to-one correspondence model but that new algorithms incorporating more representative transformation models are required to achieve sufficiently accurate registrations for this application.