Insight into Images

Noisy multimodal brain image registration using markov random field model

Integrating segmentation methods from the Insight Toolkit into a visualization application

Habib Zaidi, PhD This is an exciting time for molecular imaging because molecular medicine is expected to lead to a revolutionary paradigm shift in health care. While the nuclear medicine community is witnessing a revolution in the practice of this specialty with the introduction of dual-modality positron emission tomography (PET) and computed tomography (CT), many groups in academic and corporate settings are focusing their efforts on the development of promising multimodality imaging technologies that will combine magnetic resonance (MR) imaging and PET components in a single gantry. One such preclinical system is described by Judenhofer et al (1) in this issue of Radiology, and different design trends have been reported in the literature (2–8). Similar design concepts are also being investigated for combined single photon emission CT (SPECT) and MR imaging instrumentation (9). The recent interest in simultaneous PET/MR imaging is not the consequence of controversies surrounding the role and clinical benefits of PET/CT (10,11) but will likely be the subject of similar critique and debates. The development of PET/MR imaging has been motivated by various factors and has several important incentives (12). First, MR imaging is used to obtain anatomic and structural images with submillimeter spatial resolution that offers better soft-tissue contrast sensitivity than that offered by CT. MR imaging has excellent contrast between white and gray matter and allows for functional imaging in brain studies, and, more importantly, it can be used to assess flow, diffusion, perfusion, and cardiac motion in one examination (13). In addition, MR imaging can be combined with MR spectroscopy to measure the regional biochemical content and to assess the metabolic status or the presence of neoplasia and other diseases in specific tissue areas. Finally, MR imaging does not involve the use of ionizing radiation; thus, it can be used without restrictions in serial studies, for pediatric studies, and in many other situations where radiation exposure is a concern. PET imaging is used to record the regional distribution of radiolabeled tracers; however, unlike MR spectroscopy, it cannot be used to distinguish the specific molecular species to which the radionuclide is attached, and, unlike MR imaging, it provides little anatomic information. While many technical problems have recently been solved, there are still several important challenges to the implementation and operation of a PET/MR imaging system that must be overcome. In comparison with x-ray CT, MR imaging typically is more costly, involves longer examination times, and produces anatomic images from which it is more difficult or at least not as straightforward to derive attenuation maps for photon correction of the emission data (14). The latter issue was not described by Judenhofer et al (1), as it has received only limited attention in the scientific literature and few investigators have addressed the problem of the use of segmented MR data to construct an attenuation map for attenuation correction purposes with PET (15). However, interest in PET/MR imaging has been the driving force behind many worthwhile research efforts recently undertaken by different research groups. The major difficulty lies in the fact that the MR signal or tissue intensity level is not directly related to electronic attenuation, which renders conversion of MR images to attenuation maps less obvious when compared with CT. It is worth emphasizing that the optimal transmission scanning technique for PET/CT (ie, CT versus radionuclide sources) is still an open issue that remains to be addressed (16). This is not an issue for PET/MR imaging because of the limited space available, and thus, placement of external Published online 10.1148/radiol.2443070092

Post-implant dosimetric analysis for permanent implant of the prostate benefits from the use of a computed tomography (CT) dataset for optimal identification of the radioactive source (seed) positions and a magnetic resonance (MR) dataset for optimal description of the target and normal tissue volumes. The CT/MR registration process should be fast and sufficiently accurate to yield a reliable dosimetric analysis. Since critical normal tissues typically reside in dose gradient regions, small shifts in the dose distribution could impact the prediction of complication or complication severity. Standard procedures include the use of the seed distribution as fiducial markers (seed match), a time consuming process that relies on the proper identification of signals due to the same seed on both datasets. Mutual information (MI) is more efficient because it uses image data requiring minimal preparation effort. A comparison of MI registration and seed-match registration was performed for twelve patients. MI was applied to a volume limited to the prostate and surrounding structures, excluding most of the pelvic bone structures (margins around the prostate gland were approximately 2 cm right-left, approximately 1 cm anterior-posterior, and approximately 2 cm superior-inferior). Seeds were identified on a 2 mm slice CT dataset using an automatic seed identification procedure on reconstructed three-dimensional data. Seed positions on the 3 mm slice thickness T2 MR data set were identified using a point-and-click method on each image. Seed images were identified on more than one MR slice, and the results used to determine average seed coordinates for MR images and matched seed pairs between CT and MR images. On average, 42% (19%-64%) of the seeds (19-54 seeds) were identified and matched to their CT counterparts. A least-squares method applied to the CT and MR seed coordinates was used to produce the optimum seed-match registration. MI registration and seed match registration angle differences averaged 0.5 degrees, which was not significantly different from zero. Translation differences averaged 0.6 (1.2 standard deviation) mm right-left, -0.5(1.5) mm posterior-anterior, and -1.2(2.0) mm inferior-superior. Registration error estimates were approximately 2 mm for both the MI and seed-match methods. The observed standard deviations in the offset values were consistent with propagation of error. Registration methods as applied here using mutual information and seed matching are consistent, except for a small systematic difference in the inferior-superior axis for a minority of cases (approximately 15%). Cases registered with mutual information and with bony anatomy misregistration of greater than approximately 5 mm should be evaluated for rescan or seed-match registration. The improvement in efficiency of use for the MI registration method is substantial, approximately 30 min compared to several hours using seed match registration.

This paper proposes a new methodology for image segmentation based on the integration of deformable and Markov Random Field models. Our method makes use of Markov Random Field theory to build a Gibbs Prior model of medical images with arbitrary initial parameters to estimate the boundary of organs with low signal to noise ratio (SNR). Then we use a deformable model to fit the estimated boundary. The result of the deformable model fit is used to update the Gibbs prior model parameters, such as the gradient threshold of a boundary. Based on the updated parameters we restart the Gibbs prior models. By iteratively integrating these processes we achieve an automated segmentation of the initial images. By careful choice of the method used for the Gibbs prior models, and based on the above method of integration with deformable model our segmentation solution runs in close to real time. Results of the method are presented for several examples, including some MRI images with significant amount of noise.

X-ray computed tomography (CT) imaging plays an essential role in disease diagnosis due to its noninvasive, painless mode and superior penetration depth. However, the resolution of the soft tissue ...

\nNowadays, heart diseases are the leading cause of death. Left ventricle segmentation of a human heart in magnetic resonance images (MRI) is a crucial step in both cardiac diseases diagnostics and heart internal structure reconstruction. It allows estimating such important parameters as ejection faction, left ventricle myocardium mass, stroke volume, etc. In addition, left ventricle segmentation helps to construct the personalized heart computational models in order to conduct the numerical simulations. At present, the fully automated cardiac segmentation methods still do not meet the accuracy requirements. We present an overview of left ventricle segmentation algorithms on short-axis MRI. A wide variety of completely different approaches are used for cardiac segmentation, including machine learning, graph-based methods, deformable models, and low-level heuristics. The current state-of-the-art technique is a combination of deformable models with advanced machine learning methods, such as deep learning or Markov random fields. We expect that approaches based on deep belief networks are the most promising ones because the main training process of networks with this architecture can be performed on the unlabelled data. In order to improve the quality of left ventricle segmentation algorithms, we need more datasets with labelled cardiac MRI data in open access.\n

In this study, a Markov Random Field (MRF) approach is used to locate source boundary positions which are difficult to identify from Bouguer gravity and magnetic maps. As a generalized form of Markov Chains, the MRF approach is an unsupervised statistical model based algorithm and is applied to the analysis of images, particularly in the detection of visual patterns or textures. Here, we present a dynamic programming based on the MRF approach for boundary detection of noisy and super-positioned potential anomalies, which are produced by various geological structures. In the MRF method, gravity and magnetic maps are considered as two-dimensional (2-D) images with a matrix composed of N1 × N2 pixels. Each pixel value of the matrix is optimized in real time with no a priori processing by using two parameter sets; average steering vector (θ) and quantization level (M). They carry information about the correlation of neighboring pixels and the locality of their connections. We have chosen MRF as a processing approach for geophysical data since it is an unsupervised, efficient model for image enhancement, border detection and separation of 2-D potential anomalies. The main benefit of MRF is that an average steering vector and a quantization level are enough in evaluation of the potential anomaly maps. We have compared the MRF method to noise implemented synthetic potential field anomalies. After satisfactory results were found, the method has been applied to gravity and magnetic anomaly maps of Gelibolu Peninsula in Western Turkey. Here, we have observed Anafartalar thrust fault and another parallel fault northwest of Anafartalar thrust fault. We have modeled a geological structure including a lateral fault, which results in a higher susceptibility and anomaly amplitude increment. We have shown that the MRF method is effective to detect the broad-scale geological structures in the Gelibolu Peninsula, and thus to delineate the complex tectonic structure of Gelibolu Peninsula.

A number of surgical procedures are planned and executed based on medical images. Typically, x-ray computed tomography (CT) and magnetic resonance (MR) images are acquired preoperatively for diagnosis and surgical planning. In the operating room, execution of the surgical plan becomes feasible due to registration between preoperative images and surgical space where patient anatomy lies. In this paper, we present a new automatic algorithm where we use ultrasound (US) 2D B-mode images to register the preoperative MR image coordinate system with the surgical space which in our experiments is represented by the reference coordinate system of a DC magnetic position sensor. The position sensor is also used for tracking the position and orientation of the US images. Furthermore, we simulated patient anatomy by using custom-built phantoms. Our registration algorithm is a hybrid between fiducial- based and image-based registration algorithms. Initially, we perform a fiducial-based rigid-body registration between MR and position sensor space. Then, by changing various parameters of the rigid-body fiducial-based transformation, we produce an MR-sensor misregistration in order to simulate potential movements of the skin fiducials and/or the organs. The perturbed transformation serves as the initial estimate for the image-based registration algorithm, which uses normalized mutual information as a similarity measure, where one or more US images of the phantom are automatically matched with the MR image data set. By using the fiducial- based registration as the gold standard, we could compute the accuracy of the image-based registration algorithm in registering MR and sensor spaces. The registration error varied depending on the number of 2D US images used for registration. A good compromise between accuracy and computation time was the use of 3 US slices. In this case, the registration error had a mean value of 1.88 mm and standard deviation of 0.42 mm, whereas the required computation time was approximately 52 sec. Subsampling the US data by a factor of 4 X 4 and reducing the number of histogram bins to 128 reduced the computation time to approximately 6 sec. with a small increase in the registration error.

Proposes a methodology for 3D medical image segmentation based on the integration of 3D deformable and Markov random field models. Our method makes use of Markov random field theory to build Gibbs Prior models for the 3D medical image with arbitrary initial parameters to estimate the organ boundary. Then we use a 3D deformable model to fit the estimated boundary under the influence of gradient information in the initial 3D image and the balloon force. The result of the deformable model fit is used to update the Gibbs Prior model parameters, such as the gradient threshold of a boundary. Based on the updated parameters we restart the Gibbs Prior models. By integrating these processes recursively we achieve an automated segmentation of the initial 3D images. Our segmentation solution greatly reduces the time for 3D segmentation process and is capable of getting out of local minim. Results of the method are presented for several examples, including some MRI images with significant amount of noise.

We present an in vitro study of rigid registration methods for computer-aided surgery. The goals of the study were to obtain accuracy measures empirically under optimal laboratory conditions, and to identify the weak links in the registration chain. Specifically, we investigated two common registration methods (contact-based registration and image-based landmark registration) and established a framework for comparing the accuracy of both methods. The phantoms, protocols, and algorithms for tool tip calibration, contact-based registration with an optical tracker, fluoroscopic X-ray camera calibration, and fluoroscopic X-ray image-based landmark registration are described. Average accuracies of 0.5 mm (1.5 mm maximum) and 2.75 mm (3.4 mm maximum) were found for contact-based and image-based landmark registration, respectively. Based on these findings, the camera calibration was identified as being the main source of error in image-based landmark registration. Protocol improvements and algorithmic refinements to improve the accuracy of image-based landmark registration are proposed.

The aim of this paper is to present a new approach for improving deformable models and especially the Level Set Method (LSM) for segmentation and extraction of brain tumors in MRI (Magnetic Resonance Imaging) with more accuracy, the contribution of this work is to exploit thermal behavior of brain tumors for correcting level set contours. Human Body temperature distribution is an indicator of health condition, the brain tumor cells generate more heat than normal brain cells due to their higher metabolism and their vascular dilation. Heat distribution in human body is modeled using Pennes BioHeat Transfer Equation (PBHTE) solved by Finite Volume Method (FVM), and with the inverse analysis using Genetic Algorithm (GA) we will estimate the size and location of brain tumor, with this way Level Set Method extracts tumors contours with more accuracy and efficiency. To our knowledge, this is the first approach which introduces thermal analysis to improve the accuracy of segmentation and extraction of tumors in MRI images.

A non-rigid registration method is presented for the alignment of pre-procedural magnetic resonance (MR) images with delineated suspicious regions to intra-procedural 3D transrectal ultrasound (TRUS) images in TRUS-guided prostate biopsy. In the first step, 3D MR and TRUS images are aligned rigidly using six pairs of manually identified approximate matching points on the boundary of the prostate. Then, two image volumes are non-rigidly registered using a finite element method (FEM)-based linear elastic deformation model. A vector of observation prediction errors at some points of interest within the prostate volume is computed using an intensity-based similarity metric called the modality independent neighborhood descriptor (MIND). The error vector is employed in a classical state estimation framework to estimate prostate deformation between MR and TRUS images. The points of interests are identified using speeded-up robust features (SURF) that are scale and rotation-invariant descriptors in MR images. The proposed registration method on 10 sets of prostate MR and TRUS images yielded a target registration error of 1.99±0.83 mm, and 1.97±0.87 mm in the peripheral zone (PZ) and whole gland (WG), respectively, using 68 manually-identified fiducial points. The Dice similarity coefficient (DSC) was 87.9±2.9, 82.3±4.8, 93.0±1.7, and 84.2±6.2 percent for the WG, apex, mid-gland and base of the prostate, respectively. Moreover, the mean absolute distances (MAD) between the WG surfaces in the TRUS and registered MR images was 1.6±0.3 mm. Registration results indicate effectiveness of the proposed method in improving the targeting accuracy in the TRUS-guided prostate biopsy.

Multifunctional BaYbF5: Gd/Er upconversion nanoparticles for in vivo tri-modal upconversion optical, X-ray computed tomography and magnetic resonance imaging

The U S. National Library of Medicine (NLM) has long been a world leader in the archiving and distribution of the print-based images of biology and medicine. NLM has also been a pioneer in the use of computer systems to encode and distribute textual knowledge of the life sciences. NLM's Long Range Planning effort of 1985-86 foresaw a coming era where NLM's Bibliographic and factual database services would be complemented by libraries of digital images, distributed over high speed computer networks. The NLM Planning Panel on Electronic Imaging recommended that NLM should undertake the building a digital image library consisting of computerized tomography (CT) and magnetic resonance (MR) images, and cryosection images of a representative, carefully selected and prepared male and female cadaver -- the “Visible Human Project®.” The male and female Visible Human data sets are now being made available through a license agreement with the NLM. The data sets are supporting a wide range of educational, diagnostic, treatment planning, and commercial uses. The NLM, in partnership with other U.S. government research agencies has begun a three prong effort within the Visible Human Project to address: the creation of a new online, interactive, digital head-and-neck atlas; the development of a tool kit of computational programs capable of automatically performing many of the basic data handling functions required for using Visible Human data in applications; and the improved resolution of future Visible Human data sets through the reduction of the anatomical artifacts introduced by the methods used to stabilize and section the anatomical materials and the development of staining and wide-spectrum methods for increasing tissue contrast.