Fast data acquisition in Magnetic Resonance Imaging (MRI) is vastly in demand and scan time directly depends on the number of acquired k-space samples. Recently, the deep learning-based MRI reconstruction techniques were suggested to accelerate MR image acquisition. The most common issues in any deep learning-based MRI reconstruction approaches are generalizability and transferability. For different MRI scanner configurations using these approaches, the network must be trained from scratch every time with new training dataset, acquired under new configurations, to be able to provide good reconstruction performance. Here, we propose a new generalized parallel imaging method based on deep neural networks called NLDpMRI to reduce any structured aliasing ambiguities related to the different k-space undersampling patterns for accelerated data acquisition. Two loss functions including non-regularized and regularized are proposed for parallel MRI reconstruction using deep network optimization and we reconstruct MR images by optimizing the proposed loss functions over the network parameters. Unlike any deep learning-based MRI reconstruction approaches, our method doesn’t include any training step that the network learns from a large number of training samples and it only needs the single undersampled multi-coil k-space data for reconstruction. Also, the proposed method can handle k-space data with different undersampling patterns, and the different number of coils. Experimental results show that the proposed method outperforms the current state-of-the-art GRAPPA method and the deep learning-based variational network method.
The hippocampus and the insula are responsible for episodic memory formation and retrieval. Hence, visualization of the cytoarchitecture of such structures is of primary importance to understand the underpinnings of conscious experience. Magnetic Resonance Imaging (MRI) offers an opportunity to non-invasively image these crucial structures. However, current clinical MR imaging operates at the millimeter scale while these anatomical landmarks are organized into sub-millimeter structures. For instance, the hippocampus contains several layers, including the CA3-dentate network responsible for encoding events and experiences. To investigate whether memory loss is a result of injury or degradation of CA3/dentate, spatial resolution must exceed one hundred micron, isotropic, voxel size. Going from one millimeter voxels to one hundred micron voxels results in a 1000× signal loss, making the measured signal close to or even way below the precision of the receiving coils. Consequently, the signal magnitude that forms the structural images will be biased and noisy, which results in inaccurate contrast and less than optimal signal-to-noise ratio (SNR).
In this paper, we propose a strategy to perform high spatial resolution MR imaging of the hippocampus and insula with 3T scanners that enables accurate contrast (no systematic bias) and arbitrarily high SNR. This requires the collection of additional repeated measurements of the same image and a proper averaging of the k-space data in the complex domain. This comes at the cost of additional scan time, but long single-session scan times are not practical for obvious reasons. Hence, we also develop an approach to combine k-space data from multiple sessions, which enables the total scan time to be split into arbitrarily short sessions, where the patient is allowed to move and rest in-between. For validation, we hereby illustrate our multi-session complex averaging strategy by providing high spatial resolution 3T MR visualization of the hippocampus and insula using an ex-vivo specimen, so that the number of sessions and the duration of each session are not limited by physiological motion or poor subject compliance.
KEYWORDS: Brain, Diffusion, Tissues, Scanners, Neuroimaging, Distortion, Computer programming, Magnetic resonance imaging, Signal to noise ratio, In vivo imaging
Diffusion-weighted magnetic resonance imaging (DW-MRI) provides a novel insight into the brain to facilitate our understanding of the brain connectivity and microstructure. While in-vivo DW-MRI enables imaging of living patients and longitudinal studies of brain changes, post-mortem ex-vivo DW-MRI has numerous advantages. Ex-vivo imaging benefits from greater resolution and sensitivity due to the lack of imaging time constraints; the use of tighter fitting coils; and the lack of movement artifacts. This allows characterization of normal and abnormal tissues with unprecedented resolution and sensitivity, facilitating our ability to investigate anatomical structures that are inaccessible in-vivo. This also offers the opportunity to develop today novel imaging biomarkers that will, with tomorrow’s MR technology, enable improved in-vivo assessment of the risk of disease in an individual. Post-mortem studies, however, generally rely on the fixation of specimen to inhibit tissue decay which starts as soon as tissue is deprived from its blood supply. Unfortunately, fixation of tissues substantially alters tissue diffusivity profiles. In addition, ex-vivo DW-MRI requires particular care when packaging the specimen because the presence of microscopic air bubbles gives rise to geometric and intensity image distortion. In this work, we considered the specific requirements of post-mortem imaging and designed an optimized protocol for ex-vivo whole brain DW-MRI using a human clinical 3T scanner. Human clinical 3T scanners are available to a large number of researchers and, unlike most animal scanners, have a bore diameter large enough to image a whole human brain. Our optimized protocol will facilitate widespread ex-vivo investigations of large specimen.
Subdivision surfaces and parameterization are desirable for many algorithms that are commonly used in Medical Image Analysis. However, extracting an accurate surface and parameterization can be difficult for many anatomical objects of interest, due to noisy segmentations and the inherent variability of the object. The thin cartilages of the knee are an example of this, especially after damage is incurred from injuries or conditions like osteoarthritis. As a result, the cartilages can have different topologies or exist in multiple pieces. In this paper we present a topology preserving (genus 0) subdivision-based parametric deformable model that is used to extract the surfaces of the patella and tibial cartilages in the knee. These surfaces have minimal thickness in areas without cartilage. The algorithm inherently incorporates several desirable properties, including: shape based interpolation, sub-division remeshing and parameterization. To illustrate the usefulness of this approach, the surfaces and parameterizations of the patella cartilage are used to generate a 3D statistical shape model.
A major challenge in neurosurgery oncology is to achieve maximal tumor removal while avoiding postoperative neurological deficits. Therefore, estimation of the brain deformation during the image guided tumor resection process is necessary. While anatomic MRI is highly sensitive for intracranial pathology, its specificity is limited. Different pathologies may have a very similar appearance on anatomic MRI. Moreover, since fMRI and diffusion tensor imaging are not currently available during the surgery, non-rigid registration of preoperative MR with intra-operative MR is necessary. This article presents a translational research effort that aims to integrate a number of state-of-the-art technologies for MRI-guided neurosurgery at the Brigham and Women's Hospital (BWH). Our ultimate goal is to routinely provide the neurosurgeons with accurate information about brain deformation during the surgery. The current system is tested during the weekly neurosurgeries in the open magnet at the BWH. The preoperative data is processed, prior to the surgery, while both rigid and non-rigid registration algorithms are run in the vicinity of the operating room. The system is tested on 9 image datasets from 3 neurosurgery cases. A method based on edge detection is used to quantitatively validate the results. 95% Hausdorff distance between points of the edges is used to estimate the accuracy of the registration. Overall, the minimum error is 1.4 mm, the mean error 2.23 mm, and the maximum error 3.1 mm. The mean ratio between brain deformation estimation and rigid alignment is 2.07. It demonstrates that our results can be 2.07 times more precise then the current technology. The major contribution of the presented work is the rigid and non-rigid alignment of the pre-operative fMRI with intra-operative 0.5T MRI achieved during the neurosurgery.
We present a new method for modeling organ deformations due to successive resections. We use a biomechanical model of the organ, compute its volume-displacement solution based on the eXtended Finite Element Method (XFEM). The key feature of XFEM is that material discontinuities induced by every new resection can be handled
without remeshing or mesh adaptation, as would be required by the conventional Finite Element Method (FEM). We focus on the application of preoperative image updating for image-guided surgery. Proof-of-concept demonstrations are shown for synthetic and real data in the context of neurosurgery.
A new fast non rigid registration algorithm is presented. The algorithm estimates a dense deformation field by optimizing a criterion that measures image similarity by mutual information and regularizes with a linear elastic energy term. The optimal deformation field is found using a Simultaneous Perturbation Stochastic Approximation to the gradient. The implementation is parallelized for symmetric multi-processor architectures.
This algorithm was applied to capture non-rigid brain deformations that occur during neurosurgery. Segmentation of the intra-operative data is not required but preoperative segmentation of the brain allows the algorithm to be robust to artifacts due to the craniotomy.
Simon Warfield, Florin Talos, Corey Kemper, Eric Cosman, Alida Tei, Matthieu Ferrant, Benoit Macq, William Wells, Peter Black, Ferenc Jolesz, Ron Kikinis
The key challenge facing the neurosurgeon during neurosurgery is to be able to remove from the brain as much tumor tissue as possible while preserving healthy tissue and minimizing the disruption of critical anatomical structures. The purpose of this work was to demonstrate the use of biomechanical simulation of brain deformation to project preoperative fMRI and DTI data into the coordinate system of the patient brain deformed during neurosurgery. This projection enhances the visualization of relevant critical structures available to the neurosurgeon. Our approach to tracking brain changes during neurosurgery has been previously described. We applied this procedure to warp preoperative fMRI and DTI to match intraoperative MRI. We constructed visualizations of preoperative fMRI and DTI, and intraoperative MRI showing a close correspondence between the matched data. We have previously demonstrated our biomechanical simulation of brain deformation can be executed entirely during neurosurgery.
We previously used a generic atlas as a substitute for patient specific data. Here we report the successful alignment of patient-specific DTI and fMRI preoperative data into the intraoperative configuration of the patient's brain. This can significantly enhance the information available to the neurosurgeon.
Common problems in medical image analysis involve surface-based registration. The applications range from atlas matching to tracking an object's boundary in an image sequence, or segmenting anatomical structures out of images. Most proposed solutions are based on deformable surface algorithms. The main problem of such methods is that the local accuracy of the matching must often be traded off against global smoothness of the surface in order to reach global convergence of the deformation process. Our contribution is to first build a Multi-Resolution (M-R) surface from a reference segmented image, and then match this surface onto the target image in an M-R fashion using a deformable surface-like algorithm. As we proceed from lower to higher resolution, the smoothing effect of the deformable surface is more and more localized, and the surface gets closer and closer to the target boundary. We present initial results of our algorithm for atlas registration onto brain MRI showing improved convergence and accuracy over classical deformable surface methods.
A statistical classification algorithm, for MRI segmentation, based on the k Nearest Neighbor rule (kNN) has been implemented with Message Passing Interface (MPI) by partitioning the dataset into similar sized subvolumes and delivering each part to one processor inside a cluster. We have tested the algorithm in two different CPU architectures (SPARC and Intel) and four different configurations including a Beowulf cluster, two Sun clusters and a symmetric multiprocessor. The experiments provide a good speedup in all the cases and show a very good performance/price ratio in the PC-Linux cluster. We present results using a three channel, high resolution original dataset in times less than two minutes in the best cases and we use the segmented maps to make clinically relevant 3D visualizations in interactive times.
During neurosurgery, the challenge for the neurosurgeon is to remove as much as possible of a tumor without destroying healthy tissue. This can be difficult because healthy and diseased tissue can have the same visual appearance. To this aim, and because the surgeon cannot see underneath the brain surface, image-guided neurosurgery systems are being increasingly used. However, during surgery, deformation of the brain occurs (due to brain shift and tumor resection), therefore causing errors in the surgical planning with respect to preoperative imaging. In our previous work, we developed software for capturing the deformation of the brain during neurosurgery. The software also allows preoperative data to be updated according to the intraoperative imaging so as to reflect the shape changes of the brain during surgery. Our goal in this paper was to rapidly visualize and characterize this deformation over the course of surgery with appropriate tools. Therefore, we developed tools allowing the doctor to visualize (in 2D and 3D) deformations, as well as the stress tensors characterizing the deformation along with the updated preoperative and intraoperative imaging during the course of surgery. Such tools significantly add to the value of intraoperative imaging and hence could improve surgical outcomes.
Arterio-venous malformations (AVMs) are a congenital disorder that affects a small percentage of the population. They are treated by blocking or reducing the blood supply followed by surgery. This paper looks in a preliminary way at visualizing the cerebral vasculature and ultimately the AVMs. These visualizations provide support for the surgeons and radiologists. Our concern is to substantiate the point that there are deficiencies in the data correctable with reference to digital subtraction angiograms and we conjecture that knowledge based processing of this data may lead to improved results. The paper explores the basis of the difficulty and it compares the performance of several algorithms. Simple geometric objects are studied and the dependence of error on several parameters is shown. A comparison is drawn between the richness of the data available from x-ray angiograms (XRAs) and magnetic resonance angiograms (MRAs). Inferences are drawn on approaches that may be appropriate for the evolution of a description of the vasculature. Comment is also made on the way in which different representations may be compared.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.