Patent Application
20070249928
These and other features, aspects, and advantages of the present invention will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:
The following detailed description is exemplary and not intended to limit the invention of the application and uses of the invention. Furthermore, there is no intention to be limited by any theory presented in the preceding background of the invention of the following detailed description of the drawings.
In a first embodiment, the invention provides a method for processing longitudinal magnetic resonance (MR) images comprising obtaining at least a first of MR images, selecting a region of interest within the first set of MR images, obtaining at least one subsequent MR image, and applying a registration relative to the first MR image and the at least one subsequent MR image wherein the region of interest in the at least one subsequent MR image is repositioned analogously to the region of interest in the first MR image. As used herein, the term “region of interest” refers to at least one voxel, or more specifically a chosen anatomical region of interest located inside the at least one voxel, which is to be chosen by the operator. Furthermore, a region of interest may comprise a metabolic region of interest, inside of which the concentration of metabolites such as glutamate or choline may be measured through a magnetic resonance spectroscopy exam. The term “region of interest overlap” refers to the percentage of the first region of interest voxel volume encompassed by the longitudinal voxel. As used herein the term “registration” refers to the correlation of two separate image regions or volumes relative to one another. For instance, registration of a first scan and any subsequent longitudinal scans. As used herein the term “analogous” or “analogously” refers to the occurrence or situation wherein the chosen regions of interest in the first and subsequent MR images are positioned to be spatially aligned in 3D (i.e. registered on all 3 Cartesian axes) with respect to one another.
In a second embodiment, the invention provides an imaging system adapted to obtain MRS images and a processor that is adapted to apply a registration relative to a first MR image and an at least one subsequent MR wherein a region of interest in the at least one subsequent MR image is repositioned analogously to a region of interest in the first MR image.
Referring to
In a well-known manner, processor 106 is configured such that there is sufficient memory for storing measured data and reconstructed images. The memory is sufficient to store the whole of N-dimensional measured data as well as reconstructed data. Also in a well-known manner, a MR image is constructed from the image or k-space data corresponding to a predetermined plurality of applications of a MRI pulse sequence initiated by a RF pulse such as from transmitter 102 of
In embodiments of the present invention, processor 106 is configured to process longitudinal MR images, which will be described in greater detail is
As a general description, magnetic resonance imaging (MRI) is a well-known imaging method in which magnetic moments are excited at specific nuclear spin precession frequencies that are proportional to the magnetic field occurring within the magnet of the MRI system. Spin is a fundamental property of nature, such as electrical charge or mass. Precession is a rotational motion about an axis of a vector whose origin is fixed at the origin. The radio-frequency (RF) signals resulting from the precession of these spins are received typically using RF coils in the MRI system and are used to generate images of a volume of interest. A pulse sequence is a selected series of RF pulses and/or magnetic field gradients applied to a spin system to produce a signal representative of some property of the spin system.
In addition, an MRI system may be adapted to obtain magnetic resonance spectroscopy (MRS) images, to obtain metabolic information (e.g. metabolite concentrations) from the anatomical regions of interest. Similar to nuclear magnetic resonance (NMR), different molecules have different signatures in the NMR spectrum, and the concentration of different molecules of interest can be quantified from their NMR spectrum. This technique takes advantage of the magnetic spin properties of biological samples to reveal metabolic information.
In a well-known manner for tracking disease progression and monitoring changes in a particular anatomy due to disease, longitudinal images may be used. As used herein, “longitudinal images” means one or more subsequent images subsequent to a baseline scan, of a single subject or a plurality of subjects, relating to a first image of a single subject or a plurality of subjects. For instance, within a cohort of subjects diagnosed with the same disease, the baseline scan of single subject within the cohort may be used as the baseline scan for all other subjects within the cohort. The longitudinal images for each subject thus may be related back to the single subjects baseline scan. In principle however, the more spatially congruent each longitudinal image is as it relates to the first image, the more accurate the interpretation of each image will be. Also, longitudinal images may be employed to track the progression of disease for other conditions and that longitudinal images may be employed to monitor response (favorable or unfavorable) to treatment.
To obtain a first MR image, also referred to herein as a baseline scan, transmitter 102 of
To obtain a subsequent brain scan of a single subject, also referred to herein as a longitudinal scan, transmitter 102 is desirably adapted to generate and apply, during MR imaging, a pulse sequence used to acquire the first or baseline image to obtain at least one subsequent MRI image. Referring to
To locate at least one region of interest, the operator may choose an anatomical region of the brain (e.g. hippocampus, or caudate nucleus) from which he/she desires metabolic information (e.g. choline concentration). Furthermore, the operator may choose to acquire the metabolic information of interest using MRS. Referring to
Referring further to
To precisely locate the metabolic region of interest in a follow-up scan, a registration procedure is generally needed. Generally, image registration is the process of aligning two images or image volumes. In its most basic form, a registration algorithm optimizes a metric function by adjusting the parameters of a transformation mapping one image to the other. It may adjust for orientation changes, transverse changes, longitudinal changes, rotational changes, or any combination thereof. More specifically, in embodiments of the present invention, the processor 106 of
Here log( ) represents the natural logarithm of a number, p(ai) is the likelihood of finding pixels with intensities ai throughout the imaging volume, and M represents the number of bins in which image intensity has been partitioned into. A digital image whose pixels are encoded in N bits, can have 2N different grayscale values in a pixel and therefore the maximum value of M can be 2N. In practice, however, choosing the maximum value for M would lead to computationally intense calculations, particularly when calculating the joint entropy of two images; M=30 is a good compromise between precision and computational cost.
The joint entropy of imaging volumes A and B, H(A,B) is then defined as:
Here p(ai,bj) is the likelihood of finding pixels with intensities ai in image A at the same time that the corresponding pixel in image B has intensity bj. Given the similarities between the two image sets to be registered, one could partition both image intensities in the same number of bins, ie M=30. The mutual information of two imaging volumes A and B, MI(A,B), is defined as
MI(A,B)=H(A)+H(B)−H(A,B) (3)
We are searching for the transformation T, that maximizes the mutual information defined as
MI(B,T(F))=H(B)+H(T(F)))−H(B,T(F)) (4)
Here B is the baseline imaging volume, F is the follow up imaging volume, and T(F) is the transformed follow up imaging volume. Because the registration may be intra-subject with relatively short times between scanning sessions, i.e. no significant changes in morphology, a six degrees of freedom rigid transformation may be used. If B=f1(r) as the baseline imaging volume (where r=[x, y, z]T is a position vector variable pointing to the space variable (x,y,z)), then the transformed follow-up imaging volume, T(F) can be expressed as T(F)=f2(r)=f1(Rr+t). Here R is a 3×3 orthonormal matrix with a determinant of 1, fully characterized by three rotation parameters (Euler angles), and t is a 3×1 translation vector comprised of three translational parameters. The matrix R and vector t represent the “motion” from time point 1 to time points 2. The objective of the registration algorithm is to estimate, as accurately as possible, the six parameters that specify R and t, by optimizing the mutual information cost function. A conjugate gradient descent method (15) may be used to search for the transformation that maximizes the mutual information MI(B, T(F)).
In a further embodiment, processor 106 is adapted to automatically measure quantitative changes between baseline scans and longitudinal scans, specifically in reference to each corresponding region of interest. Additionally, processor 106 is adapted to automatically send detailed exam information to display 107 and storage 108.
In an exemplary embodiment, a longitudinal exam protocol started with a short localizer, followed by an 86 slice axial brain volume with parameters identical acquisition to the baseline scan. Two types of protocols were used to compare the eye and automatic repositioning procedure for the regions of interest. In the first protocol, the MRS data was obtained by repositioning a voxel on the axial volume as close as possible to the initial voxel location; for each subject, screen shots showing the baseline voxel prescription were available to the operator. During the image acquisition, registration was performed between the first axial scan and the longitudinal axial scan yielding the rigid transformation that connected the head positions in the two exams. An oblique volume was then acquired in the follow up exam using the parameters output by the registration algorithm had identical acquisition parameters and brain coverage as the first exam. The images in the oblique volume were then used as support for the acquisition of another spectrum from a voxel whose positioning was given by the registration algorithm (automatic voxel repositioning procedure). The algorithm applied the inverse transformation to the position of the initial voxel location, thus mapping the first metabolic region of interest to the same anatomic location in the longitudinal scan.
In a second protocol to compare the eye and automatic repositioning procedures, the axial volumes were acquired. They were then followed by the acquisition of the oblique volume with identical orientation as the first exam. A spectrum was acquired from a voxel whose location was dictated by the output of the registration procedure. Following the completion of the first MRS exam, a longitudinal MRS scan was then prescribed on the axial series; the location of this last voxel, repositioned by eye, was chosen to match the location of the first metabolic region of interest as well as possible. These two protocols were alternated for subjects, in order to insure that the same amount of time elapsed between the automatic and eye repositioning of the voxel and the localizer scan.
Additionally, each metabolite of interest may be quantified, as shown by example in table 2. Here, the coefficient of variation (CV) and the Pearson correlation coefficients (r) are displayed for the volunteers studied, for both the eye repositioning and the automatic repositioning procedures.
In this particular exemplary embodiment, each metabolite was quantified using a linear combination of basis spectra, and version 6.0 of LCModel. Fitting was performed between 0 and 4 ppm using the basis set supplied by the program developer. NAA concentrations included the sum of the concentrations of NAA and NAA-glutamate. Similarly, choline (Cho) concentrations include the sum of all the Cho containing compounds. Additional metabolite concentrations that may be present include creatine (Cr), myo-Inositol (ml) and glutamate (Glu). Absolute quantitation was based on the reciprocity principle and signal calibration was performed with a 50 mM NAA phantom scan prior to the scans.
Referring further to table 2, the average intra-day, intra-volunteer coefficients of variation (CV's) for the metabolites recorded are shown. Consistent decreases in CV's can be noted when the automatic re-localization procedure was used, as seven out of the nine metabolites studied showed decreased CV's. More precisely, three out of nine metabolites showed consistent improvements (>10%) in CV's (Cr, Glu and Cho/Cr), while only one showed decreased (>10%) repeatability (NAA/Cr). Out of the rest of five metabolites, four showed marginal improvement (0-10%) in CV's (Glu/Cr, mI, mI/Cr, NAA), while only one other metabolite concentration showed marginally decreased (0-10%) repeatability (Cho). While both repositioning methods generally offer acceptable short term reproducibility coefficients (defined as Pearson correlation coefficients higher than 0.75) for all metabolite concentrations and concentration ratios studied (except for Glu and Glu/Cr), increased reproducibility coefficients can be noted when the automatic repositioning procedure is used. Most notably, the reproducibility becomes borderline to unacceptable for Cr and NAA/Cr when the manual repositioning technique is used.
The registration algorithm presented does not have any stringent requirements with respect to image contrast or brain coverage. For instance, in an exemplary embodiment, two sets of images were acquired with one volunteer in one position. The first set of images covered the whole brain (86 slices in axial orientation), had high resolution, (1.15 mm ×1.15 mm ×2 mm), and high contrast (inversion recovery preparation was used, with an inversion time of 300 ms). The second set of 18 images covered about a quarter brain, had low resolution (1.87 mm ×1.87 mm ×2 mm), and low contrast (no inversion recovery preparation). The volunteer moved to a new position and the same acquisition was performed. To estimate the accuracy of the algorithm, the parameters calculated by registering the high-resolution volumes were compared with the parameters calculated by registering the low-resolution volumes. The difference in parameters reported by the two registration operations were [0.33°, 0.22°, 0.18°] for the three Euler angles, and [0.22 mm, 0.18 mm, −0.19 mm] for the three displacements.
Additionally, by using this implementation, images with different contrast can be registered, as well as images covering different brain regions (but with significant overlap). For example, a full brain, T1 weighted data set from a baseline scan was successfully registered with a quarter brain, T2 weighted data set in the follow-up exam. In this instance, subjective evaluation was performed by visual inspection. This additional flexibility might be needed when two different protocols are selected for the baseline and follow-up exams. Moreover, the first image of a metabolic region of interest may be compared to the metabolic region of interest in each subsequent longitudinal image. The operator may compare each images optically, and track metabolic regions of interest over time. Furthermore, the operator can track disease progression or response to therapy by tracking, for example, lesion growth or reduction.
In another exemplary embodiment, four healthy, normal volunteers were scanned on four different days during the course of six months, three times each day. The subjects were then removed from the scanner room between the daily scanning sessions. Care was taken for consistent repositioning of the subjects in repeat scans. By moving the subject's head, the longitudinal landmark line was aligned with the subject nose, while the transverse landmark light was aligned with the inter-pupillary line. If slight rotations were discovered following the acquisition of the scout, however, no subject repositioning in the magnet was performed. To assess the performance of the registration algorithm, one of the high-resolution baseline volumes for one of the volunteers was rotated and translated by a known amount, with rotation angles limited to ±8 degrees and translation limited to ±10 mm in all dimensions. The baseline volume was then registered to the transformed volume, and the errors (defined as the difference between the known and calculated parameters) for the three translations and three rotations recorded. The average realignment error (for all voxels) was also computed as
Here N represents the number of voxels in the imaging volume, R and t are the known rotation and translation matrices, respectively, {circumflex over (R)} and {circumflex over (t)} are their estimates obtained from the registration algorithm. Note that, if {circumflex over (R)}=R and {circumflex over (t)}=t, then the realignment error becomes zero. This process was repeated 100 times, by randomly choosing a set of rotation and translation parameters bounded by the limits specified above.
In the same exemplary embodiment, metabolic region of interest overlap calculations were measured for both automatic and eye repositioning. The longitudinal axial was registered to the first metabolic region of interest axial using the algorithm described above; the longitudinal voxel (placed by eye) was transformed into the space of the baseline scan. Samples, regularly spaced at 0.1 mm intervals inside the baseline voxel were compared against the transformed follow up voxel. Samples were considered overlapping if they are contained in both the first metabolic region of interest voxel and the transformed longitudinal voxel. The procedure was repeated for the triple oblique exam and automatically placed voxel.
While only certain features of the invention have been illustrated and described herein, many modifications and changes will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.
This application is related to Provisional Application U.S. Ser. No. 60/793,058, entitled “Automatic Repositioning Of Single Voxels In Longitudinal 1H MRS Studies”, filed Apr. 19, 2006, the contents of which are herein incorporated by reference and the benefit of priority to which is claimed under 35 U.S.C. 119(e).
| Number | Date | Country | |
|---|---|---|---|
| 60793058 | Apr 2006 | US |
