Patent Application
20190197686
The disclosure of this application relates, in general, to techniques by which images can be created, analyzed and presented to health care professionals or other individuals or systems.
Diagnostic imaging of the spine of a patient is often useful for identifying disease or an injury to the spine itself or as a readily locatable landmark for other tissues. Present practice is to take and digitally store lots of data on a patient, including MR and CT images. You want to both compare each patient's data to his/her own data, and “pools” of data from other people. However, making sense of pictures taken at different times, and using different types of machines presents a problem which is presently beyond the reach of most automated systems. Additional complications are presented by variations in quality between images, incomplete images of the spine or neuro-axis (e.g., an image might not include top or bottom vertebrae), failure to adequately capture portions of the neuro-axis due to congenital defect, disease, injury or surgery, exceptional curvature of the spine exhibited in some individuals, and the existence of individuals having more or less than the highly prevalent 23 mobile pre-sacral vertebrae. Thus, analysis of images and prescription of additional data collection and treatment requires an extensively trained technician.
Unfortunately, even for skilled technicians, human error may occur due to the variability in the patient population or due to an oversight. The mistake may arise in incorrectly identifying vertebrae and discs in a diagnostic image. The mistake may arise in incorrectly visually identifying the corresponding vertebrae under the skin before performing a surgical or therapeutic (e.g., radiation) treatment. The mistake may arise in improperly identifying a normal, benign, or malignant condition because an opportunity is missed to correctly correlate information from a plurality of imaging systems (e.g., a type of tissue maybe determined if an MRI and a CT image could be properly correlated and analyzed).
Assuming the vertebrae and discs may be accurately identified in the diagnostic image, it is often helpful to be able to obtain a one-to-one correspondence between the readily visible and markable skin/surface and underlying structures or pathology detectable by a variety of imaging modalities. This may facilitate clinical correlation XRT, image guided biopsy or surgical exploration, multimodality or interstudy image fusion, motion correction/compensation, and 3D space tracking. However, current methods, (e.g. bath oil/vitamin E capsules for MRI), have several limitations including single image modality utility requiring completely different and sometimes incompatible devices for each modality, complicating the procedure and adding potential error in subsequent multimodality integration/fusion. They require a separate step to mark the skin/surface where the localizer is placed and when as commonly affixed to the skin by overlying tape, may artificially indent/compress the soft tissue beneath the marker or allow the localizer to move, further adding to potential error. Sterile technique is often difficult to achieve. Furthermore, it may be impossible to discriminate point localizers from each other or directly attain surface coordinates and measurements with cross sectional imaging techniques. In regards to the latter, indirect instrument values are subject to significant error due to potential inter-scan patient motion, non-orthogonal surface contours, and technique related aberrations which may not be appreciated as current multipurpose spatial reference phantoms are not designed for simultaneous patient imaging.
Limited coverage, resolution and contrast of conventional MRI localizers coupled with a high prevalence of spinal variance make definitive numbering difficult and may contribute to the risk of spinal intervention at the wrong level. Only 20% of the population exhibit the classic 7 cervical, 12 thoracic, 5 lumbar, 5 sacral, and 4 coccygeal grouping. For instance, 2-11% of individuals have a cephalad or caudal shift of lumbar-sacral transition, respectively resulting in 23 or 25 rather than the typical 24 mobile presacral vertebrae. Numbering difficulties are often heightened in patients referred for spine Mill. Such patients are more likely than the general population to have anomalies, acquired pathology, or instrumentation that distorts the appearance of vertebrae and discs. Moreover, these patients are often unable to he still within the magnet for more than a short period of time due to a high prevalence of back pain and spasms. Resultant intrascan motion confounds image interpretation and interscan motion renders scan coordinates and positional references unreliable. Those difficulties and inadequacies in presently available technology can lead to significant problems. While data remains somewhat limited, various authors report an approximately 2-5% incidence of wrong level approach spinal intervention, with most cases involving the lower lumbar interspaces. Such surgical misadventures may lead to needless pain and suffering, as well as contribute to accelerating medical malpractice costs. The first multi-million-dollar jury verdict for such a wrong level approach was awarded in 2002. In addition to potentially introducing errors, the use of human technicians to analyze images in present technology leads to significant delays and cost increases in diagnosis.
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention, and, together with the general description of the invention given above, and the detailed description of the embodiments given below, serve to explain the principles of the present invention.
Correctly identifying each spinal structure, that is, each vertebra or disc, in the spine 20 is complicated in certain instances when the skull 18 and/or the pelvic bones 22 are not imaged. In some instances, a vertebra or disc (depending on whether CT, MRI, or some other imaging modality, is being used) will fail to image properly, depicted at 24. In addition, the highly predominant number of twenty-four mobile pre-sacral vertebrae may not be the case for certain individuals. Having the correct vertebrae references may be important in localizing a suspicious lesion 26, which might be later targeted by some type of therapeutic instrument, that is any tool or device which can be used to help improve or restore health. A non-limiting example of such a therapeutic instrument is the radiation device 101 depicted in
As shown in
Several approaches to implementing systems which are operable to obtain and label medical diagnostic images of patients have been found to be effective. For example, in imaging system 12, MRI studies were performed on a clinical 1.5 T magnet with standard 4-channel quadrature array 6-element spine coil and surface coil correction algorithm. Contiguous two-station 35 centimeter (cm) field of view (FOV) sagittal fast gradient-recalled echo (FGRE) sequences were pre-programmed, providing full cervical, thoracic and lumbar (C-TL) spine coverage. Combined sagittal FOV of 70 cm., 7 sections, scans from 15 millimeters left of midline to 15 millimeters right of midline (L15-R15), 4 millimeter (mm) section thickness with 1 mm intersection gap; 512×352 matrix, 1 excitation/image acquisition (nex), repetition time (in milliseconds) (TR) 58, echo time (in milliseconds) (TE) 2.0, 30° flip, Bandwidth (BW) 15.6 kHz; 21 sec×2 stations=42 sec. To facilitate and standardize auto-prescriptions, a line was drawn on the spine coil for the technologists to landmark (set as scanner's 0 coordinate) rather than have the technologist use a superficial anatomic feature. The coil has a contoured head/neck rest assuring grossly similar positioning of the cranial-cervical junction of each patient relative to this landmark. The semi-automated disc detection and numbering algorithm of the spine image processor 40 was iteratively developed, tested and refined on batches of consecutive de-identified patient studies and results compared to neuroradiologist assignments. The spine image processor 40 was implemented in MATLAB 7.
In another instance, images were prepared using out-of-phase FGRE sequencing (512×352, TR 57-62 msec, TE 2.2 msec at 1.5 T and TE 1.4 msec or 3.2 msec at 3.0 T) and experimental FSE IDEAL sequencing, the latter with T1 and/or T2 weighting. In this instance, for T1 weighting, TR ranged from 350-750 msec, TE 14-33 msec, ET 1-4, frequency encoding 320-512 and phase encoding 224-256. For T-2 weighting, TR ranged from 1867-2600 msec, TE 20-70 msec, ET 8-10, frequency 320-512 and phase 224-352. On different types of hardware, 7, 9 or 11 sections were obtained in a single acquisition. Acquisition time for FGRE sequences was 21-22 seconds to accommodate breath holding, and IDEAL sequencing was limited to less than five minutes. In this instance, the spine image processor 40 was configured to accommodate variable number of sections; integrate age, gender and weight information available in DICOM header with a database to constrain search; compare candidate disc properties with estimated values derived from the database; use a search algorithm constraining subsequent discs based on previous (more cephalad) disc characteristics; and review intensity statistics of candidate discs. Of course, it should be understood that other approaches are also possible, and that the approaches set forth herein are intended to be illustrative only, and not limiting.
Returning the discussion of
These analyses performed by the spine image processor 40 in the illustrative version key upon one or more sources of information to identify a unique definable reference landmark for characterization and unambiguous labeling of one or more spinal structures. In the illustrative MRI embodiment, the centroid of the top (most cephalad) intervertebral disc (C2-3) is utilized as a key landmark. Alternatively, the top vertebrae (C1 or C2), or other unique identifiable structure, or other landmark may be employed.
In particular, in block 46, the top intervertebral disc is identified, which may be automated in block 48 by interfacing with an automated cranium identification system such as described in U.S. patent application Ser. No. 10/803,700, “AUTOMATED BRAIN MRI AND CT PRESCRIPTIONS IN TALAIRACH SPACE” to Dr. Weiss, filed 18 Mar. 2004 the disclosure of which is hereby incorporated by reference in its entirety. Alternatively, logic may be incorporated into the spine image processor 40 wherein a spine algorithm in block 50 recognizes a top vertebra or intervertebral disc. As a further alternative and as fall back option should automated means fail, in block 52 a technologist input is received to identify the top vertebra or intervertebral disc.
During testing, the neuroradiologist could readily visualize and definitively number all cervical-thoracic-lumbar (C-T-L) levels on ASSIST localizers. 6/50 patients had a congenital shift in lumbar-sacral transition; 3 with 23 mobile pre-sacral vertebrae (as shown in
With a labeled disc image 64, correct relative location to other imaged tissue of interest 65 may be used for further diagnostic procedures or therapeutic intervention. For instance, data fusion (block 66) might allow images taken with the same type of imaging modality at different times to be correlated, which could facilitate monitoring growth progression of a suspicious lesion or changes due to an intervening injury. In addition, images taken with different imaging modalities might be cross referenced to perform multi-spectral diagnoses (block 68), wherein information on a type of tissue may be gained by its different responses to various types of electromagnetic spectra. With tissue diagnosis complete, in block 70 a guided therapeutic procedures algorithm may correctly orient a therapeutic agent, such as the radiation device 101 or a guided minimally invasive surgical instrument (not shown). Alternatively, for an externally oriented procedure, a surgeon may reference either the relative location to a known spinal structure (e.g., vertebrae) and/or reference the skin/surfacing marking system 16.
During testing of the illustrative algorithm, patients received a rapid total spine ASSIST localizer with pre-set parameters. All studies were performed on a 1.5 T GE Excite MRI system with standard 4-channel, 6-element quadrature spine coil and cm field of view surface coil correction algorithm. Contiguous two-station 35 cm field of view (FOV) sagittal fast gradient-recalled echo (FGRE) sequences were pre-programmed, providing full cervical, thoracic and lumbar (C-T-L) spine coverage. Combined sagittal FOV was 70 cm., 7 sections were obtained at each station, L15-R15, 4 mm section thickness with 1 mm intersection gap; 512×352, 1 nex, TR 58, TE 2.0, 30° flip, BW 15.6; 21 sec×2 stations=42 sec. These ASSIST studies were de-identified in consecutive batches and copied to CD for subsequent off-line review, computer algorithm development and testing. A semi-automated disc detection and numbering algorithm was iteratively developed and results compared to neuroradiologist assignments.
During testing, a first batch of 27 cases was initially run with an algorithm 100 developed using previously obtained, non-surface-coil corrected ASSIST images. As the first step 102, we input cervical thoracic (top half), and thoracic lumbar (bottom half) spines. A threshold and a median spatial filter are applied to the search region. Then, additional disc constraints are applied to identify longest chain in top and bottom images. Candidate discs must extend onto at least two adjacent slices. As used in this application, adjacent slices should be understood to refer to two slices for which the space between those slices is not represented in a medical diagnostic image. Objects at the boundary or touching the boundary of the search region are excluded. Different threshold values, and candidate discs' constraints are applied to the top, and bottom half image.
In
Initially, the two data sets (top and bottom half) are processed to obtain putative discs separately utilizing different threshold values and disc constraint parameters (block 104). This results in upper and lower images (106, 108) depicted in
There are a variety of approaches to the application of these constraints to the voxels. In some implementations, voxels can be subjected to constraints which are appropriate for normal spinal morphologies (e.g., default constraints), which may or may not be preprogrammed for the algorithm. Thus, in some cases, an algorithm can be designed so that acceptable voxels are selected to extend onto at least two adjacent sagittal sections but not touch the boundary of the search region 112, 114. Similarly, such an algorithm might include a requirement that the acceptable voxels must lie 6-80 mm in the cervical thoracic (C-T), and 8-100 mm in the thoracic lumbar (T-L) region from adjacent acceptable voxels. Of course, other approaches are also possible. For example, in some cases, an algorithm might be configured to utilize a database of medical diagnostic images which can be used to derive constraints in a way which is specific to an image under study. In such an approach, where a database is utilized to customize the search for acceptable voxels, an algorithm such as described herein could be augmented through machine learning. That is, as more images are received and processed, those images could be included in the database, so that more fine-grained and accurate customizations could be made based on the increasing data set.
Continuing with the discussion of how images can be processed, in some implementations, after the acceptable voxels are identified, the algorithm can continue with identification of centroids of candidate discs. These centroids, that is, voxels with coordinates determined by a weighted mean of the coordinates of the voxels in a voxel cluster, the weight of the coordinates being determined by an attribute such as intensity of the voxel at that coordinate, of these acceptable voxel clusters 120 (candidate discs) are then connected by curved line 122. Additionally, the identification of centroids can be facilitated by considering the possible relationships between discs. For example, in some cases the algorithm might assume that the angle subtended by the line 122 connecting the centroid of adjacent candidate discs 120 and the major axis of these discs will be between 45° and 135° in both the C-T and T-L spine. Similarly, in some cases where a database of medical images is used to modify processing, the algorithm might consider the relationship between candidate discs in archival images in the database which have attributes which are similar to those of the image under study (e.g., if the image under study is from a patient with scoliosis, then scoliotic images from the database can be used to determine the relationship between adjacent discs for purposes of identifying candidates). In
Continuing with the illustrative algorithm depicted in
In the illustrative algorithm, if 23 discs are not selected in block 146, search criteria and thresholds can be refined. There are a variety of methods by which such refinement can take place. For example, in some implementations, the following formula for inter-disc height for each level can be used for the refinement (Equation 1):
h=0.6(M) for L=1,2, or 3 else
h=M+(L−1.2)*0.05(M) for L>3
where h is inter-disc height, L is the disc level, and M is the distance along the line through the centroids/23 (mean). It is also possible that a database of archived medical diagnostic images such as described above could be used in which the estimated inter-disc height could be determined based on the observed inter-disc height from archival images which have attributes which are similar to the attributes of the image under study (e.g., patients having a similar height).
Continuing with the illustrative algorithm depicted in
E(xj,yj)=(xj−1,yj−1)+ha(xj−1,yj−1)=Σ1j−1hs(x,y)/ha(x,y) (Equation 2)
In equation 2, ha is the average vertebral height from a 22 subject training set, hs is the vertebral height from the subject in question. The program searches for local maxima along this line extension and 24 adjacent parallel lines. A further determination is made in block 152 as to whether 23 discs have been found. Iteration continues as long as twenty-three (23) discs are not selected, as represented by block 154 that extends the search and the further determination of block 156, which labels the vertebrae if 23 are selected (block 157). Depending on the resources available in a particular implementation, the value ha in the equation above can be an average value taken from some set of training data, or could be a more customized value for a particular patient. As an example of how a customized value might be used, in cases where there is available a database of previously processed images, the program employing the algorithm of
If not 23 discs in block 156, then in block 158 a further determination is then made that twenty-two (22) discs are selected. If so, the algorithm 100 will determine in block 160 whether the last identified level (L4-5) satisfies criteria for the terminal presacral interspace suggesting a congenital variant with 23 rather than the typical 24 mobile presacral vertebrae. In some implementations, to be considered a terminal disc, the centroid of the 22nd disc must lie within 25% of its expected location E(x,y) relative to the 21st disc. Additionally, an algorithm can be implemented such that the centroid lies posterior to the centroid of the 21st disc and the slope of the 22nd disc is positive and greater than that of the 21st disc. If those criteria are met, then twenty two discs will be labeled (block 162).
If 23 discs are found in block 166, the discs are labeled in block 168. Else if the terminal disc criteria are not met in block 166, the position of the 23rd (L5-S1) disc is estimated using Equation 2 (block 164), and search constraints refined. If the 23rd disc is still not identified in block 166, the disc is presumed to be severely degenerated or post-surgical and the estimated position for L5-S1 will be accepted in block 170 and the discs thus labeled (block 168).
If less than 22 discs are identified by the algorithm in block 158, then the algorithm depicted in
The algorithm was run on an INTEL (San Jose, Calif.) personal computer with a 2.8 Ghz Xeon processor. Computer auto-labeling was compared to independent neuroradiologist assignments for each patient's study. The automated spine MRI sequencing provided a robust survey of the entire C-T-L spine in 42 seconds total acquisition time. In all patients (50/50), the neuroradiologist could readily visualize and definitively number all C-T-L levels on the ASSIST localizers. These included six patients with a congenital shift in their lumbar-sacral transition; three with 23 mobile pre-sacral vertebrae (
During testing, the exemplary algorithm was accurate in all 50/50 cases (100%) of the patients to include the original 27 patients plus 23 subsequent cases, despite the presence of congenital variations (
Additional variations on an algorithm such as described above could be implemented to extend its applicability to a wider section of the population. For example, in some cases where the medical diagnostic images analyzed are sagittal sections, the algorithm could be modified so that, in scoliotic patients, additional sagittal sections could be obtained, resulting in complete spinal coverage even given the abnormal curvature of those patients' spines. Similar modifications could be made in an implementation configured to analyze three-dimensional medical diagnostic images, where a larger portion of the three dimensional image could be analyzed to locate spinal structures.
Although the ASSIST algorithm 100 was successful in all 50 patients studied, the 7 section sagittal acquisition might be insufficient for some patients. For examples, it might be expected to fail in subjects with severe scoliosis due to insufficient spine coverage. As such, if significant scoliosis is suspected, more sagittal sections could be auto-prescribed, the cost being a proportionate increase in scan time. The automated disc detection/numbering algorithm 100 was designed to accept any number of sagittal sections, and so is potentially applicable to other populations in addition to the adult population which was used to test the exemplary embodiment described above. An alternative method of accommodating additional populations is to modify the disc constraints and parameters, described above, before using the algorithm to identify spinal structures in those populations.
While the illustrative disc detection algorithm 100 described above requires manual input of the most cephalad disc, C2-3, to achieve maximal accuracy, it should be appreciated that automated computer detection of this interspace may be implemented. The C2-3 disc may be readily discerned on midline sagittal head images with a 22-24 cm FOV, (as described in 2004 at pages 233-37 of volume 25 of the American Journal of Neuroradiology and in 2003 at pages 922-29 of volume 24 of the American Journal of Neuroradiology, the teachings of which are incorporated by reference) or ASSIST 35 cm FOV cervico-thoracic spine images based on several unique features. These include the characteristic shape of the C2 vertebrae and relatively constant relationship to the skull base and cervico-medullary junction. Alternatively, in some instances, such as implementations using MRI sequencing and disc detection algorithms as described herein the most cephalad disc space identified may simply be labeled C2-3.
Additionally, it is possible that, in some instances, rapid volumetric MRI and CT imaging of the bring, spine or entire body and rapid fat-water MRI decomposition techniques can be exploited to make disc detection and other algorithms more robust. For example, as the calvarium, spine and their contents are nearly symmetric bilaterally, if volumetric MRI or CT images of these structures are obtained, a plane of symmetry can be automatically derived using a cost minimization function designed to reduce differences between the right and left side of the body. Weighted functions might emphasize structures near the midline or with particular MRI or CT characteristics. Volume data reconstructed along this plane of symmetry corresponds to the patient optimized midline sagittal image. This image could then be utilized by the AC-PC searching algorithm or ASSIST algorithms previously described. Having the midline of these structures defined would also facilitate computer detection of asymmetric regions, which might signify pathologic processes and thus be worthy of further image interrogation.
With a head/neck/spine (HNS) array coil, ASSIST can be extended to include the head and neck as well as the spine. Depending on patient height and the magnet's vertical field coverage, the entire head, neck, cervical, thoracic, and lumbar-sacral spine can now be auto-imaged in 2 or 3 contiguous or slightly overlapping 3D sagittal stations. The head can be imaged with either straight sagittal 3D sequencing and subsequently reformatted or with optimized oblique sagittal 3D sequencing, the latter potentially using the method described in Weiss K L, H. P. Computer Automated Three-Step (CATS) Program for Brain MRI Prescription in Talairach Space for rapidly obtaining a 2D midline sagittal section of the brain. Further, as an alternative to acquiring 7 sagittal 2D sections through the spine with a single GRE sequence, the entire spine can now in the same time be volumetrically imaged with a dual-echo 3D sequence affording subsequent fat-water decomposition. As an example of how fat-water decomposition techniques can be utilized to help identify spinal structures (or other regions of interest in a medical diagnostic image), consider that, with fat-water decomposition techniques, candidate structures of interest can be further classified by % water (W/W+F) signal×100%. Intervetebral discs, brain, cord, and surrounding CSF should approach 100% water signal whereas body fat should approach 0% water signal and bone marrow within skull and vertebrae should have an intermediate % water allowing improved tissue characterization over the initially described single echo out-of-phase GRE technique (as shown in
The illustrative algorithm above may be modified to enhance robustness over a wider range of populations, modalities and scan sequences. For example, rather than initially searching for discontinuous structures, (intervertebral discs or vertebral bodies) a MRI algorithm might instead search for the continuous relatively midline and predominately vertically oriented tube-like spinal cord extending from the brainstem to the conus and ending at approximately the L1 level; and a CT-optimized algorithm might search for the non-bone containing spinal canal extending from the foramen magnum (large opening at the skull through which the brainstem passes as it transitions to the cervical cord) to the sacrum. For Mill, the algorithm could continue caudal from the conus by searching for the fluid-filled spinal canal extending to the sacrum. A region growing algorithm such as “regiongrow” (MATLAB) might be employed starting from the readily identifiable brainstem (with MRI) or foramen magnum (with CT) and followed caudal to the sacrum. The idea is to start with a set of “seed” points (which may be automatically placed) and from these grow regions by appending to each seed neighboring voxels with predetermined properties similar to the seed.
Once the course of the spinal cord or canal is identified the search for intervertebral discs or vertebral bodies could be favorably constrained to the appropriate region immediately anterior to the cord/canal and the orientation of the putative intervertebral discs and bodies could be constrained so that they are nearly perpendicular to the long axis of the cord/canal at each level.
Alternatively or additionally, the algorithm could iteratively search for and identify structures from the head caudally to the sacrum. In
After auto-locating the tip of the clivus, the algorithm could then search for C2 (dens and body) situated in very close proximity beneath the clival tip. After identifying and perhaps outlining the C2 body, the algorithm could search for the adjacent disc (C2-3) or vertebra (C3) and continue this iterative process caudally until the sacrum is identified. As each vertebra and/or disc is consecutively identified, search parameters for the next level can be modified based on the size and orientation of previously identified structures (vertebrae and discs).
As disclosed for MRI and CT of the brain in the cross-referenced application, direct scanner integration and related algorithms for computer assisted diagnosis could eventually be used in “real-time” automated spine image analysis and iterative scan prescriptions. For example, with MRI, optimally angled axial oblique sequencing could be auto-prescribed through all interspaces or those discs demonstrating abnormal morphology or signal characteristics on the ASSIST or subsequent sagittal sequences. Specific Absorption Rate (SAR) limitations set by manufactures, FDA, or similar international regulatory bodies to reduce body or body part heating may also be mitigated by an algorithm to optimize the order of auto-prescriptions (regions to be covered) so as to allow time for cooling of one region while scanning another. By streamlining and converting Matlab code to C++, algorithm processing time might be significantly shortened. Coupled with an integrated head and spine array coil, rapid computer automated iterative prescription and analysis of the entire neuro-axis may be possible.
Alternatively, or additionally, 3-D sagittal spine sequencing may be obtained. Thus, a single 3-D sequence could be substituted for multiple-multi-planar 2-D sequences, which, in the aggregate, are far more time consuming. Using this approach, the volumetric MRI data set or a volumetric x-ray (e.g., CT) data set could be labeled and then automatically reformatted in optimized planes for visualization of anatomy and pathology. Typically, the three chosen orthogonal planes would be oriented relative to the patient's spinal axis as derived with ASSIST algorithm 100 or modification thereof, rather than conventional auto-generated sagittal, axial, and coronal planes oriented relative to the scanner.
In regards to CT or other potentially volumetric imaging techniques, after acquisition of the initial volumetric data set, the ASSIST computer algorithm could be used to prescribe additional sections through specific areas of interest, with different imaging parameters (e.g. kVp, mAs, slice thickness or helical acquisition technique) or in a dynamic mode post contrast administration. Moreover, ASSIST may be used to optimize the reconstruction algorithms (block 6) or field-of-view (FOV) utilized after acquisition of the volumetric data set. For example, in relation to spine imaging, FOV centering or size may be adjusted so as to focus on the spinal elements. A bone-optimized algorithm could be chosen for sections containing bone (e.g. axial oblique sections through the vertebral bodies) and a soft-tissue algorithm could be chosen for sections containing soft tissues (e.g., discs, or the brain as shown in
Another use for detection and labeling algorithms such as described above is the automated production of reports. For example, at the present time, a neuroradiologist examining a picture of a patient's neuro axis can write notes such as that the canal is stenotic (tight). However, utilizing the algorithms described above, a computer can automatically identify various features of the neuro axis, and then write summary reports regarding those features which are much more detailed and clear than is currently possible. For example, the algorithms described above could identify the location of the canal, and produce a report which includes both the precise canal volume and an indication of its relationship to the average canal volume in terms of standard deviations. Similar information could be produced and expressed for any portion of, or structure in, the neuro axis.
In conclusion, the entire spine can be effectively surveyed with sub-millimeter in-plane resolution MRI or volumetric multi-detector CT in less than 1 minute. All C-T-L vertebrae and discs can be readily identified and definitively numbered by visual inspection or semi-automated computer algorithm. We advocate ASSIST for all thoracic and lumbar spine MRI studies, as well as all MRI or volumetric CT studies which include the entire cervical, thoracic, and lumbar spine. This rapid technique should facilitate accurate vertebral numbering, improve patient care, and reduce the risk of surgical misadventure. The ASSIST computer algorithm may be modified to provide iterative patient-optimized MRI or CT scan prescriptions through regions of suspected pathology as well as optimized multiplanar reconstructions, display and labeling of volumetric MRI or CT data sets.
Yet a further technique which can be beneficially applied to the creation, presentation and processing of images is the utilization of a skin/surface marking system to provide spatial reference information. With or without internal structures labeled, there are a number of advantages to providing an external skin/surface marking system. There are three major configurations of the device as follows: (1) a point localizer, (2) cross-shaped localizer grid, and (3) full planar localizer grid/phantom.
In one version, the point localizer 500 of
In the second major detectable region configuration, a unique cross shaped grouping of prefilled or fillable tubing is utilized as a grid for cross sectional imaging with the number and position of tubes imaged in the axial or sagittal planes corresponding respectively to the slice's z or y distance from the center. In this configuration, each tubing might be constructed as a different section of the detectable region. For planar radiography, a flexible radiopaque ruled cross shaped grid can be employed as a detectable region. Both grids are removable from similarly ruled cross shaped adhesive strips after patient application and imaging.
As a third example of a detectable region, a unique essentially planar grid/phantom is described which may be of flexible construction and reversibly affixed to an adhesive/plastic sheet with corresponding grid pattern for skin marking and to serve as a sterile interface between patient and reusable grid/phantom. The grid/phantom may also be directly adherent to the skin for guided aspiration or biopsy with the cross sectionally resolvable spatial reference in place. A diagonally oriented prefilled or finable tube overlies a grid like lattice of regularly spaced tubing so that slice location and thickness can be readily determined in the axial and sagittal planes. Additionally, the spatial accuracy of the imaging modality could be assessed and, if individual tubes are filled with different solutions, as might be the case in instances where the different tubes are different sections, multipurpose references for MR/CT, and nuclear imaging could be achieved. Furthermore, if the tubing is surrounded by a perfluorocarbon or other uniform substance without magnetic susceptibility, MR imaging could be improved by reducing skin/air susceptibility and motion artifact. Additionally, the grid/phantom could be incorporated in routinely utilized pads and binding devices with or without skin adhesion and marking.
Returning to the drawings,
The tube lumen can be filled with an appropriate concentration of an iodinated contrast agent for optimal CT visualization doped with a paramagnetic relaxant such as CuSO4, MgSO4, or GdDTPA to maximize MRI signal via preferential T1 shortening. Alternatively, the tubing itself may be optimally radiopaque for CT, obviating the iodinated contrast. If desired, for optical imaging, thermography, light based 3D space tracking, or improved visibility in a darkened environment, one could add an activatable chemiluminescent mixture whose critical reagents are separated by a membrane readily breached by external force applied to the ring.
A slightly redundant backing 512 is provided for the adhesive side of the fabric 511 to facilitate peeling (
A port 515 may be directly integrated into the tubular ring 510 as in
To increase spatial reference capability and allow multiple localizers to be discriminated from each other, the detectable regions such as the ring and underlying fabric marking may be modified as in
Multiple localizers may also be dispensed on a single sheet rather than individually packaged.
Embodiments of a prefilled or Tillable cross shaped localizer grid 600 are illustrated in
The cross-shaped grid of tubing is reversibly affixed to a medical grade adhesive fabric 626 with corresponding markings and backing. The fabric 626 is illustrated in
To add further utility and integration with the previously described point localizers, a modified ring may serve as the hub of the cross-shaped grid with associated modification of the limbs as illustrated in
For planar radiography, the detectable region might comprise a cross-shaped grid of radiopaque (e.g. lead or aluminum) dots at 1 cm intervals interposed by 5 cm spaced dashes (
In a planar localizer grid/phantom 700, such as shown in
The GP may be reversibly affixed to an adhesive/plastic sheet 713 with a corresponding graph paper-like grid for skin marking and to serve as a sterile interface between the patient and GP. Perforations 733 may be placed in the sheet shown in
The square GP can preferably be constructed to have outer dimensions equal to a multiple of 10 cm (e.g. 30 cm as illustrated) to allow for simple computation if GPs are placed side to side for expanded surface coverage. Adapters could be provided to ensure secure and precise positioning of adjacent GPs either in plane or at right angles to each other. The GPs can be flexible or rigid in construction and be utilized with or without skin adhesion and marking.
Tubes may be filled uniformly or with a variety of known solutions having different imaging properties to serve as multipurpose references. For the latter, the 5 cm spaced tubes and slice indicator may be filled with the same optimized solution as previously described, while each set of 4 intervening tubes could be filled with different solutions in similar sequence. In this fashion, identical series of 5 reference solutions would repeat every 5 cm, allowing intra-slice signal homogeneity to be assessed as well. If 9 rather than 5 different solutions are desired, sequences could instead be repeated every 10 cm. For MRI, the central tubes may also be surrounded by an oil/lipid within a larger lumen tube to serve as both a lipid signal reference and allow for measurement of the fat/water spatial chemical shift. Furthermore, if the GP tubing is surrounded by a perfluorocarbon or other substance without magnetic susceptibility, MR imaging could be improved by reducing skin/air susceptibility artifact and dampening motion. The GP may also be incorporated into a variety of nonmodality specific pads (including the ubiquitous scanner table pad(s)), binders, compression plates, biopsy grids and assorted stereotaxic devices.
Two additional variations are now described, potentially replacing the somewhat complex cross design (
In the first instance, rather than packaging the ring localizers in a sheet as illustrated in
In the second instance, by slightly modifying the square design illustrated in
Cross sectional distance from isocenter would be read by adding the distance determined by each square's diagonally oriented slice indicator to 10 times the numberline position of the largest diameter tube.
Alternatively, the caliber of all tubes could be kept constant and instead an additional diagonal indicator tube passing through isocenter added for each elongated axis (vertical with slope of 10 and horizontal with slope of 1/10). Cross-sectional distance from isocenter would then be determined by looking at the position of both the additional and original diagonal indicator tubes in reference to the cross sectionally-created number line.
It should also be noted that localizer grids similar to those illustrated in
It should be further noted that applications consistent with the present invention may be modified to include a sheath for and inclusion of a flexible array MRI surface coil. Positioned vertically, this device could be closely applied to the entire cervical, thoracic, and lumbar-sacral spine. Additionally, the quantity of tubes which need to be filled in the planar configuration to uniquely denote cross-sectional positioning, has been substantially reduced from the original embodiment.
Phased array surface coils significantly increase signal to noise in MRI and are commonly employed for spine imaging. With rigid coils, patients can only be effectively scanned in the supine position, lying with the back against the coil. This results in signal drop-off in regions where the spine/back is not in close proximity to the planar coil, particularly the lumbar and cervical lordotic regions. By contrast, a flexible array surface coil would reduce the signal drop-off and allow patients to be scanned in any position. The prone position for example, would facilitate interventional spine procedures that could not be performed with the patient supine. Patients could also be more readily scanned in flexion or extension; or with traction compression devices. Current surface coils also lack an integrated spatial reference and skin marking system. Inclusion of the proposed spatial reference and skin marking system would facilitate multi-modality image fusion and registration as well as the performance of diagnostic or therapeutic spine procedures, such as biopsies, vertebroplasty, or XRT. Alternatively, if a coil is not constructed in a flexible fashion, it could be constructed in a contoured fashion so as to follow the natural curves of a patient's body, though this is not as preferred as a flexible array coil.
A grid-localizer sheath overlaying a coil could be adhered to the patient's spine. Tubing could be filled with a MRI readily-visible solution such as water doped with CuSO4. The grid itself would typically be 10 cm wide and 70-90 cm in length to cover the entire spine. An attached clear plastic sheath would allow introduction of a flexible array coil such as illustrated in
The configuration of tubing would allow unambiguous determination of the MRI scan plane (axial or sagittal) in reference to the patient's back/skin surface. The number of thin caliber tubes could denote the distance from (0, 0) in multiples of 10 cm as illustrated in
Using the described reference/marking system affixed to the patient's back, a diagnostic or therapeutic procedure could be performed under direct MR guidance. Alternatively, with a corresponding radiopaque grid affixed to patient's back, the patient could be taken out of MRI scanner and have the procedure done with CT or fluoroscopic guidance. In either case, procedures could be performed by hand or with a robotic arm.
In regards to MRI-guided intervention: the spine array coil, tubular grid, and grid imprinted/affixed sheath may be designed so that the tubular grid and/or coil can be independently repositioned within the sheath to allow for unencumbered instrument insertion into the desired spine region.
The aforementioned array coil—localizer grid or parts thereof in contact with the surgical site may be designed so that they are disposable or reusable after appropriate sterilization.
It should be noted that all aforementioned localizers may be made with non-liquid CT and/or MRI-visible materials rather than liquid-fillable tubing. Moreover, rather than actual tubing, corresponding virtual spaces imbedded in plastic may be advantageously employed so they can be subsequently insufflated/filled with an imaging modality visible gas or liquid.
Algorithms characterizing discs, vertebrae, and other spinal structures. Once identified with ASSIST, spinal structures can be further auto-characterized in terms of their morphology and tissue characteristics; and subsequently compared to population or individual norms to determine which structures are abnormal or in need of further imaging. Such characterization and comparison could be used in the generation of reports as described above. Labeling could be used to correlate available image data, and facilitate multi-parametric (spectral) analysis of the identified structures (regions-of-interest) to evaluate tissue characteristics. The technique can facilitate the automated scan prescriptions, real-time lesion detection, and examination tailoring such as discussed previously.
Advances in MRI to include the clinical implementation of phased array-coils including multi-channel combined head and spine coils and parallel sensitivity encoded imaging when combined with the algorithms described herein offer the potential for time and cost effective non-invasive holistic screening and detailed assessment of neuro-axis pathology, to include stroke and back pain—both leading causes of disability in the U.S.
Based on the teachings of U.S. provisional patent application No. 60/552,332, Weiss 2003, and Weiss 2004, Talairach referenced axial oblique diffusion-weighted images (DWI), whether prescribed by a technologist or a computer, are obtained following the initial roll and yaw corrected sagittal T2 sequence. If computer image analysis of the initial DWI sequence suggests regions of acute infarction, the basic brain protocol may be streamlined and modified to include MR angiography and perfusion sequencing. This respectively permits evaluation of the underlying vascular lesion and the detection of potential perfusion/diffusion mismatches directing emergent neuro-vascular intervention. Stroke is the leading cause of disability in this country. Because the time to emergent therapy strongly inversely correlates with morbidity and mortality, the successful implementation of the proposed computer algorithms may significantly improve patient outcome. [0111] Automated MRI Detection, Analysis, and Guided-Intervention of Spine Pathology involving vertebrae (fractures, osteoporosis and related fracture risk malignancy, endplate degeneration—Modic Type I, II, and III changes, facet degeneration) intervertebral discs (degeneration and herniation), spinal canal (spinal stenosis, extramedullary lesions, cord lesions or tethering) and nerve roots (compression or neoplasm) Spine pathology is a leading cause of morbidity and disability in this country. The techniques described herein will improve detection and assessment of disco-vertebral degeneration, osteoporotic and pathologic compression fractures-all potential underlying causes of ubiquitous back/neck pain in this country.
Using advanced MR imaging techniques, the entire spinal axis can be interrogated in the sagittal plane in less than a minute. With this screening data, the vertebral bodies and inter-vertebral discs can be identified and subsequently analyzed with software implementing the algorithms described herein. Based on this initial assessment, regions of suspected pathology to include vertebral fractures and disc herniations, could be further interrogated with more dedicated computer driven prescriptions to confirm and better characterize pathology. If for example, a fracture is identified, additional multiparametric sequencing through the involved vertebrae could be obtained to determine whether the fracture was related to osteoporosis or underlying malignancy. In some embodiments, the screening data might be obtained by the use of a scout scan, or initial scan. In embodiments where a scout scan is used, regions which were further interrogated would be referred to as regions subjected to detailed scans, that is scans which obtain information not available from the scout scan.
Automated Spine MRI Assessment and Guided Intervention: The techniques described herein can be used to improve current spine screening, assessment, surveillance, and therapeutic intervention. As is known to those of skill in the art, MRI-derived parameters hold promise for improved fracture risk prediction and evaluation of degenerative spine disease (see Wehrli F W, Hopkins J A, Hwang S N, Song H K, Snyder P J, Haddad J G. Cross-sectional study of osteopenia with quantitative MR imaging and bone densitometry. Radiology 2000; 217:527-38). Unfortunately, MRI has been too time intensive and costly to justify as an osteoporosis-screening instrument. It also has lacked accurate parametric quantification of key markers related to spine degeneration and back pain. These shortcomings can be addressed by integrating MRI-derived parameters of bone quality with a novel rapid high-resolution total spine screen (ASSIST).
In particular, the ASSIST technique described above can be integrated with fast dual GRE (e.g., as described in: Ma J. Breath-hold water and fat imaging using a dual-echo two-point Dixon technique with an efficient and robust phase-correction algorithm. Magn Reson Med 2004; 52:415-419), fast spin-echo and GRASE (Gradient and Spin-Echo Imaging) versions of an imaging technique termed iterative decomposition of water and fat with echo asymmetric and least square estimation (IDEAL). (
This integration promotes patient care by providing improved risk assessment, identification and characterization of vertebral fractures; improved assessment of spine degeneration-back pain; and metastatic disease; and improved image-guided therapeutic management as well as facilitate related image-guided procedures in conjunction with a MRI-compatible robotic arm and aforementioned interventional spine array coil. The technique of integrating MRI Automated Spine Survey Iterative Scan Technique (ASSIST) can be used to provide quantitative assessment of subendplate degenerative/stress related Modic Type 1, 2, and 3 changes. Additionally, the integrated MRI Automated Spine Survey Iterative Scan Technique (ASSIST) can be used to provide quantitative assessment of fatty bone marrow replacement for early detection of metastatic disease in the spine and differentiation of osteoporotic from pathologic fractures. Further, ASSIST adaptations will enable real-time MRI-guided vertebroplasty, biopsy, and epidural injections.
A further technique which can be applied with beneficial effects in the presentation of images is the combination of multiple images in such a manner as to obtain a single resulting image. For example, in the case of a CT scan of a patient's head, it is possible to obtain a image which accurately depicts details of both the patient's skull and brain by combining two co-registered images derived from the same CT source data but created with two different convolution kernels: a first convolution kernel with a high pass filter algorithm for the skull, and a second convolution kernel with a low pass filter algorithm for the brain. This combination might take place by beginning with two images (one created with the first convolution kernel and one created with the second convolution kernel) and, in the image created with the second convolution kernel, replacing pixels having HU values outside the range of −150 and 150 with corresponding pixels from the image created with the first convolution kernel, thereby creating a third image which displays the details from the skull (captured using the first convolution kernel) and from the brain (captured using the second convolution kernel). An illustration of this is provided in
It should be understood that the description of image combination set forth above is not intended to be an exhaustive recitation of techniques which could be used to combine medical diagnostic or other types of images. For example, while the above discussion focused on the combination of images for the head, the same techniques can be successfully applied to combine images taken of the lungs and spine, and could be applied to other areas of the body as well. Similarly, while the discussion above disclosed replacement of pixels with HU values outside the range of −150 and 150, other ranges could be used as well. For example, ranges with a minimum value of −150+/−75 HU, and maximum values of 150+/−75 HU could also be used (e.g., from −200 to 100 HU), for example, depending on the material depicted (e.g., type of tissues) in the images being imaged, and the particular hardware used to obtain the images. Also, while the discussion above focused on the combination of images created with different convolution kernels, combining images is not limited to combining images created with different convolution kernels. For example, when a CT machine obtains data sets at different energy levels, the data sets obtained at different energy levels can be combined as well. For instance, if the CT machine obtains data at 140 and 80 kvp levels, then those data sets can be combined at a factor of 0.70/0.30. It is also possible that more than two images could be combined. For example, in some cases there might be images created using three (or more) different reconstruction kernels, in which case a composite image could be created the pixels from the first image having a first set of values, with pixels from a second image having a second set of values, and pixels from a third image having a third set of values. It is also possible that images could be combined simply by averaging pixel values in a series of adjacent slices to get a composite slice reflecting data collected from the entire scan area.
Of course, the approaches described above are not mutually exclusive. For example, in one application of image combination, volumetric dual energy CTs obtained from hard palate to vertex at 80 and 140 kvp, approximately 15 cm coverage, were reconstructed in the axial plane at 0.6 mm intervals using both high pass and low pass convolution algorithms. A hybrid convolution kernel (HCK) which combined high and low pass algorithms as described above was applied to reduce data sets by a factor of two and the 80 and 140 kVp data sets were combined at a 0.30/0.70 ratio to further reduce data sets by a factor of two and optimize contrast vs SN. The reduced volumetric data set was then semi-auto or auto-reformatted to approximate tri-planar Talairach space manually, identifying two similar contralateral points (e.g. round window basal turn of cochlea) on source axial images, generating the sagittal plane equidistant from these points, identifying the hard palate on this roll and yaw corrected midline sagittal image, and angling 12 degrees; or having the computer auto-identify the sagittal plane optimally bisecting the volumetric data set and auto-identifying the hard palate. Similarly, 2 mm coronal maxillofacial CTs were auto-generated perpendicular to the hard palate. Using volumetric dual energy HCK data sets stored in the computer, ROIs placed on the reconstructed data sets could then be deconvolved and displayed as 2-D HU scatter plots, the two axes being 80 and 140 kVp, with each point corresponding to a voxel within the selected 3D ROI. The ROI itself could be determined in a number of manners, such as by the selection of a single voxel in a presented image, the selection of a region in a presented image, the selection of a voxel and its closest neighbors in an image, the selection of a voxel and its closest neighbors and their closest neighbors, etc. . . . Conversely, in some implementations a scatter plot presentation interface could be designed so that, in response to a user selecting a point on the scatter plot, the computer would interactively display its origin on orthogonal triplanar reformations passing through this voxel. Examples of scatter plots which could be shown using this approach are shown in
While the technologies described were initially developed for medical imaging, they may also be applied to industrial or other usage. For example, analysis of dual-energy x-rays or CT scans of inanimate objects, such as luggage, may be enhanced by the HCK and deconvolvable 2-D scatter plot technology described herein. For example, using the combination of images created with different convolution kernels in the context of luggage screening, there might be a first convolution kernel which is particularly adapted to detecting and displaying dense objects, such as firearms, while there could be a second convolution kernel which is particularly adapted to detecting and displaying organic material, such as might be found in an explosive. By the adaptation of techniques such as described above, the process of screening luggage for weapons and explosives could be made more efficient, thereby improving the security and convenience of air travel. Of course, it should be understood that this example is provided as an illustration only, and that other approaches described herein (e.g., two dimensional scatterplots with mapping back to original slices) could also be applied in areas outside of medical imaging. Thus, in no event should the examples and explanations provided be treated as implying limitations on the scope of any claims included in this or any related applications.
The present invention may be embodied in other specific forms without departing from its spirit or characteristics. The described embodiments are to be considered in all respects only as illustrative and not restrictive. The scope of the invention is, therefore, indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.
This application is a continuation of U.S. application Ser. No. 15/694,940 filed Sep. 4, 2017, which is a continuation-in-part of U.S. application Ser. No. 14/948,209 filed Nov. 20, 2015 now U.S. Pat. No. 9,754,369 issued on Sep. 5, 2017, which is a continuation of U.S. application Ser. No. 13/848,638 filed Mar. 21, 2013 now U.S. Pat. No. 9,196,035 issued Nov. 24, 2015, which is a continuation of U.S. application Ser. No. 12/155,803 filed Jun. 10, 2008 now U.S. Pat. No. 8,805,042 issued on Aug. 12, 2014, and is also a continuation-in-part of U.S. application Ser. No. 13/136,165 filed Jul. 25, 2011 now U.S. Pat. No. 8,457,377 issued Jun. 4, 2013, which is a continuation-in-part of U.S. application Ser. No. 10/598,764 filed Sep. 11, 2006 now U.S. Pat. No. 8,014,575 issued Sep. 6, 2011, which is a continuation-in-part of United States PCT Application No. PCT/US05/08311 filed Mar. 11, 2005, which claims the benefit of U.S. Provisional Application No. 60/552,332 filed Mar. 11, 2004. This application is also a continuation of U.S. application Ser. No. 14/948,209 filed Nov. 20, 2015 now U.S. Pat. No. 9,754,369 issued on Sep. 6, 2017, which is a continuation of abandoned U.S. application Ser. No. 14/288,080 filed May 27, 2014, which is a continuation of U.S. application Ser. No. 12/155,803 filed Jun. 10, 2008 now U.S. Pat. No. 8,805,042 issued on Aug. 12, 2014, all of which are hereby incorporated herein by reference in their entirety.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14288080 | May 2014 | US |
| Child | 16292316 | US | |
| Parent | 15694940 | Sep 2017 | US |
| Child | 14288080 | US | |
| Parent | 12155803 | Jun 2008 | US |
| Child | 14288080 | US |
