Correlative image registration

Correlative image registration

Correlative Image Registration D.A. Weber and M. Ivanovic Image registration in nuclear medicine and radiology refers to the spatial matching or mergi...

4MB Sizes 5 Downloads 91 Views

Correlative Image Registration D.A. Weber and M. Ivanovic Image registration in nuclear medicine and radiology refers to the spatial matching or merging of two or more images from the same or different imaging modalities. The coordinates of the corresponding picture elements (pixels) from different images are transformed to align and equate their positions and spatial coordinates. Correlative image registration is a more restrictive term that applies to the matching of spatial coordinates of images coming from different imaging modalities, The registration of correlative images provides a useful approach to combine the best sensitivities and specificities of complementary procedures to detect, locate, monitor, and measure pathological and other physical changes, Here we review the registra-

tion of images from nuclear medicine (single-photon emission computed tomography, positron emission tomography and planar imaging) with those from other imaging modalities (magnetic resonance imaging, computed tomography, digital subtraction angiography and ultrasound) to closely correlate changes in metabolism, blood flow, receptor density, and other functional measurements with regional anatomy and morphological changes. The types of image registration applications, techniques, and terminology associated with image registration and examples of applications are presented. Copyright 9 1994 by W.B. Saunders Company

HE REGISTRATION of images from several modalities can provide a highly advantageous approach to identifying, correlating, and quantifying regional changes in anatomy and function. The interpretation and review of single-photon emission computed tomography (SPECT) or positron emission tomography (PET) images registered with magnetic resonance imaging (MRI), computed tomography (CT), digital subtraction angiography (DSA), or ultrasound (US) images frequently contribute additional and new information to the workup of subjects beyond that obtained from the individual procedures. An example of this is seen in the merging of fine anatomical detail from MR images of the brain with functional PET images to provide the information needed to more accurately and reproducibly measure regional cerebral function. Comparing like images acquired on film in a side-by-side configuration has been a common means of (1) evaluating regional radiopharmaceutical uptake and clearance from serial studies, (2) evaluating the localization properties of a tracer before and after therapy, and (3) comparing nuclear medicine images with radiographs. However, the use of computer-assisted image registration techniques to size, align, and merge images from the same or different imaging modalities is a relatively recent approach that has become more practical as technology has advanced and as more imaging modalities use digital image acquisition, storage, processing, and display techniques. The image-processing procedures used to register images operate on the quantitative

data associated with these procedures. The processed registered images maintain the same quantitative format of the original images and can be stored, networked, and displayed by conventional picture archiving communication and storage (PACS) techniques. Applied to a variety of clinical and investigational problems, image registration offers a major advance in diagnostic imaging (Fig 1). Here we review the scope of correlative image registration applications that have been used to merge individual nuclear medicine images from SPECT, PET, and planar scintillation camera imaging with images from CT, MRI, other radiographic modalities, and anatomic atlases.

T

TYPES OF IMAGE REGISTRATION

The registration of nuclear medicine images with other nuclear medicine images, with images from other imaging modalities, or with images derived from a standard anatomic atlas provides the basis for merging complementary information that can yield additional information about the function, structure, or position of tissues or organs. Although image registration is the term most widely used to reference this From the Department of Radiology, University of California Davis Medical Center, Sacramento, CA. Supported in part by US DOE Contract DE-ACO276CH00016. Address reprint requests to David A. Weber, PhD, Department of Radiology, University of California Davis Medical Center, Radiology Research, FOLBII-E, 2421-45th St, Sacramento, CA 95817. Copyright 9 1994 by W.B. Saunders Company 0001-2998/94/2404-0004505. 00/0

Seminars in Nuclear Medicine, Vol XXIV, No 4 (October), 1994:pp 311-323

311

312

WEBER AND IVANOVIC

Nuclear Medicine

m~

Diagnostic Radiology

f

|

L

<-

Correlative Image Registration

Functional Images

Anatomical Images Function/Anatomy

Radiation as.

,

v "q~

Therapy

Therapy

"~

Applications

procedure, the terms image coregistration, superimposition, geometric correlation, matching, or fusion, are also used interchangeably and are found in the nuclear medicine and radiology literature. 1-7Types of image registration applications range from the comparison and correlation of regional function observed in serial images of a single radiopharmaceutical to the accurate identification of anatomic regions with MRI or CT for use in conjunction with the investigation of functional measurements on SPECT or PET images. Specifically, these applications include: (1) serial SPECT, PET, or scintillation camera images of the same organ or region of the body to observe regional clearance of a single radiopharmaceutical or repeat images of the same radiopharmaceutical for imaging before and after surgery or other treatment to evaluate the effects of therapy; (2) SPECT, PET, or scintillation camera images of an organ with two different radiopharmaceuticals to compare different functional assays; (3) a comparison of SPECT with PET functional measurements designed to map the same or similar function, such as images of blood flow [99mTC hexamethyl propylene amine oxime ([99mTc]HMPAO) and H2150] or metabolism [18F-2-fluoro-2deoxyglucose (18F FDG) and 1231iodo-oL-methyl-

Metabolism

Diagnosis

Fig 1. Correlative image registration. Schematic representation of registration of high-resolution images of function and anatomy from nuclear medicine and radiology procedures to correlate regional changes in function and anatomy.

tyrosine ([123I]IMT)] or the evaluation of complementary functional assays such as metabolism and receptor density [18F FDG, 123I iodobenzamide ([123I] IBZM)]; (4) functional imaging with SPECT or PET and anatomical imaging with MRI or CT to improve the anatomic resolution of measuring function; (5) SPECT or PET images with a composite normal control image or anatomic atlas to structurally localize and detect alterations in function; and (6) functional and/or anatomic images with three-dimensional (3-D) treatment plans for dosimetry and treatment planning. Although each of these image registration applications place similar demands on image processing, correlative image registration places the greatest demands on data processing as a result of variations in image orientation and the variations in image data format presented by the different imaging modalities. 7,8 Emphasis here is placed on discussing the 3-D nuclear medicine imaging applications that appear to have received the greatest benefit from correlative image registration. The majority of these applications involve registration of SPECT or PET with MRI or CT with anatomic atlases, and with images of control subjects. The choice between CT or MRI examination for the ana-

CORRELATIVE IMAGE REGISTRATION

tomic images depends in large part on the objectives of the registration procedure. If the intent is to perform dose distribution calculations for therapy, CT is usually the examination of choice. CT maps x-ray photon linear attenuation coefficients and provides high-resolution images for identifying and outlining bone and soft-tissue structures. The mapping of the attenuation coefficients provides the basis for deriving the electron density of structures needed for the calculation of radiation dose distributions. On the other hand, when the main objective of the procedure is to improve the definition of soft-tissue structures on the functional examination, MR is frequently selected as a result of its better contrast of soft-tissue structures. The registration of MR images with PET or SPECT is particularly useful for characterizing the gray and white matter structures and lesions in the brain. The relaxation time images of soft-tissue structures provide highcontrast images that delineate organs, tumor, and other structures and can be used to develop detailed mappings of surface contours and textures of tissue structures. In these applications, image registration is used (1) to correlate complementary or competitive examinations of function; (2) to obtain anatomical information to more accurately define the location of regional changes in function on SPECT, PET, or other examinations; and (3) to evaluate structural changes in tissues and organs. Frequently involving paired image sets of differing matrix size, format, and patient positioning, correlative image registration requires a complex pixel coordinate transformation to merge the two images. Typically these may include translation, rotation, and isotropic and/or anisotropic scaling of pixel coordinates. IMAGE REGISTRATION TECHNIQUES

Many different image processing techniques are applied to registering images from the various imaging modalities used in nuclear medicine and radiology. Regardless of the method used, image registration should map each point in one image onto the corresponding point in the second image. This is illustrated schematically in Fig 2. The coordinates (x,y) and (x',y') of the two corresponding points, P and P', from the two digital images, I and I', can be related

313

by the equation (x,y) = T(x', y', p). T is the registration (geometric mapping) function and p is a set or vector of unknown parameters of the mapping function. The image registration process can be divided into three general steps: 1. Selecting corresponding points (control points or control features) in I and I'. Only a limited number of control-points is needed to reconstruct the mapping function because it can be assumed that the mapping function is at least piecewise smooth. 2. Determining the parameters p (translations, rotations, scaling, etc.) of the mapping function T. 3. Computing (resampling) I"(x',y') = I(T(x',y')) in order to register I with respect to I'. In practice, methods used for image registration range from the use of positioning devices to align the patient and restrict movement to the use of relatively complex mathematical algorithms that accurately register sets of imaging data with and without use of internal or external markers. In the former, special holders, frames, molds, masks, markers, or laser alignment devices are used to position the patients as accurately as possible in the same orientation. The methods rely on identical patient positioning and orientation for image acquisition in all imaging modalities and assume that reconstructed and viewing planes are identical. 9al Methods designed to have less restrictions on image acquisition variables often use positioning devices and/or extrinsic markers in conjunction with computation methods to register images. 2,3,6,12-28 Extrinsic markers and/or patient alignment devices are sometimes used to identify related reconstruction planes (in axial direction) to reduce the 3-D registration problem to a computationally less intensive 2-D problem. 1316,23 Other image registration methods have been developed that are independent of acquisition variables. 4,5,29-52 The types of transformations used in image registration algorithms depend on the degree of distortions and variability in geometric structures between the matching images. The most commonly used are rigid, affine, and polynomial (curved) transformations. 7,8 Rigid transformations are composed of translations, rotations,

314

WEBER AND IVANOVIC

]>

>

reflection, and linear scaling. They are applied when there are no spatial distortions or variability in geometric structures between images. Affine transformations include uniform and nonuniform scaling and shearing in addition to rigid transformations. Polynomial (or curved) transformations are used to remove the spatial distortions between images caused by differences in acquisition parameters and variability in geometric structures. This technique is often referred to as warping. The majority of algorithms used for medical image registration are based on one or more Of the following four approaches: (1) matching anatomical and/or externally placed landmarks, (2) matching surfaces (surface fitting method), (3) matching spatial moments (centroid and principal axes), and (4) maximizing the value of the correlation coefficient between the two images.

Fig 2. Principle of image registration: corresponding points P and P' selected on images I and I' determine the mapping function T that applied to image I' (T(I']) maps one image onto another.

MatchingLandmarks The landmarks are used to identify corresponding points on correlated images and to determine the parameters of the mapping function (type of coordinate transformations) for image matching. The use of landmarks or fiducials as reference sites for image registration is widely used in radiation therapy and stereotactic surgery, although it is not limited to these applications. Extrinsic, intrinsic, or a combination of both landmarks are used. Extrinsic markers are applied externally (eg, head frames, molds, and markers added to the skin or f~ame) and internally (eg, contrast agents or radiopharmaceuticals administered to the subject). The use of the 99mTc methylene diphosphate ([99mTc]MDP) bone Scan to define landmarks for registration of nlln antibody images with CT is an example of internally applied extrinsic markers. 2 Intrinsic markers include anatomic

315

CORRELATIVE IMAGE REGISTRATION

landmarks, pixel intensities, and geometrical features (skin or bone surfaces, ventricles, rib cage, etc). The complexity of the intrinsic landmarks varies widely and may consist of points, lines, curves, planes, surfaces, and volumes. 12'16'31'35'46 When one of the matching image modalities is CT, MRI, or x:ray imaging, the internal organ structures can be used as the anatomical landmarks and extrinsic markers are then applied only on the functional imaging modality (PET, SPECT, planar images, etc). la Regardless of the type of fiducials, the detection of the markers can be accomplished manually, semiautomatically, or automatically. The majority of methods are semiautomatic, although the degree of interaction can vary considerably. Although the repeated precise application of external markers and/or identifying internal anatomical landmarks requires considerable operator expertise, landmark-based techniques are the most frequently applied to matching multimodality extracranial images. 12,13,16,46,47

Surface Fitting Method The surface fitting (SFIT) method (known as the head and hat method) for 3-D image correlation is based on minimizing the root meansquare (RMS) distance between the surfaces of an anatomical structure that is visible on all modalities. 29,3~ Tlae surfaces are obtained by outlining contours on the serial slices of each image set, either manually or by using a semiautomatic edge detection algorithm. 41,42The head represents the surface model extracted from the higher resolution scan, and the hat represents a set of independent points from the surface of lower resolution scan. The RMS distance between two surfaces is expressed as a function of independent variables that include rigid body translation and rotation and linear scaling along the three orthogonal axes. The hat is fitted on the head using an iterative nonlinear leastsquares search that minimizes the RMS distance between them. The algorithm allows operator intervention to prealign the surfaces to prevent the search from converging to a local minimum instead of to the global minimum and to readjust transformation parameters to speed up convergence. Currently this application is limited to regions that satisfy rigid-body conditions. The primary application thus far has been

in cranial imaging. The best results are achieved by matching PET with c T or MRI when transmission images are used. The major weakness of this method is its sensitivity to the local deformations in surface contours. Using the edge of the brain instead of the skull for surface fitting limits the application to blood flow and metabolism studies. The accuracy that can be achieved with this method is On the order Of one to two image pixels.

Matching Spatial Moments Image registration b y matching spatial moments is based on aligning centroid and principal axes (PAX). The PAX are orthogonal axes about which the moments of inertia are minimum. The registration consists of calculating acentroid and three PAX for each study and then translating the centroid and rotating it to align the PAX and scaling. The PAX algorithms either treat the objects as surface contours 32 or as filled interiors (volumes). 33 The latter improves the accuracy of registration to approximately 1 mml The PAX method is an analytical method and does not depend on convergence. It is much faster than the SFIT method, but it is not as robust against missing data. The method is sensitive to incomplete scan coverage in either of the data sets and to the differences in the surface contours because the method is based on the assumption that volumes modeled from the two scans are identical.

Cross-Correlation Methods The cross-correlation method for image alignment is based on maximizing the Value of the correlation coefficient between the two image data sets, The method is search based and the cross-c0rrelation function is recalculated after each translational and rotational correction of o n e image relative to another. The method works well with images from the same modality, for example, in registering cardia c PET attenuation scans to correct for changes in patient positioning in repeat studies. 34 It is not as useful generally for registering images from different modalities because the pixel intensity values in different modalities are usually not related,

316

WEBER AND IVANOVIC

Prospective and Retrospective Image Registration These four approaches of image registration can be divided into the two broad categories: (1) prospective image matching methods, which include special measures in the image acquisition protocol to enable registration (methods based on using alignment devices and extrinsic markers) and (2) retrospective image registration methods that do not require special patient setup procedures or adjustments in the image acquisition protocol. The retrospective nature of SFIT, PAX, cross-correlation methods, and methods based on intrinsic markers make them attractive for applications in a routine clinical practice. Unfortunately, no single method works well on all image data sets or satisfies the specific requirements of all applications. Currently, access to several different registration algorithms is necessary to use correlative image registration in routine clinical practice. APPLICATIONS OF CORRELATIVE IMAGE REGISTRATION IN NUCLEAR MEDICINE

Applications of correlative image registration address both diagnosis and therapy. The majority of applications reported in the literature have been directed toward merging functional SPECT or PET images with MR and CT images to improve anatomic mapping of functional changes and to correlate pathology and anatomy with functional abnormalities in diagnostic and investigative studies of the brain. Because familiarity with registration techniques is evolving, however, the number and types of applications are greatly expanding. Image registration is progressively playing an increasingly larger role in image interpretation in many different procedures. Table 1 lists several examples of different nuclear medicine and other radiologic imaging procedures that have benefited from correlative image registration. Major areas of application include registration studies of SPECT or PET with CT or MRI in procedures of the brain involving blood flow, metabolism, and receptor density 4,5,23,29,41,43,5~ and in procedures involving the detection of primary or recurrent tumor. 1"3,26,35T h e registration of MR with SPECT images of the brain, for example, allows the SPECT functional images of regional blood flow to be viewed and inter-

preted for the anatomic structures seen on MR. 4~ T2-weighted ~ a I MR images regiStered with HMPAO SPECT images provide the means to accurately identify alterations in regional blood flow in the thalamus, caudate nucleus, midbrain nuclei, and cortex on the HMPAO SPECT examination. Image registration is also useful in discriminating recurrent tumor. An example of this is shown in the same study. 41 In a subject who had undergone prior surgery and stereotactic interstitial radiation therapy for recurrent glioblastoma, an MRI/SPECT registration study provided the means to discriminate necrosis from recurrent tumor and was successfully used to direct surgeons to the site of recurrent disease. Registration studies involving SPECT and MR images also have been applied to evaluating comparative uptake properties of different tracers in laboratory animals.55,5a,59Here, MRI plays an important role in providing detailed images of soft-tissue structures in the dog's brain and of tumor in the rat's brain that can be used to select for regions of interest (ROIs) on the SPECT functional examinations. Recent clinical studies show that registration techniques also play an important role in improving anatomic definition of functional measurements in other regions of the body. Registration of MR and CT of the liver with SPECT measurements of [99mTC] red blood cells ([99mTc]RBCs) localization in hepatic hemangioma, for example, has led to improved sensitivity for the detection of Small tumors and the detection of tumor adjacent to or near regions of blood pool activity.35 The improved identification of softtissue structures obtained on nuclear medicine images resulting from the registration of MR or CT with SPECT or PET images also has led to the development of new techniques to improve the accuracy of in vivo quantitation of tracer uptake. In addition to the improvement resulting from better defined ROIs, correction routines have been developed that improved the accuracy of functional measurements. An example Of this is the use of registered MR images to identify ROIs and to correct for partial volume effects on activity quantitation in PET imaging of the brain. The partial volume correction derived from image registration studies has been used to quantify [11C]carfentanil activity in

317

CORRELATIVE IMAGE REGISTRATION

Table 1. Correlative Image Registration Applications Correlative Procedures Diagnosis SPECT, CT, MR (planar gamma camera images, radiographs)

PET, CT, MR

Therapy PET, SPECT, CT, DSA, MR

Type of Study

Reference

[~mTc]HMPAO and 2~ with MRI to compare regional function with anatomy in dementia, glioblastoma, and focal lesions in the brain

41

[99mTc]HMPAO divided by CT volume defect sizes in cerebral cortex has prognostic utility in predicting clinical recovery from a stroke involving the cerebral cortex [99mTc]HMPAO and CT defect size correlation shows prognostic value in watershed infarcts 201TI, [~mTc]HMPAO, [99r"Tc]MIBI, and pgmTc]DTPA with MRI provides means to investigate comparative tumor (gliosarcoma) uptake in rat's brain in vivo [~mTc]RBC blood pool images with CT and MR to detect hepatic hemangioma [l~qn]CEA-specific or OC-125 MoAb with CT to investigate recurrent tumor in the pelvis [1111n]CYT-103 MoAb with CT of abdomen and pelvis to detect sites of ovarian and colon carcinoma [1111n]eEA MoAb with CT to identify foci of uptake and to correlate tumor mass with uptake [99rnTc]MDP (gamma camera images) and radiographs to improve anatomical localization of focal metabolic abnormalities [~mTc]MIBI and MR to detect perfusion abnormalities and their effects on endocardial motion and myocardial thickening [18F]FDG with MR and CT permits quantification of physiologic variables in small normal and pathologic brain structures [leF]FDG with MR and CT to investigate trauma, epilepsy, brain stem, and white matter lesions [lSF]FDG mapping of surface metabolism on 3-D MR computer model of the brain to investigate epilepsy, encephalitis, and tumor MR with ~1Ccarfentanil used to correct partial volume effects in " C uptake in small gray matter structures [11C]fiuoromethane with CT and MR to quantify rCBF using computerized brain atlas program [lSF]FDG with MR and CT to create "anatometabolic images" to visualize visceral cancers and to direct biopsy

53

[124113F8antibody with MR and CT for radioimmunotherapy treatment planning for [125113F8or [131113F8antibody Skull radiographs, CT, and MR used to map subdural electrodes and correlated with [~SF]FDG before epilepsy surgery Registration of PET images with CT and MR (Gd-DTPA) to define target volumes in brain tumor treatment planning Registration of PET, MR, CT, and DSA images for planning stereotactic neurosurgery

Abbreviations: CEA, carcinoembryonic antigen; 0C-125,

111

54 55 35 2 3 1 17 49 50 4 43 23,56 57 26

60 45 46 21

In CEA-specific MoAb; CYT-103, 1111nlabeled immunoconjugate of MoAb

72.3.

mu-opiate receptor studies in the cerebral cortex of the brain. 23,56Here the corrections of the concentration of activity seen in the amygdala, hippocampus, temporal neocortex, and other gray matter structures by PET has been extremely useful in improving the accuracy of quantitation of receptor density in studies of cerebral atrophy, such as occurs in aging, dementia, and Alzheimer disease. In the majority of applications, image registration involves using MR or CT images of the

same subject (1) to improve the identification of anatomic structures on the nuclear medicine images for anatomically locating function and to enhance the information offered by each modality, (2) to correlate structural changes with functional measurements, and (3) to use the high-resolution anatomic images to correct and/or improve the quantitative measurement of function. Other applications include the use of both 2-D and 3-D registered correlative images to allow the calculation of the radiation

318

WEBER AND IVANOVIC

Fig 3. To assess the accuracy of landmark placement, an ROI is first drawn around the liver on the SPECT image (upper left). When displayed on the matching MR section (upper right), the ROI appears much smaller than the liver and overlaps it. After application of the warping algorithm, the ROI becomes registered with the MR image (bottom right) such that the ROI now clearly matches the liver in size and more accurately outlines its borders. (Courtesy of E.L. Kramer, G.Q. Maguire Jr, A.J. Megibow, M.E. Noz, D.P. Reddy, and J.J. Sanger, Image Fusion Research Group, New York University Medical Center.)

and surrounding tissue dose to tumor, to develop prognostic indicators, and to refine surgical planning. The registration of MR images with PET or SPECT images provides the means to accurately define active tumor volume and

Fig 4. An ROI drawn around the small intense focus of hemangioma blood pool activity and another ROi drawn around the aorta on the SPECT image (upper left) do not match well with the hyperintense hemangioma or aorta visible on the matching MRI section (upper right). After application of the warping algorithm, the ROI is registered on the MR image and now correlates well with the hemangioma and aorta visible on the MR image (lower right). (Courtesy of E.L. Kramer, G.Q. Maguire Jr, A.J. Megibow, M.E. Noz, D.P. Reddy, and J.J. Sanger, Image Fusion Research Group, New York University Medical Center.)

geometry of tumor in respect to other critical structures for treatment planning.46,6~ The same methodology is used to merge dose distribution data with images from the various contributing imaging methods to evaluate the accept-

CORRELATIVE IMAGE REGISTRATION

ability of treatment plans and to follow up the effects of therapy on structure and function. Recent SPECT/CT correlative imaging studies, for example, indicate that the size of cerebral blood flow defects offers prognostic utility in predicting clinical recovery.53,54In studies involving subjects presenting with a solitary stroke or with watershed infarctions, the clinical course of patients was significantly better in those subjects that showed a larger perfusion defect on the [99mTc]HMPAO SPECT image than on CT. Although the SPECT and CT images in these specific studies were separately processed to find the size distinction, the clinical correlation would readily lend itself to a registration routine. In surgical planning, applications of

319

image registration to simultaneous display of MR, CT, DSA, and PET images provide the means to localize tissue pathology with vasculature, other soft-tissue structures, bone, and functional measurements. 21,63 The information gained is important in depth electrode implantation, brain biopsy, and stereotactic microsurgery. Examples of a correlative image registration are shown in Figs 3 to 7. The first two examples of image registration shown in Figs 3 and 4 are taken from work performed to evaluate the diagnostic value of SPECT [99mTc]RBCimaging for identification of hepatic hemangiomas. 35 In a study requiring accurate definition of anatomic structures on SPECT images, a 2-D

Fig 5. Wire-cage representations (two different angles of view) showing the brain surface, midplane between hemispheres, and internal structures (thalamus) for a SPECT image set and MR image set. Structures are shown before (left) and after (right) the registration process. (Courtesy of G. Zubal, Department of Diagnostic Radiology, Yale University, School of Medicine.)

320

WEBER AND IVANOVIC

Fig 6. Registration of P23~MP SPECT images and MR scans provides the means to accurately assess changes in regional cerebral perfnsior~ betwem~ the icta~ al;(i mterictai s t a t e s it-, patie,]ts with il'Jtractable partial ~;eizures. {Top} Registered ictal and interictal rCBISPECT slices at the leve! of the, right parietal heleratopia (Bottom} Positive differm~ce image betweep~ iota[ and intmlctai slices normalized according to the total n u m b e r of counts ir~ the brair, {left}, positive difference image s u p e r i m p o s e d on MR slice of the anatomy (middle), and transaxlai MR sl~ce of the brain at the ievei of right parietat imterotopia (right}. (Courtesy of G. Zubai, Department of Diagnostic Radiology, Yale University, School of Medicine.}

polyllot/lial based \~,~t|ping algol i{hm w~l~ tlscd rcprojcct CT a n d / o r MR images to match ROls on corresponding SPECT images. Figure 3 shows an example of image registration in transferring an ROI drawn on a SPECT image to an MR image. Figure 4 shows the same registration technique applied to identifying and correlating the localization properties of [99mTc]RBC in tumor and in the aorta with the anatomy observed on MR. This type of registration study was critical in the evaluation of the sensitivity of the SPECT imaging technique to detect small tumors, especially in regions adjacent to blood pool activity. A schematic of a registration technique that uses internal image features in the head to register functional and anatomical 3-D images of the brain is shown in Fig 5. The internal Io

llan,'-;late,

iolatc,

alld

lc~/tm-cs Ih~d serve as liduci~ll~ m this rcgislration method include the brain surface ;Jlld easily identifiablc structural planes. An example of the use of this technique to register SPECT and MR images for the investigation of regional cerebral blood flow in intractable seizures is shown in Fig 6. Here, [23I N-isopropyl-piodoamphetamine ([~23I]IMP)SPECT slices of the brain obtained in the ictal and interictal states are presented and registered with MR images of the same subject. Individual SPECT slices for the two states are shown at the level of heterotopia in the right parietal region in Fig 6. A positive difference image obtained from the subtraction of the interictal from the ietal SPECT slice is shown alone and superimposed on a registered MRI slice. Preliminary studies in 12 patients with intractable partial seizures (8

CORRELATIVE IMAGE REGISTRATION

321

Fig 7. Transaxial (upper left) and sagittal (upper right) MR slices through a rat's brain showing a gliosarcoma encapsulated in the frontal region are matched with corresponding ["mTc]DTPA pinhole SPECT slices (lower left and lower right) using a 2-D registration algorithm based on manual identification of two corresponding points in each slice, ss,se Images were scaled to the same pixel size before matching and then rotated and translated to align corresponding points. Tumor and brain ROIs drawn on the MRI slice are overlaid on SPECT slices.

temporal lobe and 4 extratemporal lobe focus seizures) suggest that ictal-interictal subtraction images merged with the MRI slices enhance the usefulness of regional cerebral blood flow SPECT studies for the localization of seizure foci. Finally, Fig 7 shows an application of image registration using SPECT along with MRI to investigate the regional localization properties of radiopharmaceuticals in tumor and adjacent tissues in laboratory animals. The figure shows the method adopted to investigate the comparative localization properties of 2~ [99mTc]HMPAO, [99~Tc] methoxy isobotyl isonitrile([99~Tc]MIBI) [99mTc]diethylene triamine penta acetic acid ([99mTc]DTPA) and [123I]IMT in the rat's brain and tumor in vivo.55,58These and similar registration applications to investigate regional activity of [99mTc]HMPAO and [IZ3I]IMP in the dog's brain following acute doses of cocaine 59 have

significantly improved our ability to discriminate regional uptake in organs and tissues by in vivo imaging. CONCLUSION

Image registration provides the basis for obtaining additional and sometimes new information beyond that which can be readily seen or appreciated from inspection of the individual images. The most widely used correlative imaging applications have involved the registration of SPECT or PET images for measuring function, with high-resolution MR and CT images of anatomy. The merging of high-resolution functional images with anatomical or morphologic images provides a solid basis for improving the specificity and sensitivity of functional imaging to investigate small structures in an accurate and reproducible manner.

REFERENCES 1. Kramer EL, Noz ME: CT-SPECT fusion for analysis of radiolabeled antibodies: Applications in gastrointestinal and lung carcinoma. Int J Rad Appl Instrum [B] 18:27-42, 1991 2. Liehn JC, Loboguerrero A, Perault C, et al: Superimposition of computed tomography and single photon emission tomography immunoscintigraphic images in the pelvis:

Validation in patients with colorectal or ovarian carcinoma recurrence. Eur J Nucl Med 19:186-194, 1992 3. Loats H: CT and SPECT image registration and fusion for spatial localization of metastatic processes using radiolabeled monoclonals. J Nucl Med 34:562-566, 1993 4. Levin DN, Pelizzari CA, Chen GTY, et al: Retrospective geometric correlation of MR, CT, and PET images. Radiology 169:817-823, 1988

322

5. Kapouleas I, Alavi A, Alves WM, et al: Registration of thr~e~di-m~risi~dnaYMR and/rETitna-ges~ot'tlie~nman b t ~ n ~ without markers. Radiology 181:73i-739, 1991 6. Malison RT, Miller EG, Greene R, et al: ComputeraSsisted coregistration of multislice SPECT and MR brain images by fixed external fiducials. J Comput Assist Tomogr 17:952-960, 1993 7. van den Elsen PA, Pol E-JD, Viergever MA: Medical image matching--a review with classification. IEEE Eng Med Bio140:26-39, 1993 8. Brown LG: A survey of image registration techniques. Assoc Comput Mach 24:326-376, 1992 9. Miura S, Kanno I, Iida H, et al: Anatomical adjustments in brain positron emission tomography using CT images. J Comput Assist Tomogr 12:363-367, 1988 10. Bergstr6m M, Boethius J, Eriksson L, et al: Head fixation for reproducible position alignment in transmission CT and positron emission tomography. J Comput Assist Tomogr 5:136-141, 1981 11. Mazziotta JC, Phelps ME, Meadors AK, et al: Anatomical localization schemes for use in positron computed tomography using a specially designed headholder. J Cornput Assist Tomogr 6:848-853, 1982 12. Noz ME, Kramer EL, Maguire GQ, et al: An integrated approach to biodistribution radiation absorbed dose estimates. Eur J Nuc! Med 20:165-169, 1993 13. Kramer EL, Noz ME, Sanger J J, et al. CT-SPECT fusion to correlate radiolabeled monoclonal antibody uptake with abdominal CT findings. Radiology 172:861-865, 1989 14. Kessler ML: Computer techniques for correlating NMR and x-ray CT imaging for radiotherapy treatment planning, in Bruinvis IAD, et al (ed): Proceedings of the Ninth International Conference of the Use of Computers in Radiation Therapy. The Netherlands, Elsevier, 1987 15. Maguire GQ, Noz ME, Lee EM, et al: Correlation methods for tomographic images using two and three dimensional techniques in Bacharach SL (ed): Information Processing in Medical Imaging. Dordrecht, The Netherlands, Martinus Nijhoff, 1985, pp 266-279 16. Maguire GQ Jr, Noz ME, Rusinek H, et al: Graphics applied to medical image registration. IEEE Comput Graph Applications 11:20-28, 1991 17. Hawkes D J, Robinson L, Crossman JE, et al: Registration and display of the combined bone scan and radiograph in the diagnosis and management of wrist injuries. Eur J Nucl Med 18:752-756, 1991 18. Friston KJ, Passingham RE, Nutt JG, et al: Localization in PET images: direct fitting of the intercommissural (AC-PC) line. J Cereb Blood Flow Metab 9:690-695, 1989 19. Minoshima S, Koeppe RA, Mintun MA, et al: Automated detection of the intercommissural line for stereotactic localization of functional brain images. J Nucl Med 34:322-329, 1993 20. Fox PT, Perlmutter JS, Raichle ME: A stereotactic method of anatomical localization for positron emission tomography. J Comput Assist Tomogr 9:141-153, 1985 21. Zhang J, Levesque MF, Wilson CL, et al: Multim0daiity imaging of brain structures for stereotactic surgery. Radiology 175:435-441, 1990 22. Wilson MW, Mountz JM: A reference system for

WEBER AND IVANOVIC

neuroanatomical localization on functional reconstructed c-ere~bralfmages~ J Comput Assist Tomogr 13:174:178,1989 23. Meltzer CC, Leal JP, Mayberg HS, et al: Correction of PET data for partial volume effects in human cerebral cortex by MR imaging. J Comput Assist Tomogr 14:561-570, 1990 24. Evans AC, Marrett S, Torrescorzo J, et al: MRI-PET correlation in three dimensions using a volume-of-interest (VOI) atlas. J Cereb Blood Flow Metab 1I:A69-A78, 1991 25. Evans AC, Beil C, Marrett S, et al: Anatomicalfunctional correlation using an adjustable MRI-based region of interest atlas with positron emission tomography. J Cereb Blood Flow Metab 8:513-530, 1988 26. Wahl RL, Quint LE, Cieslak RD, et al: "Anatometabolic" tumor imaging: Fusion of FDG PET with CT or MRI to localize loci of increased activity. J Nucl Med 34:11901197, 1993 27. Schad LR, Boesecke R, Schlegel W, et al: Three dimensional image correlation of CT, MR, and PET studies in radiotherapy treatment planning of brain tumors. J Comput Assist Tomogr 11:948-954, 1987 28. Knesaurek K: Fusion of morphological MRI and CT images with functional SPECT images in brain studies. J Nucl Med 34:125, 1993 29. Pelizzari CA, Chen GTY, Spelbring DR, et al: Accurate three-dimensional registration of CT, PET, and/or MR images of the brain. J Comput Assist Tomogr 13:20-26, 1989 30. Pelizzari CA, Chen GTY: Registration of multiple diagnostic imaging scans using surface fitting, in: The Use of Computers in Radiation Therapy. 1987, pp 437-440 31. Zubal G, Tagare H, Zhang L, et al: 3-D registration of intermodality medical images, in: Proceedings of the 13th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 13:293-294, 1991 32. Gamboa-Aldeco A, Felingham LL, Chen GTY: Correlation of 3D surfaces from multiple modalities in medical imaging. Proc SPIE 626:467-473, 1986 33. Alpert NM, Bradshaw JF, Kennedy D, et al: The principal axes transformation--a method for image registration. J Nucl Med 31:1717-1722, 1990 34. Bacharach SL, Douglas MA, Carson RE, et al: Three-dimensional registration of cardiac positron emission tomography attenuation scans. J Nucl Med 34:311-321, 1993 35. Birnbaum BA, Noz ME, Chapnick J, et al: Hepatic hemangiomas: Diagnosis with fusion of MR, CT, and Tc-99m-labeled red blood cell SPECT images. Radiology 181:469-474, 1991 36. Rusinek H, Tsui W-H, Levy AV, et al: Principal axes and surface fitting methods for three-dimensional image registration. J Nucl Med 34:2019-2024, 1993 37. Rusinek H, Levy A, Noz ME: Performance of two methods for registering PET and MR brain scans, in: Conference Record of IEEE Nuclear Science Symposium and Medical Imaging Conference vol 3, 1991, pp 2159-2162 38. Barillot C, Lemoine D, Le Briquer L, et al: Data fusion in medical imaging: Merging multimodal and multipatient images, identification of structures and 3D display aspects. Eur J Radi01 17:22-27, 1993 39. Barber DC: Registration of low resolution medical images. Phys Med Bio137:1485-1498, 1992

CORRELATIVE IMAGE REGISTRATION

40. Barber DC: Automatic alignment of radionuclide images. Phys Med Bio127:387-396, 1982 41. Holman BL, Zimmerman RE, Johnson KA, et al: Technetium-99-m-HMPAO and thallium-201 SPECT images of the brain. J Nucl Med 32:1478-1484, 1991 42. Tarkington TG, Jaszczak RJ, Pelizzari CA, et al: Accuracy of registration of PET, SPECT and MR images of a brain phantom. J Nucl Med 34:1587-1594, 1993 43. Levin DN, Hu X, Tan KK, et al: The brain: Integrated three-dimensional display of MR and PET Images. Radiology 172:783-789, 1989 44. Mazziotta JC, Pelizzari CC, Che n GT, et al: Region of interest issues: The relationship between structure and function in the brain. J Cereb Blood Flow Metab ll:A51A56, i991 45. Grzeszczuk R, Tan KK, Levin DN, et al: Retrospective fusion of radiographic and M R data for localization of subdural electrodes. J Comput Assist Tomogr 16:764-773, 1992 46. Meyer CR, Leichtman GS, Brunberg JA, et al: Simultaneous usage of homologous points, lines and planes for optimal, 3D, linear registration of multimodality imaging data. IEEE Trans Med Imaging (in press) 47. Boes JL, Bland PH, Weymouth TE: Generating a normalized geometric liver model using warping. Invest Radio129:281-286, 1994 48. Dann R, Hoford J, Kova~i~ S, et al: Evaluation of elastic matching system for anatomic (CT, MR) and functional (PET) cerebral images. J Comput Assist Tomogr 13:603-611, 1989 49. Faber TL, McColl RW, Opperman RM, et al: Spatial and temporal registration of cardiac SPECT and MR images: Methods and evaluation. Radiology 179:857-861, 1991 50. Pietrzyk U, Herholz K, Heiss W-D: Three-dimensional alignment of functional and morphological tomograms. J Comput Assist Tomogr 14:51-59, 1990 51. Woods RP, Mazziotta JC, Cherry SR: MRI-PET registration with automated algorithm. J Comput Assist Tomogr 17:536,546, 1993 52. Junck L, Moen JG, Hutchins GD, et al: Correlation methods for the centering, rotation, and alignment Of functional brain images. J Nucl Med 31:1220-1226, 1990

323

53. Mountz JM. Modell JG, Foster NL, et al: Prognostication of recovery following stroke using the comparison of CT and technetium-99m HM-PAO SPECT. J Nucl Med 31:61-66, 1990 54. Sullivan T, Villanueva-Meyer J. Liu C-K, et al: Watershed infarcts, Tc-99m HMPAO SPECT and CT correlation. Clin Nucl Med 16:170-173, 1991 55. Weber DA, Ivanovic M, Franceschi D. et al: Pinhole SPECT: A new approach to in vivo high resolution SPECT imaging in small laboratory animals. J Nucl Med 35:342-348, 1994 56. Muller-Gartner HW, Links JM, Prince JL, et al: Measurement of radiotracer concentration in brain gray matter using positron emission tomography: MRI-based correction for partial volume effects. J Cereb Blood Flow Metab 12:571-583, 1992 57. Seitz RJ. Bohm C, Greitz T, et al: Accuracy and precision of the computerized brain atlas program for localization and quantification in positron emission tomography. J Cereb Blood Flow Metab 10:443-457, 1990 58. Weber DA, Franceschi D, Ivanovic M, et al: Comparative uptake properties of TI-201, Tc-99m HMPAO, -MIBI, -DTPA, and I~123 IMT in primary brain tumor. Ear J Nucl Med 20:838. 1993 (abstr). 59. Susskind H, Weber DA. Ivanovic M, et al: Quantitative Tc-99m HMPAO and 1-123 rCBF imaging in the dog using SPECT and MRI. J Nucl Med 33:1012, 1992 (abstr). 60. Sgouros G, Chiu S, Pentlow KS. et al: Threedimensional dosimetry for radioimmunotherapy treatment planning. J Nucl Med 34:1595-1601. 1993 61. Cline HE, Lorensen WE, Kikinis R, et al: Threedimensional segmentation of MR images of the head using probability and connectivity. J Comput Assist Tomogr 14:1037-1045, 1990 62. Chen GTY. Pelizzari CA, Levin DN: Image correlation in oncology, in De Vita ST, Helman S, Rosenberg SB (eds): Important Advances in Oncology. Lippincott, Philadelphia, 1990, pp 131-141 63. Cline HE, Lorenson WE, Sooza SP, et al: 3D surface rendered MR images of the brain and its vasculature. J Comput Assist Tomogr 15:344-351. 1991