Intramodality Coregistration

One of the most obvious clinical applications of coregistration is the area of serial imaging. Comparison of scans from a given patient acquired over various time intervals can be routinely performed to follow disease progression and response to treatment. Unfortunately, diagnostic imaging scans are not routinely registered in most radiology departments; in contrast, the common practice for an examiner is to do one's best to look at film montages of slices that do not match and try to assess disease change. For gross changes this method of comparison may be adequate, although it is never optimal. For subtle change, visual comparison of unmatched images is entirely inadequate. This problem is especially prevalent in assessment of disease changes in MS, brain tumors, and any time intravenous (IV) contrast is used, including evaluation of CNS inflammatory or infectious diseases.

Image coregistration is still not applied commonly in these clinical settings in spite of available fully automated, validated, highly accurate registration algorithms [1,2]. In contrast, as soon as valid methods could be developed, neuroimaging researchers quickly took advantage of coregistration techniques in serial imaging to study objective, measurable changes in image variables. Goodkin and colleagues [3] have reported on the use of coregistration to quantitatively study brain white-matter signal changes on MRI where new contrasts enhancing

Copyright © 2000 by Academic Press.

All rights of reproduction in any form reserved.

FIGURE 1 MRI to MRI coregistration — evolution of lesions in a 33-year-old woman with relapsing and remitting multiple sclerosis. Each vertical pair of T1 and proton density images represents a monthly interval acquired over a total of 6 months of disease change. Accurate alignment of images allows precise comparison of lesion changes over time. Image is courtesy of William Rooney (High-Field MRI Lab, Department of Chemistry, Brookhaven National Laboratory, Upton, New York).

FIGURE 1 MRI to MRI coregistration — evolution of lesions in a 33-year-old woman with relapsing and remitting multiple sclerosis. Each vertical pair of T1 and proton density images represents a monthly interval acquired over a total of 6 months of disease change. Accurate alignment of images allows precise comparison of lesion changes over time. Image is courtesy of William Rooney (High-Field MRI Lab, Department of Chemistry, Brookhaven National Laboratory, Upton, New York).

lesions do and do not arise in patients with relapsing and remitting MS. Figure 1 is an illustration of the dynamic nature of contrast enhancing lesions in MS examined monthly over 6 months. This figure illustrates the obvious value of simply comparing serial disease changes with accurate alignment of images alone.

To further take advantage of coregistration in serial imaging, subtraction techniques can be employed. Subtraction of one coregistered image from another produces a difference image that allows another level of visualizing changes in serial images, frequently changes that cannot be appreciated otherwise. Image subtraction is in theory a simple arithmetic subtraction of signal intensity on an image voxel-by-voxel basis. However, because global values change from one scan to another, normalization of intensity values across data sets is necessary. Usually this is accomplished by referencing individual intensity values to a mean for the entire dataset. To define subtle changes with subtraction and at the same time exclude those due to slight misalignment, especially in regions near the surface of the brain, highly accurate, subvoxel registration is required. Figure 2 illustrates coregistration of serial MRIs from a patient after surgical resection of a left temporal lobe brain tumor. Alignment of images with coregistration enabled identification of a small focal area of IV contrast enhancement. Prior to coregistration it was not possible to say confidently that the enhancement was present on one and not the other scan since precise slice by slice visual comparison was not possible.

Subvoxel coregistration and subtraction that results in complete cancellation of signals from unchanged structures in high resolution brain images has been reported by Hajinal et al. [4]. Their work demonstrated the capability to detect subtle changes in serially acquired images including normal brain maturation and development, detection of small changes in the size of tumors, and small displacements of the brain related to minor head trauma. Curati et al. [5] have demonstrated increased sensitivity to detecting changes associated with IV contrast enhancement. With coregistration and subtraction they were able to detect additional, unequivocal, but subtle changes with respect to degree and distribution of enhancement in the CNS inflammatory diseases, MS lesions, and benign and malignant tumors. They further pointed out that coregistration and subtraction were of particular value in the following: (1) recognition of small degrees of enhancement, (2) tissue or fluid enhancement with very low or high baseline signals, (3) enhancement at interfaces, boundaries, or other regions of complex anatomy, and (4) assessment of enhance

FIGURE 2 MRI to MRI coregistration — serial subtraction imaging in a 30-year-old male patient who is status post resection of a left temporal lobe glioma. MRI scan on the left was acquired 3 months prior to middle scan. Both scans were perfomed using intravenous contrast enhancement. Image alignment alone provides an excellent picture for accurate visual comparison. The right subtraction image (middle scan — left scan) leaves only the difference in tumor change (none apparent) and gadolinium enhancement.

FIGURE 2 MRI to MRI coregistration — serial subtraction imaging in a 30-year-old male patient who is status post resection of a left temporal lobe glioma. MRI scan on the left was acquired 3 months prior to middle scan. Both scans were perfomed using intravenous contrast enhancement. Image alignment alone provides an excellent picture for accurate visual comparison. The right subtraction image (middle scan — left scan) leaves only the difference in tumor change (none apparent) and gadolinium enhancement.

ment when thin slices are acquired. Also noted is the potential value of applying coregistration and subtraction in serial images with susceptibility-sensitive and diffusion-weighted sequences for serial evaluation of bleeding in the brain and disturbance of tissue blood perfusion, respectively.

Localization of focal epilepsy (an epilepsy type with seizures that start in a confined part of the brain) with single photon emission computed tomography (SPECT) is a recently recognized field for the application of coregistration and subtraction. Using radiotracers "Tcm-hexamethylpropylene amine oxide ("Tcm-HMPAO) or "Tcm-ethyl cysteinate die-thylester ("Tcm-ECD), localized cerebral blood flow (CBF) changes can be visualized during partial seizures, an imaging tool of great value to patients with medically uncontrolled partial epilepsy who are evaluated for epilepsy surgery. Both HMPAO and ECD have a very high first-pass brain extraction rate with maximal uptake achieved within 30-60 seconds of IV injection. The isotope is effectively trapped in the brain at this time, allowing a picture of CBF during a seizure. A relatively long half-life allows the image to be conveniently created up to 3-4 hours later.

Until the demonstration of coregistration and subtraction [6,7], ictal (during seizure) SPECT imaging had many limitations. First, SPECT scans are of relatively low spatial resolution. As a result, interpretation of anatomic detail is not possible without coregistration to MRI or CT. Second, interindividual differences make difficult comparison with a baseline interictal (between seizures) scan. Many patients have interictal regions of relative focal decrease in CBF, especially in regions of seizure onset. Others have baseline relative focal increases in CBF. Third, further complicating visual comparison between ictal and interictal scans are global differences in signal intensity and most importantly unmatched slice orientations. Figure 3 demonstrates these problems and their correction with coregistration, image normalization, and subtraction. O'Brien and colleagues were the first to demon-

-Jtal funcul rWtrfWnW CnWlltf

FIGURE 3 Subtraction ictal SPECT — seizure imaging of cerebral blood flow in a 9-month old-male infant with catastrophic partial epilepsy. Note that the global intensity scaling between the ictal and interictal scans is different. After normalization of scaling, subtraction of interictal from ictal scans allows visualization of focal increased blood flow in the left mesial occipital lobe, a finding that is not apparent from comparison of the ictal and interictal scans. EEC seizure recordings support localization of seizures originating from the left occipital lobe. See also Plate 97.

strate successful application [7] of subtraction and coregistration of ictal SPECT in a large series of epilepsy patients. They subsequently validated the method using phantom studies and also addressed normalization and significant difference issues [8]. Their general method, called subtraction ictal SPECT coregistered to MRI (SISCOM), is now being widely adopted by epilepsy centers around the world.

Another potential area of intrasubject serial image coregis-tration with clinical utility is enhancement of images with signal averaging. Holmes et al. [9] pioneered this use of coregistration to improve MR imaging. Increased contrast and better spatial resolution, features that allow better visibility of neuroanatomic detail, can be obtained from any given MRI scanner using signal averaging. The gain in signal is expected to increase as the root of the number of contributing scans. Although a single scan acquired at a duration equal to the total of a number of contributing scans achieves the same increase in signal-to-noise, subject movement (especially swallowing) severely limits total scan time; even the most motivated subjects cannot remain completely motionless for more than about 10 minutes. Figure 4 illustrates a twofold improvement in signal-to-noise from averaging four MRI scans in an epilepsy patient with a subtle neocortical lesion of uncertain significance (ultimately defined as epileptogenic with intracranial electrode recording of seizures). Most notable is the relative

FIGURE 4 MRI to MRI coregistration — signal averaging to improve signal-to-noise. Hoizontal pairs of T1 weighted gradient echo images were acquired from a 54-year-old patient with medically refractory focal epilepsy. The first of each horizontal pair of images (numbered in the bottom right corner of each slice) is from a single scan, while the second of the pair is the average of four coregistered scans. A subtle epileptogenic lesion, present in the right lateral occipital cortex (partially transparent white arrow), is better delineated in the averaged scan because of markedly improved contrast, a result of a twofold improvement in signal-to-noise.

FIGURE 4 MRI to MRI coregistration — signal averaging to improve signal-to-noise. Hoizontal pairs of T1 weighted gradient echo images were acquired from a 54-year-old patient with medically refractory focal epilepsy. The first of each horizontal pair of images (numbered in the bottom right corner of each slice) is from a single scan, while the second of the pair is the average of four coregistered scans. A subtle epileptogenic lesion, present in the right lateral occipital cortex (partially transparent white arrow), is better delineated in the averaged scan because of markedly improved contrast, a result of a twofold improvement in signal-to-noise.

decrease in noise, especially in the white matter, that allows better definition of the gray-matter-based lesion.

Coregistration for signal averaging in clinical neuroimaging remains to be fully exploited. This is a new application that has numerous variables to optimize, including those that allow improved spatial resolution. One example is to use signal averaging to compensate for the loss of signal with decreasing voxel sizes. Another is to compensate for decreased signal in fast acquisition scans used to image patients who cannot stay still. Optimal scan sequences to best take advantage of signal averaging still need to be developed and will be defined by the goals of the clinical application.

0 0

Post a comment