Summary – PET Working Group

Panel: Bendriem,B (CTI) , ; Charles, C (DUKE), de Leon (NYU) CHAIR, Foster, N (UM), Frank, R. (Pharmacia),  Jagust, B. (UC DAVIS), Reiman , E.(UA),  Rusinek, H. (NYU), ,Schlyer, D. (BNL), and Small, G. (UCLA)

NIA : Molchan, S and Buckholtz, N

1. PET COMPOUNDS AND ISOTOPES

The consensus of the NIH/NIA PET working group is that the standard method for the proposed Imaging Initiative should be FDG-PET. The wide availability and the previous literature using FDG make it the ideal tracer for this multi-center study.  Other probes, including but not limited to those that image amyloid plaques and tangles, are of interest but would likely be pursued at only a subset of sites. 

2. HARDWARE

The tomograph should be of the latest generation, multislice to cover the entire brain. The 3D acquisition mode will be used to accommodate lower dosimetery and to improve the statistics of the data. Measured attenuation correction using available isotope must be used on the various scanners. The image should be reconstructed with the standard clinical reconstruction including all necessary corrections (random, scatter, attenuation).

The question on the quantification was discussed and the difficulty in handling variation inherent to the use of different scanners was acknowledged. It was recommended to provide the physical characteristic of the scanner for post acquisition analysis. A set of calibration phantoms including at the minimum the Hoffman brain phantom and the uniform cylinder should be run periodically (TBD, about once per week) to assess the stability (qualitative and quantitative respectively) of the instrument over the course of the study. Phantom studies should attempt to address the potentially confounding effects of acti vity outside the field of view on the 3D PET images.

3. SUBJECT CONDITIONS

Conditions during the performance of PET scans should be fully characterized and standardized whenever possible.  Applicants are encouraged to propose PET studies during “a resting state” (e.g., eyes open, ears unoccluded in a dark room with minimal ambient noise). Procedures to minimize head movement during scans and at different scanning sessions should be fully described (e.g., using well tolerated head immobilization procedures), including evidence of reliability.  Methods for correcting for differences in head positioning should also be described.

The use of medications and the behavioral state of subjects at the time of the scan also should be carefully considered in the proposed study design.  Use of medications by subjects should fully documented. To the extent feasible, applicants are encouraged to minimize the use of centrally acting medications for at least two weeks prior to each PET scan (or longer, if dictated by the medication's elimination half-life). When this is not possible steps should be taken to minimize changes in medication.

4. TEST RE-TEST INTERVALS

The value of test-re-test data would be to facilitate power calculations in designing a clinical trial of a new medication. The applicant should explain how the proposed study would provide data adequate for statistical "power" calculations, which require an estimate of variance in the measurement (within and between subjects). These variance data would be used in combination with an estimate of the magnitude of treatment benefit, which cannot be known now and would vary depending upon the characteristics of the drug.

The number and frequency of PET scans is limited by radiation safety, budgetary constraints, and subject tolerability. The applicant should propose several scans over a 3 year period with the objective of comparing the sensitivity of imaging modalities and other potential biomarkers to disease progression.  Biological specimens, as a general rule, should be collected at same times as imaging and cognitive assessments recognizing that the value of samples can be dramatically enhanced by correlating cognitive and imaging results.  It is hoped that imaging may permit treatment effects to be recognized more rapidly than current clinical measures.  Consequently, studies as frequent as every 3 months should be considered, particularly in the first year of the study.  The type and frequency of assessments should also allow for the possibility of the linearity or other patterns of the rate of change. The performance of a diagnostic brain autopsy should be encouraged and methods for facilitating postmortem examinations should be described.

5. TRACER-KINETIC MODELS

In order to generate quantitative measurements of CMRgl (e.g., in mg/min/100 g), PET studies would require an accurate FDG input function. The best established procedure for generating this input function involves repeated blood samples from the radial artery. Since this procedure is frequently uncomfortable, may be associated with rare but significant risks, and could lead to unacceptable subject attrition, it is not recommended.  Another extensively used procedure involves repeated blood samples from “arterialized venous” samples. Because of concerns about this procedure's reliability, validity, tolerability, and inconvenience, it is not recommended. While one study suggests that it is possible to generate an accurate input function using dynamic PET scans, an image-derived input carotid artery input function, and a small number of venous samples, additional studies are needed to further standardize and validate this strategy. 

Thus, we recommend the performance of PET studies in the absence of arterial or arterialized venous measurements. Images could be generated without any quantification (i.e., in units of PET counts) or using an image-derived input function, depending on the interests and resources of the PET Centers involved.  Alternative noninvasive methods for generating the FDG input function may be proposed, but would require adequate justification for the choice of both the input function and the tracer-kinetic model, and a description of the model‘s underlying assumptions. 

The panel recognizes that this recommendation assumes that longidutinal declines in whole brain CMRgl are minimal and, thus, that declines in regional/whole brain meaurements are not significantly underestimated. In circumstance that whole brain CMRgl changes are hypothesized, it might be helpful to analyze images after they are transformed into quantitative measurements. Longitudinal declines in whole brain metabolism have been found in studies of patients with probable Alzheimer's dementia, this decline was not observed in a study of normal aging.

If quantitative measurements are proposed, applicants should justify the type of input function used and may wish to consider generating quantitative images of both CMRgl (e.g., mg/min/100g) and individual FDG rate constants (e.g., K 1 [glucose clearance rate constant from plasma into brain tissue, in ml/min/100g], k 2, [the back diffusion rate from brain tissue to plasma, in 1/min], and k 3 [the FDG phosphorylation rate constant, in 1/ min]).

6. IMAGE NORMALIZATION AND REGISTRATION

Applicants are encouraged to describe procedures to normalize PET images for the variation in absolute measurements, align sequential PET images, deform images according to the coordinates of Talairach's brain atlas and, co-register PET and MRI images.

Normalization .  The choice of whether and how to normalize FDG PET images for the variation in absolute measurements (whether the images are in units of PET counts or CMRgl) may depend on the spatial extent of PET reductions.  In a study of patients with probable AD, the whole brain was affected and the absolute data had better power than data normalized for pons or whole brain CMRgl. In studies of APOE-4 carriers and carriers who had very mild cognitive impairment or were cognitively normal, normalization for whole brain worked quite well using a ratio method. Applicants are encouraged to describe the reference regions used for image normalization (e.g., whole brain, pons, or other brain regions that are relatively spared), the procedures used to normalize data using this reference regions (e.g., a ratio method or analysis of covariance) and justify their choice (e.g, plan to demonstrate that there are minimal changes in absolute measurements from the reference region).

PET-MRI Registration, Atrophy Correction, and ROI Analyses.   While the co-registration of PET and MRI images may not be necessary for some image-analysis techniques (e.g., SPM99 and SSP), it may be helpful for some purposes, including but not limited to (a) the non-linear deformation of brain images according to the coordinates of a standard brain atlas, (b) determining the extent to which changes in regional or whole brain PET measurements are attributable to the combined effects of atrophy and partial volume averaging, and (c) characterizing declines in regional CMRgl from anatomically well characterized regions of interest (e.g., hippocampus and entorhinal cortex). In order to generate statistical brain maps of the decline in regional PET measurements from sequential MRI's, as recommended below, the applicant is encouraged to describe the image-deformation procedures used to permit image averaging and anatomical standardization.  Since the primary purpose of the proposed study is to provide the most powerful measure of disease progression, we do not recommend the use of an atrophy-correction procedure in the primary comparison; however, for secondary purposes, we would encourage applicants to describe a voxel-based correction algorithm for determining the extent to which the observed PET changes are related to brain atrophy. Since ROI analyses are less well studied than statistical brain mapping procedures for characterizing declines in regional CMRgl from sequential PET images in patients with AD and persons at risk for this disorder, and since questions remain about the optimal size, location, and anatomical landmarks for choosing the optimal ROI, approaches using a preselected ROI are not recommended for the primary analysis of declines in regional PET measurements.  As a secondary analysis , applicants may consider the possibility of comparing the power of an ROI analysis (e.g., of the hippocampus or entorhinal cortex) to the power of the statistical brain mapping procedure used in the primary comparison.

7. CENTRAL IMAGE STORAGE AND PROCESSING

We recommend the centralized storage of raw data and PET images and the centralized analysis of PET images for those comparisons that address the primary aims of the study. We also encourage applicants to describe procedures that will allow investigators from different organizations to analyze data using other methods (e.g., different image reconstruction, image deformation, or normalization techniques) in order to optimize the utility of FDG PET studies as a putative surrogate marker of AD. If data processing is done at more than one site all the data should be processed, not subsets. The data should be stored in the original and in any modified formats.

8. DATA ANALYSES

Applicants are instructed to describe data analysis procedures which address the objectives of the proposal (e.g., characterizing the rate of decline in regional FDG PET measurements in patients with MCI and normal controls, demonstrating that early declines in patients with MCI predict the subsequent onset of probable Alzheimer's dementia, and generating estimates of the power to test candidate treatments in placebo-controlled trials of patients with MCI).

Statistical brain mapping procedures have demonstrated declines in regional measurements from sequential FDG PET images in patients with probable AD, APOE-4 carriers and noncarriers with very mild cognitive impairment, and those who were cognitively normal; they are commonly used in brain mapping studies; and they have the potential to be used in a standard manner. For these reasons, a statistical brain mapping procedure is recommended for the primary purposes of the proposed PET study, while other methods (e.g., ROI analyses) may be considered for secondary purposes.

Applicants are encouraged to describe the procedures used to normalize PET images for the variation in absolute measurements (as previously noted), align sequential PET images from the same subject, deform images according to the coordinates of a standard brain atlas, generate statistical maps that address each of the proposal's specific aims, and optimize the tradeoff between statistical Type 1 and Type 2 errors (e.g., using P<.005, one-tailed and uncorrected for multiple comparisons or P<0.05 after a suitable correction for multiple comparisons).  Since SPM99 (or its most current version at the time the application is submitted) is widely available, extensively evaluated, and continuously refined, has been used in previous longitudinal PET studies of persons afflicted by and at risk for AD, has the potential to be administered in a standardized fashion, and permits many of the statistical comparisons in which the applicants are interested, it may have particular value. Alternatives to SPM99 may be proposed, or directly compared to SPM99 if adequately justified.  As previously noted, a voxel-based atrophy correction procedure is not recommended when addressing the primary aims of this study, but may be proposed to consider the extent to which the observed PET changes are related to brain atrophy. 

9. STATISTICAL POWER

Recent studies have attempted to characterize one-year decline in regional and whole brain CMRgl in patients in patients with probable AD and two-year declines in APOE-4 carriers and noncarriers with and without very mild cognitive impairment (i.e., “age-associated memory impairment”).  Unfortunately, these studies are small and do not include patients who meet criteria for MCI.  Applicants are encouraged to provide a rough estimate of their power to address the primary aims of this RFA based on the number of subjects in their proposed study.  Among other things, applicants may wish to consider the following issues in their power estimates: (1) each of the RFA's primary aims and related hypotheses (e.g., the power to characterize CMRgl declines and predict the subsequent conversion to probable AD in patients with MCI), (2) the time interval between scans, (3) the effects of subject attrition and normal aging, (4) the homogeneity of sample (e.g., MCI and age criteria as it pertains to anticipated conversion rates to probable Alzheimer's dementia), (5) freedom from potential confounds (e.g., coexisting medical disorders, medication effects, which could increase noise and numbers, but assist recruitment efforts), (6) the image analysis technique used (e.g., SPM), (7) the chosen significance level (with or without correction for multiple comparisons), (8) the region or regions postulated to demonstrate CMRgl decline in patients with MCI, (9) the effects of image normalization (e.g., reduced power if MCI patients have a decline in the reference region), (10) anticipated treatment effect sizes, (11) how the rate and variability of CMRgl decline in MCI patients compares to previously studied groups, and (12) the anticipated conversion rate to probable Alzheimer's dementia.

10. COSTS

A uniform reimbursement should be provided for PET scans for all sites.  Because no standard payment rates for brain PET images have been established, it will be important to develop a realistic budget that includes both a technical component for the performance of the scan and a component for data transfer and quality control measures.  Professional fees for image interpretation are not permitted.  Previous experience has shown that most centers have found reimbursement for the technical component of research brain FDG-PET imaging to be adequate at approximately $1500.  (If additional funds are required for the PET session (e.g., for the procedures used to generate quantitative measurements, if that is proposed), this expense would require adequate justification. It is also appropriate to consider volunteer fees for subjects consistent with their inconvenience and discomfort.  The costs associated with multiple scans must be balanced with the need to perform an adequate number of studies to achieve the project's scientific objectives.