Appendix e-1
CSF Methods and Processing
The lumbar puncture for the collection of CSF occurred a median of 1 day after the baseline clinical visit (range: -20 to 163 days, 11 missing dates). CSF was collected at each site, transferred into polypropylene transfer tubes followed by freezing on dry ice within 1 hr after collection and shipped overnight to the ADNI Biomarker Core laboratory at the University of Pennsylvania Medical Center (UPMC) on dry ice. When samples are received in the Laboratory, they are thawed and aliquots are stored in bar coded polypropylene vials at -80 oC. The complete details can be found at (http://www.adni-info.org/index.php). A standardized protocol was implemented to quantify biomarker concentrations in each of the CSF ADNI baseline aliquots using a multiplex xMAP Luminex platform (Luminex Corp, Austin, TX) with Innogenetics (INNO-BIA AlzBio3, Ghent, Belgium; for research use only reagents) immunoassay kit-based reagents which was validated in (1, 2).
MRI Methods and Preprocessing
ADNI collects 1.5T MRI scans in all subjects and 3T scans in approximately 25% of the sample; only the 1.5T MRI scans were used for this study. The MRI scan was a median of 18 days before baseline clinical exam (range: -12 days to 190 days). The images were acquired on commercial 1.5 T scanners. The nominal parameters of the morphometric T1-weighted MPRAGE can be found in (3). These images are additionally corrected for gradient non-linearity, intensity inhomogeneity and scaled geometrically using the phantom scan acquired with each subject exam (3) using a centralized MRI-processing pipeline at Mayo Clinic, Rochester. STAND-scores were estimated on these pre-processed images.
Estimation of STAND-score: MRI biomarker
MRI has shown great promise to non-invasively visualize high resolution structural brain information. However one major challenge with using MRI as a biomarker has been condensing the large amount of three-dimensional (3D) information present in the scans into a single disease relevant number. Hippocampal volumes though powerful do not take the entire 3D volume into consideration. To overcome this challenge, investigators have recently turned their attention towards multivariate and machine learning techniques in order to distinguish AD patients from CN subjects. These techniques use a library of AD and CN scans to learn about the stereotypical location and extent of atrophy in AD when compared to CN and subsequently use this information to classify new incoming patients based on their MRI scans. In a recent study, it was shown that automated techniques in general perform as well or better than radiologists (4). One such technique developed at our lab is STructural Abnormality iNDex (STAND)-scores.
ADNI Based Training Set for STAND-score Estimation: As mentioned above, STAND-score algorithm uses a training set of AD and CN scans from which the degree and extent of AD related atrophy is estimated and then the algorithm assigns scores for new incoming scans based on its knowledge from the training set. In order to obtain unbiased results, the training dataset should be independent of the cases for which STAND-scores are estimated. We accomplished this by training the algorithm with scans of ADNI CN and AD subjects who did not participate in the lumbar puncture arm of the study and then using the trained algorithm to create STAND-scores for CN, aMCI and AD ADNI subjects with CSF. A total of 221 CN, 404 MCI and 188 AD subjects had 1.5 T baseline scans that passed MRI quality control. Ninety AD and 90 CN (randomly selected from a possible 112 CN) subjects without CSF were used to train the algorithm. There were no significant demographic, cognitive, or genetic differences between CN and AD subjects used to train the STAND algorithm and the CN and AD subjects in the MRI-CSF analysis, except the education of AD subjects which was less in the training set.
The detailed description of STAND-score estimation using support vector machines (SVMs) can be found in (5). The main steps for computation of STAND-scores for each scan in the MRI-CSF set are presented here:
1) SPM5 was used for tissue segmentation and normalization (http://www.fil.ion.ucl.ac.uk/spm) of each MRI scan (6). In order to reduce any potential normalization and segmentation bias across the disease groups, customized tissue probability maps were created using all the subjects MRI scans used in this study and then and all scans were registered to this template. The gray matter (GM), white matter (WM) and CSF density probabilities of each test scan were smoothed, modulated and down-sampled to an isotropic voxel size of 8 mm by simple averaging. The GM, WM and CSF densities in voxel sizes of 8×8×8 mm3 were used as features for the development of STAND-scores.
2) 3D weight maps of pattern and severity of atrophy were created using all 180 scans (90 AD and 90 CN) in the training MRI-No CSF set of scans. A subset of voxels was selected such that accuracy is maximized in differentiating AD and CN (figure below) The weights obtained from the trained SVM classifier (for these selected voxels) are applied to the MRI-CSF set of scans to give a measure of normality or abnormality in the brain structures (typically ≥ +1 for the most abnormal and ≤ -1 for the most normal brain), which is labeled as the STAND-score. For each scan in the MRI-CSF set, the STAND-score is a single unbiased estimate of the severity of tissue density loss on a voxel-wise basis throughout the 3D MRI scan in comparison to the pattern extracted from the library of scans (MRI-No CSF set).
Note that the anatomic patterns automatically extracted as important for classification (figure below) from the MRI scans are consistent with known distribution of neurofibrillary AD pathology. For example, hippocampus and entorhinal cortex were found to be very important in all the tissue density maps because these areas are affected most severely in AD (7).
REFERENCES
1. Shaw LM, Vanderstichele H, Knapik-Czajka M, Clark C, Blennow K, Soares H, et al. Characterization of CSF Aβ1-42, Tau And P-Tau 181P concentrations at baseline In Alzheimer's Disease Neuroimaging Initiative (ADNI) study cohorts. Alzheimers & Dementia 2008;4(4 Supplement):545-546.
2. Shaw LM, Knapik-Czajka M, Clark C, Aisen PS, Petersen RC, Blennow K, et al. CSF biomarker signature in AD Neuroimaging Initiative subjects. Annals of Neurology In Press 2009.
3. Jack CR, Jr., Bernstein MA, Fox NC, Thompson P, Alexander G, Harvey D, et al. The Alzheimer's Disease Neuroimaging Initiative (ADNI): MRI methods. J Magn Reson Imaging 2008;27(4):685-691.
4. Kloppel S, Stonnington CM, Barnes J, Chen F, Chu C, Good CD, et al. Accuracy of dementia diagnosis--a direct comparison between radiologists and a computerized method. Brain %R 10.1093/brain/awn239 2008:awn239.
5. Vemuri P, Gunter JL, Senjem ML, Whitwell JL, Kantarci K, Knopman DS, et al. Alzheimer's disease diagnosis in individual subjects using structural MR images: Validation studies. Neuroimage 2008;39(3):1186-1197.
6. Ashburner J, Friston KJ. Unified segmentation. NeuroImage 2005;26(3):839-851.
7. Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol (Berl) 1991;82(4):239-259.
Figure. Anatomic patterns with maximum discriminative power between AD and controls are overlaid on the corresponding custom T1 template. Color scale used to indicate the frequency of occurrence of the voxel in multiple tissue maps. Yellow: voxel location used in all three tissues (GM, WM and CSF); orange: voxel location used in at least two tissues and red: voxel location used in one tissue only.