99mTc HMPAO SPECT prediction of conversion from mild cognitive impairment to Alzheimer's disease

99mTc HMPAO SPECT prediction of conversion from mild cognitive impairment to Alzheimer's disease

Poster Presentations P3 P372 P3-096 99MTC HMPAO SPECT PREDICTION OF CONVERSION FROM MILD COGNITIVE IMPAIRMENT TO ALZHEIMER’S DISEASE D. P. Devanand...

295KB Sizes 0 Downloads 28 Views

Poster Presentations P3

P372 P3-096

99MTC HMPAO SPECT PREDICTION OF CONVERSION FROM MILD COGNITIVE IMPAIRMENT TO ALZHEIMER’S DISEASE

D. P. Devanand, Ronald L. Van Heertum, Lawrence S. Kegeles, Xinhua Liu, Zong H. Jin, Arthur Mikhno, Gregory H. Pelton, Gnanavalli Pradhaban, Mali Pratap, Nikolaos Scarmeas, Henry Rusinek, J. John Mann, Ramin V. Parsey, Columbia University, New York, NY, USA. Contact e-mail: [email protected] Background: Parietotemporal blood flow and metabolism deficits characterize patients with Alzheimer’s disease (AD), but their utility in predicting conversion from mild cognitive impairment (MCI) to AD remains unclear. Methods: Patients with MCI (n¼131) and control subjects (n¼59) were studied. A diagnostic evaluation, neuropsychological tests, and MRI were done. A baseline 99mTc HMPAO SPECT brain scan was coregistered to each individual’s MRI scan for regions of interest (ROI) analyses. Two experts provided consensus clinical ratings of the SPECT scans in MCI patients. MCI patients were followed for 1-9 years and classified as converters and non-converters to AD. Results: Of 131 patients, 31 converted to AD during follow-up. In consensus clinical ratings, MCI converters and non-converters did not differ on any global or regional SPECT baseline measure. The three groups (MCI converters, MCI non-converters, controls) did not differ significantly in ROI values in cingulate, parietal cortex, hippocampus and parahippocampal gyrus. In MCI patients, when ROI values were dichotomized at the median, the three groups differed in parietal cortex (p¼0.04) with trend-level differences in hippocampus (p¼0.06) and parahippocampal gyrus (p¼0.06) and no significant differences in cingulate (p¼0.25). In the 3-year follow-up sample of MCI patients, in logistic regression analyses that included age and MMSE as independent variables, the ROIs were not significant when evaluated as continuous measures: cingulate (p¼0.2), parietal (p¼0.08), hippocampus (p¼0.17) and parahippocampal gyrus (p¼0.09). In similar analyses that dichotomized the ROI values at the median in MCI patients, reduced cingulate flow was not significant (p¼0.39) but reduced parietal (p¼0.01), hippocampus (p¼0.01) and parahippocampal gyrus (p¼0.03) were significant. After controlling for age and MMSE, decreased parietal, hippocampal, and parahippocampal blood flow (dichotomized at the median) retained significance (p’s < 0. 01 to 0.03) in predicting MCI conversion to AD. Conclusions: SPECT clinical ratings were not useful in predicting MCI conversion to AD. ROI analyses showed moderate predictive utility for parietal and medial temporal flow reductions in predicting conversion to AD. These caveats need to be considered in interpreting SPECT results when utilized to identify likely MCI converters to AD. P3-097

maps to estimate the clinical variables. The leave-one-out cross-validation method was used to evaluate these methods, and the results are shown in Fig. 1, demonstrating that the j-modeling method achieved better performance w.r.t. both correlation and difference between the clinically measured and the estimated variables. The regression method also detected the brain atrophy patterns correlated with the clinical variables, similar to those detected by the voxel-wise correlation analysis, as shown in Fig. 2. Conclusions: The regression method of jointly modeling clinical variables demonstrated a promising estimation performance. The differences between the clinically measured and the estimated variables are comparable to their fluctuation estimated from the longitudinal data of ADNI between 6 months (square root of mean square difference is 1.28 for MMSE and 3.12 for ADAS-Cog). We expect better results can be obtained if advanced feature extraction and feature selection techniques are used in conjunction with the regression algorithms.

Fig. 1. Leave-one-out cross-validation results of the regression modeling of MMSE and ADAD-Cog jointly and separately using RVM, with the clinically measured values shown in x-axes and the estimated values shown in y-axes. The performance of the regression is measured by correlation coefficient (CORR) and square root of mean squared error (RMSE) between clinically measured and estimated clinical scores. The estimation result are shown in red for CN, green for MCI, and blue for AD subjects, respectively.

ESTIMATING CLINICAL VARIABLES FROM BRAIN IMAGES USING BAYESIAN REGRESSION

Yong Fan, Daniel Kaufer, Dinggang Shen, UNC at Chapel Hill, Chapel Hill, NC, USA. Contact e-mail: [email protected] Background: Alzheimer’s Disease displays a continuous transition from normal to diseased state, which is often measured by clinical variables, like MMSE score. However such measurement is subject to fluctuation from one evaluation to another, because an individual’s performance can be affected by many psychophysical factors that are unrelated to the underlying pathology. Therefore, it is important to quantitatively estimate clinical variables based on the relatively objective brain images. Methods: A regression method is proposed to jointly model multiple clinical variables to capture correlations among different variables and suppress the measurement noise. Within a Bayesian framework, the regression model is built on morphological representation of structural MRI brain images by using the relevance vector machine (RVM) technique. The performance of jointly modeling multiple variables (j-modeling) is compared with that of modeling these variables separately (s-modeling). Results: The methods were used to estimate the MMSE and the ADAS-Cog scores. Structural MRI image data of 206 individuals were obtained from ADNI database, including 65 cognitively normal (CN) individuals, 87 MCI, and 55 AD patients. From structural MRI images, voxel-wise gray matter (GM) tissue density maps were computed by normalizing them to a standard template space. The regression methods were then used to build regression models on the GM tissue density

Fig. 2. Spatial patterns of brain atrophy that are significantly correlated with MMSE (left) and ADAS-Cog (right), measured by voxel-wise correction analysis (p<0.05, FDR corrected) (top) and by regression analysis (bottom). The color scales indicate, respectively, t-value for voxel-wise correlation analysis (top) and regression weight for regression analysis (bottom). Images are displayed in radiological convention. P3-098

HIPPOCAMPAL VOLUME LOSS IS ASSOCIATED WITH INCREASED A-BETA LEVELS IN THE PRECUNEUS IN MILD COGNITIVE IMPAIRMENT

Ansgar J. Furst1, Elizabeth C. Mormino1, Tyler Steed1, Ara H. Rostomian1, Chester A. Mathis2, Clifford R. Jack3, William J. Jagust1, 1HWNI, UC Berkeley, Berkeley, CA, USA; 2Department of Radiology, University of Pittsburgh, Pittsburgh, PA, USA; 3Department of Diagnostic Radiology, Mayo Clinic and Foundation, Rochester, MN, USA. Contact e-mail: bmormino@ gmail.com Background: Some patients with amnestic mild cognitive impairment (aMCI) have beta-amyloidosis and presumably will develop Alzheimer’s disease. We used PIB-PET and voxel-based morphometry (VBM) to compare amyloid burden with gray matter (GM) densities. Methods: Forty-eight