Voxel-based statistical analysis of amyloid PET scans in the J-ADNI multi-center study

Objectives: Amyloid PET imaging is useful for evaluating amyloid-β (Aβ) pathological process as a biomarker of Alzheimer's disease (AD). While visual assessment is the standard clinical practice in determination of PET positivity/negativity, objective and quantitative assessment is helpful for interpreters and is important for monitoring Aβ accumulation in clinical research and trials. Although quantification of amyloid accumulation with ROI-based standardized uptake value ratio (SUVR) has been widely performed, voxel-based statistical analysis has not been established yet. The purpose of this study was to examine the feasibility of voxel-based statistical analysis of amyloid PET scans and to compare the results with the mean cortical SUVR (mcSUVR). Methods: A total of 166 subjects who underwent 11C-PiB PET in the Japanese Alzheimer's Disease Neuroimaging Initiative (J-ADNI) multi-center study were analyzed (46 AD patients; 62 MCI subjects; 58 normal control subjects). All 11C-PiB images had been centrally classified by quantified blind readers as positive, negative and equivocal. Besides, we built up a normal database made of 18 amyloid-negative subjects acquired in another study. The 11C-PiB images were spatially normalized to the standard MNI space with an adaptive template method[asterisk]1. Z-score image was generated for each subject using the normal database, and sum of the Z-score (Z-sum) was calculated after masking out white matter areas. The mcSUVR was calculated using automated PET-only quantification method based on pre-defined ROIs[asterisk]1. A receiver-operating-characteristic (ROC) analysis was performed to evaluate the capability of Z-sum and mcSUVR to classify the scans as positive and negative, in which equivocal scans were regarded as positive. Results: Visual reading had provided positive for 88, negative for 68 and equivocal for 10 scans. ROC analysis on discrimination of positive from negative scans indicated that sensitivity and specificity were 90.8% and 100% for Z-sum, and 91.8% and 98.5% for mcSUVR, respectively. Although most of the equivocal scans were quantitatively classified as false negative both by Z-sum and mcSUVR, Z-score images correctly delineated abnormal amyloid accumulation over the regions pointed out by the visual reading. Conclusion: Voxel-based statistical analysis of amyloid PET scans provides objective Z-score images and Z-sum value. These might be helpful as an adjunct to visual interpretation, especially fo