EDITORIAL David Eidelberg, MD Wayne Martin, MD Correspondence to Dr. Eidelberg: david1@nshs.edu Neurology ® 2013;81:516–517 See page 520 Different b-amyloid binding patterns in Alzheimer and Parkinson diseases It’s the network! 11 C-Pittsburgh compound B (PiB) PET is a sensitive and increasingly popular imaging tool for assessing the deposition of fibrillar b-amyloid aggregates in the brains of living patients with Alzheimer disease (AD). Increased cortical PiB binding also occurs in dementia with Lewy bodies, but rarely in cognitively impaired individuals with Parkinson disease (PD). 1,2 To date, most PiB PET studies have focused on identifying sig- nificant group differences in radiotracer binding in sin- gle brain regions or in the cortex overall. However, recent evidence suggests that the deposition of protein aggregates in neurodegenerative disorders evolves at the systems level, with involvement of discrete sets of in- terconnected brain regions. 3,4 In this regard, techniques of pattern analysis based upon principal component analysis (PCA) and other multivariate procedures are increasingly used to quantify disease-related circuit changes in functional brain images. 5 In this issue of Neurology ® , Campbell et al. 6 used PCA to evaluate PiB binding spatial topographies in participants with AD, participants with PD with cogni- tive impairment, and healthy volunteers. PiB PET scans from the disease groups were combined with those from the healthy control subjects to create “training” sets from which the corresponding disease-related patterns were derived. The resulting topographies were validated by prospective single scan computations to quantify pattern expression in the “testing” sets that were comprised of independent patient and control scans. To assess the disease specificity of the amyloid topographies, the ex- pression of the AD pattern was evaluated in the subjects with PD and vice versa. A separate PCA was conducted on the combined AD and PD cohorts to identify addi- tional potentially discriminating patterns. Despite comparable levels of overall cognitive disa- bility, the participants with AD and PD exhibited sub- stantial differences in the topography of PiB binding involving mainly the cerebral cortex. Indeed, pattern contributions from the prefrontal cortex and posterior cingulate regions were higher in the participants with AD relative to the cognitively impaired participants with PD and healthy volunteer subjects. While cortical PiB uptake was elevated in a minority of the subjects with PD, these individuals also exhibited lower levels of pattern expression relative to their AD counterparts. Neuropathology was ultimately available in a subset of the subjects with PD and was consistent with the PCA results. Overall, the findings point to the utility of network tools in the quantitative analysis of PiB PET scans from individuals with neurodegenerative disease. The study also highlights the potential clinical applicability of this approach to improve the accuracy of diagnosis on the individual case level. Indeed, PCA-based techniques have already been applied successfully in scans of meta- bolic activity or cerebral perfusion as a means of enhanc- ing the accuracy of diagnosis in parkinsonian disorders 7 and in dementing illnesses. 8,9 The current results dem- onstrate the potential value of this analytical approach in individuals with cognitive impairment undergoing amy- loid imaging. While the initial findings are encouraging, several caveats need to be appreciated before this and related classification algorithms can be brought to the bedside. For example, how replicable are the various disease top- ographies across different populations of patients and control subjects, particularly when scanning is under- taken using different imaging platforms under a variety of physiologic conditions? Are the regional loadings on the respective disease patterns reliable across subjects or are there topographic instabilities based on outliers or heterogeneities in the “training” sets? Are prospec- tively computed quantifiers of pattern expression stable on “test-retest” evaluation? Even if the appropriate per- formance criteria are met, the ultimate value of the tech- nique can be determined only through a multicenter diagnostic study in which ambiguous cases are followed clinically by investigators blind to the results of the cat- egorization algorithm. Finally, correlation is not syn- onymous with causation. These observations do not address one of the fundamental questions related to PiB studies in degenerative dementias, namely whether the demonstrated amyloid accumulation is linked etio- logically to the cognitive decline or whether both may be secondary to some other pathologic mechanism. That said, what Campbell et al. contribute in this work is an From the Center for Neurosciences (D.E.), The Feinstein Institute for Medical Research, Manhasset, NY; and the Department of Medicine (Neurology) (W.M.), University of Alberta, Edmonton, Canada. Go to Neurology.org for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the editorial. 516 © 2013 American Academy of Neurology