Parkinsonian differentiation using PCA image correlation scores

Atypical parkinsonian syndromes are often difficult to diagnose because they present common clinical symptoms and differences in diagnostic images are subtle. Multivariate covariance analysis has been previously used in PET group data to identify neurodegenerative disease patterns. In particular, using SSM-PCA analysis, individual subject´s pattern expression of characteristic disease patterns have been shown to correlate with independent measures of disease status. These scalar subject scores, evaluated as the inner product of the unitized pattern vector and the mean centered subject data vector, can be utilized in classification algorithms to differentiate patients requiring disease specific treatment. However, diagnostic accuracy is often compromised stemming from topographic pattern similarity resulting in overlapping disease score expression. Here, we show that some improvement in classification may be achieved by utilizing the inner product of standardized pattern/patient vectors equivalent to the Pearson´s correlation coefficient to evaluate subject class scores.