The mean value of the non-displaceable binding potential (BPND) within a region of interest (ROI) is the traditionally-employed metric in neurological image analysis. The ability of the mean value to accurately track clinical disease progression may be limited since it does not capture the spatial pattern of tracer distribution. In this work, we employ the principal component analysis (PCA) to quantify the clinicallyrelevant tracer binding patterns ([11C]dihydrotetrabenazine) in high-resolution PET images of 37 Parkinson's disease subjects. The principal component (PC) scores that correspond to different binding patterns in the putamen ROIs are combined with the mean BPND and used as the input to several linear models that aim to predict the clinical severity of the disease (disease duration). Multiple regression analysis and LASSO (least absolute shrinkage and selection operator) with cross-validation are used to evaluate the contributions of the PC scores to the accuracy of the tested models. With multiple regression analysis, the value of the adjusted R2 was 0.57 when the mean BPND alone was used as the model input. When the PC scores were included as additional input variables, the value of the adjusted R2 increased to 0.70. The terms of the model representing the PC scores were statistically significant (p<0.01). In LASSO analysis, the crossvalidated accuracy improved by 25% when the PC scores were added to the input (compared to using the mean BPND alone). These results demonstrate that a) the disease- and tracer-specific binding patterns can be identified in sub-cortical brain structures from high-resolution PET images, and b) such patterns may facilitate better models of the clinical disease metrics.