Fig. 5

Mean relative loss of accuracy (a) and “maximum” relative loss of accuracy (b) of the putamen SBR z-score for differentiation between PD patients and HC subjects of the PPMI or between patients with neurodegenerative PS and patients with non-neurodegenerative PS of the clinical sample as a function of size of the normal database (DB) used to estimate mean and standard deviation of normal putamen SBR for transforming SBR values into z-scores. SBR values were Ln-transformed prior to transforming them into z-scores in all settings. The relative loss of accuracy was computed with respect to the benchmark accuracy (normal database comprising all HC subjects or all patients with non-neurodegenerative PS), that is, relative accuracy loss (%) = 100 * (benchmark accuracy − accuracy) / benchmark accuracy. The maximum accuracy loss corresponds to the 5th percentile of the accuracy estimates over the 10,000 random samples to generate the normal database.