Accurate digital quantification of tau pathology in progressive supranuclear palsy

Table 4 Classification performance with and without threshold-moving method from training for screening and subsequent tau classifiers

Classifier	Precision	Recall	F1-score	PR-AUC
Without threshold-moving approach
Screening	0.96 (± 0.044)	0.96 (± 0.053)	0.96 (± 0.060)	0.99 (± 0.011)
Cortex	0.92 (± 0.037)	0.92 (± 0.017)	0.91 (± 0.027)	0.98 (± 0.010)
PU	0.90 (± 0.046)	0.86 (± 0.068)	0.86 (± 0.061)	0.98 (± 0.015)
STN & GP	0.93 (± 0.038)	0.86 (± 0.056)	0.87 (± 0.061)	0.97 (± 0.020)
DN	0.96 (± 0.020)	0.93 (± 0.041)	0.94 (± 0.029)	0.98 (± 0.016)
With threshold-moving approach
Screening	0.96 (± 0.046)	0.96 (± 0.057)	0.95 (± 0.064)	0.99 (± 0.011)
Cortex	0.95 (± 0.025)	0.95 (± 0.027)	0.95 (± 0.021)	0.98 (± 0.010)
Putamen	0.94 (± 0.024)	0.92 (± 0.049)	0.93(± 0.037)	0.98 (± 0.015)
STN & GP	0.95 (± 0.026)	0.92 (± 0.042)	0.93 (± 0.027)	0.97 (± 0.020)
DN	0.96 (± 0.028)	0.95 (± 0.026)	0.95 (± 0.023)	0.98 (± 0.016)

Tau classifiers for the cortex, putamen (PU), subthalamic nucleus and globus pallidus (STN & GP), and dentate nucleus (DN). Classifiers were tuned for area under the precision-recall curve (PR-AUC), where precision, recall and F1-score were calculated. Mean values from cross-validation and standard deviation in brackets are reported

ISSN: 2051-5960