Fig. 2: ROC curves for dyskinesia and ON/OFF classification using different feature sets. | npj Parkinson's Disease

Fig. 2: ROC curves for dyskinesia and ON/OFF classification using different feature sets.

From: Spontaneous eye blink-based machine learning for tracking clinical fluctuations in Parkinson’s disease

Fig. 2

A ROC curves for dyskinesia classification during ON state, evaluated across patient groups including both dyskinetic and non-dyskinetic individuals, using models trained with blink-related features (“B”), blink + background features (“B + BG”), and plasma L-dopa concentration (“L”). B ROC curves for ON/OFF classification evaluated across all test patients using the same feature sets. The upper panels show AUCROC values for raw predictions, while the lower panels show AUCROC values after applying a smoothing technique on the prediction. The blue curves represent the mean ROC, with the shaded area indicating the 95% confidence interval (CI) of all ROC curves. Mean AUCROC values are displayed on each panel, with distinct letters indicating statistically significant differences (paired t-test with Bonferroni correction, α = 0.0033).

Back to article page