Table 2.

Areas under the receiver operating characteristic curve for diagnosing Parkinson’s disease with the convolutional neural network

Metrics	Cross-validation					Pooled
	1	2	3	4	5
NOS	0.688	0.767	0.764	0.887	0.843	0.761 (0.698–0.823)
DTI
Axial diffusivity	0.747	0.722	0.713	0.664	0.864	0.715 (0.649–0.781)
Mean diffusivity	0.726	0.656	0.713	0.618	0.783	0.672 (0.602–0.741)
Radial diffusivity	0.667	0.633	0.703	0.684	0.871	0.698 (0.631–0.766)
Fractional anisotropy	0.703	0.724	0.775	0.665	0.853	0.733 (0.669–0.798)
DKI
Axial kurtosis	0.839	0.941	0.972	0.911	0.957	0.891 (0.848–0.934)
Mean kurtosis	0.864	0.845	0.868	0.860	0.989	0.878 (0.833–0.923)
Radial kurtosis	0.847	0.879	0.913	0.854	0.987	0.895 (0.853–0.937)
NODDI
Intracellular volume fraction	0.784	0.847	0.798	0.747	0.896	0.801 (0.744–0.858)
Isotropic volume fraction	0.688	0.798	0.745	0.730	0.868	0.749 (0.686–0.811)
Orientation dispersion	0.561	0.760	0.743	0.662	0.817	0.691 (0.623–0.759)
g-ratio
Axon volume fraction	0.749	0.871	0.837	0.926	0.915	0.836 (0.784–0.888)
Myelin volume fraction	0.681	0.773	0.790	0.788	0.834	0.763 (0.701–0.824)
g-ratio	0.665	0.760	0.830	0.786	0.866	0.767 (0.706–0.827)

Note: The 95% confidence interval is shown in parentheses