Table 2.

Diagnostic performance of the AI models with different pipelines

Test dataset	XM-pipeline	Baseline	HE	CLAHE	UM
	AUC (95% CI)	AUC (95% CI) (p)	AUC (95% CI) (p)	AUC (95% CI) (p)	AUC (95% CI) (p)
VHDR1	0.970 (0.967–0.972)	0.966 (0.961–0.971) (< 0.001)	0.962 (0.958–0.965) (< 0.001)	0.965 (0.963–0.968) (0.097)	0.965 (0.963–0.966) (0.002)
IHDR2	0.948 (0.944–0.951)	0.898 (0.887–0.909) (< 0.001)	0.934 (0.929–0.939) (< 0.001)	0.931 (0.927–0.935) (< 0.001)	0.914 (0.902–0.926) (< 0.001)
SZDR3	0.956 (0.951–0.960)	0.908 (0.895–0.921) (< 0.001)	0.949 (0.944–0.954) (0.005)	0.933 (0.923–0.943) (< 0.001)	0.917 (0.901–0.932) (< 0.001)
IHCR1	0.945 (0.940–0.950)	0.899 (0.889–0.909) (< 0.001)	0.933 (0.925–0.941) (0.063)	0.920 (0.909–0.931) (< 0.001)	0.906 (0.887–0.924) (< 0.001)
IHCR2	0.944 (0.939–0.948)	0.658 (0.622–0.692) (< 0.001)	0.917 (0.908–0.926) (0.001)	0.705 (0.662–0.749) (< 0.001)	0.544 (0.520–0.567) (< 0.001)
MGCR4	0.977 (0.974–0.981)	0.918 (0.899–0.936) (< 0.001)	0.966 (0.957–0.974) (0.372)	0.958 (0.952–0.963) (0.038)	0.946 (0.922–0.970) (0.003)
IHCR3,Mobile	0.949 (0.940–0.957)	0.937 (0.927–0.947) (0.043)	0.933 (0.922–0.944) (0.042)	0.932 (0.923–0.942) (0.009)	0.925 (0.912–0.938) (0.001)

AUC Area under ROC curve, CI Confidence interval, CLAHE Contrast-limited histogram equalization, HE Histogram equalization, UM Unsharp masking