Table 3. Quantitative evaluation of VCModule for six classes in the Luhe and Xuanwu Hospital datasets.

Dataset	Class	Sensitivity	Specificity	PPV	NPV	F1 score
Lehe dataset (Train)	OVCF	0.984	0.991	0.939	0.998	0.961
	OF	0.886	0.998	0.954	0.994	0.919
	SN	0.713	0.998	0.781	0.997	0.745
	KD	0.917	1.000	0.963	0.999	0.939
	PS	0.993	1.000	1.000	1.000	0.997
Lehe dataset (Test)	OVCF	0.958	0.986	0.910	0.994	0.934
	OF	0.808	0.996	0.911	0.991	0.856
	SN	0.756	0.995	0.532	0.998	0.624
	KD	0.913	0.999	0.913	0.999	0.913
	PS	0.997	0.999	0.986	1.000	0.992
Xuanwu dataset (Test)	OVCF	0.980	0.957	0.775	0.997	0.866
	OF	0.741	0.997	0.860	0.994	0.796
	SN	0.213	1.000	0.842	0.994	0.340
	KD	0.874	1.000	0.982	0.998	0.925
	PS	0.970	1.000	0.991	0.999	0.980

VCModule, vertebra classification module; PPV, positive predictive value; NPV, negative predictive value; OVCF, osteoporotic vertebral compression fractures; OF, old fracture; SN, Schmorl’s node; KD, Kummell’s disease; PS, previous surgery.