Cervical Lesion Classification Method Based on Cross-Validation Decision Fusion Method of Vision Transformer and DenseNet

Table 3

Comparison results between this fusion model and other models.

Model	Classes	ACC	F1	PPV	SEN	SPEC	NPV
Vision Transformer	Cancer	0.6000	0.8350	0.8136	0.8600	0.9780	0.9846
	HSIL		0.6184	0.5726	0.6744	0.6210	0.7190
	LSIL		0.4366	0.4984	0.3912	0.8778	0.8282
	Negative		0.6018	0.6556	0.5584	0.9080	0.8674
ShuffleNetV2	Cancer	0.6040	0.7722	0.7892	0.7600	0.9780	0.9738
	HSIL		0.6076	0.5744	0.6466	0.6384	0.7062
	LSIL		0.4628	0.5486	0.4086	0.8986	0.8364
	Negative		0.6386	0.6284	0.6498	0.8790	0.8884
MobileNetV3	Cancer	0.6260	0.7910	0.7490	0.8400	0.9692	0.9824
	HSIL		0.6402	0.5960	0.6930	0.6456	0.7372
	LSIL		0.4686	0.5376	0.4176	0.8934	0.8370
	Negative		0.6564	0.7052	0.6166	0.9184	0.8840
EfficientNetV2	Cancer	0.6360	0.8160	0.8406	0.8000	0.9824	0.9780
	HSIL		0.6346	0.6218	0.6512	0.6982	0.7268
	LSIL		0.5446	0.6158	0.4956	0.9064	0.8584
	Negative		0.6390	0.6096	0.6750	0.8632	0.8944
DenseNet161	Cancer	0.6560	0.8672	0.8500	0.9000	0.9802	0.9890
	HSIL		0.6514	0.6684	0.6420	0.7544	0.7388
	LSIL		0.5666	0.5798	0.5654	0.8778	0.8724
	Negative		0.6486	0.6370	0.6666	0.8790	0.8938
Ours	Cancer	0.6800	0.9000	0.9000	0.9000	0.9890	0.9890
	HSIL		0.6890	0.6600	0.7210	0.7190	0.7740
	LSIL		0.6050	0.6500	0.5650	0.9090	0.8750
	Negative		0.6380	0.6520	0.6250	0.8950	0.8830