Table 6. The ablation study on three different backbones: DenseNet121, ViT, and our network (DenseNet121 + ViT).

T(IoU)	Model	Atelectasis	Cardiomegaly	Effusion	Infiltration	Mass	Nodule	Pneumonia	Pneumothorax
0.1	DenseNet121	0.596	0.993	0.678	0.707	0.674*	0.113	0.786	0.418*
	ViT	0.365	0.937	0.392	0.586	0.306	0.000	0.429	0.202
	DenseNet121+ViT	0.673*	1.000*	0.713*	0.808*	0.633	0.127*	0.833*	0.330
0.2	DenseNet121	0.321	0.923	0.566	0.636	0.469	0.000	0.691	0.259*
	ViT	0.179	0.887	0.147	0.374	0.102	0.000	0.286	0.149
	DenseNet121+ViT	0.372*	1.000*	0.615*	0.737*	0.510*	0.000	0.738*	0.245
0.3	DenseNet121	0.160*	0.606	0.364	0.414	0.326	0.000	0.571	0.145
	ViT	0.100	0.740	0.063	0.242	0.061	0.000	0.071	0.096
	DenseNet121+ViT	0.160*	0.958*	0.420*	0.616*	0.347*	0.000	0.643*	0.160*
0.4	DenseNet121	0.064*	0.232	0.217	0.303	0.225*	0.000	0.500	0.103
	ViT	0.058	0.486	0.014	0.172	0.020	0.000	0.071	0.032
	DenseNet121+ViT	0.064*	0.704*	0.259*	0.535*	0.204	0.000	0.524*	0.106*
0.5	DenseNet121	0.013	0.070	0.105	0.222	0.123	0.000	0.309	0.053
	ViT	0.032*	0.225	0.007	0.091	0.020	0.000	0.024	0.021
	DenseNet121+ViT	0.006	0.409*	0.112*	0.424*	0.163*	0.000	0.381*	0.085*
0.6	DenseNet121	0.000	0.021	0.028	0.111	0.032	0.000	0.095	0.043*
	ViT	0.006*	0.070	0.007	0.040	0.000	0.000	0.000	0.000
	DenseNet121+ViT	0.006*	0.190*	0.042*	0.232*	0.041*	0.000	0.262*	0.043*
0.7	DenseNet121	0.000	0.007	0.007	0.061	0.020*	0.000	0.024	0.011
	ViT	0.006*	0.014	0.006	0.010	0.000	0.000	0.000	0.000
	DenseNet121+ViT	0.006*	0.049*	0.028*	0.162*	0.020*	0.000	0.119*	0.021*

*, the best results of our experiments. ViT, vision transformer; DenseNet121+ViT, our proposed method which includes two network branches, DenseNet121 and ViT; T(IoU), different IoU thresholds.