Table 3.

Cancer detection stage model architecture.

Module	Layers	Number of filter	Filter size	Output size
Input		-	-	3 × 128 × 128
VGG11bn	Conv + BN + ReLU	64	3 × 3	64 × 128 × 128
	Max Pool	64	2 × 2	64 × 64 × 64
	Conv + BN + ReLU	128	3 × 3	128 × 64 × 64
	Max Pool	128	2 × 2	128 × 32 × 32
	Conv + BN + ReLU	256	3 × 3	256 × 32 × 32
	Conv + BN + ReLU	256	3 × 3	256 × 32 × 32
	Max Pool	256	2 × 2	256 × 16 × 16
	Conv + BN + ReLU	512	3 × 3	512 × 16 × 16
	Conv + BN + ReLU	512	3 × 3	512 × 16 × 16
	Max Pool	512	2 × 2	512 × 8 × 8
	Conv + BN + ReLU	512	3 × 3	512 × 8 × 8
	Conv + BN + ReLU	512	3 × 3	512 × 8 × 8
	Max Pool	512	2 × 2	512 × 4 × 4
Instance feature embedding	Conv	256	1 × 1	256 × 4 × 4
	FC + ReLU + Dropout	-	-	256
Attention module	FC + Tanh + Dropout	-	-	512
	FC	-	-	1
Classifier	FC	-	-	1