TSHVNet: Simultaneous Nuclear Instance Segmentation and Classification in Histopathological Images Based on Multiattention Mechanisms

Table 1

Parameter number and the tensor dimension of each layer of TSHVNet.

Name	Parameters	Output shape
Residual block 1	145 K	[None, 256, 264, 264]
Residual block 2	938.5 K	[None, 512, 132, 132]
Residual block 3	4.86 M	[None, 1024, 66, 66]
Residual block 4	10.5 M	[None, 2048, 33, 33]
Transformer (×12)	85 M	[None, 1089, 768]
DecoderCup	7.08 M	[None, 1024, 33, 33]
Nuclear pixel (NP) branch
Upsample, Conv()	9.72 M	[None, 256, 62, 62]
Dense unit (×8)		[None, 32, 30, 30]
SimAM, Conv(), upsample, Conv()		[None, 128, 56, 56]
Dense unit (×4)		[None, 32, 40, 40]
SimAM, Conv(), upsample, Conv(), Conv()		[None, 2, 80, 80]
HoVer (HV) branch
Upsample, Conv()	9.72 M	[None, 256, 62, 62]
Dense unit (×8)		[None, 32, 30, 30]
SimAM, Conv(), upsample, Conv()		[None, 128, 56, 56]
Dense unit (×4)		[None, 32, 40, 40]
SimAM, Conv(), upsample, Conv(), Conv()		[None, 2, 80, 80]
Nuclear classification (NC) branch
Upsample, Conv()	9.72 M	[None, 256, 62, 62]
Dense unit (×8)		[None, 32, 30, 30]
SimAM, Conv(), upsample, conv ()		[None, 128, 56, 56]
Dense unit (×4)		[None, 32, 40, 40]
SimAM, Conv(), upsample, Conv (), Conv()		[None, 5, 80, 80]