Table 1.

Epithelium detection network architecture

Layers	Configurations	Size
Input	-	3 × 250 × 250
Convolution block 1	[k: 3 × 3, s: 1, p: 1] × 2	64 × 250 × 250
Pool 1	mp: 2 × 2, s: 2	64 × 125 × 125
Convolution block 2	[k: 3 × 3, s: 1, p: 1] × 2	128 × 125 × 125
Pool 2	mp: 2 × 2, s: 2	128 × 62 × 62
Convolution block 3	[k: 3 × 3, s: 1, p: 1] × 4	256 × 62 × 62
Pool 3	mp: 2 × 2, s: 2	256 × 31 × 31
Convolution block 4	[k: 3 × 3, s: 1, p: 1] × 4	512 × 31 × 31
Pool 4	mp: 2 × 2, s: 2	512 × 15 × 15
Convolution block 5	[k: 3 × 3, s: 1, p: 1] × 4	512 × 15 × 15
Pool 5	mp: 2 × 2, s: 2	512 × 7 × 7
Flatten	-	25088 × 1
FC 1	nh: 1024	1024 × 1
Dropout	prob: 0.5	1024 × 1
FC 2	nh: 1024	1024 × 1
FC 3	nh: 2	2 × 1
Output	softmax	2 × 1

k, s, p, mp, nh, prob are kernel, stride size, padding size, max pooling, number of neurons, and probability, respectively. FC: Fully connected single-layer neural network