Table 1. Structures of networks. Global average pooling (GAP) is short for the global average pooling layer.

Block names	Information map extractor F	3D ResNet10 (or 34)
Conv1	kernel size =7*7*7, channel =64, stride = (2, 2, 2)
Max pooling	kernel size =3*3*3, stride = (2, 2, 2)
Block1 (residual structure)	$\left[\begin{array}{c}3\text{*}3\text{*}3,64\\ 3\text{*}3\text{*}3,64\end{array}\right]\times 2$	$\left[\begin{array}{c}3\text{*}3\text{*}3,64\\ 3\text{*}3\text{*}3,64\end{array}\right]\times 1\left(3\right)$
Block2 (residual structure)	$\left[\begin{array}{c}3\text{*}3\text{*}3,128\\ 3\text{*}3\text{*}3,128\end{array}\right]\times 2$	$\left[\begin{array}{c}3\text{*}3\text{*}3,128\\ 3\text{*}3\text{*}3,128\end{array}\right]\times 1\left(4\right)$
Block3 (residual structure)	$\left[\begin{array}{c}3\text{*}3\text{*}3,256\\ 3\text{*}3\text{*}3,256\end{array}\right]\times 2$	$\left[\begin{array}{c}3\text{*}3\text{*}3,256\\ 3\text{*}3\text{*}3,256\end{array}\right]\times 1\left(6\right)$
Block4 (residual structure)	$\left[\begin{array}{c}3\text{*}3\text{*}3,512\\ 3\text{*}3\text{*}3,512\end{array}\right]\times 2$	$\left[\begin{array}{c}3\text{*}3\text{*}3,512\\ 3\text{*}3\text{*}3,512\end{array}\right]\times 1\left(3\right)$
Block5	$\left[\begin{array}{l}\begin{array}{c}1\text{*1*1},256\\ 1\text{*1*1},128\end{array}\\ 1\text{*1*1},1\end{array}\right]\times 1$	–
	–	GAP and FC

FC, fully connected layer.