Cascade and Fusion of Multitask Convolutional Neural Networks for Detection of Thyroid Nodules in Contrast-Enhanced CT

Table 2

Architecture of CNN-1 and CNN-2 in this study.

Layer	Input size	CNN-1	Input size	CNN-2
Conv_1	256 × 256 × 1	3 × 3, 32	256 × 256 × 1	3 × 3, 32
Dense-block_1	256 × 256 × 32		256 × 256 × 32
Max-pooling_1	256 × 256 × 64	2 × 2, stride 2	256 × 256 × 64	2 × 2, stride 2
Dense-block_2	128 × 128 × 64		128 × 128 × 64
Max-pooling_2	128 × 128 × 128	2 × 2, stride 2	128 × 128 × 128	2 × 2, stride 2
Dense-block_3	64 × 64 × 128		64 × 64 × 128
Max-pooling_3	64 × 64 × 256	2 × 2, stride 2	64 × 64 × 256	2 × 2, stride 2
Dense-block_4	32 × 32 × 256		32 × 32 × 256
Max-pooling_4	32 × 32 × 512	2 × 2, stride 2	32 × 32 × 384	2 × 2, stride 2
Dense-block_5	16 × 16 × 512		16 × 16 × 384
Average-pooling	16 × 16 × 1024	16 × 16	16 × 16 × 512	16 × 16
Fully connected layer	1024	2	512	2
Output	2		2

Here, “conv” denotes convolutional layer. Number formats of CNN-1 and CNN-2 are all: convolution kernel size, number of convolution kernels.