Skip to main content

Table 2 Architecture of the proposed convolutional neural network

From: Comparative clinical evaluation of atlas and deep-learning-based auto-segmentation of organ structures in liver cancer

Block typeIngredientsSize of feature maps
Input256 × 256 ×1
Down layer (D1)conv+res + drop+conv+batchnorm+max128 ×128 × 64
Down layer (D2)conv+res + drop+conv+batchnorm+max64 ×64 × 128
Down layer (D3)conv+res + drop+conv+batchnorm+max32 ×32× 256
Down layer (D4)conv+res + drop+conv+batchnorm+max16 ×16 × 512
Bridge layer (B)conv+res + conv16 ×16 × 1024
Upscaling layer (U1)deconv+merge+conv+res + conv32 × 32 × 512
Upscaling layer (U2)deconv+merge+conv+res + conv64 × 64 × 256
Upscaling layer (U3)deconv+merge+conv+res + conv128 × 128 × 128
Upscaling layer (U4)deconv+merge+conv+res + conv256 × 256 × 64
Outputconv256 × 256 ×1