Research Article

Pixel-Label-Based Segmentation of Cross-Sectional Brain MRI Using Simplified SegNet Architecture-Based CNN

Table 1

Simplified SegNet network.

S. No.Layer nameTypeDescription

1“Image input”Image208 × 1761 images with “zero center” normalization
2“encoder1_conv1”Convolution64 3 × 3 × 1 convolutions with stride [1 1] and padding [1 1 1 1]
3“encoder1_bn_1”Batch normalizationBatch normalization with 64 channels
4“encoder1_relu_1”ReLUReLU
5“encoder1_conv2”Convolution64 3 × 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
6“encoder1_bn_2”Batch normalizationBatch normalization with 64 channels
7“encoder1_relu_2”ReLUReLU
8“encoder1_maxpool”Max pooling2 × 2 max pooling with stride [2 2] and padding [0 0 0 0]
9“encoder2_conv1”Convolution64 3 × 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
10“encoder2_bn_1”Batch normalizationBatch normalization with 64 channels
11“encoder2_relu_1”ReLUReLU
12“encoder2_conv2”Convolution64 3× 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
13“encoder2_bn_2”Batch normalizationBatch normalization with 64 channels
14“encoder2_relu_2”ReLUReLU
15“encoder2_maxpool”Max pooling2 × 2 max pooling with stride [2 2] and padding [0 0 0 0]
16“decoder2_unpool”Max unpoolingMax unpooling
17“decoder2_conv2”Convolution64 3× 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
18“decoder2_bn_2”Batch normalizationBatch normalization with 64 channels
19“decoder2_relu_2”ReLUReLU
20“decoder2_conv1”Convolution64 3 × 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
21“decoder2_bn_1”Batch normalizationBatch normalization with 64 channels
22“decoder2_relu_1”ReLUReLU
23“decoder1_unpool”Max unpoolingMax unpooling
24“decoder1_conv2”Convolution64 3 × 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
25“decoder1_bn_2”Batch normalizationBatch normalization with 64 channels
26“decoder1_relu_2”ReLUReLU
27“decoder1_conv1”Convolution4 3 × 3 × 64 convolutions with stride [1 1] and padding [1 1 1 1]
28“decoder1_bn_1”Batch normalizationBatch normalization with 4 channels
29“decoder1_relu_1”ReLUReLU
30“Softmax”SoftmaxSoftmax
31“Pixel_classify”Pixel classification layerClass weighted cross-entropy loss with “background,” “CSF,” “GM,” and “WM” classes