Deep Learning Models to Predict Fatal Pneumonia Using Chest X-Ray Images

Figure 2

Network architectures of the ResNet model. The revised template of the network (tutorial.basics. 12_residual_learning) (https://dl.sony.com/) was used to provide the structure of the deep neural network. The square box indicates the function of the layer. The numbers to the right of the box indicate the specifications of each layer. For example, the three numbers to the right of the first input layer indicate the number of colors and size (height and width) of the input image, respectively. In the second convolutional layer, the same format is used to indicate the number of outputs and size (height and width) of the feature map, respectively. “ReLU” stands for rectified linear unit. “Kernel shape” indicates the pixel size of each filter for convolution of the input. The NNC models were trained with a batch size of 16, epochs of 100, and Adam optimization (learning rate 0.001). For other specifications, please refer to the reference (Sony Network Communications Inc. 2020).