3-1. Convolutional model
In our research, we proposed a simple convolutional neural network model that processed 229x229 MRI image data. The model incorporated RGB color channels with a batch size of 64 for the input size. To recognize curves, arcs, and borders, we added a 16-filter CNN layer with a filter size of 3x3, followed by a max-pooling layer with a filter size of 2x2. To capture a comprehensive view of the image, we included convolution layers and filters with increasing numbers (16, 32, 64, 128, 256, and 512) and 3x3 filter sizes, based on the Resnet and VGG networks. The model was able to identify larger patterns by increasing the number of filters and merging the layers, and the max-pooling layers applied to the CNN layers provided the most benefits. Finally, we combined the Softmax output layer with a fully connected dense layer of 1024 neurons, which calculated the probability for each class and determined whether the input MRI image contained cancer or not. To improve the efficiency of the proposed model, we also applied the Drop Out technique for network regularization and pruning. The proposed convolution architectural plan is shown in Figure 3-3.