Abstract:

The research is devoted to the convolutional neural network model construction for image recognition of X-ray of patients with an established brain tumor diagnosis. The architecture of the proposed model consists of two convolution layers and one fully connected layer. The architecture of the proposed model makes it possible to achieve high accuracy and significantly reduce the training time of the model compared to pre-trained models. To compare the quality of training of the proposed model, the pre-trained models VGG-16, VGG-19, Inception-V3, InceptionResNet-V2, ResNet-50, ResNet-152 and Xception are considered. It is shown that for the models VGG-16, VGG-19, Xception and the proposed model, accuracy is equal to 98.84%, but the F-score for the proposed model was higher and amounted to 98.96%. The training time of all the models considered differs depending on the the neural network architecture. For the proposed CNN model, the training time is more than 4 times less than for the training of the Inception-V3 model and more than 32 times less than the training time of the ResNet-152 model. There were no detected signs of the disease among the samples of X-ray images of patients with an established diagnosis of brain tumor was 0.783%. The proposed neural network model can act as an additional tool of a doctor in the brain tumor diagnosis.