Brain Tumour Classification Algorithm Based on Multi-scale Convolution and Attention Mechanism

This paper proposes a classification algorithm that integrates a multi-scale convolutional structure and adaptive attention mechanism (SEblock) for the automatic classification technology of MRI brain tumors, which is used to extract high-dimensional features and recognize images that cannot be observed by the human eye. The model mentioned in this paper is improved on the pre-trained resnet50 network, and two smaller and larger datasets are used to improve the generalization ability and anti-interference data ability of the model, and the convolutional kernel of 3*3, 5*5, and 7*7 is fused to obtain receptive fields of different scales and increase the feature extraction ability of different fine-grained features. The SEblock module is added to increase the perception of key feature channels and improve the accuracy of classification. On the two datasets published by Kaggle, the model achieves an accuracy of 79.93% and 85.35%, respectively, indicating that the model has a certain degree of accurate classification ability. The existing models still have many deficiencies in accuracy, and more improvements will be introduced in the attention mechanism in the future.