A new ordered pooling network based on multi-scale fusion feature for medical image recognition

Aiming at the imbalance problem of Computed Tomography(CT) images diagnosis model under artificial intelligence assisted method, a new ordered pooling network based on multi-scale fusion feature for medical image recognition is proposed. Based on traditional CNN(Convolutional Neural Networks) architecture, the new overall neural network mainly consists of three parts: Multi-scale feature fusion layer, deep feature ordered pooling layer and feature recognition layer. Multi-scale feature fusion layer upsamples and merges different scale CNN features to achieve multi-scale deep feature fusion. Deep feature ordered pooling layer regards the fusion spatial deep convolution features as a continuous Weisfeiler-Lehman(WL) color, which makes a meaningful ordering to weaken the excessive spatial information feature, and improves the generalization capability. Feature recognition layer combines final sorted deep features with the softmax layer to realize CT images recognition for disease diagnosis. The experimental results show that the proposed method can extract the multi-dimensional features of the image better and effectively improve the accuracy of the approximate image recognition.