Method for evaluating chest DR image quality based on feature fusion

Abstract

The application discloses a chest DR image quality evaluation method based on feature fusion, first, collecting chest DR images and segmenting to obtain a region of interest; then, using an imageomics feature extraction network and a deep learning feature extraction network to extract imageomics features and deep learning features respectively; at the same time, using a double-feature contrast prediction coding and a contrast logarithmic upper limit combination method to optimize the deep learning feature extraction network through complementary contrast learning; finally, performing feature fusion through a deep learning classification head to classify the image quality grade. On the one hand, the application considers the interaction of imageomics features and deep learning features, realizes comprehensive interaction between features by using feature fusion technology, improves the representation ability of the fused features, improves the prediction performance and ensures the accuracy of the quality grade evaluation; on the other hand, the application maximizes and minimizes the mutual information between features through a complementary contrast learning strategy, enhances the complementarity between features and improves the model interpretability.

Images