abstract |
The invention discloses a three-dimensional medical image unsupervised feature extraction system, comprising the following steps: step 1, acquiring a CT image data set, and decomposing 3D image data including a prediction target into 2D views of different viewing angles; The unsupervised feature extraction model of view comparison learning extracts features from 2D view data to generate features corresponding to each view; each view builds an independent network structure, and each network structure in the feature extraction model is composed of An encoder and a mapping head are composed; the similarity of views is calculated; step 3, the extracted features are fused and classified and predicted: the representation is extracted from the encoder for feature fusion, the fusion representation is input into the classifier, and the dimension of the input representation is , to obtain the probability of benign and malignant nodules. The present invention has the advantages that the quality of representation learning can be improved and the accuracy of target acquisition can be improved. |