CT image contrast feature learning method for new coronal pneumonia clinical typing

Abstract

The invention discloses a CT image contrast feature learning method for new coronal pneumonia clinical typing. The method comprises the following steps: S1, carrying out an FPN-based full-automatic lung segmentation algorithm; S2, constructing a feature learning network; S3, constructing a sample pair; and S4, carrying out comparative feature learning. According to the method, a convolutional neural network model based on feature comparison learning is adopted, and the feature distance and a cross entropy loss function are combined, so that the deep learning features of the samples of the samecategory are similar, the deep learning feature difference of the samples of different categories is large, the features are optimized, and the classification precision is improved. The CT image contrast learning method provided by the invention can be used for carrying out full-automatic image processing on new coronal pneumonia CT images so as to realize clinical typing diagnosis of the new coronal pneumonia.

Description

technical field [0001] The invention relates to a medical technology, in particular to a CT image contrast feature learning method for clinical classification of new coronary pneumonia. The present invention also relates to the use of the CT image contrast feature learning method for fully automatic image processing of CT images of new coronary pneumonia. Background technique [0002] The clinical classification of new coronary pneumonia can reflect the severity of new coronary pneumonia. According to the different clinical classifications, the treatment strategies adopted are different; in the management process of patients with new coronary pneumonia, the clinical classification of new coronary pneumonia is also to judge whether the patients meet the One of the criteria for discharge conditions. [0003] The gold standard for clinical classification of new coronary pneumonia needs to be judged by doctors' interpretation of CT images, combined with biochemical indicators s...

AI Technical Summary

Problems solved by technology

At present, the method of classifying CT images using only a single neural network is commonly used, which mainly mines high-dimensional information from a single CT image, ignoring the difference between different samples. The main defect of this method is that the intensive reading is relatively low. The information obtained by this method assists the clinical classification of new coronary pneumonia, which is not only inefficient, but also easily leads to misdiagnosis of subsequent clinical classification of new coronary pneumonia

Images