The invention discloses a
lung disease multi-
subtype classification method,
system and device based on chest images, and belongs to classification of
lung diseases in the technical field of
artificial intelligence medicine. The method aims at solving the technical problem that in the prior art, due to the fact that a
loss function cannot enable a model to pay attention to main
pathology types and all
pathology types in data at the same time, the multi-
subtype classification accuracy of the model is low, the multi-
subtype classification model is constructed, and three different losses are creatively synthesized during model training, so that the multi-subtype classification accuracy of the model is improved. The # imgabs0 # is used for enabling the model to predict all existing
pathological types, the # imgabs1 # is used for enabling the model to mainly learn to predict the main
pathological types, and the # imgabs2 # is used for enabling the model to restrain prototype vectors of various types to have distinction degrees, so that the model can better capture correlation among different subtypes while learning
image representation of main
pathological subtypes, and the model can be used for predicting the main pathological types. The generalization ability of the model is improved, the robustness of model training is improved, the
overfitting problem of the model is relieved, and the accuracy of the model for multi-subtype classification is improved.