Chest image diagnosis report automatic generation method capable of balancing cross-modal information

The invention provides a chest image diagnosis report automatic generation method for balancing cross-modal information, and relates to the field of medical information. The network fully considers the relation between an image and a text by optimizing an encoder-decoder framework, the network focuses on image feature processing during encoding, and multi-dimensional feature extraction is performed on an X-ray image region through a DFC-MSPCNN and a multi-receptive field residual network focus region information extraction module; during decoding, the network focuses on generation of a medical report, searching of an intermediate state between an image and a text through an image and text alignment module CMITAM, introduction of the intermediate state into a report generation module based on a Transform decoding layer, validity verification on IUX-Ray and MIMIC-CXR reference data sets, and results show that the network has the advantages that the network can be used for medical report generation, and the netw