The invention relates to the technical field of
atrial fibrillation recognition, in particular to an
artificial intelligence atrial fibrillation recognition system based on an
MRI image,
computer equipment and a storage medium, and the method comprises the steps: inputting an MRI-DWI
image sequence collected at the early stage of admission of a
stroke patient; performing standardized preprocessing on the image; entering a double-
branch architecture of an AI model, and extracting semantic features of a
stroke focus area; meanwhile, multi-dimensional image
omics features are extracted from the
focus area; the semantic features and the multi-dimensional image
omics features are fused, and an AF
classification result is output through a full connection layer Softmax; in the training stage, Dice loss and a
cross entropy loss function are used for joint optimization, and in the testing stage, an AF
prediction probability is output. The method can be expanded to other high-
risk screening fields of cardiac
stroke, atrial
myopathy and the like through transfer learning. The
system can also be integrated in a PACS
system, automatic uploading, analysis and result pushing of images are achieved, and a clinical auxiliary basis is provided for a post-stroke anticoagulation strategy.