The invention relates to the technical field of mode recognition, in particular to a cardiac image
data processing system based on
deep learning, which comprises an initial map construction module, an artifact
branch pruning module, a fracture path repairing module and a topological smooth output module. According to the method,
initial topology containing node coordinates is constructed, the maximum inscribed circle area is calculated to generate circulation attributes, virtual
blood flow injection
simulation and flow resistance calculation are utilized to accurately quantify
branch traffic capacity, artifact interference is eliminated according to the matching relation between the
blood flow proportion and the flow resistance, and suspended node geometric features and a path texture sequence are combined, so that the
branch traffic capacity is accurately quantified. A graph neural network is utilized to predict a connection probability to repair a fracture path, a node coordinate weighted updating mechanism is matched to guarantee smoothness and continuity of a topological structure, the problems of artifact
confusion and path fracture in a traditional method are effectively solved, and the construction accuracy and
automation level of a
heart structure model are remarkably improved.