This invention relates to the field of medical
image processing technology, specifically to an automatic detection and classification method and
system for intracranial hemorrhage based on three-window CT fusion and
label-dependent graph neural network, including the following steps: S100, acquiring the original CT slice image of the head to be detected, performing
grayscale mapping on the original slice
image based on preset brain window, subdural window and bone window respectively, and stacking the three mapped images along the channel dimension to generate a three-channel fused image; This invention greatly enriches the model's global
perception ability of multi-level physiological structures such as brain
parenchyma, microhematomas and
skull lesions through innovative three-window
physical mapping and channel fusion; it introduces a
hybrid adjacency matrix graph neural network that combines prior knowledge of neuroradiology with data-driven approach, explicitly establishing pathophysiological causal relationships between various hemorrhage subtypes at the
model architecture level, effectively eliminating logical contradictions in predictions that violate clinical common sense.