This invention discloses an electromagnetic registration method, device, and medium for a bronchial
navigation system, relating to the fields of medical navigation and electromagnetic
positioning technology. The method first acquires CT image data of the target
bronchus, constructs a three-dimensional image space, and completes
airway image segmentation. Then, based on the
airway image segmentation results, it extracts key points of the
bronchus and the
airway centerline in the image space, while simultaneously acquiring the corresponding key points and centerline point sets in the actual space using an electromagnetic
positioning system. Subsequently, it uses the dual-space key points to complete coarse registration of the discrete
point set, obtaining an initial
rigid body transformation matrix. Finally, it uses this matrix to initialize the
pose of the image centerline, and employs a
point cloud iterative
algorithm to complete fine registration of the centerline, outputting the final
spatial transformation matrix. This invention employs a
deep learning key point extraction and a two-stage coarse and fine registration strategy, significantly improving the robustness and accuracy of bronchial navigation registration, and adapting to precise navigation of equipment with complex airway structures.