公开(公告)号：US20240203108A1

公开(公告)日：2024-06-20

申请号：US17802953

申请日：2022-02-28

Applicant: ZHEJIANG UNIVERSITY

Inventor： JING WANG ,  BO DONG ,  HONGJIAN HE ,  XIUJUN CAI

IPC: G06V10/80 ,  A61B34/10 ,  G06V10/26 ,  G06V10/74 ,  G06V10/77 ,  G06V10/776 ,  G06V20/70

CPC classification number: G06V10/806 ,  A61B34/10 ,  G06V10/26 ,  G06V10/761 ,  G06V10/7715 ,  G06V10/776 ,  G06V20/70 ,  A61B2034/107 ,  G06V2201/03

Abstract: The present invention discloses a decoupling divide-and-conquer facial nerve segmentation method and device. As for the characteristics of a small facial nerve structure and a low contrast, a facial nerve segmentation model including a feature extraction module, a rough segmentation module, and a fine segmentation module is constructed. The feature extraction module is configured to extract a low-level feature and a plurality of different-and high-level features. The rough segmentation module is configured to globally search the different-and high-level features for facial-nerve features and fuse them. The fine segmentation module is configured to decouple a fused feature to obtain a central body feature. After the central body feature is combined with the low-level feature to obtain an edge-detail feature, a space attention mechanism is used to extract attention features from the central body feature and the edge-detail feature, to obtain a facial nerve segmentation image. The method improves the precision and speed of automatic facial nerve segmentation, and meets the needs of preoperative path planning for robotic cochlear implantation.