Application of segmented neural network intelligent model in nasal-skull base endoscopy surgery
XU Wei
ZHAO Hao
CAO Ke
QIU Yongming
ZHOU Zhiyi
MIAO Yifeng
Abstract：Objective To address complex anatomy and high complications risks from misoperation in skull base neurosurgery by means of a semantic segmentation neural network to realize real-time recognition/segmentation of the sellar floor surgical area.Methods Clinical data of 20 patients(10 males,mean age(55±18)years;10 females,mean age(45±19)years)who underwent transsphenoidal endoscopic sellar surgery at Shanghai Renji Hospital and Fujian Fuding Hospital(Aug.2023-Aug.2024)were chosen as study subjects.Endoscopic videos were used to build datasets.ResNet101 was applied for sellar floor classification.RefineNet(ResNet-based,multi-level feature fusion)was used for sellar floor segmentation.The two networks formed a segmented neural network with shared ResNet feature extraction layer,receiving gradient optimization from both tasks during backpropagation.Results Based on endoscopic videos of these 20 patients,a classification dataset of over 1 million images and a segmentation dataset of over 100 000 images(a subset of the classification dataset)were constructed.The segmented neural network achieved a classification accuracy of 94.73%and a segmentation accuracy of 92.41%,both superior to those of the networks trained individually.Conclusions Neural networks can be successfully applied to skull base endoscopic surgery.The segmented architecture can improve computational efficiency and recognition accuracy via feature sharing,aiding intraoperative sellar floor localization.Future integration with surgical instrument recognition and preoperative navigation will optimize real-time anatomical annotation.
Keywords：neural endoscopyskull base surgerysegmented neural networksemantic segmentationsurgical assistance
Publication Date：2025-12-28
Online Publishing Date：2025-12-02(First online date of this platform, not the publication date of the document)
Pages：5( 59-63 )