兰州大学学报（医学版）2025，Vol.51Issue(11)：16-21,28,7.DOI:10.13885/j.issn.2097-681X.2025.11.003
深度学习在坐骨神经超声图像识别中的应用研究
Application of deep learning in ultrasound image recognition of the sciatic nerve
摘要
Abstract
Objective To provide a segmentation model to identify the anatomy of the sciatic nerve in the piri-formis region in ultrasound images.Methods From March to May 2024,60 patients(34 female and 26 male)who underwent ultrasound examination of the sciatic nerve in the piriformis region in The Second Hospital of Lanzhou University were retrospectively collected.Their ultrasound images were collected and the locations of the glute maximus,piriformis,bone structure and sciatic nerve labeled to establish a data set.The U-Mam-ba ultrasound imaging automatic segmentation model was built using the mamba network,and the data set was used to train and verify the model.Mean Intersection over Union(mIoU),mean Dice Similarity Coeffi-cient(mDice),accuracy,sensitivity,specificity and precision were used to evaluate the performance of the model.Results Using the 949 labeled ultrasound images obtained as the data set,including an 852 as the training set and 97 as the test set.The mIoU of 0.984 and mDice coefficient of 0.992 were achieved on the training set.In addition,the mIoU of 0.741 and mDice coefficient of 0.841 were achieved on the test set,with an average accuracy of 0.986.After a 5-fold cross-validation,the median and upper and lower quartiles of IoU and Dice were 0.784(0.678,0.847)and 0.879(0.808,0.917).Conclusion Based on U-mamba framework,the deep learning model had automatically identified the anatomical structure of the sciatic nerve in the piriformis region,and obtained an ideal segmentation effect it could have a good clinical application.关键词
深度学习/U-Mamba/梨状肌/坐骨神经/超声图像Key words
deep learning/U-Mamba/piriformis/sciatic nerve/ultrasound image分类
医药卫生引用本文复制引用
张明,程偲,黄生辉,胡万均..深度学习在坐骨神经超声图像识别中的应用研究[J].兰州大学学报（医学版）,2025,51(11):16-21,28,7.基金项目
甘肃省自然科学基金资助项目(22JR5RA954) （22JR5RA954）
兰州市科技计划资助项目(2024-3-37) （2024-3-37）
