Institution: | 1. Department of Otorhinolaryngology Head and Neck Surgery, The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi 530000, China;2. School of Computer Science and Cyber Engineering, Guangzhou University, Guangzhou, China;1. Department of Otorhinolaryngology, Nagoya University Graduate School of Medicine, Nagoya, Japan;2. Department of Radiology, Nagoya University Graduate School of Medicine, Nagoya, Japan;1. Department of Otolaryngology, The University of Tokyo, Tokyo, Japan;2. Department of Neurology, The University of Tokyo, Tokyo, Japan;1. Department of Radiology, Gifu University, 1-1 Yanagido, Gifu 501-1194, Japan;2. Department of Otolaryngology, Gifu University, Gifu, Japan;1. Division of Otolaryngology, Tohoku Medical and Pharmaceutical University Hospital, 1-12-1, Fukumuro, Miyaginoku, Sendai 983-8512, Japan;2. Department of Otolaryngology, Yamagata City Hospital Saiseikan, Yamagata, Japan;3. Division of Anatomy and Cell Biology, Tohoku Medical and Pharmaceutical University, 1-12-1, Fukumuro, Miyaginoku, Sendai 983-8512, Japan;4. Division of Pathology, Tohoku Medical and Pharmaceutical University, 1-12-1, Fukumuro, Miyaginoku, Sendai 983-8512, Japan |
Abstract: | ObjectiveTo investigate the feasibility of a deep learning method based on a UNETR model for fully automatic segmentation of the cochlea in temporal bone CT images.MethodsThe normal temporal bone CTs of 77 patients were used in 3D U-Net and UNETR model automatic cochlear segmentation. Tests were performed on two types of CT datasets and cochlear deformity datasets.ResultsThrough training the UNETR model, when batch_size=1, the Dice coefficient of the normal cochlear test set was 0.92, which was higher than that of the 3D U-Net model; on the GE 256 CT, SE-DS CT and Cochlear Deformity CT dataset tests, the Dice coefficients were 0.91, 0.93, 0 93, respectively.ConclusionAccording to the anatomical characteristics of the temporal bone, the use of the UNETR model can achieve fully automatic segmentation of the cochlea and obtain an accuracy close to manual segmentation. This method is feasible and has high accuracy. |