摘要
基于脑组织分割的精确头模型有助于提升经颅磁刺激的治疗效果,但由于人脑的复杂性,很难实现精确的脑组织分割。为此,本文提出了基于迁移学习的多尺度注意网络,该网络可以学习多模态数据之间的互补信息,采用迁移学习方法解决小样本数据引起的过拟合问题,利用膨胀卷积提取多尺度特征,加入注意力机制提高脑组织分割的准确性。通过MRBrainS挑战赛验证了网络的有效性,在多项指标中取得了最好成绩。多尺度注意网络可以为个性化头模型的建立提供一个较好的分割结果,进而优化经颅磁刺激的治疗效果。
The personalized transcranial magnetic stimulation can improve the treatment effect, and the reconstruction of the head model based on brain tissue segmentation helps to optimize the personalized treatment plan. Therefore, in order to achieve automated brain tissue segmentation and improve segmentation accuracy, a multi-scale attention network(MSAN) based on transfer learning are proposed.In the network, the method of transfer learning is used to solve the overfitting problem caused by small sample data, and the complementary information of the multi-modality data can be learned. At the same time, the extraction and fusion of multi-scale features are used to improve the performance of the network, and the attention mechanism is used to focus the important information after the fusion of multi-scale features, which improves the accuracy of brain tissue segmentation. Extensive experiments have been conducted on the database provided by the MRBrainS Challenge to verify the effectiveness of our model.Currently, this method ranks second in the MRBrainS Challenge and has achieved the best results in multiple indicators. MSAN can provide a better brain tissue segmentation result for the reconstruction of personalized head models, and optimize the stimulation effect of transcranial magnetic stimulation.
作者
刘近贞
高国辉
熊慧
Liu Jin-Zhen;Gao Guo-Hui;Xiong Hui(School of Control Science and Engineering,Tiangong University,Tianjin 300387,China;Key Laboratory ofIntelligent Control of Electrical Equipment,Tiangong University,Tianjin 300387,China)
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2023年第2期576-583,共8页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(61871288)
天津市高等学校创新团队培养计划项目(TD13-5036)
天津市自然科学基金项目(18JCYBJC90400,18JCQNJC84000)
天津市教委科研计划项目(2019KJ014)。
关键词
计算机应用
神经网络
脑组织分割
多尺度注意网络
迁移学习
computer application
neural network
brain tissue segmentation
multi-scale attention network
transfer learning
