摘要
目的实现脑部MRI图像中海马体区域的自动精准分割。方法本研究提出了一种基于深度学习的海马体自动分割框架AF-VNet,通过在网络输出层添加注意力机制模块,提升海马体分割精度。此外,引入Focal Loss损失函数和增加通道数,避免训练样本不均衡。结果采用AF-VNet分割左右海马得到的分割Dice系数为0.8787±0.0281,相对于基准VNet提升了2.37%。结论该方法提高了海马体分割的准确性,具有一定的临床应用价值,可为神经退行性疾病的诊断和治疗提供辅助工具。
Objective To achieve automatic and accurate segmentation of hippocampal region in brain MRI images.Methods AF-VNet,an automatic hippocampus segmentation framework based on deep learning,was proposed to improve the accuracy of hippocampus segmentation by adding attention mechanism module to the output layer of the network.In addition,the Focal Loss function was introduced and the number of channels was increased to avoid unbalanced training samples.Results The Dice score obtained by AF-VNet was 0.8787±0.0281,which was 2.37%higher than the benchmark VNet.Conclusion This method improves the accuracy of hippocampus segmentation,has certain clinical application value,and can provide an auxiliary tool for the diagnosis and treatment of neurodegenerative diseases.
作者
许晶晶
黄殿
吴雪
胡敏
吕昊
韩世鹏
XU Jingjing;HUANG Dian;WU Xue;HU Min;LYU Hao;HAN Shipeng(No.5 Cadet Regiment,School of Basic Medical Sciences,Air Force Medical University,Xi'an 710032,China;College of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China;Shaanxi Provincial Key Laboratory of Bioelectromagnetic Detection and Intelligent Perception,Air Force Medical University,Xi'an 710032,China;Department of Medical Electronics,School of Military Biomedical Engineering,Air Force Medical University,Xi'an 710032,China)
出处
《空军军医大学学报》
CAS
2024年第11期1227-1232,共6页
Journal of Air Force Medical University
基金
国家自然科学基金青年科学基金(62303473)。
关键词
海马体分割
深度学习
注意力机制
损失函数
hippocampus segmentation
deep learning
attention mechanism
loss function
