解剖学报

解剖学报 ›› 2024, Vol. 55 ›› Issue (1): 73-81.doi: 10.16098/j.issn.0529-1356.2024.01.011

• 解剖学 • 上一篇    下一篇

 基于空间域与频域特征自适应融合和类间边界区域增强的三维海马分割

  白贺1,2 滕野1,2  冯蕾1,2 孟海伟1,2汤煜春1,2刘树伟1,2*   

  1. 1.山东大学基础医学院解剖学与神经生物学系，断层解剖学与数字人研究院，济南 250012; 2.山东大学脑与类脑科学研究院，济南 250012
  • 收稿日期:2023-07-10 修回日期:2023-08-25 出版日期:2024-02-06 发布日期:2024-02-06
  • 通讯作者: 刘树伟 E-mail:Liusw@sdu.edu.cn
  • 基金资助:
    注意网络的影像遗传学研究;大脑皮质折叠的早期形态发生;结构脑网络组的出生前发育与可视化;中国人4D数字化脑图谱可视化系统;数字人体研究成果在临床手术规划的应用与产业化

 3D hippocampal segmentation based on spatial and frequency domain features adaptive fusion and inter-class boundary region enhancement

 BAI  He1,2   TENG  Ye1,2  FENG  Lei1,2  MENG  Hai-wei1,2  TANG  Yu-chun1,2  LIU  Shu-wei1,2*   

  1. 1.Institute of Sectional Anatomy and Digital Human, Department of Anatomy and Neurobiology, School of Basic Medical Sciences, Shandong University， Ji’nan 250012,China;  2.Institute of Brain and Brain-Inspired Sciences, Shandong University，Ji’nan 250012, China
  • Received:2023-07-10 Revised:2023-08-25 Online:2024-02-06 Published:2024-02-06
  • Contact: LIU Shu-wei E-mail:Liusw@sdu.edu.cn

PDF

摘要：

目的  海马萎缩是诊断阿尔茨海默病等诸多精神疾病的临床重要标志，因此准确分割海马是一个重要的科学问题。随着深度学习的发展，人们提出了大量先进的自动分割方法。然而，由于MRI中各种噪声的影响以及海马不同类别之间不清晰的边界，三维海马分割仍然具有挑战性。因此本文旨在提出新的自动分割方法来更精确地分割海马头、体、尾。   方法  为了克服这些挑战，本文提出了两个策略。一种是空间域与频域特征自适应融合策略，通过快速傅立叶变换和卷积自动选择合适的频率组合，减少噪声对特征提取的影响。另一种是类间边界区域增强策略，它允许网络通过加权每个类之间边界区域的损失函数来增强对边界区域的学习，以达到精确定位边界和调节海马头、体、尾大小的目的。   结果  在50例青少年大脑MRI数据集上进行的实验表明，我们的方法实现了较先进的海马分割，海马头、体、尾相较于现有的方法都取得了一定的提升。消融实验证明我们提出的两种策略有效，我们还在260例Task04_Hippocampus数据集上验证了网络具有强大的泛化能力，说明本文提出的方法可用于更多的海马分割场景。   结论 我们提出的方法可以帮助临床医生更清楚地观测海马萎缩，并完成更精确的病情诊断和追踪。

关键词:  空间域与频域特征自适应融合, 类间边界区域增强, 医学图像分割, 海马分割, 人

Abstract:  Objective  Hippocampal atrophy is a clinically important marker for the diagnosis of many psychiatric disorders such as Alzheimer’s disease, so accurate segmentation of the hippocampus is an important scientific issue. With the development of deep learning, a large number of advanced automatic segmentation method  have been proposed. However, 3D hippocampal segmentation is still challenging due to the effects of various noises in MRI and unclear boundaries between various classes of the hippocampus. Therefore, the aim of this paper is to propose new method  to segment the hippocampal head, body, and tail more accurately.   Methods  To overcome these challenges, this paper proposed two strategies. One was the spatial and frequency domain features adaptive fusion strategy, which reduced the influence of noise on feature extraction by automatically selecting the appropriate frequency combination through fast Fourier transform and convolution. The other was an inter-class boundary region enhancement strategy, which allowed the network to focus on learning the boundary regions by weighting the loss function of the boundary regions between each class to achieve the goal of pinpointing the boundaries and regulating the size of the hippocampal head, body and tail.

   Results  Experiments performed on a 50-case teenager brain MRI dataset show that our method  achieves state-of-the-art hippocampal segmentation. Hippocampal head, body and tail had been improved compared to the existing method. Ablation experiments demonstrated the effectiveness of our two proposed strategies, and we also validated that the network had a strong generalization ability on a 260-case Task04_Hippocampus dataset. It was shown that the method  proposed in this paper could be used in more hippocampal segmentation scenarios.    Conclusion  The method  proposed in this paper can help clinicians to observe hippocampal atrophy more clearly and accomplish more accurate diagnosis and follow-up of the condition.

Key words: Spatial and frequency domain features adaptive fusion, Inter-class boundary region enhancement, Medical image segmentation, Hippocampal segmentation, Human

中图分类号: