[go: up one dir, main page]

CN110473202B - Time sequence invariant feature extraction method for high-order dynamic function connection network - Google Patents

Time sequence invariant feature extraction method for high-order dynamic function connection network Download PDF

Info

Publication number
CN110473202B
CN110473202B CN201910528161.0A CN201910528161A CN110473202B CN 110473202 B CN110473202 B CN 110473202B CN 201910528161 A CN201910528161 A CN 201910528161A CN 110473202 B CN110473202 B CN 110473202B
Authority
CN
China
Prior art keywords
correlation
dynamic
brain
time
sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910528161.0A
Other languages
Chinese (zh)
Other versions
CN110473202A (en
Inventor
赵峰
陈红瑜
安志勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Technology and Business University
Original Assignee
Shandong Technology and Business University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Technology and Business University filed Critical Shandong Technology and Business University
Priority to CN201910528161.0A priority Critical patent/CN110473202B/en
Publication of CN110473202A publication Critical patent/CN110473202A/en
Application granted granted Critical
Publication of CN110473202B publication Critical patent/CN110473202B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30016—Brain

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Biophysics (AREA)
  • Medical Informatics (AREA)
  • Theoretical Computer Science (AREA)
  • Radiology & Medical Imaging (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种高阶动态功能连接网络的时序不变特征提取方法，主要步骤为：(1)对于给定的滑动窗窗宽和步长，将整个磁共振序列分割为多个子序列；(2)计算每个子序列内的各个脑区之间相关性，得到动态功能连接网络，进而计算任意两个脑区的动态功能连接序列的中心矩特征；(3)对于动态功能连接网络的每一个子序列，将每个脑区与其他脑区的连接序列视为一个一维随机序列，计算任意一对脑区的相关性，进而构建了一个高阶动态功能连接网络并获取它的中心矩特征。本发明利用中心矩特征作为动态功能连接网络及高阶动态功能连接网络的时序不变特征，能够捕获脑功能连接的深层次关联关系，为后续的医学图像处理提供了稳定特征。

Figure 201910528161

The invention discloses a time-series invariant feature extraction method of a high-order dynamic functional connection network. The main steps are: (1) for a given sliding window width and step size, the entire magnetic resonance sequence is divided into multiple subsequences; (2) Calculate the correlation between each brain region in each subsequence to obtain the dynamic functional connectivity network, and then calculate the central moment characteristics of the dynamic functional connectivity sequence of any two brain regions; (3) for each dynamic functional connectivity network A subsequence that treats the connection sequence between each brain region and other brain regions as a one-dimensional random sequence, calculates the correlation of any pair of brain regions, and then constructs a high-order dynamic functional connection network and obtains its central moment feature. The present invention uses the central moment feature as the time-series invariant feature of the dynamic functional connectivity network and the high-order dynamic functional connectivity network, which can capture the deep-level correlation of brain functional connectivity and provide stable features for subsequent medical image processing.

Figure 201910528161

Description

一种高阶动态功能连接网络的时序不变特征提取方法A temporal invariant feature extraction method for high-order dynamic functional connectivity networks

◆ 技术领域◆ Technical field

本发明涉及神经影像和机器学习的技术领域，具体涉及一种高阶动态功能网络的时序不变特征提取方法。The present invention relates to the technical field of neuroimaging and machine learning, and in particular to a method for extracting time-invariant features of a high-order dynamic functional network.

◆ 背景技术◆ Background technology

静息态功能磁共振图像（rs-FMRI，resting-state functional magneticresonance imaging）是在被试者避免系统性思维活动环境下，基于脑神经活动所引起的血氧变化而产生的磁振造影技术，是人类探知大脑活动规律的一种十分有效的神经影像方式。由于rs-FMRI具有无创性和无辐射损害性以及较高的时空分辨率，目前已在认知科学、神经科学、药理学和精神疾病等领域的应用十分广泛。Resting-state functional magnetic resonance imaging (rs-FMRI) is a magnetic resonance imaging technique based on the changes in blood oxygen caused by brain nerve activity in an environment where the subject avoids systematic thinking activities. It is a very effective neuroimaging method for humans to explore the laws of brain activity. Because rs-FMRI is non-invasive, non-radiation-damaging, and has a high temporal and spatial resolution, it has been widely used in cognitive science, neuroscience, pharmacology, and mental illness.

为了有效提取和利用rs-FMRI所包含的信息，需要采用一定的策略和方法对rs-FMRI数据进行处理，用以捕获rs-FMRI所蕴含的重要特征。其中基于滑动窗策略的动态功能连接网络(D-FCN，dynamic functional connectivity network)是捕获rs-FMRI细微动态变化特征的重要方式。对于一个个体，令

Figure DEST_PATH_IMAGE001
表示第i个脑功能区的平均rs-FMRI时间序列，其中M表示整个扫描时间段内的采样数目，N表示大脑皮层的功能区数目。则D-FNC构建的基本步骤可表述为：In order to effectively extract and utilize the information contained in rs-FMRI, it is necessary to adopt certain strategies and methods to process rs-FMRI data to capture the important features contained in rs-FMRI. Among them, the dynamic functional connectivity network (D-FCN) based on the sliding window strategy is an important way to capture the subtle dynamic changes of rs-FMRI. For an individual, let
Figure DEST_PATH_IMAGE001
represents the average rs-FMRI time series of the i -th brain functional area, where M represents the number of samples in the entire scanning time period, and N represents the number of functional areas in the cerebral cortex. The basic steps of D-FNC construction can be expressed as:

1）将整个扫描时间段内的rs-FMRI划分为多个rs-FMRI子序列。1) The rs-FMRI in the entire scanning time period is divided into multiple rs-FMRI subsequences.

利用滑动窗技术，将rs-FMRI时间序列

Figure 575033DEST_PATH_IMAGE002
分割为K个子序列，即：Using sliding window technology, rs-FMRI time series
Figure 575033DEST_PATH_IMAGE002
Divide into K subsequences, namely:

Figure DEST_PATH_IMAGE003
(1)
Figure DEST_PATH_IMAGE003
(1)

其中

Figure 921701DEST_PATH_IMAGE004
表示rs-FMRI的子序列数目，T表示滑动窗宽，S表示滑动步长。in
Figure 921701DEST_PATH_IMAGE004
represents the number of subsequences of rs-FMRI, T represents the sliding window width, and S represents the sliding step size.

）D-FCN构建。)D-FCN construction.

对第

Figure DEST_PATH_IMAGE005
个子窗口的rs-FMRI序列，计算序列之间相关性。即For
Figure DEST_PATH_IMAGE005
rs-FMRI sequences of sub-windows, and calculate the correlation between sequences.

Figure 396545DEST_PATH_IMAGE006
(2)
Figure 396545DEST_PATH_IMAGE006
(2)

显然

Figure DEST_PATH_IMAGE007
是一个相关性矩阵，描述了任意一对脑区
Figure 486861DEST_PATH_IMAGE008
在一个短时间内的相关性。基于公式（1），我们可以得到
Figure DEST_PATH_IMAGE009
个相关性序列，即Apparently
Figure DEST_PATH_IMAGE007
is a correlation matrix describing any pair of brain regions
Figure 486861DEST_PATH_IMAGE008
Correlation over a short period of time. Based on formula (1), we can get
Figure DEST_PATH_IMAGE009
A correlation sequence, that is

Figure 730760DEST_PATH_IMAGE010
(3)
Figure 730760DEST_PATH_IMAGE010
(3)

其中，

Figure DEST_PATH_IMAGE011
描述了任意一对脑区随着扫描时间而发生的变化。in,
Figure DEST_PATH_IMAGE011
Describes changes in any pair of brain regions over scanning time.

目前，已有众多证据表明：基于滑动窗策略所构建的D-FCN，能够较好反应脑区之间连接的动态变化，而且这种变化可以揭示不同个体之间脑组织以及脑功能之间的差异性，因此在认知科学、神经科学、药理学和精神疾病等领域存在巨大的应用前景。At present, there is a lot of evidence showing that D-FCN constructed based on the sliding window strategy can better reflect the dynamic changes in connections between brain regions, and this change can reveal the differences in brain tissue and brain function between different individuals. Therefore, it has great application prospects in the fields of cognitive science, neuroscience, pharmacology and mental illness.

然而，尽管D-FCN为我们了解大脑活动提供了一个新渠道，但是D-FCN存在以下几个明显的缺点：（1）D-FCN对时间顺序特别敏感。即在静息态环境下，不同个体在整个磁共振扫描时间区间的脑神经活动规律，并不具备时间上的一致性。也就是说，不同个体在同一个磁共振扫描时间段内的脑区之间的相关性并不一致。因此，对不同个体而言，沿着扫描时间的动态功能连接子序列（即D-FCN）存在时间上的错匹配。（2）D-FCN只是反映了任意两个脑区之间的动态连接特性，无法反映多个脑区之间的关联关系。所以，采用何种方法获取D-FCN时序不变特征和多个脑区之间的关联关系，仍是当今有待解决的难题。However, although D-FCN provides a new channel for us to understand brain activity, it has the following obvious shortcomings: (1) D-FCN is particularly sensitive to time sequence. That is, in a resting state, the brain neural activity patterns of different individuals in the entire MRI scanning time interval do not have temporal consistency. In other words, the correlation between brain regions of different individuals in the same MRI scanning time period is not consistent. Therefore, for different individuals, the dynamic functional connection subsequence (i.e., D-FCN) along the scanning time has temporal mismatches. (2) D-FCN only reflects the dynamic connection characteristics between any two brain regions, and cannot reflect the correlation between multiple brain regions. Therefore, what method to use to obtain the temporal invariant features of D-FCN and the correlation between multiple brain regions is still a difficult problem to be solved today.

◆ 发明内容◆ Summary of the invention

为了解决上述问题，本发明提供一种获取D-FCN以及高阶D-FCN时序不变特征的提取方法。该特征反映了功能连接的高阶信息，可以捕获多个脑区之间的功能连接关系，可用于后续的脑损伤检测、康复效果观察等方面。In order to solve the above problems, the present invention provides a method for extracting D-FCN and high-order D-FCN time-invariant features. This feature reflects the high-order information of functional connectivity, can capture the functional connectivity relationship between multiple brain regions, and can be used for subsequent brain injury detection, rehabilitation effect observation, etc.

◆ 本发明的具体步骤如下：◆ The specific steps of the present invention are as follows:

第1步）初始参数设定。设定滑动窗宽度T, 滑动步长S。Step 1) Initial parameter setting. Set the sliding window width T and sliding step size S.

第2步）功能磁共振子序列获取。利用给定的滑动窗和步长，沿着扫描时间轴，将整个磁共振序列分割为多个子序列。Step 2) Acquisition of functional magnetic resonance subsequences: Using the given sliding window and step size, the entire magnetic resonance sequence is divided into multiple subsequences along the scanning time axis.

第3步）获取D-FCN。计算每个子窗口内各个脑区之间相关性，进而得到D-FCN。Step 3) Obtain D-FCN. Calculate the correlation between each brain region in each sub-window to obtain D-FCN.

第4步）获取D-FCN的中心矩特征。计算任意两个脑区的动态功能连接序列的中心矩特征。由于一个随机序列的中心矩具有时序不变性，因此D-FCN的中心矩特征可以作为D-FCN的时序不变特征。Step 4) Obtain the central moment feature of D-FCN. Calculate the central moment feature of the dynamic functional connection sequence of any two brain regions. Since the central moment of a random sequence is time-invariant, the central moment feature of D-FCN can be used as the time-invariant feature of D-FCN.

第5步）获取高阶动态功能连接网络（Ho-D-FCN, high-order dynamicfunctional connectivity network）。即对于D-FCN的每一个子序列，将每个脑区与其他脑区的连接序列视为一个一维随机序列，计算任意一对脑区的相关性，即“相关性的相关性”，它反映了多个脑区之间的关联关系，进而构建了一个高阶动态功能连接网络，即Ho-D-FCN。Step 5) Obtain a high-order dynamic functional connectivity network (Ho-D-FCN). That is, for each subsequence of D-FCN, the connection sequence between each brain region and other brain regions is regarded as a one-dimensional random sequence, and the correlation between any pair of brain regions is calculated, that is, the "correlation of correlation", which reflects the correlation between multiple brain regions, and then a high-order dynamic functional connectivity network, namely Ho-D-FCN, is constructed.

第6步）获取Ho-D-FCN的中心矩特征。类似步骤4，获取Ho-D-FCN的高阶中心矩特征，作为D-FCN的高阶时序不变特征。Step 6) Obtain the central moment feature of Ho-D-FCN. Similar to step 4, obtain the high-order central moment feature of Ho-D-FCN as the high-order time-series invariant feature of D-FCN.

◆ 本发明的有益效果：◆ Beneficial effects of the present invention:

1）D-FCN的中心矩特征具有时序不变性，能够有效克服D-FCN的时序敏感性问题。1) The central moment feature of D-FCN is time-invariant and can effectively overcome the time-sensitivity problem of D-FCN.

2）Ho-D-FCN的中心矩特征保持时序不变性的同时，能够反映多个脑区的关联关系，为深度探索脑神经活动规律提供了更丰富的信息。2) The central moment feature of Ho-D-FCN can maintain temporal invariance while reflecting the correlation between multiple brain regions, providing richer information for in-depth exploration of the laws of brain neural activity.

◆ 附图说明◆ Description of the attached figure

图1是发明的流程示意图FIG. 1 is a schematic diagram of the process of the invention

◆ 具体实施方式◆ Specific implementation method

下面结合附图（图1）与实施例对本发明作进一步说明。如图1所示，包括以下步骤：The present invention is further described below in conjunction with the accompanying drawings (Fig. 1) and embodiments. As shown in Fig. 1, the process comprises the following steps:

第1步）初始参数设定。设定滑动窗窗宽T, 滑动步长S。Step 1) Initial parameter setting. Set the sliding window width T and sliding step length S.

第2步）功能磁共振子序列获取。令

Figure 186012DEST_PATH_IMAGE001
表示某个个体的第i个脑功能区的平均rs-FMRI时间序列，其中M表示整个扫描时间段内的采样数目，N表示大脑皮层的功能区数目。利用滑动窗技术，将整个扫描时间段内的rs-FMRI时间序列
Figure 616993DEST_PATH_IMAGE002
分割为K个子序列，即Step 2) Acquire functional magnetic resonance subsequences.
Figure 186012DEST_PATH_IMAGE001
represents the average rs-FMRI time series of the i- th brain functional area of an individual, where M represents the number of samples in the entire scanning time period, and N represents the number of functional areas of the cerebral cortex. Using the sliding window technique, the rs-FMRI time series in the entire scanning time period are
Figure 616993DEST_PATH_IMAGE002
Divide into K subsequences, that is,

Figure 714262DEST_PATH_IMAGE003
(1)
Figure 714262DEST_PATH_IMAGE003
(1)

其中

Figure 343827DEST_PATH_IMAGE004
表示rs-FMRI的子序列数目。in
Figure 343827DEST_PATH_IMAGE004
Indicates the number of subsequences of rs-FMRI.

第3步）D-FCN（动态功能连接网络）的构建。对第

Figure 235560DEST_PATH_IMAGE005
个子窗口的rs-FMRI序列，采用皮尔逊相关系数计算序列之间相关性。即Step 3) Construction of D-FCN (Dynamic Functional Connection Network).
Figure 235560DEST_PATH_IMAGE005
The Pearson correlation coefficient is used to calculate the correlation between the rs-FMRI sequences of the sub-windows.

Figure 153837DEST_PATH_IMAGE006
(2)
Figure 153837DEST_PATH_IMAGE006
(2)

显然

Figure 258059DEST_PATH_IMAGE007
是一个相关性矩阵，描述了任意一对脑区
Figure 679813DEST_PATH_IMAGE008
在一个短时间内的相关性。基于公式（1），我们可以得到
Figure 539185DEST_PATH_IMAGE009
个相关性序列，即Apparently
Figure 258059DEST_PATH_IMAGE007
is a correlation matrix describing any pair of brain regions
Figure 679813DEST_PATH_IMAGE008
Correlation over a short period of time. Based on formula (1), we can get
Figure 539185DEST_PATH_IMAGE009
A correlation sequence, that is

Figure 679179DEST_PATH_IMAGE010
(3)
Figure 679179DEST_PATH_IMAGE010
(3)

其中，

Figure 383830DEST_PATH_IMAGE011
描述了任意一对脑区之间相关性随着扫描时间而发生的变化。in,
Figure 383830DEST_PATH_IMAGE011
Describes how the correlation between any pair of brain regions changes over scanning time.

第4步）获取D-FCN（动态功能连接网络）的中心矩特征。由公式（3），我们可以得到一个功能连接序列

Figure 722408DEST_PATH_IMAGE012
，即Step 4) Get the central moment feature of D-FCN (Dynamic Functional Connection Network). From formula (3), we can get a functional connection sequence
Figure 722408DEST_PATH_IMAGE012
,Right now

Figure DEST_PATH_IMAGE013
= [
Figure 487101DEST_PATH_IMAGE014
] (
Figure DEST_PATH_IMAGE015
), (4)
Figure DEST_PATH_IMAGE013
= [
Figure 487101DEST_PATH_IMAGE014
] (
Figure DEST_PATH_IMAGE015
), (4)

其中

Figure 645550DEST_PATH_IMAGE016
反映了第i个脑区与第j 个脑区之间沿着时间轴的动态连接关系。事实上，不同个体相应脑区之间的动态连接关系
Figure 888313DEST_PATH_IMAGE016
不存在时间一致性，即不同个体的同一时刻相应脑区的动态连接关系存在很大差异性。为此，为了获取动态连接序列的时序不变特征，我们求取每个动态序列的中心矩特征，即in
Figure 645550DEST_PATH_IMAGE016
It reflects the dynamic connection relationship between the i -th brain area and the j -th brain area along the time axis. In fact, the dynamic connection relationship between the corresponding brain areas of different individuals
Figure 888313DEST_PATH_IMAGE016
There is no temporal consistency, that is, the dynamic connection relationship of the corresponding brain regions of different individuals at the same time is very different. Therefore, in order to obtain the temporal invariant characteristics of the dynamic connection sequence, we calculate the central moment feature of each dynamic sequence, that is,

Figure DEST_PATH_IMAGE017
(
Figure 815817DEST_PATH_IMAGE018
), (8)
Figure DEST_PATH_IMAGE017
(
Figure 815817DEST_PATH_IMAGE018
), (8)

其中

Figure 751412DEST_PATH_IMAGE018
表示中心矩阶次，
Figure DEST_PATH_IMAGE019
表示
Figure 600420DEST_PATH_IMAGE013
= [
Figure 381294DEST_PATH_IMAGE020
]的均值。显然，由于一维随机信号序列的中心矩具有时序不变心，所以
Figure DEST_PATH_IMAGE021
(
Figure 694463DEST_PATH_IMAGE022
)可以作为动态功能连接序列
Figure 535380DEST_PATH_IMAGE013
= [
Figure 137263DEST_PATH_IMAGE020
] (
Figure 721828DEST_PATH_IMAGE015
)的时序不变特征，用于后续的影像分类、辅助诊断等。in
Figure 751412DEST_PATH_IMAGE018
represents the central moment order,
Figure DEST_PATH_IMAGE019
express
Figure 600420DEST_PATH_IMAGE013
= [
Figure 381294DEST_PATH_IMAGE020
]. Obviously, since the central moment of a one-dimensional random signal sequence has a time-series invariant center,
Figure DEST_PATH_IMAGE021
(
Figure 694463DEST_PATH_IMAGE022
) can be used as a dynamic functional connection sequence
Figure 535380DEST_PATH_IMAGE013
= [
Figure 137263DEST_PATH_IMAGE020
] (
Figure 721828DEST_PATH_IMAGE015
)’s time-series invariant features are used for subsequent image classification, auxiliary diagnosis, etc.

第5步）Ho-D-FCN（高阶动态功能连接网络）的构建。考虑到

Figure 92767DEST_PATH_IMAGE016
（见公式（4））仅仅反映了两个脑区之间的动态连接关系，并不能捕获多个脑区之间的动态连接关系。为了获取多个脑区之间的动态连接关系，我们按照如下方式定义一个高阶动态网络Ho-D-FCN。Step 5) Construction of Ho-D-FCN (Higher-order Dynamic Functional Connection Network). Considering
Figure 92767DEST_PATH_IMAGE016
(See formula (4)) only reflects the dynamic connection relationship between two brain regions, and cannot capture the dynamic connection relationship between multiple brain regions. In order to obtain the dynamic connection relationship between multiple brain regions, we define a high-order dynamic network Ho-D-FCN as follows.

Figure DEST_PATH_IMAGE023
，
Figure 370164DEST_PATH_IMAGE024
），（9）
Figure DEST_PATH_IMAGE023
,
Figure 370164DEST_PATH_IMAGE024
), (9)

其中

Figure DEST_PATH_IMAGE025
, 它表示了第i个脑区在第k个时间段内与其他脑区之间的相关性，所以，
Figure 990501DEST_PATH_IMAGE026
表示了第i个脑区与第j个脑区之间“相关性的相关性”，是多个脑区之间相关性的一种体现，我们称之为“高阶相关性”。基于公式（9），我们可以得到
Figure 378757DEST_PATH_IMAGE009
个相关性序列，即in
Figure DEST_PATH_IMAGE025
, which represents the correlation between the i- th brain region and other brain regions in the k- th time period, so,
Figure 990501DEST_PATH_IMAGE026
It represents the "correlation of correlation" between the i -th brain region and the j-th brain region, which is a manifestation of the correlation between multiple brain regions. We call it "high-order correlation". Based on formula (9), we can get
Figure 378757DEST_PATH_IMAGE009
A correlation sequence, that is

Figure DEST_PATH_IMAGE027
(10)
Figure DEST_PATH_IMAGE027
(10)

显然，

Figure 338623DEST_PATH_IMAGE028
描述了多个脑区之间相关性随着扫描时间而发生的变化。Obviously,
Figure 338623DEST_PATH_IMAGE028
Describes changes in correlations between multiple brain regions over scanning time.

第6步）获取Ho-D-FCN的中心矩特征。依据公式(9), 我们可以得到一个高阶功能连接序列

Figure DEST_PATH_IMAGE029
，即Step 6) Obtain the central moment feature of Ho-D-FCN. According to formula (9), we can get a high-order functional connection sequence
Figure DEST_PATH_IMAGE029
,Right now

Figure 52501DEST_PATH_IMAGE030
= [
Figure DEST_PATH_IMAGE031
] (
Figure 437433DEST_PATH_IMAGE015
), (11)
Figure 52501DEST_PATH_IMAGE030
= [
Figure DEST_PATH_IMAGE031
] (
Figure 437433DEST_PATH_IMAGE015
), (11)

其中

Figure 363800DEST_PATH_IMAGE030
反映了第i个脑区（与其它脑区）的相关性与第j 个脑区（与其他脑区）的相关性之间沿着时间轴的动态连接关系, 即“相关性的相关性”的动态变化关系。类似第4步，为了获取这种动态连接序列的时序不变特征，我们求取每个动态序列的中心矩特征，即in
Figure 363800DEST_PATH_IMAGE030
It reflects the dynamic connection relationship between the correlation of the i -th brain region (with other brain regions) and the correlation of the j -th brain region (with other brain regions) along the time axis, that is, the dynamic change relationship of "correlation of correlation". Similar to step 4, in order to obtain the time-invariant characteristics of this dynamic connection sequence, we calculate the central moment feature of each dynamic sequence, that is,

Figure 443752DEST_PATH_IMAGE032
(
Figure 266214DEST_PATH_IMAGE018
), (12)
Figure 443752DEST_PATH_IMAGE032
(
Figure 266214DEST_PATH_IMAGE018
), (12)

其中

Figure 64406DEST_PATH_IMAGE018
表示中心矩阶次，
Figure DEST_PATH_IMAGE033
表示
Figure 591202DEST_PATH_IMAGE013
= [
Figure 525660DEST_PATH_IMAGE020
]的均值。显然，由于一维随机信号序列的中心矩具有时序不变心，所以
Figure 315762DEST_PATH_IMAGE034
(
Figure 804512DEST_PATH_IMAGE022
)可以作为高阶动态功能连接序列
Figure 72682DEST_PATH_IMAGE030
= [
Figure 127226DEST_PATH_IMAGE031
] 的时序不变特征，用于后续的影像分类、辅助诊断等。in
Figure 64406DEST_PATH_IMAGE018
represents the central moment order,
Figure DEST_PATH_IMAGE033
express
Figure 591202DEST_PATH_IMAGE013
= [
Figure 525660DEST_PATH_IMAGE020
]. Obviously, since the central moment of a one-dimensional random signal sequence has a time-series invariant center,
Figure 315762DEST_PATH_IMAGE034
(
Figure 804512DEST_PATH_IMAGE022
) can be used as a high-order dynamic functional connection sequence
Figure 72682DEST_PATH_IMAGE030
= [
Figure 127226DEST_PATH_IMAGE031
]’s time-invariant features are used for subsequent image classification, auxiliary diagnosis, etc.

由公式(8)和公式(12)，我们分别获取了D-FCN（动态功能连接网络）与Ho-D-FCN（高阶动态功能连接网络）的时序不变特征

Figure 88229DEST_PATH_IMAGE021
(
Figure 861013DEST_PATH_IMAGE022
)和
Figure 932874DEST_PATH_IMAGE034
(
Figure 107503DEST_PATH_IMAGE022
)，并将其用于后续的脑影像处理，得到较好的结果。From formula (8) and formula (12), we obtain the time-invariant features of D-FCN (dynamic functional connection network) and Ho-D-FCN (high-order dynamic functional connection network) respectively.
Figure 88229DEST_PATH_IMAGE021
(
Figure 861013DEST_PATH_IMAGE022
)and
Figure 932874DEST_PATH_IMAGE034
(
Figure 107503DEST_PATH_IMAGE022
) and used it for subsequent brain image processing to obtain better results.

Claims (3)

1.一种高阶动态功能连接网络的时序不变特征提取方法，其特征是，主要包含以下几个步骤：步骤1，初始参数设定；设定滑动窗窗宽尺寸，滑动步长长度；步骤2，获取功能磁共振子序列； 令xi＝(xi1,xi2,…,xiM)(i＝1,2…,N)表示某个个体的第i个脑功能区的平均rs-FMRI时间序列，其中M表示整个扫描时间段内的采样数目，N表示大脑皮层的功能区数目；利用滑动窗技术，将整个扫描时间段内的rs-FMRI时间序列
Figure FDA0003993933960000011
分割为K个子序列，即1. A method for extracting time-invariant features of a high-order dynamic functional connection network, characterized in that it mainly includes the following steps: step 1, initial parameter setting; setting the sliding window width and sliding step length; step 2, obtaining functional magnetic resonance imaging subsequences; let x i = (x i1 , x i2 , …, x iM ) (i = 1, 2 …, N) represent the average rs-FMRI time series of the i-th brain functional area of an individual, where M represents the number of samples in the entire scanning time period, and N represents the number of functional areas of the cerebral cortex; using the sliding window technology, the rs-FMRI time series in the entire scanning time period is obtained.
Figure FDA0003993933960000011
Divide into K subsequences, that is, xi(k)＝(xi1(k),xi2(k),…,xiT(k))(k＝1,2…,K) (1)x i (k) = (x i1 (k), x i2 (k),..., x iT (k)) (k = 1, 2..., K) (1) 其中K＝[(M-T)/S]+1表示rs-FMRI的子序列数目，T表示滑动窗宽，S表示滑动步长；步骤3.D-FCN(动态功能连接网络)的构建；步骤4.获取D-FCN(动态功能连接网络)的中心矩特征；步骤5.Ho-D-FCN(高阶动态功能连接网络)的构建；步骤6.获取Ho-D-FCN(高阶动态功能连接网络)的中心矩特征；Where K = [(M-T)/S] + 1 represents the number of subsequences of rs-FMRI, T represents the sliding window width, and S represents the sliding step size; Step 3. Construction of D-FCN (dynamic functional connection network); Step 4. Obtaining the central moment feature of D-FCN (dynamic functional connection network); Step 5. Construction of Ho-D-FCN (high-order dynamic functional connection network); Step 6. Obtaining the central moment feature of Ho-D-FCN (high-order dynamic functional connection network); 所述步骤5包括：The step 5 comprises: 定义一个高阶相关性序列：Define a higher-order correlation sequence: HP(k)＝(hρij(k))1≤i,j≤N(k＝1,2,…,K)， (10)HP(k)＝(hρ ij (k)) 1≤i,j≤N (k＝1,2,…,K), (10) 这里，HP(k)描述了多个脑区之间相关性随着扫描时间而发生的变化；其中hρij(k)的定义如下：Here, HP(k) describes the change in the correlation between multiple brain regions over the scanning time; where hρ ij (k) is defined as follows: hρij(k)＝corr(ci(k),cj(k))(1≤i,j≤N,k＝1,2,…,K) (9)hρ ij (k)＝corr(c i (k),c j (k))(1≤i,j≤N,k＝1,2,…,K) (9) 其中ci(k)＝[ci1(k),ci2(k),ciN(k)]，它表示了第i个脑区在第k个时间段内与其他脑区之间的相关性，所以，hρij(k)表示了第i个脑区与第j个脑区之间“相关性的相关性”，是多个脑区之间相关性的一种体现，我们称之为“高阶相关性”；Among them, c i (k) = [c i1 (k), c i2 (k), c iN (k)], which represents the correlation between the ith brain region and other brain regions in the kth time period. Therefore, hρ ij (k) represents the "correlation of correlation" between the ith brain region and the jth brain region, which is a manifestation of the correlation between multiple brain regions. We call it "high-order correlation"; 所述步骤6包括：The step 6 comprises: 对任意一对脑区(i,j)(1≤i,j≤N)定义一个高阶功能连接序列：For any pair of brain regions (i, j) (1≤i,j≤N), a high-order functional connection sequence is defined: hρij＝[hρij(1),hρij(2),…,hρij(K)](1≤i,j≤N)， (11)hρ ij =[hρ ij (1),hρ ij (2),…,hρ ij (K)](1≤i,j≤N), (11) 其中hρij反映了第i个脑区与其它脑区的相关性与第j个脑区与其它脑区的相关性之间沿着时间轴的动态连接关系，即“相关性的相关性”的动态变化关系；为了获取这种动态连接序列的时序不变特征，我们求取每个动态序列的中心矩特征，即Among them, hρ ij reflects the dynamic connection relationship between the correlation between the ith brain region and other brain regions and the correlation between the jth brain region and other brain regions along the time axis, that is, the dynamic change relationship of "correlation of correlation"; in order to obtain the time-series invariant characteristics of this dynamic connection sequence, we obtain the central moment feature of each dynamic sequence, that is,
Figure FDA0003993933960000021
Figure FDA0003993933960000021
 其中d＝1,2,…D表示中心矩阶次，
Figure FDA0003993933960000022
表示ρij＝[ρij(1),ρij(2),…,ρij(K)]的均值；显然，由于一维随机信号序列的中心矩具有时序不变性，所以hcij(d)(1≤i,j≤N,d＝1,2,…D)可以作为高阶动态功能连接序列hρij＝[hρij(1),hρij(2),…,hρij(K)]的时序不变特征，用于后续的影像分类、辅助诊断。Where d = 1, 2, ... D represents the central moment order,
Figure FDA0003993933960000022
represents the mean of ρ ij =[ρ ij (1),ρ ij (2),…,ρ ij (K)]; obviously, since the central moment of the one-dimensional random signal sequence is time-invariant, hc ij (d)(1≤i,j≤N,d＝1,2,…D) can be used as the time-invariant feature of the high-order dynamic functional connection sequence hρ ij =[hρ ij (1),hρ ij (2),…,hρ ij (K)] for subsequent image classification and auxiliary diagnosis. 2.如权利要求1所述的一种高阶动态功能连接网络的时序不变特征提取方法，其特征是，所述步骤3包括：相关性序列P(k)(k＝1,2,…,K)的定义：即2. A method for extracting time-invariant features of a high-order dynamic functional connection network as claimed in claim 1, characterized in that step 3 comprises: the definition of the correlation sequence P(k) (k=1,2,…,K): P(k)＝(ρij(k))1≤i,j≤N(k＝1,2,…,K)， (3)P(k)＝(ρ ij (k)) 1≤i,j≤N (k＝1,2,…,K), (3) 这里，P(k)描述了任意一对脑区之间相关性随着扫描时间而发生的变化；其中ρij(k)(1≤i,j≤N)是一个相关性矩阵，描述了任意一对脑区(i,j)在一个短时间内的相关性；其定义为：对第k(k＝1,2…,K)个子窗口的rs-FMRI序列，采用皮尔逊相关系数计算序列之间相关性，即Here, P(k) describes the change of the correlation between any pair of brain regions with the scanning time; where ρ ij (k) (1≤i,j≤N) is a correlation matrix, which describes the correlation between any pair of brain regions (i,j) in a short time; it is defined as: for the rs-FMRI sequence of the kth (k＝1,2…,K) subwindow, the Pearson correlation coefficient is used to calculate the correlation between the sequences, that is, ρij(k)＝corr(xi(k),xj(k)) (2)。ρ ij (k)=corr (x i (k), x j (k)) (2). 3.如权利要求1所述的一种高阶动态功能连接网络的时序不变特征提取方法，其特征是，所述步骤4包括：3. The method for extracting time-invariant features of a high-order dynamic functional connection network according to claim 1, wherein step 4 comprises: 对任意一对脑区(i,j)(1≤i,j≤N)，定义一个功能连接序列：For any pair of brain regions (i, j) (1≤i,j≤N), define a functional connection sequence: ρij＝[ρij(1),ρij(2),…,ρij(K)](1≤i,j≤N)， (4)ρ ij =[ρ ij (1),ρ ij (2),…,ρ ij (K)](1≤i,j≤N), (4) 其中pij反映了第i个脑区与第j个脑区之间沿着时间轴的动态连接关系；事实上，不同个体相应脑区之间的动态连接关系pij不存在时间上的一致性，即不同个体的同一时刻相应脑区的动态连接关系存在很大差异性；为此，为了获取动态连接序列的时序不变特征，我们求取每个动态序列的中心矩特征，即：Where p ij reflects the dynamic connection relationship between the i-th brain area and the j-th brain area along the time axis; in fact, the dynamic connection relationship p ij between the corresponding brain areas of different individuals does not have temporal consistency, that is, the dynamic connection relationship between the corresponding brain areas of different individuals at the same time is very different; therefore, in order to obtain the time-series invariant characteristics of the dynamic connection sequence, we obtain the central moment feature of each dynamic sequence, that is:
Figure FDA0003993933960000031
Figure FDA0003993933960000031
 其中d＝1,2,…D表示中心矩阶次，
Figure FDA0003993933960000032
表示ρij＝[ρij(1),ρij(2),…,ρij(K)]的均值；显然，由于一维随机信号序列的中心矩具有时序不变性，所以cij(d)(1≤i,j≤N,d＝1,2,…D)可以作为动态功能连接序列ρij＝[ρij(1),ρij(2),…,ρij(K)](1≤i,j≤N)的时序不变特征，用于后续的影像分类、辅助诊断。Where d = 1, 2, ... D represents the central moment order,
Figure FDA0003993933960000032
represents the mean of ρ ij =[ρ ij (1),ρ ij (2),…,ρ ij (K)]; obviously, since the central moment of the one-dimensional random signal sequence is time-invariant, c ij (d)(1≤i,j≤N,d＝1,2,…D) can be used as the time-invariant feature of the dynamic functional connection sequence ρ ij =[ρ ij (1),ρ ij (2),…,ρ ij (K)](1≤i,j≤N) for subsequent image classification and auxiliary diagnosis.
CN201910528161.0A 2019-06-18 2019-06-18 Time sequence invariant feature extraction method for high-order dynamic function connection network Active CN110473202B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910528161.0A CN110473202B (en) 2019-06-18 2019-06-18 Time sequence invariant feature extraction method for high-order dynamic function connection network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910528161.0A CN110473202B (en) 2019-06-18 2019-06-18 Time sequence invariant feature extraction method for high-order dynamic function connection network

Publications (2)

Publication Number Publication Date
CN110473202A CN110473202A (en) 2019-11-19
CN110473202B true CN110473202B (en) 2023-03-31

Family

ID=68506926

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910528161.0A Active CN110473202B (en) 2019-06-18 2019-06-18 Time sequence invariant feature extraction method for high-order dynamic function connection network

Country Status (1)

Country Link
CN (1) CN110473202B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325288B (en) * 2020-03-17 2022-02-25 山东工商学院 Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104700120A (en) * 2015-03-23 2015-06-10 南京工业大学 fMRI feature extraction and classification method based on adaptive entropy projection clustering algorithm
US9633430B2 (en) * 2011-12-28 2017-04-25 Institute Of Automation, Chinese Academy Of Sciences Method for analyzing functional MRI brain images
CN106618575A (en) * 2017-01-09 2017-05-10 天津大学 Method for collecting cerebral function connection spontaneous fluctuation variability
US10034645B1 (en) * 2017-04-13 2018-07-31 The Board Of Trustees Of The Leland Stanford Junior University Systems and methods for detecting complex networks in MRI image data

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8340437B2 (en) * 2007-05-29 2012-12-25 University Of Iowa Research Foundation Methods and systems for determining optimal features for classifying patterns or objects in images
WO2016094375A1 (en) * 2014-12-08 2016-06-16 Rutgers, The State University Of New Jersey Methods for measuring physiologically relevant motion
EP3592224A4 (en) * 2017-03-07 2021-01-13 The Regents of The University of California METHOD OF MONITORING THE TREATMENT OF NEUROPSYCHIATRIC DISEASES
US10769791B2 (en) * 2017-10-13 2020-09-08 Beijing Keya Medical Technology Co., Ltd. Systems and methods for cross-modality image segmentation

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9633430B2 (en) * 2011-12-28 2017-04-25 Institute Of Automation, Chinese Academy Of Sciences Method for analyzing functional MRI brain images
CN104700120A (en) * 2015-03-23 2015-06-10 南京工业大学 fMRI feature extraction and classification method based on adaptive entropy projection clustering algorithm
CN106618575A (en) * 2017-01-09 2017-05-10 天津大学 Method for collecting cerebral function connection spontaneous fluctuation variability
US10034645B1 (en) * 2017-04-13 2018-07-31 The Board Of Trustees Of The Leland Stanford Junior University Systems and methods for detecting complex networks in MRI image data

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Diagnosis of Autism Spectrum Disorder Using Central-Moment;Zhao F 等;《Frontiers in Neuroence》;20200428;第1-15页 *
基于深度学习的胃癌病理图像分类方法;张泽中等;《计算机科学》;20181115;第273-278页 *
基于静息态功能磁共振成像的动态功能连接分析及临床应用研究进展;袁悦铭等;《磁共振成像》;20180813(第08期);第28-37页 *
非线性脑区相关性分析及动态脑网络构建方法;龙雨涵等;《信号处理》;20180825(第08期);第85-95页 *

Also Published As

Publication number Publication date
CN110473202A (en) 2019-11-19

Similar Documents

Publication Publication Date Title
JP7524279B2 (en) Method and system for personalizing brain atlas based on magnetic resonance and siamese graph neural network
CN111863244B (en) Functional connection mental disease classification method and system based on sparse pooling graph convolution
Wang et al. A new method for EEG-based concealed information test
Bayram et al. Deep learning methods for autism spectrum disorder diagnosis based on fMRI images
CN111523617B (en) Epilepsy detection system based on white matter fusion characteristic diagram and residual error attention network
Kuntzelman et al. A practical comparison of algorithms for the measurement of multiscale entropy in neural time series data
CN119295473B (en) A multimodal dynamic structure-function multi-layer brain network computing method and terminal
CN114947883A (en) A deep learning EEG noise reduction method based on time-frequency domain information fusion
CN111387936B (en) A sleep stage identification method, device and equipment
Tosun et al. Novel eye‐blink artefact detection algorithm from raw EEG signals using FCN‐based semantic segmentation method
Zheng et al. Novel linear and nonlinear features for the analysis of dynamic brain functional connectivity
CN119920436B (en) Brain function dynamic prediction device, equipment and medium based on rTMS
Paquola et al. The cortical wiring scheme of hierarchical information processing
CN110473202B (en) Time sequence invariant feature extraction method for high-order dynamic function connection network
Ali et al. ConVnet BiLSTM for ASD Classification on EEG Brain Signal.
Sun et al. MEEG-Transformer: transformer Network based on Multi-domain EEG for emotion recognition
CN110322969A (en) A kind of fMRI data classification method based on width study
Wen et al. Pulvinar response profiles and connectivity patterns to object domains
CN115937129A (en) Method and device for processing left-right half-brain relation based on multi-modal magnetic resonance image
Ma et al. Probabilistic Boolean network analysis of brain connectivity in Parkinson's disease
CN116649902A (en) Space-time characteristic mapping system and method for brain electrical signals of schizophrenia
Li et al. Research on recognition of O-MI based on CNN combined with SST and LSTM
Shameedha Begum et al. EEG-Based Identification of Schizophrenia Using Deep Learning Techniques
CN118986319B (en) Quantitative method and device for brain structure-function coupling based on brain network gradients
Zhang et al. An Explainable Diagnostic Method for Autism Spectrum Disorder Using Neural Network.

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant