WO2017210873A1 - 基于磁共振影像的脑疾病个体化预测方法和系统 - Google Patents

基于磁共振影像的脑疾病个体化预测方法和系统 Download PDF

Info

Publication number
WO2017210873A1
WO2017210873A1 PCT/CN2016/085217 CN2016085217W WO2017210873A1 WO 2017210873 A1 WO2017210873 A1 WO 2017210873A1 CN 2016085217 W CN2016085217 W CN 2016085217W WO 2017210873 A1 WO2017210873 A1 WO 2017210873A1
Authority
WO
WIPO (PCT)
Prior art keywords
brain
spatial
feature
magnetic resonance
module
Prior art date
Application number
PCT/CN2016/085217
Other languages
English (en)
French (fr)
Inventor
隋婧
姜荣涛
孟醒
Original Assignee
中国科学院自动化研究所
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 中国科学院自动化研究所 filed Critical 中国科学院自动化研究所
Priority to PCT/CN2016/085217 priority Critical patent/WO2017210873A1/zh
Publication of WO2017210873A1 publication Critical patent/WO2017210873A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
    • G16B40/30—Unsupervised data analysis
    • 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding

Definitions

  • Embodiments of the present invention relate to the field of biological information and computational medical technology, and in particular, to a method and system for predicting individualized brain diseases based on magnetic resonance images.
  • the invention will make full use of the existing high-efficiency and high-quality algorithm theory such as machine learning and pattern recognition, and discover potential biomarkers that are hidden in the image data and have specificity for specific diseases, and identify differences in brain structure and function. Helps the physician's clinical decision-making and improve people's quality of life. This is the hotspot and difficulty of research by researchers all over the world, and it is expected to produce breakthrough scientific results.
  • the present invention has been made in order to provide a magnetic resonance imaging-based individualized prediction of brain diseases that overcomes the above problems or at least partially solves the above problems. method.
  • a brain disease individualized prediction system based on magnetic resonance imaging is also provided.
  • a method for predicting individualized brain diseases based on magnetic resonance images may include:
  • Step 1 Obtain brain magnetic resonance images of patients with mental illness
  • Step 2 performing denoising and dimensionality reduction on the brain magnetic resonance image of the patient
  • Step 3 using the ReliefF algorithm to perform feature selection on the processing result of the step 2;
  • Step 4 adaptively obtaining the spatial brain region using the spatial clustering analysis method based on the result of the step 3;
  • Step 5 Based on the spatial brain region obtained in step 4, using a correlation-based feature selection algorithm to remove redundant features and obtain an optimal feature subset;
  • Step 6 Perform multiple linear regression analysis based on the optimal feature subset to identify potential predictive biomarkers.
  • a brain disease individualized prediction system based on magnetic resonance images is further provided, and the system may include:
  • An acquisition module configured to acquire brain magnetic resonance images of a mentally ill patient
  • a pre-processing module configured to perform denoising and dimensionality reduction on the magnetic resonance image of the patient
  • a feature selection module configured to perform feature selection on a processing result of the processing module by using a ReliefF algorithm
  • a clustering module configured to adaptively obtain a spatial brain region using a data-driven spatial clustering analysis method based on a result of the feature selection module
  • a screening module configured to remove a redundant feature based on a spatial brain region obtained by the clustering module, and obtain an optimal feature subset by using a correlation-based feature selection algorithm
  • An identification module is configured to perform a multiple linear regression analysis based on the optimal subset of features to identify potential predictive biomarkers.
  • the feature selection of the patient's brain magnetic resonance image subjected to denoising and dimensionality reduction processing is performed by the whole brain-based voxel search, thereby avoiding the model-based feature extraction in the traditional algorithm to be subjected to the predefined fixed template.
  • Limitations on the number and size of brain regions promote the discovery of more accurate brain regions.
  • the spatial brain region is obtained using a data-driven spatial clustering analysis method, so that the process of automatically clustering the brain region of interest (ROI) is not limited by parameters. Whether it is for a variety of modal brain imaging features, or different types of mental illness have a good generalization performance, not tied to specific image features and combinations of disease types.
  • the correlation feature-based selection algorithm is used to remove redundant features and obtain the optimal feature subset. Then, based on the optimal feature subset, multiple linear regression analysis is performed to identify potential predictive organisms. Sign.
  • the quantitative identification of disease biomarkers can more accurately locate abnormal brain regions closely related to the pathogenesis of mental illness.
  • the embodiment of the invention integrates a plurality of machine learning methods, and can quickly and conveniently perform quantitative and accurate prediction of individualized characteristics (cognitive level or symptom performance) of mental diseases based on image data of a plurality of different modal types, thereby further identifying Clinically meaningful biomarkers provide valuable clues for understanding brain structure, dysfunction, and underlying pathogenesis of disease.
  • FIG. 1 is a flow chart showing a method for predicting individualized brain diseases based on magnetic resonance images according to an exemplary embodiment
  • FIG. 2 is a schematic diagram showing adaptively obtaining several brain regions using data-driven spatial clustering analysis, according to another exemplary embodiment
  • FIG. 3 is a flow chart showing the prediction of the effect of pre-treatment treatment for patients with major depression by using the method proposed by the embodiment of the present invention according to an exemplary embodiment
  • FIG. 4a is a schematic diagram showing a prediction result of a symptom reduction state of a patient with severe depression after receiving electric shock treatment and a related brain region identified by using the multiple linear regression model proposed by the embodiment of the present invention
  • 4b is a diagram illustrating driving with multiple linear regression models, according to an exemplary embodiment Schematic diagram of the predicted results of the AAL template for symptom relief in patients with major depression after treatment;
  • FIG. 5 is a schematic structural diagram of a brain disease individualized prediction system based on magnetic resonance images according to an exemplary embodiment.
  • the basic idea of the embodiment of the present invention is based on images such as functional magnetic resonance imaging (fMRI), structural magnetic resonance imaging (sMRI) or diffusion tensor imaging (DTI) scanned by a mental patient, which are segmented, registered, and standardized.
  • the pre-processing operations result in high-dimensional heterogeneous three-dimensional spatial data that can be used for computer system processing analysis.
  • Next use features such as ReliefF
  • the algorithm selects all the features to sort the distinguishing ability of the target to be predicted, selects some voxel features with higher weight values in all whole brain voxels, completes the dimensionality reduction operation of the original data, and removes the voxel features of the spatial sparse distribution. It constitutes a spatial brain region with physiological significance and intuitive visualization effect.
  • the data-driven whole brain voxel search performs adaptive clustering operation through spatial neighborhood connectivity rules, and uses its mean value as the brain region feature representation, using correlation-based
  • the feature selection (CFS) algorithm removes the redundant features and finally combines multiple cross-validation and multiple linear regression analysis to obtain the relationship between the characteristic brain region and the patient's personalized characteristics (cognitive level or symptom performance), so as to identify potential organisms.
  • the purpose of the logo is to identify the characteristic brain region and the patient's personalized characteristics (cognitive level or symptom performance), so as to identify potential organisms.
  • Embodiments of the present invention provide a method for predicting individualized brain diseases based on magnetic resonance images. As shown in FIG. 1, the method may include steps S1 to S6.
  • Step S1 Obtaining a brain magnetic resonance image of a mentally ill patient.
  • mental illness refers to a series of schizophrenia (SZ), major depression (MDD) and bipolar disorder (BP), characterized by severe psychological disorders and behavioral abnormalities, accompanied by serious Cognitive impairment of mental illness.
  • SZ schizophrenia
  • MDD major depression
  • BP bipolar disorder
  • Magnetic resonance images can be a variety of images including functional magnetic resonance imaging (fMRI), structural magnetic resonance imaging (sMRI), and diffusion tensor imaging (DTI).
  • fMRI functional magnetic resonance imaging
  • sMRI structural magnetic resonance imaging
  • DTI diffusion tensor imaging
  • Step S2 Denoising and reducing the dimensionality of the patient's brain magnetic resonance image.
  • Embodiments of the present invention provide human brain features obtained by preprocessing various image modalities including functional magnetic resonance imaging (fMRI), structural magnetic resonance imaging (sMRI), and diffusion tensor imaging (DTI) of a mentally ill patient (such as low-frequency oscillation amplitude ALFF, segmentation gray matter map GM and white matter integrity and consistency, etc.), as a data basis for studying cognitive or symptom manifestations of patients with mental illness.
  • image modalities including functional magnetic resonance imaging (fMRI), structural magnetic resonance imaging (sMRI), and diffusion tensor imaging (DTI) of a mentally ill patient (such as low-frequency oscillation amplitude ALFF, segmentation gray matter map GM and white matter integrity and consistency, etc.), as a data basis for studying cognitive or symptom manifestations of patients with mental illness.
  • fMRI functional magnetic resonance imaging
  • sMRI structural magnetic resonance imaging
  • DTI diffusion tensor imaging
  • the magnetic resonance image of the mental patient group is scanned, and preprocessing operations such as gray matter segmentation, smoothing, and normalization are performed to remove the interference factors such as noise in the image, and the most primitive data of the prediction process is obtained.
  • the image usually has hundreds of thousands of dimensions, each dimension being a specific voxel feature of the spatial brain region (ie, brain voxel features).
  • each dimension being a specific voxel feature of the spatial brain region (ie, brain voxel features).
  • Step S3 Using the ReliefF algorithm, feature selection is performed on the processing result of step 2.
  • the target predictive attribute for a specific type of mental illness, according to the professional rating scale, quantitatively evaluate the patient's cognitive ability or symptom performance, and represent the level of cognitive level or symptom severity in the form of final score. .
  • the step may specifically include: using the ReliefF algorithm, calculating the ability of each voxel feature to distinguish the target prediction attribute; using the weight value as the division ability quantization standard, selecting a predetermined number of voxel features in descending order.
  • the good voxel feature should make the samples with small differences in target prediction properties as close as possible, and the different types of samples with different target prediction properties should be far away as far as possible.
  • Such voxel features are better. Distinguish performance.
  • the ability to distinguish each voxel feature from the target prediction attribute is calculated, and the weight size is used as the quantization standard of all the division ability, and all the attributes are sorted in descending order. According to actual needs, select a number of high-weight, target prediction genus The voxel features with large sexual impact factors are representative of the original data set. According to the actual situation, repeated experiments are repeated to select the best threshold size.
  • Step S4 Based on the result of step 3, the spatial brain region is adaptively obtained using a data-driven spatial clustering analysis method.
  • a spatial cube is constructed with each voxel as a center, and all voxels adjacent to the central voxel are gathered by a spatial neighborhood connection rule. Form a spatial brain area.
  • the process is realized by setting the whole brain space voxel selected by reliefF to 1, and the other positions without voxels to 0, and then constructing a 3 ⁇ 3 ⁇ 3 spatial cube with each voxel as the center, and collecting All voxels adjacent to the central voxel are connected into a spatial brain region with physiological significance and an intuitive visualization effect, thereby completing the transformation of the sparse voxel into the spatially connected brain region.
  • Taking the mean value of all the voxels in each cluster of brain regions as the characteristic attribute value of the brain region it becomes the representative of the brain region of interest (ie, the overall feature of the brain region), and finally obtains a space brain with a number of about 100. Area.
  • FIG. 2 exemplarily shows a schematic diagram of obtaining several brain regions using automatic spatial cluster analysis.
  • Step S5 based on the spatial brain region obtained in step 4, using correlation-based feature selection The algorithm is selected to remove the redundant features and obtain the optimal feature subset.
  • the mean values of the voxels in the spatial brain region are represented, and the correlation feature selection (CFS) algorithm is used to remove redundant features, and the optimal extraction is performed.
  • CFS correlation feature selection
  • the optimal feature subset refers to the brain region set with the smallest correlation and the greatest correlation with the target prediction attribute among all possible number of brain regions.
  • Step S6 Perform multiple linear regression analysis based on the selected optimal feature subset to identify potential biomarkers.
  • regression analysis refers to combining multiple linear regression, Pace regression or other multiple regression analysis algorithms.
  • the step may specifically include establishing a regression relationship between the optimal feature subset and the target prediction attribute; calculating a correlation coefficient between the predicted value and the true value of the target predicted attribute by using the regression relationship; determining the best according to the correlation coefficient
  • the degree of fitting; the brain region in the regression relationship corresponding to the degree of best fit is determined to be a predictive biomarker.
  • the identification of the potential biomarker refers to the correlation coefficient between the predicted value and the real value of the target predictive property calculated by the regression equation, and the brain region variable appearing in the regression equation is used as the recognized biomarker. If a better correlation coefficient and a more concentrated characteristic brain region are obtained, it indicates that the detected brain region can be an important biomarker for the mental illness.
  • the biomarker can be applied to other independent sample populations for test analysis to assist the physician in early diagnosis and treatment of the disease.
  • the selected optimal feature subset is used as an independent variable
  • the cognitive level score or the symptom performance score is used as a dependent variable
  • a regression function between the independent variable and the dependent variable is established by means of multiple linear regression analysis.
  • the symptom scores of patients before and after electroshock treatment were measured by the Hamilton Depression Scale (HDRS), and the difference represents the degree of symptom reduction ( ⁇ HDRS).
  • HDRS Hamilton Depression Scale
  • ⁇ HDRS degree of symptom reduction
  • Step S21 Acquire original data.
  • the raw data can be structural magnetic resonance imaging (sMRI) before and after treatment in 38 patients with major depression.
  • sMRI structural magnetic resonance imaging
  • Step S22 Perform denoising and dimensionality reduction processing on the original data.
  • each image of the sample is subjected to gray matter segmentation, smoothing, and normalization by SPM professional processing software to remove noise interference, and a 10e+4 magnitude voxel feature suitable for experimental analysis is obtained.
  • the dimensionality reduction of each feature can be utilized for dimensionality reduction.
  • Step S23 Perform feature selection using the ReliefF algorithm.
  • the original high-dimensional heterogeneous data dimension is reduced by selecting thousands of voxel features in all features that have a large influence factor on the degree of symptom reduction ⁇ HDRS score.
  • Step S24 adaptively obtaining a spatial brain region based on data-driven spatial clustering analysis.
  • the spatially sparsely distributed voxel features are clustered based on the spatial 26 neighborhood connectivity, and the brain regions containing less than 5 voxels are removed, and finally dozens of spatially interested brain regions are generated.
  • Step S25 Selecting the optimal feature subset by CFS to remove the redundant feature.
  • the spatial brain region is removed from the redundant features by CFS feature selection, and finally simplified to a dozen.
  • Step S26 Multiple linear regression analysis.
  • a multivariate linear regression model between the above-mentioned region of interest (spatial brain region) and ⁇ HDRS is established, and the brain region appearing in the regression equation is a potentially predictive brain region identified.
  • the individual prediction is performed under the nested-one-cross-validation framework by using the flowchart shown in FIG.
  • the samples in the training set are subjected to feature selection, spatial clustering analysis, adaptively obtaining spatial brain regions, calculating cluster mean values, feature subset selection, and regression analysis.
  • the target set of interest indicators is predicted and calculated for the test set samples.
  • a ten-fold cross-validation of the results of the regression analysis is performed to verify the validation set obtained by the feature subset selection.
  • the prediction results of the model for all samples will be finally obtained, and the predictive image markers for the treatment effect will be determined according to the number of repeated occurrences of brain regions in the 38 regression equations.
  • the characteristic brain regions with higher frequency (preferably, more than 20 times) appearing in the six brain regions are obtained, including brain regions such as the left hippocampus, the hippocampus, the inferior temporal gyrus, the middle and the right gyrus, and the right horn.
  • Figure 4a exemplarily shows the spatial brain regions associated with the recovery of symptoms of severe depression receiving electroshock treatment.
  • FIG. 4a exemplarily shows the prediction result of the symptom reduction state of the patients with major depression using the multiple linear regression model proposed by the embodiment of the invention, and the correlation between the symptom reduction value and the true value predicted by the model is shown.
  • the coefficient reached 0.89, and the predictive sensitivity to the patient's recovery was 88.9%, the specificity was 90.9%, and the prediction of treatment effect reached 97.4%.
  • Figure 4b exemplarily shows the prediction of symptom reduction in patients with major depression using the AAL template driven by multiple linear regression models. From the comparison of results, the prediction method based on whole brain search can significantly improve the prediction accuracy, and can identify the characteristic brain regions with clinical significance, which can be used as a biomarker for auxiliary medical diagnosis, and the prediction accuracy of the disease recovery is as high as 89.5%. It can truly realize the goal of predicting the treatment effect before the patient receives treatment and achieving the goal of precision medicine.
  • the embodiment of the present invention further provides a brain disease individualized prediction system based on magnetic resonance images.
  • the system may include an acquisition module 51, a pre-processing module 52, a feature selection module 53, a clustering module 54, a screening module 55, and an identification module 56.
  • the acquisition module 51 is configured to acquire a brain magnetic resonance image of a mentally ill patient.
  • the pre-processing module 52 is configured to perform denoising and dimensionality reduction processing on the patient's brain magnetic resonance image.
  • the feature selection module 53 is configured to perform feature selection on the processing result of the processing module using the ReliefF algorithm.
  • the clustering module 54 is configured to adaptively derive the spatial brain region using a data-driven spatial clustering analysis method based on the results of the feature selection module.
  • the screening module 55 is configured to remove the redundant features based on the spatial brain regions obtained by the clustering module by using a correlation-based feature selection algorithm to obtain an optimal feature subset.
  • the identification module 56 is configured to perform a multiple linear regression analysis based on the optimal subset of features to identify potential biomarkers.
  • the magnetic resonance imaging-based brain disease individualized prediction system provided by the above embodiments is only illustrated by the division of the above functional modules when performing individualized prediction of brain diseases.
  • the above function assignment is performed by different functional modules, that is, the internal structure of the system is divided into different functional modules to complete all or part of the functions described above.
  • the above-mentioned magnetic resonance image-based brain disease individualized prediction system further includes some other well-known structures, such as a processor, a memory, etc., in order to unnecessarily obscure the embodiments of the present disclosure, these well-known structures are not in the This is shown in Figure 5.
  • module may refer to a software object or routine that is executed on a computing system.
  • the different modules described herein can be implemented as an object or process executing on a computing system (eg, as a separate thread).
  • systems and methods described herein are preferably implemented in software, implementation in hardware or a combination of software and hardware is also possible and can be envisioned.
  • the various steps of the present invention can be implemented with a general-purpose computing device, for example, they can be centralized on a single computing device, such as a personal computer, a server computer, a handheld device or a portable device, a tablet device, or a multi-processor device. It may be distributed over a network of computing devices, which may perform the steps shown or described in a different order than the ones described herein, or separate them into individual integrated circuit modules, or multiple of them. Or the steps are made into a single integrated circuit module. Thus, the invention is not limited to any specific hardware or software or combination thereof.
  • the methods provided by the present invention can be implemented using programmable logic devices, or can be implemented as computer program software or program modules (including routines, programs, objects, components or data structures that perform particular tasks or implement particular abstract data types, etc.
  • an embodiment in accordance with the invention may be a computer program product that is executed to cause a computer to perform the method for the demonstration.
  • the computer program product comprises a computer readable storage medium having computer program logic or code portions for implementing the method.
  • the computer readable storage medium may be a built-in medium installed in a computer or a removable medium detachable from a computer main body (for example, a storage device using hot plug technology).
  • the built-in medium includes but is not limited to rewritable non-volatile memory Memory, such as RAM, ROM, flash memory, and hard disk.
  • the removable medium includes, but is not limited to, optical storage media (eg, CD-ROM and DVD), magneto-optical storage media (eg, MO), magnetic storage media (eg, magnetic tape or mobile hard disk), with built-in weight
  • a medium for example, a memory card
  • a medium for example, a ROM box

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Epidemiology (AREA)
  • Evolutionary Computation (AREA)
  • Bioethics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biotechnology (AREA)
  • Evolutionary Biology (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

一种基于磁共振影像的脑疾病个体化预测方法和系统（50）。该方法包括步骤1：获取精神疾病病人脑磁共振影像（S1）；步骤2：对病人脑磁共振影像进行去噪和降维处理（S2）；步骤3：利用ReliefF算法，进行特征选择（S3）；步骤4：使用空间聚类分析方法自适应地得到空间脑区（S4）；步骤5：利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集（S5）；步骤6：基于最优特征子集，进行多元线性回归分析，识别出潜在的生物标志（S6）。该系统（50）包括获取模块（51）、预处理模块（52）、特征选择模块（53）、聚类模块（54）、筛选模块（55）和识别模块（56）。该方法和系统（50）基于多种不同模态类型的影像数据，能对精神疾病的兴趣特征进行定量化、个体化地预测，有利于找出潜在的发病机制。

Description

基于磁共振影像的脑疾病个体化预测方法和系统 技术领域
本发明实施例涉及生物信息及计算医学技术领域，具体涉及一种基于磁共振影像的脑疾病个体化预测方法和系统。
背景技术
伴随经济、卫生、医疗水平的发展，世界各国人民的平均寿命得到延长。但与此同时，由于竞争压力增大等诸多因素，在全球范围内，精神疾病的发病率逐年增长，并成为导致死亡的主要原因之一[1](van Waarde et al.A functional MRI marker may predict the outcome of electroconvulsive therapy in severe and treatment-resistant depression.Mol Psychiatry 20,609-614,2015)。在临床上，人们对精神疾病的认识主要是通过磁共振影像(MRI)。作为一种非侵入性的成像技术，它大大加深了人们对精神疾病复杂的发病机理和多变的临床生物学差异的理解，已经成为认知科学、神经科学和神经精神病学研究不可或缺的工具。同时，伴随人类进入高科技信息化时代，数据挖掘、机器学习等一系列先进技术的发展成熟成为推动社会进步的重要因素，并已成功地应用在金融、通讯、地理、电子工程、航天等诸多领域，随之产生了各种尖端的算法理论体系模型。然而，这些尖端算法理论在神经影像数据上的应用在一定程度上受到了限制。
在临床上，尚未发现可用于评定精神疾病严重性与认知水平且具 有一定稳定性的生物标志，目前流行的大多数准则主要依赖于病人行为表现和医生的经验进行定性分析及在此基础上的合理推测。此外，由于多种精神疾病的认知或者行为表现之间存在较大程度的表征重叠交叉，单纯的只依赖于症状或者行为表现的诊断存在明显不足，因此急需一种能够及时辅助诊断或治疗，并具有客观性的神经影像标志。利用数据挖掘、机器学习的算法思想从已有的高维异构神经影像中提取潜在的特征信息，建立影像特征与目标度量(如症状表现、认知水平等)之间的关系模型，并识别出潜在的生物标志，进行个性化预测，已成为当前国际神经影像学研究的热点和前沿，并在教育、医疗、刑侦等诸多领域取得了一定成果[2-7](Demos et al,Individual differences in nucleus accumbens activity to food and sexual images predict weight gain and sexual behavior.J Neurosci 32,5549-5552,2012；Hoeft et al.,Functional and morphometric brain dissociation between dyslexia and reading ability.Proc Natl Acad Sci U S A 104,4234-4239,2007；Lener and Iosifescu,In pursuit of neuroimaging biomarkers to guide treatment selection in major depressive disorder:a review of the literature.Ann N Y Acad Sci 1344,50-65,2015；Mahmood et al.,Adolescents'fMRI activation to a response inhibition task predicts future substance use.Addict Behav 38,1435-1441,2013；Risacher et al.,Alzheimer's Disease Neuroimaging,I.,2009.Baseline MRI predictors of conversion from MCI to probable AD  in the ADNI cohort.Curr Alzheimer Res 6,347-361,2009；Sui et al.,In search of multimodal neuroimaging biomarkers of cognitive deficits in schizophrenia.Biol Psychiatry 78,794-804,2015)。
然而，已有算法往往基于固定的、预定义的模板(如AAL template)来进行脑区划分，并以此作为预测模型的输入特征，这在很大程度上限制了寻找更加精准的、具有预测性能的兴趣脑区(predictor，regions of interest，ROI).因为实际情况中起到关键预测作用的脑区(ROI)很可能是由预定模板划分得到的多个分区的部分组成。当前已有的研究模型多采用单一的数据类型或单一特征选择算法，不能有效地推广到多种模态影像数据类型及不同目标度量等不确定因子的预测任务中，这在很大程度上限制了其在多种类型的疾病诊断及影像模态上的推广，对于复杂的疾病研究，其重复性较差。因此，亟待产生一种具有一定良好的通用性和自适应性，并能在多种个性化预测任务中产生稳定结果的预测模型。本发明将充分利用现有机器学习、模式识别等高效优质算法理论，挖掘出隐藏在影像数据内的、对特定疾病具有针对性的潜在生物标志，识别出大脑结构与功能中存在的差异，有助于辅助医师的临床决策，提高人们的生活质量。这是世界各国科研人员研究的热点与难点所在，有望产生突破性的科学成果。
发明内容
鉴于上述问题，提出了本发明以便提供一种克服上述问题或者至少部分地解决上述问题的一种基于磁共振影像的脑疾病个体化预测 方法。此外，还提供一种基于磁共振影像的脑疾病个体化预测系统。
为了实现上述目的，根据本发明的一个方面，提供了以下技术方案：
一种基于磁共振影像的脑疾病个体化预测方法，所述方法可以包括：
步骤1：获取精神疾病病人脑磁共振影像；
步骤2：对所述病人脑磁共振影像进行去噪和降维处理；
步骤3：利用ReliefF算法，对所述步骤2的处理结果进行特征选择；
步骤4：基于所述步骤3的结果，使用空间聚类分析方法自适应地得到空间脑区；
步骤5：基于所述步骤4得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集；
步骤6：基于所述最优特征子集，进行多元线性回归分析，识别出潜在的具有预测性的生物标志。
为了实现上述目的，根据本发明的另一个方面，还提供了一种基于磁共振影像的脑疾病个体化预测系统，该系统可以包括：
获取模块，被配置为获取精神疾病病人脑磁共振影像；
预处理模块，被配置为对所述病人脑磁共振影像进行去噪和降维处理；
特征选择模块，被配置为利用ReliefF算法，对所述处理模块的处理结果进行特征选择；
聚类模块，被配置为基于所述特征选择模块的结果，使用基于数据驱动的空间聚类分析方法自适应的得到空间脑区；
筛选模块，被配置为基于所述聚类模块得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集；
识别模块，被配置为基于所述最优特征子集，进行多元线性回归分析，识别出潜在的具有预测性的生物标志。
与现有技术相比，上述技术方案至少具有以下有益效果：
本发明实施例通过基于全脑的体素搜索，对进行了去噪和降维处理的病人脑磁共振影像进行特征选择，避免了传统算法中基于模型的特征提取需在预定义固定模板下受脑区数量、大小等的限制，促进更加精确的脑区异常的发现。使用基于数据驱动的空间聚类分析方法得到空间脑区，使得自动聚类生成感兴趣脑区(ROI)的过程不受参数的限制。无论是对于多种模态脑影像特征，还是不同类型的精神疾病都有很好的泛化性能，不拘泥于特定的影像特征及疾病类型的组合。基于得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集；然后基于最优特征子集，进行多元线性回归分析，识别出潜在的具有预测性的生物标志。由此，对疾病生物标志的定量识别可以更加准确地定位与精神疾病发病机制密切相关的异常脑区。本发明实施例综合了多种机器学习方法，可以快速、便捷地基于多种不同模态类型的影像数据实现对精神疾病个体化特征(认知水平或症状表现)进行定量化精准预测，进而识别出具有临床意义的生物标识，为理解疾病的脑结构、功能异常与潜在的发病机制提供有价值的线索。
当然，实施本发明的任一产品不一定需要同时实现以上所述的所有优点。
本发明的其它特征和优点将在随后的说明书中阐述，并且，部分地从说明书中变得显而易见，或者通过实施本发明而了解。本发明的目的和其它优点可通过在所写的说明书、权利要求书以及附图中所特别指出的方法来实现和获得。
附图说明
附图作为本发明的一部分，用来提供对本发明的进一步的理解，本发明的示意性实施例及其说明用于解释本发明，但不构成对本发明的不当限定。显然，下面描述中的附图仅仅是一些实施例，对于本领域普通技术人员来说，在不付出创造性劳动的前提下，还可以根据这些附图获得其他附图。在附图中：
图1为根据一示例性实施例示出的基于磁共振影像的脑疾病个体化预测方法的流程示意图；
图2为根据另一示例性实施例示出的使用基于数据驱动的空间聚类分析自适应地得到若干脑区的示意图；
图3为根据一示例性实施例示出的利用本发明实施例所提出的方法对重度抑郁症患者接受治疗前治疗效果进行预测的流程示意图；
图4a为根据一示例性实施例示出的利用该发明实施例所提出的多元线性回归模型对重度抑郁症患者接受电击治疗后症状减轻状况的预测结果及鉴定出的相关脑区示意图；
图4b为根据一示例性实施例示出的利用多元线性回归模型驱动 的AAL模板对重度抑郁症患者接受治疗后症状减轻状况的预测结果的示意图；
图5为根据一示例性实施例示出的基于磁共振影像的脑疾病个体化预测系统的结构示意图。
这些附图和文字描述并不旨在以任何方式限制本发明的构思范围，而是通过参考特定实施例为本领域技术人员说明本发明的概念。
具体实施方式
下面结合附图以及具体实施例对本发明实施例解决的技术问题、所采用的技术方案以及实现的技术效果进行清楚、完整的描述。显然，所描述的实施例仅仅是本申请的一部分实施例，并不是全部实施例。基于本申请中的实施例，本领域普通技术人员在不付出创造性劳动的前提下，所获的所有其它等同或明显变型的实施例均落在本发明的保护范围内。本发明实施例可以按照权利要求中限定和涵盖的多种不同方式来具体化。
需要说明的是，在下面的描述中，为了方便理解，给出了许多具体细节。但是很明显，本发明的实现可以没有这些具体细节。
需要说明的是，在没有明确限定或不冲突的情况下，本发明中的各个实施例及其中的技术特征可以相互组合而形成技术方案。
本发明实施例的基本思想是以精神病人扫描的功能磁共振影像(fMRI)、结构磁共振影像(sMRI)或者弥散张量成像(DTI)等影像为基础，对其进行分割、配准、标准化等预处理操作得到可用于计算机系统处理分析的高维异构三维空间数据。接下来利用诸如ReliefF特征 选算法计算所有特征对目标待预测属性的区分能力进行排序，选取所有全脑体素中具有较高权重值的若干体素特征，完成原始数据的降维操作；将空间稀疏分布的体素特征构成具有生理意义并具有直观可视化效果的空间脑区，基于数据驱动的全脑体素搜索通过空间邻域连通规则进行自适应聚类操作，并以其均值作为脑区特征代表，利用基于相关的特征选择(CFS)算法去除冗余特征，并最终结合多重交叉验证和多元线性回归分析，得到特征脑区与病人个性化特征(认知水平或者症状表现)之间的关系，从而达到鉴定潜在生物标志的目的。
本发明实施例提供一种基于磁共振影像的脑疾病个体化预测方法。如图1所示，该方法可以包括步骤S1至步骤S6。
步骤S1：获取精神疾病病人脑磁共振影像。
其中，精神疾病是指以精神分裂症(SZ)、重度抑郁症(MDD)和躁郁症(BP)等为主的一系列以患者严重的心理障碍和行为异常为特征，并伴有严重的认知损害的精神疾病。
磁共振影像可以为功能磁共振影像(fMRI)、结构磁共振影像(sMRI)及弥散张量成像(DTI)在内的多种影像。
步骤S2：对病人脑磁共振影像进行去噪和降维处理。
本发明实施例从精神疾病的病人的功能磁共振影像(fMRI)、结构磁共振影像(sMRI)及弥散张量成像(DTI)在内的多种影像模态预处理后得到的人脑特征(如低频振荡振幅ALFF、分割灰质图GM及脑白质完整性与一致性等)入手，将其作为研究精神疾病患者认知或者症状表现的数据依据。
扫描精神病人群组磁共振影像，对其进行灰质分割、平滑、标准化等预处理操作，以去除影像中的噪声等干扰因素后，得到预测流程最原始的数据。该图像通常拥有高达数十万数量的维度，每一个维度是空间脑区的一个具有特异性的体素特征(也即脑体素特征)。为了降其维度，通常可以根据每个体素的差异大小，去除掉所有病人样本中不具有显著组内差异的特征。通常意义上，该具有较小差异的特征在疾病的认知或症状表现上不具有有效的区分能力，由此可实现对原始高维数据的降维操作。
步骤S3：利用ReliefF算法，对步骤2的处理结果进行特征选择。
定义目标预测属性为：针对特定类型的精神疾病，根据专业的评定量表，对病人的认知能力或者症状表现进行量化评测，并以最终得分的形式代表其认知水平高低或症状严重性程度。
本步骤具体可以包括：利用ReliefF算法，计算每个体素特征对目标预测属性的区分能力；以权值大小作为划分能力量化标准，按照从大到小的顺序，选择预定数量的体素特征。
对于每一个体素特征属性而言，好的体素特征应该使得目标预测属性差别小的样本尽量靠近，而使目标预测属性差别较大的不同类样本尽量远离，这样的体素特征有着较优质的区分性能。基于被预测属性的相对距离，计算每个体素特征对目标预测属性的区分能力，并以权值大小作为所有其划分能力量化标准，按照从大到小的顺序对所有属性进行排序。根据实际需要，选择若干拥有高权值、对目标预测属 性影响因子较大的体素特征作为原始数据集的代表，根据实际情况，重复多次试验选择出最好的阈值大小。
步骤S4：基于步骤3的结果，使用基于数据驱动的空间聚类分析方法自适应地得到空间脑区。
具体地，基于步骤3的结果，针对空间稀疏分布的全脑体素，以每一体素为中心，构建空间正方体，通过空间邻域连通规则聚集与该中心体素相邻接的所有体素，构成空间脑区。
在实际实施过程中，对于与认知或者症状表现相关的生物标志的鉴定，通常需要获取空间紧密连接、具有临床意义的联通区域。通过基于诸如空间26(3×3×3-1)邻域连通规则将稀疏的3D全脑体素聚集成连接紧密的团簇，实现超高维数据特征的降维操作的同时，构成具有生理意义并具有直观可视化效果的空间脑区，并去除掉包含体素数量小于5的那些脑区。由于包含体素较少，其不具有可解释的生理意义。该过程的实现是通过将reliefF选出的全脑空间体素置为1，其他没有体素的位置置为0，然后以每一个体素为中心，构建3×3×3的空间正方体，聚集与该中心体素相邻接的所有体素，将其连接成具有生理意义并具有直观可视化效果的空间脑区，从而完成稀疏体素到空间连通脑区的转化。以每个聚集脑区簇中所有的体素的均值作为该脑区的特征属性值，成为该感兴趣脑区的代表(即该脑区整体特征代表)，最终得到数量在100左右的空间脑区。图2示例性地示出了使用自动空间聚类分析得到若干脑区的示意图。
步骤S5：基于步骤4得到的空间脑区，利用基于相关的特征选 择算法，去除冗余特征，得到最优特征子集。
具体地，以步骤4空间聚类后得到的空间脑区为基础特征，以该空间脑区体素的均值为特征代表，利用基于相关的特征选择(CFS)算法去除冗余特征，提取最优特征子集。
其中，最优特征子集是指在所有可能数量的脑区组合中，各脑区间相关性最小、而与目标预测属性相关性最大的脑区集合。
通过计算每个特征与目标预测属性的相关性以及各特征之间的相关性大小，选择出所有可能数量的特征组合中能产生最优效果的特征子集，从而使得所选特征集与目标预测属性拥有最大的相关程度而特征属性之间冗余度最小，从而减少了特征数量。
步骤S6：基于所选择出的最优特征子集，进行多元线性回归分析，识别出潜在的生物标志。
其中，回归分析是指结合多元线性回归、Pace回归或者其他多种回归分析算法。
本步骤具体可以包括建立最优特征子集与所述目标预测属性之间的回归关系；通过回归关系，计算目标预测属性的预测值与真实值之间的相关系数；根据相关系数，确定最佳拟合程度；将得到最佳拟合程度所对应的回归关系中的脑区，确定为具有预测性的生物标志。
识别出潜在的生物标志是指通过回归方程计算得到目标预测属性的预测值与真实值间的相关系数，并以回归方程中出现的脑区变量作为识别出的生物标志。如果能得到较好的相关系数以及较为集中的特征脑区，说明检测到的脑区可以作为该精神疾病的重要生物标记， 并可以通过将该生物标记应用到其他的独立样本群体中进行检验分析，辅助医师进行疾病的早期诊断与治疗。
具体地，将所选择出的最优特征子集作为自变量，将认知水平得分或者症状表现得分作为因变量，借助多元线性回归分析建立自变量与因变量之间的回归函数。通过回归方程计算目标预测属性的预测值与真实值之间的皮尔逊相关系数，可以得到多元线性回归模型的预测能力，越接近于1，多元线性回归模型的拟合程度越好。回归方程中出现的脑区即可作为从影像中鉴定出的与该疾病密切相关联的生物标志。
下面以预测重度抑郁症(MDD)电击治疗(ECT)效果并鉴定相关生物标志为例来说明本发明的实现过程。
其中，病人经电击治疗前后的症状表现得分由汉密顿抑郁量表(HDRS)进行测量，其差值代表症状减轻程度(△HDRS)。
步骤S21：获取原始数据。
其中，原始数据可以为38名重度抑郁症患者治疗前后的结构磁共振影像(sMRI)。
步骤S22：对原始数据进行去噪和降维处理。
具体地，每个被试样本的影像经过SPM专业处理软件进行灰质分割、平滑、标准化等过程去除掉噪声干扰，得到适用于实验分析的10e+4量级体素特征。同时，鉴于数据的高维特性，可以利用各特征的组内差异大小进行降维处理。
步骤S23：利用ReliefF算法进行特征选择。
其中，通过选取所有特征中对症状减轻程度△HDRS得分影响因子较大的几千个体素特征，原始的高维异构数据维度得到降低。
步骤S24：基于数据驱动的空间聚类分析自适应得到空间脑区。
其中，将空间稀疏分布的体素特征基于空间26邻域连通进行聚类操作，去除掉包含体素数量少于5的脑区，最终产生数十个空间感兴趣脑区。
步骤S25：利用CFS选择最优特征子集去除冗余特征。
本步骤中，将上述空间脑区通过CFS特征选择去除掉冗余特征，最终简化到十几个。
步骤S26：多元线性回归分析。
建立上述感兴趣区域(空间脑区)与△HDRS之间的多元线性回归模型，出现在回归方程中的脑区为识别出的具有潜在预测性的脑区。
在本发明实施例所述方法的基础之上，利用图3所示流程图在嵌套留一交叉验证框架下进行个体预测。每次通过将一个样本作为测试集，其他样本(37个样本)作为训练集。训练集中的样本经由本发明所提方法进行特征选择、空间聚类分析自适应得到空间脑区、计算聚类均值、特征子集选择以及回归分析等过程。然后利用得到的回归方程对测试集样本进行目标兴趣指标预测计算。最后，对回归分析的结果进行十折交叉验证，验证通过特征子集选择得到的验证集。通过重复该过程38次，将最终得到模型对所有样本的预测结果，并根据38次回归方程中脑区重复出现的次数，确定对治疗效果具有预测性的影像标志。用选择出的脑区与产生的回归模型对测试集中的样本进 行临床特征个性化预测。最终得到6个脑区出现频次较高(优选地，频次大于20次)的特征脑区，其包括左侧海马、海马旁回、颞下回、额中回及右角回等脑区。通过检验所得脑区对应的脑结构名称及相应的功能特性，图4a示例性地示出了与重度抑郁症接受电击治疗症状恢复相关的空间脑区。该空间脑区与已有的科研结果中关于电击治疗重度抑郁症的相关发现具有很大的交叉性，保持相一致，说明了该预测模型(即多元线性回归模型)的有效性与可靠性。图4a示例性地示出了利用该发明实施例所提出的多元线性回归模型对重度抑郁症患者接受治疗后症状减轻状况的预测结果，通过模型预测得到的症状下降值与真实值之间的相关系数达到了0.89，并且对病人的恢复情况的预测敏感性为88.9％，特异性为90.9％，而对治疗效果的预测达到了97.4％。所有38名病人中仅有一名病人分类错误；图4b示例性地示出了利用多元线性回归模型驱动的AAL模板对重度抑郁症患者接受治疗后症状减轻状况的预测结果。由结果对比可知，该基于全脑搜索的预测方法能够显著提高预测精度，并且能够识别出具有临床意义的特征脑区，可作为辅助医疗诊断的生物标志，对病情恢复情况高达89.5％的预测精度能够真正地实现在病人接受治疗前就对其治疗效果进行预测，实现精准医疗的目标。
本实施例中虽然将各个步骤按照上述先后次序的方式进行了描述，但是本领域技术人员可以理解，为了实现本实施例的效果，不同的步骤之间不必按照这样的次序执行，其可以同时(并行)执行或以颠倒的次序执行，这些简单的变化都在本发明的保护范围之内。
基于与方法实施例相同的技术构思，本发明实施例还提供一种基于磁共振影像的脑疾病个体化预测系统。如图5所示，该系统可以包括：获取模块51、预处理模块52、特征选择模块53、聚类模块54、筛选模块55和识别模块56。其中，获取模块51被配置为获取精神疾病病人脑磁共振影像。预处理模块52被配置为对病人脑磁共振影像进行去噪和降维处理。特征选择模块53被配置为利用ReliefF算法，对处理模块的处理结果进行特征选择。聚类模块54被配置为基于特征选择模块的结果，使用基于数据驱动的空间聚类分析方法自适应地得到空间脑区。筛选模块55被配置为基于聚类模块得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集。识别模块56被配置为基于最优特征子集，进行多元线性回归分析，识别出潜在的生物标志。
需要说明的是，上述实施例提供的基于磁共振影像的脑疾病个体化预测系统在进行脑疾病个体化预测时，仅以上述各功能模块的划分进行举例说明，在实际应用中，可以根据需要而将上述功能分配由不同的功能模块来完成，即将系统的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。
本领域技术人员可以理解，上述基于磁共振影像的脑疾病个体化预测系统还包括一些其他公知结构，例如处理器、存储器等，为了不必要地模糊本公开的实施例，这些公知的结构未在图5中示出。
应该理解，图5中的各个模块的数量仅仅是示意性的。根据实际需要，可以具有任意数量的各模块。
上述系统实施例可以用于执行上述方法实施例，其技术原理、所解决的技术问题及产生的技术效果相似，所属技术领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
应指出的是，上面分别对本发明的系统实施例和方法实施例进行了描述，但是对一个实施例描述的细节也可应用于另一个实施例。对于本发明实施例中涉及的模块、步骤的名称，仅仅是为了区分各个模块或者步骤，不视为对本发明的不当限定。本领域技术人员应该理解：本发明实施例中的模块或者步骤还可以再分解或者组合。例如上述实施例的模块可以合并为一个模块，也可以进一步拆分成多个子模块。
以上对本发明实施例所提供的技术方案进行了详细的介绍。虽然本文应用了具体的个例对本发明的原理和实施方式进行了阐述，但是，上述实施例的说明仅适用于帮助理解本发明实施例的原理；同时，对于本领域技术人员来说，依据本发明实施例，在具体实施方式以及应用范围之内均会做出改变。
需要说明的是，本文中涉及到的流程图或框图不仅仅局限于本文所示的形式，其还可以进行划分和/或组合。
需要说明的是：附图中的标记和文字只是为了更清楚地说明本发明，不视为对本发明保护范围的不当限定。
术语“包括”或者任何其它类似用语旨在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备/装置不仅包括那些要素，而且还包括没有明确列出的其它要素，或者还包括这些 过程、方法、物品或者设备/装置所固有的要素。
如本文中所使用的，术语“模块”可以指代在计算系统上执行的软件对象或例程。可以将本文中所描述的不同模块实现为在计算系统上执行的对象或过程(例如，作为独立的线程)。虽然优选地以软件来实现本文中所描述的系统和方法，但是以硬件或者软件和硬件的组合的实现也是可以的并且是可以被设想的。
本发明的各个步骤可以用通用的计算装置来实现，例如，它们可以集中在单个的计算装置上，例如：个人计算机、服务器计算机、手持设备或便携式设备、平板型设备或者多处理器装置，也可以分布在多个计算装置所组成的网络上，它们可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。因此，本发明不限于任何特定的硬件和软件或者其结合。
本发明提供的方法可以使用可编程逻辑器件来实现，也可以实施为计算机程序软件或程序模块(其包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件或数据结构等等)，例如根据本发明的实施例可以是一种计算机程序产品，运行该计算机程序产品使计算机执行用于所示范的方法。所述计算机程序产品包括计算机可读存储介质，该介质上包含计算机程序逻辑或代码部分，用于实现所述方法。所述计算机可读存储介质可以是被安装在计算机中的内置介质或者可以从计算机主体上拆卸下来的可移动介质(例如：采用热插拔技术的存储设备)。所述内置介质包括但不限于可重写的非易失性存 储器，例如：RAM、ROM、快闪存储器和硬盘。所述可移动介质包括但不限于：光存储介质(例如：CD－ROM和DVD)、磁光存储介质(例如：MO)、磁存储介质(例如：磁带或移动硬盘)、具有内置的可重写非易失性存储器的媒体(例如：存储卡)和具有内置ROM的媒体(例如：ROM盒)。
本发明并不限于上述实施方式，在不背离本发明实质内容的情况下，本领域普通技术人员可以想到的任何变形、改进或替换均落入本发明的保护范围。

Claims (6)

  1. 一种基于磁共振影像的脑疾病个体化预测方法，其特征在于，所述方法至少包括：
    步骤1：获取精神疾病病人脑磁共振影像；
    步骤2：对所述病人脑磁共振影像进行去噪和降维处理；
    步骤3：利用ReliefF算法，对所述步骤2的处理结果进行特征选择；
    步骤4：基于所述步骤3的结果，使用基于数据驱动的空间聚类分析方法自适应地得到空间脑区；
    步骤5：基于所述步骤4得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集；
    步骤6：基于所述最优特征子集，进行多元线性回归分析，识别出潜在的生物标志。
  2. 根据权利要求1所述的方法，其特征在于，所述步骤3具体包括：
    基于所述步骤2的处理结果，利用ReliefF算法，计算每个体素特征对目标预测属性的区分能力；
    以权值大小作为划分能力量化标准，按照从大到小的顺序，选择预定数量的体素特征。
  3. 根据权利要求2所述的方法，其特征在于，所述步骤4具体包括：
    基于所述步骤3的结果，针对空间稀疏分布的全脑体素，以每一体素为中心，构建空间正方体，通过空间邻域连通规则聚集与该中心 体素相邻接的所有体素，构成空间脑区。
  4. 根据权利要求3所述的方法，其特征在于，所述步骤5具体包括：
    以所述步骤4所得的空间脑区为基础特征，以该空间脑区体素的均值为特征代表，利用基于相关的特征选择算法去除冗余特征，提取最优特征子集。
  5. 根据权利要求4所述的方法，其特征在于，所述步骤6具体包括：
    建立所述最优特征子集与所述目标预测属性之间的回归关系；
    通过所述回归关系，计算所述目标预测属性的预测值与真实值之间的相关系数；
    根据所述相关系数，确定最佳拟合程度；
    将得到最佳拟合程度所对应的回归关系中的脑区，确定为具有预测性的生物标志。
  6. 一种基于磁共振影像的脑疾病个体化预测系统，其特征在于，所述系统至少包括：
    获取模块，被配置为获取精神疾病病人脑磁共振影像；
    预处理模块，被配置为对所述病人脑磁共振影像进行去噪和降维处理；
    特征选择模块，被配置为利用ReliefF算法，对所述处理模块的处理结果进行特征选择；
    聚类模块，被配置为基于所述特征选择模块的结果，使用基于数 据驱动的空间聚类分析方法自适应地得到空间脑区；
    筛选模块，被配置为基于所述聚类模块得到的空间脑区，利用基于相关的特征选择算法，去除冗余特征，得到最优特征子集；
    识别模块，被配置为基于所述最优特征子集，进行多元线性回归分析，识别出潜在的生物标志。
PCT/CN2016/085217 2016-06-08 2016-06-08 基于磁共振影像的脑疾病个体化预测方法和系统 WO2017210873A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/085217 WO2017210873A1 (zh) 2016-06-08 2016-06-08 基于磁共振影像的脑疾病个体化预测方法和系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/085217 WO2017210873A1 (zh) 2016-06-08 2016-06-08 基于磁共振影像的脑疾病个体化预测方法和系统

Publications (1)

Publication Number Publication Date
WO2017210873A1 true WO2017210873A1 (zh) 2017-12-14

Family

ID=60578352

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/085217 WO2017210873A1 (zh) 2016-06-08 2016-06-08 基于磁共振影像的脑疾病个体化预测方法和系统

Country Status (1)

Country Link
WO (1) WO2017210873A1 (zh)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443798A (zh) * 2018-12-25 2019-11-12 电子科技大学 一种基于磁共振图像的自闭症检测方法、装置及系统
CN110880008A (zh) * 2018-09-06 2020-03-13 刘艳 基于脑核磁影像数据的结构特征提取和分类方法
CN111000571A (zh) * 2019-12-26 2020-04-14 上海市精神卫生中心（上海市心理咨询培训中心） 创伤后心理疾病的风险预测方法，相关设备及存储装置
CN111415324A (zh) * 2019-08-09 2020-07-14 复旦大学附属华山医院 基于磁共振成像的脑病灶图像空间分布特征的分类鉴别方法
CN111414579A (zh) * 2020-02-19 2020-07-14 深圳市儿童医院 基于多角度相关关系获取脑区关联信息的方法和系统
CN111949812A (zh) * 2020-07-10 2020-11-17 上海联影智能医疗科技有限公司 脑图像分类方法和存储介质
CN112837274A (zh) * 2021-01-13 2021-05-25 南京工业大学 一种基于多模态多站点数据融合的分类识别方法
CN112990266A (zh) * 2021-02-07 2021-06-18 西安电子科技大学 多模态脑影像数据处理的方法、装置、设备及存储介质
CN113610751A (zh) * 2021-06-03 2021-11-05 迈格生命科技（深圳）有限公司 图像处理方法、装置及计算机可读存储介质
US11523761B2 (en) 2019-06-06 2022-12-13 Tata Consultancy Services Limited Method and system for assessment of cognitive workload using breathing pattern of a person
CN115984266A (zh) * 2023-03-20 2023-04-18 首都医科大学附属北京天坛医院 一种脑区中靶点定位方法及系统
CN117058471A (zh) * 2023-10-12 2023-11-14 之江实验室 基于正常脑影像数据库的疾病脑影像分型系统
CN117476247A (zh) * 2023-12-27 2024-01-30 杭州深麻智能科技有限公司 一种疾病多模态数据智能分析方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855491A (zh) * 2012-07-26 2013-01-02 中国科学院自动化研究所 一种基于网络中心性的脑功能磁共振图像分类方法
CN103034778A (zh) * 2012-09-28 2013-04-10 中国科学院自动化研究所 适合多被试脑功能数据分析的个体脑功能网络提取方法
CN103116764A (zh) * 2013-03-02 2013-05-22 西安电子科技大学 一种基于多线性主元分析的大脑认知状态判定方法
CN103886328A (zh) * 2014-03-19 2014-06-25 太原理工大学 基于脑网络模块结构特征的功能磁共振影像数据分类方法
US20150018664A1 (en) * 2013-07-12 2015-01-15 Francisco Pereira Assessment of Traumatic Brain Injury

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855491A (zh) * 2012-07-26 2013-01-02 中国科学院自动化研究所 一种基于网络中心性的脑功能磁共振图像分类方法
CN103034778A (zh) * 2012-09-28 2013-04-10 中国科学院自动化研究所 适合多被试脑功能数据分析的个体脑功能网络提取方法
CN103116764A (zh) * 2013-03-02 2013-05-22 西安电子科技大学 一种基于多线性主元分析的大脑认知状态判定方法
US20150018664A1 (en) * 2013-07-12 2015-01-15 Francisco Pereira Assessment of Traumatic Brain Injury
CN103886328A (zh) * 2014-03-19 2014-06-25 太原理工大学 基于脑网络模块结构特征的功能磁共振影像数据分类方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MENG, XING: "Predicting individualized clinical measures by a generalized prediction framework and multimodal fusion of MRI data", NEUROLMAGE, 10 May 2016 (2016-05-10), XP029856191, ISSN: 1053-8119 *

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110880008A (zh) * 2018-09-06 2020-03-13 刘艳 基于脑核磁影像数据的结构特征提取和分类方法
CN110880008B (zh) * 2018-09-06 2023-10-17 刘艳 基于脑核磁影像数据的结构特征提取和分类方法
CN110443798A (zh) * 2018-12-25 2019-11-12 电子科技大学 一种基于磁共振图像的自闭症检测方法、装置及系统
US11523761B2 (en) 2019-06-06 2022-12-13 Tata Consultancy Services Limited Method and system for assessment of cognitive workload using breathing pattern of a person
CN111415324A (zh) * 2019-08-09 2020-07-14 复旦大学附属华山医院 基于磁共振成像的脑病灶图像空间分布特征的分类鉴别方法
CN111415324B (zh) * 2019-08-09 2024-03-08 复旦大学附属华山医院 基于磁共振成像的脑病灶图像空间分布特征的分类鉴别方法
CN111000571A (zh) * 2019-12-26 2020-04-14 上海市精神卫生中心（上海市心理咨询培训中心） 创伤后心理疾病的风险预测方法，相关设备及存储装置
CN111414579A (zh) * 2020-02-19 2020-07-14 深圳市儿童医院 基于多角度相关关系获取脑区关联信息的方法和系统
CN111414579B (zh) * 2020-02-19 2023-05-23 深圳市儿童医院 基于多角度相关关系获取脑区关联信息的方法和系统
CN111949812A (zh) * 2020-07-10 2020-11-17 上海联影智能医疗科技有限公司 脑图像分类方法和存储介质
CN112837274B (zh) * 2021-01-13 2023-07-07 南京工业大学 一种基于多模态多站点数据融合的分类识别方法
CN112837274A (zh) * 2021-01-13 2021-05-25 南京工业大学 一种基于多模态多站点数据融合的分类识别方法
CN112990266B (zh) * 2021-02-07 2023-08-15 西安电子科技大学 多模态脑影像数据处理的方法、装置、设备及存储介质
CN112990266A (zh) * 2021-02-07 2021-06-18 西安电子科技大学 多模态脑影像数据处理的方法、装置、设备及存储介质
CN113610751A (zh) * 2021-06-03 2021-11-05 迈格生命科技（深圳）有限公司 图像处理方法、装置及计算机可读存储介质
CN115984266A (zh) * 2023-03-20 2023-04-18 首都医科大学附属北京天坛医院 一种脑区中靶点定位方法及系统
CN115984266B (zh) * 2023-03-20 2023-07-04 首都医科大学附属北京天坛医院 一种脑区中靶点定位方法及系统
CN117058471A (zh) * 2023-10-12 2023-11-14 之江实验室 基于正常脑影像数据库的疾病脑影像分型系统
CN117058471B (zh) * 2023-10-12 2024-01-09 之江实验室 基于正常脑影像数据库的疾病脑影像分型系统
CN117476247A (zh) * 2023-12-27 2024-01-30 杭州深麻智能科技有限公司 一种疾病多模态数据智能分析方法
CN117476247B (zh) * 2023-12-27 2024-04-19 杭州乐九医疗科技有限公司 一种疾病多模态数据智能分析方法

Similar Documents

Publication Publication Date Title
WO2017210873A1 (zh) 基于磁共振影像的脑疾病个体化预测方法和系统
Altinkaya et al. Detection of Alzheimer’s disease and dementia states based on deep learning from MRI images: a comprehensive review
Li et al. Alzheimer's disease diagnosis based on multiple cluster dense convolutional networks
Basheer et al. Computational modeling of dementia prediction using deep neural network: analysis on OASIS dataset
Wang et al. Multi-site clustering and nested feature extraction for identifying autism spectrum disorder with resting-state fMRI
CN106156484A (zh) 基于磁共振影像的脑疾病个体化预测方法和系统
Direito et al. Modeling epileptic brain states using EEG spectral analysis and topographic mapping
Yan et al. Cortical surface biomarkers for predicting cognitive outcomes using group l2, 1 norm
Lei et al. Multi-scale enhanced graph convolutional network for mild cognitive impairment detection
JP2022507861A (ja) 脳機能地図のサル－ヒト種間移行に基づいて精神疾患の個別的予測を行う方法およびシステム
Liu et al. An enhanced multi-modal brain graph network for classifying neuropsychiatric disorders
Cao et al. Generalized fused group lasso regularized multi-task feature learning for predicting cognitive outcomes in Alzheimers disease
CN115662576B (zh) 关联认知障碍病症的神经反馈训练范式的生成方法和系统
Liu et al. Fused group lasso regularized multi-task feature learning and its application to the cognitive performance prediction of Alzheimer’s disease
Baker et al. [Retracted] Implementing Critical Machine Learning (ML) Approaches for Generating Robust Discriminative Neuroimaging Representations Using Structural Equation Model (SEM)
Gamal et al. Automatic early diagnosis of Alzheimer’s disease using 3D deep ensemble approach
Rashid et al. Biceph-Net: A robust and lightweight framework for the diagnosis of Alzheimer's disease using 2D-MRI scans and deep similarity learning
Fareed et al. ADD-Net: an effective deep learning model for early detection of Alzheimer disease in MRI scans
Jiang et al. Interpreting Brain Biomarkers: Challenges and solutions in interpreting machine learning-based predictive neuroimaging
Qiu et al. Multi-channel sparse graph transformer network for early alzheimer’s disease identification
Taghavirashidizadeh et al. WTD‐PSD: Presentation of Novel Feature Extraction Method Based on Discrete Wavelet Transformation and Time‐Dependent Power Spectrum Descriptors for Diagnosis of Alzheimer’s Disease
Shanmugavadivel et al. Advancements in computer-assisted diagnosis of Alzheimer's disease: A comprehensive survey of neuroimaging methods and AI techniques for early detection
Mehmood et al. Early Diagnosis of Alzheimer's Disease Based on Convolutional Neural Networks.
Rao et al. A Review on Alzheimer’s disease through analysis of MRI images using Deep Learning Techniques
KR102439639B1 (ko) 주요 우울증에 대한 정보 제공 방법 및 이를 이용한 주요 우울증에 대한 정보 제공용 디바이스

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16904326

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16904326

Country of ref document: EP

Kind code of ref document: A1