CN117954080A - Depression characteristic analysis method and system based on LSTM effective connection brain network model - Google Patents
Depression characteristic analysis method and system based on LSTM effective connection brain network model Download PDFInfo
- Publication number
- CN117954080A CN117954080A CN202410128454.0A CN202410128454A CN117954080A CN 117954080 A CN117954080 A CN 117954080A CN 202410128454 A CN202410128454 A CN 202410128454A CN 117954080 A CN117954080 A CN 117954080A
- Authority
- CN
- China
- Prior art keywords
- connection
- effective connection
- network model
- brain
- brain network
- 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.)
- Pending
Links
- 210000004556 brain Anatomy 0.000 title claims abstract description 100
- 238000004458 analytical method Methods 0.000 title claims abstract description 29
- 239000013598 vector Substances 0.000 claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 34
- 230000000694 effects Effects 0.000 claims abstract description 33
- 230000015654 memory Effects 0.000 claims abstract description 24
- 230000004927 fusion Effects 0.000 claims abstract description 20
- 230000003925 brain function Effects 0.000 claims description 28
- 238000000034 method Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 10
- 230000033001 locomotion Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000007787 long-term memory Effects 0.000 claims description 5
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 13
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 7
- 201000010099 disease Diseases 0.000 abstract description 6
- 238000002599 functional magnetic resonance imaging Methods 0.000 description 16
- 238000012360 testing method Methods 0.000 description 11
- 238000013145 classification model Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 7
- CIWBSHSKHKDKBQ-JLAZNSOCSA-N Ascorbic acid Chemical compound OC[C@H](O)[C@H]1OC(=O)C(O)=C1O CIWBSHSKHKDKBQ-JLAZNSOCSA-N 0.000 description 5
- 208000020401 Depressive disease Diseases 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 208000020016 psychiatric disease Diseases 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 230000001771 impaired effect Effects 0.000 description 2
- 238000002595 magnetic resonance imaging Methods 0.000 description 2
- 238000002610 neuroimaging Methods 0.000 description 2
- 229910052760 oxygen Inorganic materials 0.000 description 2
- 230000001936 parietal effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 208000024891 symptom Diseases 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 208000019901 Anxiety disease Diseases 0.000 description 1
- 208000027559 Appetite disease Diseases 0.000 description 1
- 206010003805 Autism Diseases 0.000 description 1
- 208000020706 Autistic disease Diseases 0.000 description 1
- 208000019454 Feeding and Eating disease Diseases 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000036506 anxiety Effects 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000994 depressogenic effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013399 early diagnosis Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 210000001061 forehead Anatomy 0.000 description 1
- 230000005714 functional activity Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 210000004326 gyrus cinguli Anatomy 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 230000004886 head movement Effects 0.000 description 1
- 230000005923 long-lasting effect Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 230000004630 mental health Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000036651 mood Effects 0.000 description 1
- 230000000926 neurological effect Effects 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000009323 psychological health Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
- 238000005201 scrubbing Methods 0.000 description 1
- 230000007958 sleep Effects 0.000 description 1
- 208000019116 sleep disease Diseases 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a depression feature analysis method and a depression feature analysis system based on an LSTM (least squares) effective connection brain network model, which are characterized in that firstly, active vectors with time stamps in each interested region of each sample in a standard brain image data set are extracted; then determining the functional connection strength of the sample; secondly, an effective connection brain network model is constructed based on a long-period memory network, an activity vector with a time stamp is used as training data, the effective connection brain network model is trained, and weight parameters of the trained effective connection brain network model are used as the effective connection of an input gate, a forgetting gate and an output gate to be tested; the input door, the forgetting door and the output door are effectively connected and the functional connection strength are fused, so that the connection fusion strength of the sample brain network is obtained, the connection fusion strength is adopted to judge the depression of the testee, the interference of images among different patients can be effectively eliminated, the precision of depression classification is improved, the time required by disease diagnosis is greatly reduced, and the diagnosis efficiency is improved.
Description
Technical Field
The invention relates to the technical field of medical treatment, in particular to a depression characteristic analysis method and a depression characteristic analysis system based on an LSTM (least squares) effective connection brain network model.
Background
With the acceleration of life rhythm, social competitiveness is increasing, and people stress anxiety, so that patients suffering from depression are increasing. Depression is a mental disorder disease characterized by significantly long-lasting mood, impaired interest, impaired cognitive function, sleep and appetite disorders, and has the characteristics of high prevalence, high misdiagnosis rate, and high recurrence rate. About 20% of people are statistically suffering from depression during life. Depression has higher disability rate and mortality rate, not only seriously affects the personal life quality of patients, but also brings great burden to the families of the patients and the whole society. Despite centuries of psychiatric and neurological studies, people have had very limited knowledge of the physiological and pathological mechanisms of depression.
Early diagnosis of depression is critical for rehabilitation of depressed patients. However, depression is a heterogeneous disease, clinical symptoms are very complex and similar to some symptoms of other mental diseases, so that depression diagnosis mainly depends on evaluation of mental health scales and subjective experience of psychiatrists, and has poor consistency of results, high misdiagnosis rate and lack of objective diagnosis means. If depression can be diagnosed early, patients can have a better quality of life through continued treatment and management of the disease. Accordingly, a great deal of research has been devoted to innovative approaches to identify early depression and to supplement current diagnostic approaches.
In recent years, non-invasive Magnetic Resonance Imaging (MRI) techniques have been widely used in clinical practice to provide objective evidence for the diagnosis of depression. Resting state functional magnetic resonance imaging (fMRI) is one of the brain imaging modalities commonly used in mental science, the principle of which is to reflect spontaneous brain functional activity through fluctuations in Blood Oxygen Level Dependent (BOLD) signals. fMRI is noninvasive and free of radiation damage to the human body, and the tested person is easy to cooperate, so that the change of brain function activities of mental disorder patients can be explored. Through fMRI scanning and analysis, the accuracy of mental disease diagnosis is improved, and therefore, neuroimaging methods are popular in research. Although depression can be predicted by fMRI, there are still problems: first, fMRI analysis is performed manually, and medical image visual analysis is a subjective, time-consuming process; second, fMRI requires manual extraction of specified features based on a priori knowledge of the physician, which results in significant limitations in the representation of image features; finally, most fMRI analyses need to be focused on image analysis at a certain point in time for a certain patient, which is easily disturbed by different patients. Therefore, there is a strong need for an improved depression feature analysis method and system that fuses multiple features of fMRI images to enable diagnosis of autism.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a depression feature analysis method and a depression feature analysis system based on an LSTM (least squares) effective connection brain network model, which effectively integrate various features of tested fMRI (magnetic resonance imaging) images, are beneficial to the time required by doctors for diagnosing depression, and can improve the accuracy of diagnosis.
The invention is realized by the following technical scheme:
A depression feature analysis method based on LSTM effective connection brain network model comprises the following steps:
Step 1, extracting activity vectors with time stamps of each interested region of each sample in a set according to a standard brain image data set;
Step 2, determining the functional connection strength of the sample according to the interrelation between any two vectors with time stamps of the same sample;
step 3, constructing an effective connection brain network model based on a long-period memory network, taking an activity vector with a time stamp as training data, training the effective connection brain network model, and taking weight parameters of the trained effective connection brain network model as input gate effective connection, forgetting gate effective connection and output gate effective connection of a sample;
And 4, fusing the input door effective connection, the forget door effective connection, the output door effective connection and the functional connection strength of each sample to obtain the connection fusion strength of the sample brain network.
Preferably, the method of the standard brain image dataset of step 1 is as follows:
and acquiring an original brain function image data set and preprocessing to obtain a standard brain function image data set.
Preferably, the preprocessing method of the original brain function image dataset is as follows:
Converting the original brain function image data into Nifti data, then removing the first n images in Nifti data, removing various noises from the rest images, and enhancing the characteristic information to obtain standard brain image data.
Preferably, the method for extracting the activity vector with the timestamp in the step 1 is as follows:
Selecting the interested regions of each sample, extracting the time sequence of the corresponding brain region of each interested region by using a brain function template, taking the time sequence as an activity signal of the interested region, determining the average value of the current time point of each interested region according to the activity signal, taking the average value as the activity intensity of the interested region at the current time point, and further obtaining the activity vector with the time stamp of each interested region.
Preferably, the method for determining the functional connection strength in step 2 is as follows:
And calculating the average value of the active signals of the two regions of interest by adopting a correlation coefficient method to obtain a correlation coefficient, and converting the correlation coefficient into functional connection strength.
Preferably, the correlation coefficient method is Pearson correlation coefficient, spearman correlation coefficient or Kendall rank correlation coefficient.
Preferably, the training method for the effective connection brain network model is as follows:
Taking the activity vector with the time stamp as the input of the effective connection brain network model, according to the current time step input and the last time step hidden state of each door of the effective connection brain network model, and calculating an input door, a forget door and an output door by combining an activation function; calculating candidate memory cells effectively connected with the brain network model by using the tanh function as an activation function;
And iteratively updating weights of the input gate, the forgetting gate, the output gate and the candidate memory cells by adopting an error back propagation and gradient descent algorithm until the maximum training iteration number is reached or the error between the hidden states H t-1 and H t of two adjacent time steps is smaller than a set value, so as to obtain the trained effective connection brain network model.
Preferably, in the step 4, the input door effective connection, the forget door effective connection, the output door effective connection and the functional connection strength of the sample are fused, so as to obtain the connection fusion strength of the brain network.
Preferably, the input door effective connection, the forgetting door effective connection, the output door effective connection and the functional connection strength are respectively unfolded and spliced into one-dimensional vectors, so as to obtain the connection fusion strength of the brain network.
A system for LSTM-based method of profiling depression in an active linked brain network model, comprising:
The motion vector module is used for extracting motion vectors with time stamps of the interesting areas of each sample in the collection according to the standard brain image data set;
the functional connection strength module is used for determining the functional connection strength of the sample according to the interrelation between any two vectors with time stamps of the same sample;
The effective connection module is used for constructing an effective connection brain network model based on the long-term memory network, training the effective connection brain network model by taking the activity vector with the time stamp as training data, and generating input gate effective connection, forgetting gate effective connection and output gate effective connection of a sample by the trained effective connection brain network model;
And the connection fusion strength module is used for fusing the input door effective connection, the forgetting door effective connection, the output door effective connection and the functional connection strength of each sample to obtain the connection fusion strength of the brain network of the sample. Compared with the prior art, the invention has the following beneficial technical effects:
According to the depression feature analysis method based on the LSTM effective connection brain network model, the LSTM model is used for generating effective connection of the tested fMRI image, the effective connection is effectively fused with the functional connection strength, the feature vector suitable for a machine learning classification model is generated, the LSTM model realizes memory of different brain area states in the information transmission process through the memory cell C t and the hidden state H t, and the effective connection relation between the brain areas is embodied. Secondly, training a tested person based on an LSTM (least squares) effective connection brain network model, inputting the connection fusion strength of the tested brain network, effectively mining the brain network characteristics related to depression, and judging the depression of the tested person by adopting the connection fusion strength; in addition, the method for classifying the depression provided by the invention effectively extracts the characteristics of the depression and the fMRI images of the control group through machine learning, and can apply the trained model to a new tested, so that the interference of images among different patients can be effectively eliminated, the precision of classifying the depression is improved, the time required by disease diagnosis is greatly reduced, and the diagnosis efficiency is improved.
Drawings
FIG. 1 is a flow chart of a depression feature analysis method based on an LSTM operatively connected brain network model of the present invention;
FIG. 2 is a block diagram of a long and short term memory network of the present invention;
FIG. 3 is a graph of training time for the classification model of the present invention.
Detailed Description
The invention will now be described in further detail with reference to the accompanying drawings, which illustrate but do not limit the invention.
Referring to fig. 1, a depression feature analysis method based on an LSTM active connection brain network model includes the following steps:
Step 1, preprocessing an original brain function image dataset to be tested to obtain a standard brain image dataset.
Converting each original DICOM brain function magnetic resonance data acquired by a test into Nifti-format data; and then, preprocessing operations such as removing the first n images, SLICE TIMING, head movement correction, standardization, space smoothing and the like are carried out on the Nifti format data obtained through conversion, irrelevant information such as various noises and the like in the original image is removed, the detectability of useful information is enhanced, and a brain function image with more ideal quality is obtained.
In this embodiment, the collected brain function magnetic resonance data of each test is preprocessed, and each test is taken as a sample, that is, the brain function magnetic resonance data of the sample is preprocessed, and the preprocessing steps that may be used include, but are not limited to: DICOM conversion to Nifti, removal of the first n images, SLICE TIMING, head correction, normalization, spatial smoothing, de-linearities, regression covariates, frequency domain filtering, scrubbing. Let the operation set for preprocessing be A, the original brain function image data set beStandard brain image dataset d= { subj i |i e [1, n ] }, after preprocessing.
Wherein N represents the number of the tested, subj i represents the standard brain image data of the tested i, and the standard brain image data is represented as Standard brain image data representing the subject i at time t. In operation, firstly, determining a preprocessing operation to be executed, constructing a preprocessed operation set A= { act j |j epsilon [1, K ] }, then sequentially taking out an operation act j from the set A, and defining parameters of the operation and applying the parameters to a tested/>And then the processed standard brain image data subj i is obtained, namely/>
And 2, extracting the activity vector with the time stamp in each tested region of interest in the data set according to the standard brain image data set.
According to each tested in the standard brain function signal set obtained in the step 1, selecting each tested region of interest, extracting a time sequence of the brain region corresponding to each region of interest by adopting a brain function template, taking the time sequence as an activity signal of the region of interest, determining the average value of the current time point of each region of interest according to the activity signal, taking the average value as the activity intensity of the region of interest at the current time point, and further obtaining an activity vector with a time stamp of each region of interest.
In this embodiment, brain function signal of subject iLet the brain function signal/>, at time t, of the tested iM represents the number of the interested regions, calculates the average value of the activity signals of each interested region of the tested i at the moment t, and is taken as the activity intensity of the interested region of the tested i at the moment t and is recorded as/>Then the brain function signal mean of the tested i at time t can be expressed as/>The mean value of brain function signals of the tested i in the [1, T ] time period is/>The mean of the N brain function signals tested over the [1, T ] time period can be expressed as/>
And step 3, obtaining the functional connection strength of the same tested according to the interrelation between any two motion vectors with time stamps.
Brain function signal mean value of test i in [1, T ] time periodWherein M, T represent the number of regions of interest and the length of time, respectively. Let/> Representing the mean value of the activity signal of the mth region of interest of the tested i in the [1, T ] time period. At/>The mean value/>, of the activity signals of two different regions of interest of m a、mb is takenAnd/>Calculate its correlation coefficient/>Calculation methods include, but are not limited to, pearson correlation coefficients, spearman correlation coefficients, or Kendall rank correlation coefficients, which are commonly used in statistics. For correlation coefficient/>Conversion of Fisher Z transforms into functional ligation Strength/>The data is made to approximate a normal distribution for further analysis, and the calculation formula is as follows:
The functional connection strength of test i over the [1, T ] period can be expressed as The functional connection strength of N tested time periods 1, t may be expressed as z= { Z i |i e 1, N }.
And 4, constructing an effective connection brain network model based on the long-term and short-term memory network, training the effective connection brain network model by taking each tested activity vector with a time stamp as training data, and taking the weight parameters of the trained effective connection brain network model as three effective connection of each tested input gate, forgetting gate and output gate.
Each tested active connection is generated using the LSTM-based active connection brain network model. LSTM incorporates 3 gates, i.e., an input gate, a forget gate, and an output gate, and memory cells of the same dimensions as the hidden state for recording additional information, as shown in fig. 2.
The inputs of each gate of LSTM are the current time step input X t and the last time step hidden state H t-1, and the output is calculated by the fully connected layer whose activation function is the sigmoid function (sigma). The value thresholds for these 3 gate elements are all 0, 1. The input gate I t, the forget gate F t and the output gate O t of time step t are calculated as follows:
It=σ(XtWxi+Ht-1Whi+bi) (2)
Ft=σ(XtWxf+Ht-1Whf+bf) (3)
Ot=σ(XtWxo+Ht-1Who+bo) (4)
Where h is the number of hidden units, W xi、Wxf、Wxo、Whi、Whf、Who is the weight parameter, and b i、bf、bo is the bias parameter corresponding to each gate.
Candidate memory cellsSimilar to the calculation of 3 gates, a tanh function with a value range of [ -1,1] is used as the activation function. Candidate memory cells for time step t/>The calculation formula of (2) is as follows:
Where W xc and W hc are weight parameters and b c is a bias parameter.
The calculation of the current time step memory cell C t needs to combine the information of the previous time step memory cell and the current time step candidate memory cell, and the flow of the information is controlled through a forgetting gate and an input gate, and the calculation formula is as follows:
wherein the symbol +.is that the matrices are multiplied by element.
After the memory cells are present, the flow of information from the memory cells to the hidden state H t is controlled by the output gate, and the calculation formula is:
Ht=Ot⊙tanh(Ct) (7)
n tested brain function signal average values obtained according to the step 2 For each tested i, its brain function signal mean/>As training data, M, T denotes the number of regions of interest and the length of time, respectively. Assuming that the number of hidden units in the LSTM model is h, the batch size is n (n < T), and the input/>, of a given time step TThe hidden state H t-1∈Rn×h at the previous time step, the memory cell C t-1∈Rn×h at the previous step, the input gate I t∈Rn×h, the forgetting gate F t∈Rn×h and the output gate O t∈Rn×h are respectively calculated according to the formulas (2) (3) (4), the parameter dimension is Wxi、Wxf、Wxo∈RM×h,Whi、Whf、Who∈Rh×h,bi、bf、bo∈R1×h;, and the candidate memory cell/>, is calculated according to the formula (5)The parameter dimensions are W xc∈RM×h,Whc∈Rh×h,bc∈R1×h, respectively; calculating a memory cell C t∈Rn×h according to formula (6); according to equation (7), the hidden state H t∈Rn×h is calculated. The weights of the input gate, the forgetting gate, the output gate and the candidate memory cells are then updated using an error back propagation and gradient descent algorithm.
The training process is repeated until the maximum number of training iterations or the error between the hidden states H t-1 and H t of two adjacent time steps is less than a certain value. The weight parameters W hi、Whf、Who of the input gate I t, the forgetting gate F t and the output gate O t of the tested I are square matrixes, and the number of rows and columns is equal to the number of the interested areas. Therefore, the three square matrixes can be used as three effective connection models between the interested areas and are respectively marked as I i、Fi、Oi, wherein each value in each square matrix represents the effective connection strength, and the effective connection direction is that the brain area corresponding to the column where the value is located points to the brain area corresponding to the row where the value is located. Training each test is repeated, then the three active connections of the N test may be represented as I= { I i|i∈[1,N]}、F={Fi|i∈[1,N]}、O={Oi |i ε [1, N ] }, respectively.
And 5, fusing the three effective connection strengths and the functional connection strengths of each tested to be used as training data, constructing a depression classification model, training, and outputting judgment data by the trained depression classification model.
And (3) fusing the functional connection strength Z obtained in the step (3) with the three effective connections I, F, O obtained in the step (4) to obtain N tested data features, namely training data. Specifically, for the tested I, three effective connection I i、Fi、Oi and the functional connection strength Z i are respectively unfolded and spliced into a one-dimensional vector, so as to obtain the connection fusion strength of the brain network, and the connection fusion strength is used as the data characteristic of the tested I.
Selecting a model as a classifier for training, carrying out normalization operation on all tested data features to improve convergence speed and precision of the classification model, setting each tested label as a category of a group (a disease group and a comparison group), and carrying out model verification by using K-fold cross verification; and finally, calculating the accuracy, recall and F value of the model to evaluate the performance of the classification model.
In this embodiment, the functional connection strength and the three effective connections are used as the data features to be tested, and are divided into a training set and a testing set after normalization. Constructing a depression classification model, training on a training set by using a K-fold cross validation method, and validating the classification effect of the model on a test set.
Referring to fig. 3, a number of trials was selected for the experiment, increasing from 100 to 1000 for each increment of 100. The running program computer is configured to: CPU Intel to strong Gold 5218, frequency 2.30GHZ, 16GB memory. Using a support vector machine as a classifier, run in the Python 3.11.4 environment and the time required for training is shown in FIG. 3. It can be seen that the training time increases gradually with increasing number of subjects, but does not exceed 0.385 seconds, greatly improving the diagnostic efficiency.
Example 1
Taking a specific problem of depression classification as an example, two groups of brain function images with the same number are selected for pretreatment, and the interested areas are set as four brain areas of a default network (DMN), namely an inner forehead cortex (mPFC), a left rear parietal cortex (LPC), a right Rear Parietal Cortex (RPC) and a Posterior Cingulate Cortex (PCC). For each test, the dimension was calculated in turn to be 1×6 functional connection strength, three 4×4 effective connections were generated using the LSTM-based effective connection brain network model, and the functional connection strength and three effective connections were unfolded and stitched into one 1×54 data feature. And carrying out normalization operation on all tested data features, using a support vector machine as a classifier, adopting 10 times of 10-fold cross validation during classification, namely dividing all tested data features into 10 parts, wherein 9 parts are used as training sets, 1 part is used as a test set, training and evaluating a model, and finally calculating the average value of 10 evaluation indexes to obtain the model with the accuracy of 92.5%, the recall rate of 92.3% and the F value of 92.0%.
Example 2
A system for LSTM-based method of profiling depression in an active linked brain network model, comprising:
The motion vector module is used for extracting motion vectors with time stamps of the interesting areas of each sample in the collection according to the standard brain image data set;
the functional connection strength module is used for determining the functional connection strength of the sample according to the interrelation between any two vectors with time stamps of the same sample;
The effective connection module is used for constructing an effective connection brain network model based on the long-term memory network, training the effective connection brain network model by taking the activity vector with the time stamp as training data, and taking the weight parameters of the trained effective connection brain network model as the effective connection of the input gate, the effective connection of the forgetting gate and the effective connection of the output gate to be tested;
and the connection fusion strength module is used for fusing the input door effective connection, the forgetting door effective connection, the output door effective connection and the functional connection strength of each sample to obtain the connection fusion strength of the brain network of the sample.
According to the depression feature analysis method based on the LSTM effective connection brain network model, traditional depression diagnosis requires subjective experience of a psychiatrist according to experience, psychological health scales and fMRI images are relied on in the judging process, the fMRI images need to be manually extracted based on priori knowledge of doctors, the representation of the image features is limited greatly, and secondly, the analysis of the fMR I images by the psychiatrist often needs to be concentrated on the image analysis of a certain time point of a certain patient, and the analysis is easily interfered by different individuals. The invention effectively fuses functional connection and effective connection information of the tested fMRI image, generates the feature vector suitable for the machine learning classification model by using the LSTM model, realizes the memory of different brain area states in the information transmission process by the LSTM model through the memory cell C t and the hidden state H t, and reflects the effective connection relation between the brain areas. According to the invention, the tested person is trained based on the LSTM effective connection brain network model to generate the effective connection of the tested person, and the effective connection is fused with the functional connection strength of the tested person to obtain the connection fusion strength of the brain network, so that the effective representation of the brain network related to the depression is obtained, and the depression can be judged by adopting the connection fusion strength. The depression feature analysis method and the depression feature analysis system provided by the invention can effectively mine the features of the tested fMRI image, and apply the trained model to a new tested, so that the interference of images among different patients can be effectively eliminated, the precision of depression classification is improved, the time required by disease diagnosis is greatly reduced, and the diagnosis efficiency is improved.
The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (10)
1. The depression characteristic analysis method based on the LSTM effective connection brain network model is characterized by comprising the following steps of:
Step 1, extracting activity vectors with time stamps of each interested region of each sample in a set according to a standard brain image data set;
Step 2, determining the functional connection strength of the sample according to the interrelation between any two vectors with time stamps of the same sample;
step 3, constructing an effective connection brain network model based on a long-period memory network, taking an activity vector with a time stamp as training data, training the effective connection brain network model, and taking weight parameters of the trained effective connection brain network model as input gate effective connection, forgetting gate effective connection and output gate effective connection of a sample;
And 4, fusing the input door effective connection, the forget door effective connection, the output door effective connection and the functional connection strength of each sample to obtain the connection fusion strength of the sample brain network.
2. The method for analyzing depression characteristics based on LSTM operatively connected brain network model according to claim 1, wherein the method for standard brain image dataset of step 1 is as follows:
and acquiring an original brain function image data set and preprocessing to obtain a standard brain function image data set.
3. The depression feature analysis method based on the LSTM active connection brain network model according to claim 2, wherein the preprocessing method of the original brain function image dataset is as follows:
Converting the original brain function image data into Nifti data, then removing the first n images in Nifti data, removing various noises from the rest images, and enhancing the characteristic information to obtain standard brain image data.
4. The depression feature analysis method based on the LSTM active connection brain network model according to claim 1, wherein the extraction method of the activity vector with the timestamp in the step 1 is as follows:
Selecting the interested regions of each sample, extracting the time sequence of the corresponding brain region of each interested region by using a brain function template, taking the time sequence as an activity signal of the interested region, determining the average value of the current time point of each interested region according to the activity signal, taking the average value as the activity intensity of the interested region at the current time point, and further obtaining the activity vector with the time stamp of each interested region.
5. The depression characterization method based on the LSTM active connection brain network model according to claim 1, wherein the functional connection strength determining method in step 2 is as follows:
And calculating the average value of the active signals of the two regions of interest by adopting a correlation coefficient method to obtain a correlation coefficient, and converting the correlation coefficient into functional connection strength.
6. The method for analyzing depression characteristics based on an LSTM operatively connected brain network model according to claim 5, wherein the correlation coefficient method is Pearson correlation coefficient, spearman correlation coefficient or Kendall rank correlation coefficient.
7. The depression feature analysis method based on an LSTM active connection brain network model according to claim 1, wherein the training method of the active connection brain network model is as follows:
Taking the activity vector with the time stamp as the input of the effective connection brain network model, according to the current time step input and the last time step hidden state of each door of the effective connection brain network model, and calculating an input door, a forget door and an output door by combining an activation function; calculating candidate memory cells effectively connected with the brain network model by using the tanh function as an activation function;
And iteratively updating weights of the input gate, the forgetting gate, the output gate and the candidate memory cells by adopting an error back propagation and gradient descent algorithm until the maximum training iteration number is reached or the error between the hidden states H t-1 and H t of two adjacent time steps is smaller than a set value, so as to obtain the trained effective connection brain network model.
8. The depression feature analysis method based on the LSTM effective connection brain network model according to claim 1, wherein in step 4, the input gate effective connection, the forgetting gate effective connection, the output gate effective connection and the functional connection strength of the sample are fused to obtain the connection fusion strength of the brain network.
9. The depression feature analysis method based on the LSTM effective connection brain network model according to claim 8, wherein the input gate effective connection, the forgetting gate effective connection, the output gate effective connection and the functional connection strength are respectively unfolded and spliced into one-dimensional vectors to obtain the connection fusion strength of the brain network.
10. A system for performing the LSTM-based method of profiling depression of an operatively connected brain network model as claimed in any one of claims 1-9, comprising:
The motion vector module is used for extracting motion vectors with time stamps of the interesting areas of each sample in the collection according to the standard brain image data set;
the functional connection strength module is used for determining the functional connection strength of the sample according to the interrelation between any two vectors with time stamps of the same sample;
The effective connection module is used for constructing an effective connection brain network model based on the long-term memory network, training the effective connection brain network model by taking the activity vector with the time stamp as training data, and generating input gate effective connection, forgetting gate effective connection and output gate effective connection of a sample by the trained effective connection brain network model;
and the connection fusion strength module is used for fusing the input door effective connection, the forgetting door effective connection, the output door effective connection and the functional connection strength of each sample to obtain the connection fusion strength of the brain network of the sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410128454.0A CN117954080A (en) | 2024-01-30 | 2024-01-30 | Depression characteristic analysis method and system based on LSTM effective connection brain network model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410128454.0A CN117954080A (en) | 2024-01-30 | 2024-01-30 | Depression characteristic analysis method and system based on LSTM effective connection brain network model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117954080A true CN117954080A (en) | 2024-04-30 |
Family
ID=90804703
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410128454.0A Pending CN117954080A (en) | 2024-01-30 | 2024-01-30 | Depression characteristic analysis method and system based on LSTM effective connection brain network model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117954080A (en) |
-
2024
- 2024-01-30 CN CN202410128454.0A patent/CN117954080A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Khare et al. | PDCNNet: An automatic framework for the detection of Parkinson’s disease using EEG signals | |
Nogay et al. | Detection of epileptic seizure using pretrained deep convolutional neural network and transfer learning | |
Mumtaz et al. | Review of challenges associated with the EEG artifact removal methods | |
AU2019201766B2 (en) | Systems and methods for diagnosis of depression and other medical conditions | |
CN109447183B (en) | Prediction model training method, device, equipment and medium | |
US20110028827A1 (en) | Spatiotemporal pattern classification of brain states | |
Hssayeni et al. | Multi-modal physiological data fusion for affect estimation using deep learning | |
Shaban | Automated screening of parkinson's disease using deep learning based electroencephalography | |
Tawhid et al. | Automatic and efficient framework for identifying multiple neurological disorders from EEG signals | |
Soni et al. | Graphical representation learning-based approach for automatic classification of electroencephalogram signals in depression | |
Malviya et al. | A novel technique for stress detection from EEG signal using hybrid deep learning model | |
Rahman et al. | Extended ICA and M-CSP with BiLSTM towards improved classification of EEG signals | |
CN112426161A (en) | Time-varying electroencephalogram feature extraction method based on domain self-adaptation | |
CN115691794A (en) | Auxiliary analysis method and system for neural diagnosis | |
Zhang et al. | DWT-Net: Seizure detection system with structured EEG montage and multiple feature extractor in convolution neural network | |
Singh et al. | Detection of Parkinson's Disease using Machine Learning Algorithm | |
Yang et al. | A gated temporal-separable attention network for EEG-based depression recognition | |
CN116662736A (en) | Human body state assessment method based on deep learning hybrid model | |
CN117954080A (en) | Depression characteristic analysis method and system based on LSTM effective connection brain network model | |
Zhu | Early diagnosis of Parkinson's Disease by analyzing magnetic resonance imaging brain scans and patient characteristic | |
Khan et al. | Deep convolutional neural network based sleep apnea detection scheme using spectro-temporal subframes of EEG signal | |
Ranjan et al. | A machine learning framework for automatic diagnosis of schizophrenia using eeg signals | |
Li et al. | Cross-subject EEG linear domain adaption based on batch normalization and depthwise convolutional neural network | |
Harpale et al. | Brain Seizure Detection and Classification Using EEG Signals | |
Samiei et al. | A complex network approach to time series analysis with application in diagnosis of neuromuscular disorders |
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 |