Parkinson's disease deep brain electrical stimulation postoperative curative effect prediction system based on brain network similarity

Abstract

The invention discloses a Parkinson's disease DBS postoperative curative effect prediction system based on group horizontal structure brain network similarity, and belongs to the technical field of biomedical image mode recognition. According to the system, on the basis of brain structure magnetic resonance image data of healthy subjects and Parkinson's disease patients before operation, grey matter volumes in 16 networks defined by AAL are used as feature vectors, Spearman rank correlation among the vectors is calculated, and similar matrixes among the subjects are constructed in two groups respectively. Secondly, after the matrix is averaged, the matrix is input into a binary logistic regression classification framework as features, and a classification model related to the DBS operation curative effect is established; according to the system, heterogeneity between subjects is quantified by constructing a group level structure brain network similarity matrix, classification of Parkinson's disease patients and normal people and intelligent prediction of DBS operation curative effects are achieved by using the heterogeneity as characteristics, and more comprehensive brain image information reference can be provided for preoperative accurate evaluation.

Description

Technical Field

[0001] This method belongs to the technical field of biomedical image pattern recognition, and specifically involves the processing of structural magnetic resonance brain image data, the construction of brain network similarity matrices, and the prediction method for the efficacy of deep brain stimulation for Parkinson's disease. Background Art

[0002] Parkinson's Disease (PD) is a progressive neurodegenerative disease mainly characterized by motor symptoms. The motor symptom triad (rest tremor, bradykinesia, and muscle rigidity) and axial symptoms (postural reflex disorder) constitute the diagnostic cornerstone. In the clinical management of Parkinson's disease, the initial treatment mainly uses dopaminergic neurotransmitter replacement therapy. Among them, levodopa, as a prodrug, can cross the blood-brain barrier and be converted into functional dopamine, and most patients show significant improvement in motor symptoms in the initial stage of treatment. However, when the disease progresses to the middle and late stages (Hoehn-Yahr stage III-IV), some patients develop motor complications after dopaminergic treatment. At this stage, deep brain stimulation (DBS), as a representative means of neuromodulation therapy, can reshape the functional connections of the basal ganglia-thalamus-cortex circuit through high-frequency electrical stimulation targeting the subthalamic nucleus (STN) or the internal segment of the globus pallidus (GPi). However, the actual postoperative improvement of DBS varies among individuals.

[0003] There are significant differences in the progression rate of motor symptoms and treatment response in PD, which reveals the heterogeneity within the PD group. However, the "heterogeneity" reported in previous studies (which can be attributed to heterogeneity in methodological factors rather than the true attributes of the population) may be caused by changes in the overall distribution of one group relative to another group, or by differential sampling of distributions with similar variances. Therefore, in addition to evaluating the average difference, more attention should be paid to the differences between individuals, which helps to improve the accuracy of predicting the efficacy of DBS for Parkinson's disease. Although there have been improvements in the preoperative evaluation of DBS in clinical practice, there is an urgent need for innovation in new technologies and new methods to promote the development of precise diagnosis and treatment of PD. Summary of the Invention

[0004] The present invention first develops a method for characterizing the morphological network similarity at the group level of PD based on structural magnetic resonance imaging, and uses this as a feature to establish a prediction system for predicting the motor improvement after DBS based on the preoperative brain structure information of patients.

[0005] The technical solution of the present invention is as follows: A system for predicting the efficacy of deep brain stimulation (DBS) for Parkinson's disease based on brain network similarity, which system comprises 4 modules: a magnetic resonance (MR) structural image acquisition and surgical prognosis evaluation module, a structural image data preprocessing module, a structural brain network similarity matrix construction module, and a surgical efficacy classification prediction module;

[0006] The said MR structural image acquisition and surgical prognosis module: This module is for the pre-operative MR structural images of Parkinson's disease patients who have received the same type of DBS surgery and the clinical scale data before and after the surgery collected;

[0007] In the said preprocessing module, the following method is used to process the sample data:

[0008] Step A1: Manually adjust the anterior commissure - posterior commissure to make them on the same horizontal line;

[0009] Step A2: Use the Cat12 software package to remove the information of the skull from the images, and the images of all subjects are segmented into three major categories: gray matter, white matter, and cerebrospinal fluid;

[0010] Step A3: Through affine transformation and non - linear transformation, register the gray matter image and the white matter image into the standard Montreal Neurological Institute (MNI) brain space;

[0011] Step A4: Perform non - linear modulation on the segmented gray matter image to correct the individual differences in brain size;

[0012] Step A5: Based on the image quality report generated after segmentation by the Cat12 software, calculate the quality index interquartile range (IQR), display the image slices of all subjects, and calculate the uniformity of the generated gray matter (GM) images between subjects. The larger the value, the more complete the data. The GM image refers to the gray matter map;

[0013] Step A6: Perform Gaussian smoothing on the segmented GM image to eliminate noise and artifacts;

[0014] The said structural brain network similarity matrix construction module:

[0015] Step B1: For each subject, extract the average gray matter volume in the prior template to generate a feature vector;

[0016] Step B2: For each subject, calculate the rank - order correlation coefficient between the feature vector of this subject and the feature vectors of other subjects in the same group; Obtain the similarity matrix between subjects:

[0017]

[0018] where rg i = rank_order(v i ), v iThe gray matter vectors of 16 networks for subject i, Cov represents the covariance of vector rg, sd represents the standard deviation; rank_order represents the rank of v i of;

[0019] Step B3: For each subject, calculate the average of the correlation coefficients, that is, the PBSI score, as a measure of the similarity between this subject and other subjects in the same group; the higher the PBSI score, the greater the similarity between the morphological networks of this subject and other subjects in the same sample; the low level of morphological similarity within the patient group supports the heterogeneity of the disease in terms of neuroanatomical profiles, and the calculation method is as follows:

[0020]

[0021] where N is the number of subjects in this group;

[0022] Surgical efficacy classification prediction module:

[0023] Step C1: Adopt five-fold cross-validation to divide the dataset into a training set and a test set;

[0024] Step C2: Use a logistic regression model, put the test set into the trained model, and classify the normal control group and the patient group; the input of the logistic regression model is the PBSI score, and the output is the probability of classification;

[0025] Step C3: Classify and predict the efficacy of Parkinson's disease after deep brain stimulation surgery.

[0026] Furthermore, divide the dataset into 5 folds, where 4 folds are the training set and 1 fold is the test set.

[0027] Furthermore, use the Z-test statistic in the structural brain network similarity matrix module to compare the differences in correlation coefficients between groups;

[0028]

[0029] where z1 and z2 are the corresponding correlation coefficients of the two groups of subjects of the Fisher-z transformation values, and n1 and n2 are the numbers of subjects in the two groups.

[0030] Furthermore, the 16 networks in step B2 include: caudate nucleus, putamen, globus pallidus, thalamus, hippocampus, amygdala, parahippocampal gyrus, cingulate cortex, frontal lobe, parietal lobe, occipital lobe, temporal lobe, insula, sensorimotor cortex, Broca's area, cerebellar network.

[0031] Based on the pre-operative brain structural magnetic resonance imaging data of healthy subjects and Parkinson's disease patients, the gray matter volume in 16 networks defined by AAL is used as the feature vector, and the Spearman rank correlation between vectors is calculated to construct the similarity matrix between subjects in both groups. Secondly, after averaging this matrix, it is used as a feature to input into the binary logistic regression classification framework to establish a classification model related to the efficacy of DBS surgery. This system quantifies the heterogeneity between subjects by constructing a group-level structural brain network similarity matrix, and uses it as a feature to achieve the classification of Parkinson's disease patients and normal people, as well as the intelligent prediction of the efficacy of DBS surgery, which can provide more comprehensive brain imaging information reference for pre-operative precise evaluation. Description of the Drawings

[0032] Figure 1 It is the overall flowchart of the present invention.

[0033] Figure 2 It is the ROC curve graph of the PBSI score difference and logistic regression classification prediction between the normal control group and the PD group.

[0034] Figure 3 It is the ROC curve graph of the PBSI score difference and logistic regression classification prediction between the group with good DBS efficacy and the group with poor DBS efficacy in PD. Detailed Implementation Modes

[0035] The implementation process of the present invention is described in detail below with reference to the drawings and examples. First, pre-operative magnetic resonance structural images of 48 normal controls and 76 Parkinson's disease patients who were gender- and age-matched and received STN-DBS surgery were clinically collected, followed by a one-month postoperative follow-up, and the improvement ratio of the UPDRS-III score was calculated as the clinical DBS efficacy quantification score. The patients were divided into a group with good efficacy (DBS improvement rate greater than or equal to 60%) and a group with poor efficacy (DBS improvement rate less than 60%) according to the DBS efficacy.

[0036] Then the following treatments are carried out:

[0037] A. Data preprocessing:

[0038] Step A1: Manually correct the origin in the SPM software, adjust the anterior commissure and posterior commissure so that they are on the same horizontal line and save;

[0039] Step A2: Use Cat12 to remove the skull, and then use the prior tissue probability map and the naive Bayes algorithm to identify different tissue types, and segment the image into gray matter image and white matter image;

[0040] Step A3: Register the gray matter image to the standard MNI space, and resample the image using the default template to obtain a gray matter image with a resolution of 1.5mm×1.5mm×1.5mm;

[0041] Step A4: Select gray matter images with an interquartile range (IQR) of the quality index less than 80 and low homogeneity (below the mean - 2*sd) for subsequent analysis;

[0042] Step A4: Perform isotropic Gaussian smoothing on the gray matter images with a smoothing kernel size of 8 mm, and finally obtain the gray matter images of smwp*.nii;

[0043] B. Construct a structural brain network similarity matrix:

[0044] Step B1: Based on the automated anatomical labeling (AAL) template, divide the gray matter images of each subject into 16 classic morphological brain networks, including the caudate nucleus, putamen, globus pallidus, thalamus, hippocampus, amygdala, parahippocampal gyrus, cingulate cortex, frontal lobe, parietal lobe, occipital lobe, temporal lobe, insula, sensorimotor cortex, Broca's area, and cerebellar network, and extract the average gray matter volume of each network;

[0045] Step B2: For each subject, use the gray matter volumes of the 16 networks as vectors to calculate the Spearman rank correlation coefficient between the vectors of this subject and other subjects in the same group, and obtain an N×N similarity matrix for each group of subjects;

[0046] Step B3: Calculate the PBSI score for each subject.

[0047] Step B4: Use the Z-test statistic to compare the differences in the correlation coefficients between groups.

[0048] C. Surgical efficacy classification prediction module:

[0049] Step C1: Considering the sample size, use five-fold cross-validation to equally divide the dataset into 5 folds, with 4 folds as the training set and 1 fold as the validation set;

[0050] Step C2: Use the PBSI score as the input to train the model in a binary logistic regression classification framework. Put the test set into the trained model to classify the normal control group and the patient group. Train 1000 times, with an average accuracy of 93.9% and an area under the curve (AUC) of 0.97;

[0051] Step C3: Recalculate the PBSI score for the group with good DBS efficacy and the group with poor DBS efficacy respectively, and use this as a feature to predict the efficacy of Parkinson's disease after DBS surgery. Train 1000 times, with an average accuracy of 74.4% and an AUC of 0.83.

Claims

1. A Parkinson's disease deep brain stimulation postoperative efficacy prediction system based on brain network similarity, which system comprises 4 modules: a magnetic resonance structural image acquisition and surgical prognosis evaluation module, a structural image data preprocessing module, a structural brain network similarity matrix construction module, and a surgical efficacy classification prediction module; The magnetic resonance structural image acquisition and surgical prognosis module: This module is for the preoperative magnetic resonance structural images of Parkinson's disease patients who have received the same type of DBS surgery and the clinical scale data before and after the surgery; In the preprocessing module, the following method is used to process the sample data: Step A1: Manually adjust the anterior commissure - posterior commissure to make them on the same horizontal line; Step A2: Use the Cat12 software package to remove the information of the skull from the images, and the images of all subjects are segmented into three major categories: gray matter, white matter, and cerebrospinal fluid; Step A3: Through affine transformation and non - linear transformation, register the gray matter image and the white matter image into the standard MNI brain space; Step A4: Perform non - linear modulation on the segmented gray matter image to correct the individual differences in brain size; Step A5: Calculate the quality index IQR based on the image quality report generated after segmentation by the Cat12 software, display the image slices of all subjects, and calculate the uniformity of the generated GM images among the subjects. The larger the value, the more complete the data. The GM image represents the gray matter map; Step A6: Perform Gaussian smoothing on the segmented GM image to eliminate noise and artifacts; The structural brain network similarity matrix construction module: Step B1: For each subject, extract the average gray matter volume in the prior template to generate a feature vector; Step B2: For each subject, calculate the rank correlation coefficient between the eigenvector of this subject and the eigenvectors of other subjects in the same group to which this subject belongs. Obtain the similarity matrix among the subjects: where rg i = rank_order(v i ), v i is the gray matter vector of 16 networks of subject i, Cov represents the covariance of vector rg, sd represents the standard deviation; rank_order represents the rank of v i ; Step B3: For each subject, calculate the average value of the correlation coefficients, that is, the PBSI score, as a measure of the similarity between this subject and other subjects in the same group; The higher the PBSI score, the greater the similarity between the morphological network of this subject and other subjects in the same sample; A low level of morphological similarity within the patient group supports the heterogeneity of the disease in terms of neuroanatomical profiles. The calculation method is as follows: Where N is the number of subjects in this group; The surgical efficacy classification prediction module: Step C1: Adopt five - fold cross - validation to divide the data set into a training set and a test set; Step C2: Use a logistic regression model, put the test set into the trained model to classify the normal control group and the patient group; The input of the logistic regression model is the PBSI score, and the output is the classification probability; Step C3: Classify and predict the efficacy of Parkinson's disease after deep brain stimulation surgery.

2. The efficacy prediction system for deep brain stimulation after Parkinson's disease surgery based on brain network similarity according to claim 1, wherein The data set is divided into 5 folds, where 4 folds are the training set and 1 fold is the test set.

3. The efficacy prediction system for deep brain stimulation after Parkinson's disease surgery based on brain network similarity according to claim 1, characterized in that, In the structural brain network similarity matrix module, the Z - test statistic is used to compare the differences in the correlation coefficients between groups; where z1 and z2 are the correlation coefficients corresponding to two groups of subjects , and n1 and n2 are the numbers of subjects in the two groups.

4. A system for predicting the efficacy after deep brain stimulation for Parkinson's disease based on brain network similarity according to claim 1, wherein, The 16 networks in Step B2 include: caudate nucleus, putamen, globus pallidus, thalamus, hippocampus, amygdala, parahippocampal gyrus, cingulate cortex, frontal lobe, parietal lobe, occipital lobe, temporal lobe, insula, sensorimotor cortex, Broca's area, cerebellar network.

Images