阅读说明：本技术 一种基于聚类和分类算法的ECoG术中脑功能定位方法 (ECoG intraoperative brain function positioning method based on clustering and classification algorithm ) 是由 姜涛 许永川 于 2020-12-14 设计创作，主要内容包括：本发明公开了一种基于聚类和分类算法的ECoG术中脑功能定位方法,包括以下步骤：首先采用功能区边界聚类算法,提取六层小波分解重构单子频带的能量占比为特征量,并使用凝聚式层次聚类模型,聚类各导联的类簇；其次采用功能区属性分类算法,以6维时域统计量和5维节律能量为特征量,并使用PCA对特征降维,并使用SVM分类导联的功能区属性；最后,集成上述算法建立ECoG术中脑功能定位集成算法,采用自定义的相似度算法,将聚类的类簇匹配功能区属性,完成功能区定位。本发明的算法模型具有良好的泛化性,可实现准确快速的术中脑功能定位临床应用,可广泛应用于基于脑电分析的术中脑功能定位等神经科学的研究。(The invention discloses an ECoG intraoperative brain function positioning method based on clustering and classification algorithms, which comprises the following steps of: firstly, extracting the energy ratio of six layers of wavelet decomposition reconstruction single sub-bands as characteristic quantity by adopting a functional region boundary clustering algorithm, and clustering the clusters of each lead by using an agglomeration type hierarchical clustering model; secondly, a functional area attribute classification algorithm is adopted, 6-dimensional time domain statistics and 5-dimensional rhythm energy are taken as feature quantities, PCA is used for reducing the dimension of the features, and SVM is used for classifying the functional area attributes of leads; and finally, integrating the above algorithms to establish an ECoG intraoperative brain function positioning integration algorithm, and matching the clustered clusters with the functional area attributes by adopting a self-defined similarity algorithm to complete functional area positioning. The algorithm model has good generalization, can realize accurate and rapid clinical application of intraoperative brain function localization, and can be widely applied to neuroscience researches such as intraoperative brain function localization based on electroencephalogram analysis.)

1. An ECoG intraoperative brain function positioning method based on clustering and classification algorithms is characterized by comprising the following steps:

conducting lead category clustering: carrying out clustering pretreatment on ECoG data of each lead of a patient, and clustering functional areas and non-functional areas to obtain two clustered clusters;

conducting lead attribute classification: carrying out classification pretreatment on ECoG data of each lead of a patient, extracting time domain statistics and energy of common rhythms in electroencephalogram signals as features for each sample, using PCA (principal component analysis) to reduce dimensions, and using a pre-trained SVM (support vector machine) classification model to calculate lead attributes to obtain two categories of a functional area and a non-functional area;

matching the clustered clusters with the attributes of the functional areas by adopting an ECoG intraoperative brain function positioning integration method: obtaining two clusters of the boundary cluster and two attribute categories of the attribute classification, respectively calculating the similarity of the two clusters and the two attribute categories of the two categories, regarding the attribute of the category with higher similarity as the attribute of the cluster, and taking the result of the cluster with the attribute as the final brain function positioning result.

2. The ECoG intraoperative brain function localization method based on clustering and classification algorithm according to claim 1, wherein the conducting lead category clustering adopts a functional area boundary clustering algorithm, specifically:

s11, sampling and preprocessing ECoG data of each lead of a patient;

s12, performing 6-layer wavelet decomposition on the preprocessed ECoG data of each lead by using a db3 wavelet, and extracting the energy ratio of each single sub-band as a feature;

s13, performing two-class clustering on all lead data of the patient by using a hierarchical clustering method to obtain two cluster classes.

3. The method for locating brain function in an ECoG operation based on a clustering and classifying algorithm as claimed in claim 2, wherein in the step S11, the ECoG data sampling specifically comprises: implanting a platinum electrode array having a diameter of 4mm and an inter-electrode distance of 10mm under the dura of the patient; the sampling frequency of the ECoG data is 500 Hz; after craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; awakening the patient after anesthesia, and acquiring an ECoG signal acquired by the intraoperative neuroelectrophysiology detector in the resting state of the patient to obtain the resting ECoG of the patient;

the ECoG data preprocessing specifically comprises the following steps: and carrying out 1Hz high-pass filtering treatment, carrying out 49-51Hz notch filtering treatment, and finally carrying out ICA filtering treatment to obtain preprocessed ECoG data of each lead.

4. The method for locating brain function in an ECoG operation based on a clustering and classifying algorithm according to claim 2, wherein the step S12 is specifically as follows: selecting db3 wavelet base, performing 6-layer wavelet decomposition on the EEG signal, extracting detail coefficients and approximation coefficients of each layer, reconstructing single subband signals corresponding to the coefficients of each layer, and calculating the energy ratio of each layer signal as a boundary clustering feature.

5. The method for locating brain function in an ECoG operation based on a clustering and classifying algorithm according to claim 2, wherein the step S13 is specifically as follows: the used clustering method is a hierarchical clustering algorithm, and the optimal parameters of the clustering algorithm found by grid search are as follows: definition and linkage are "euclidean" and "ward".

6. The method for locating brain function in ECoG operation based on clustering and classifying algorithm as claimed in claim 1, wherein said classifying lead category is specifically:

s21, sampling and preprocessing ECoG data of each lead of a patient;

s22, dividing each lead data into a sample every 2 seconds;

s23, extracting time domain statistics and rhythm energy as features for each sample, and using PCA to reduce dimension, wherein the dimension-reduced features retain variance information of 85% of original features;

s24, judging the attributes of the leads by using the pre-trained SVM classification model, and when the proportion of the samples of a certain lead to the functional region is more than or equal to 0.5, considering the lead to belong to the functional region, otherwise, belonging to the non-functional region, and obtaining two classification categories.

7. The method for locating brain function in ECoG operation based on clustering and classifying algorithm according to claim 6, wherein the step S21 is specifically: the ECoG data sampling specifically comprises the following steps: implanting a platinum electrode array having a diameter of 4mm and an inter-electrode distance of 10mm under the dura of the patient; the sampling frequency of the ECoG data is 500 Hz; after craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; awakening the patient after anesthesia, and acquiring an ECoG signal acquired by the intraoperative neuroelectrophysiology detector in the resting state of the patient to obtain the resting ECoG of the patient;

the ECoG data preprocessing specifically comprises the following steps: firstly, carrying out 1Hz high-pass filtering processing, secondly, carrying out 49-51Hz notch filtering processing, and finally, carrying out ICA filtering processing.

8. The ECoG intraoperative brain function localization method based on clustering and classification algorithm as claimed in claim 6, wherein the SVM classification model is pre-trained in the step S24, specifically: using ECoG data collected in advance as data of a training model;

dividing the data into training sets and testing sets according to patients;

preprocessing and dividing samples of training set data and testing set data;

extracting time domain statistics and rhythm energy characteristics from all samples, and using PCA to reduce dimension, wherein the dimension-reduced characteristics retain variance information of 85% of original characteristics;

and finally, optimizing the key parameters of the SVM by using grid search, finding the optimal parameters of the SVM, and finishing the training of the classification model.

9. The method of claim 8, wherein the time domain statistics comprise the sum, mean, variance, standard deviation, maximum, minimum of sample points; the rhythmic energy characteristics include α, β, γ, θ, δ; the optimal parameters are as follows: the optimal parameters are C ═ 1, gamma ═ auto ', and Kernel ═ rbf'.

Technical Field

The invention relates to the field of neuroscience research and clinical application research of electroencephalogram analysis, in particular to an ECoG intraoperative brain function positioning method based on clustering and classification algorithms.

Background

The brain is the central nervous system of human beings, and brain pathological changes will influence a plurality of functions of human beings such as memory, language, motion, and seriously influence the survival and the quality of life of human beings, wherein, epilepsy and brain glioma are one of the most common brain pathological changes. Currently, the main treatment mode of such brain diseases is treatment through neurosurgery, and the most important step in neurosurgery is brain function localization during operation. The traditional intraoperative brain function positioning method (anatomy, image, cortical electrical stimulation and the like) has the problems of speed, accuracy, possible wound and the like, and the intraoperative brain function positioning method using cortical electroencephalogram (ECoG) is expected to realize quick, accurate and noninvasive positioning, but has the problem of poor generalization performance of samples and positioning models, namely one positioning model is not suitable for different crowds and is only suitable for specific crowds. An important embodiment of poor generalization performance of the positioning model is that the positioning model established based on the self sample has higher self-detection accuracy, which indicates that the individual specificity exists between the individual functional region and the non-functional region. However, the low accuracy of the detection of the foreign body indicates that the brain functional regions of the population have no common specificity or are difficult to find out common specificity. Therefore, in order to find a fast and accurate brain function area positioning method with good generalization performance, the invention provides an ECoG intraoperative brain function positioning method based on clustering and classification algorithms.

Disclosure of Invention

The invention mainly aims to overcome the defects in the prior art and provide an ECoG intraoperative brain function positioning method based on clustering and classification algorithms.

The purpose of the invention is realized by the following technical scheme:

an ECoG intraoperative brain function positioning method based on clustering and classification algorithms is characterized by comprising the following steps:

firstly, a functional area boundary clustering algorithm is adopted for conducting lead category clustering: preprocessing each lead ECoG data of a patient, performing 6-layer wavelet decomposition on the data by using a db3 wavelet, extracting the energy ratio of each single subband as a characteristic, and performing functional area and non-functional area clustering on the data of all leads of the patient to obtain two clustered clusters;

secondly, classifying the lead categories by adopting a functional region attribute classification algorithm: preprocessing each lead ECoG data of a patient, regarding the data of every two seconds as a sample, extracting time domain statistics (sum, mean, variance, standard deviation, maximum value and minimum value) and energy of common rhythm (alpha, beta, gamma, theta and delta) in an electroencephalogram signal as features for each sample, and using PCA to reduce dimension, wherein the feature after dimension reduction retains variance information of 85% of original features; calculating the lead attribute by using a pre-trained SVM classification model, and when the ratio of the sample of a certain lead to the functional region is more than or equal to 0.5, considering the lead to belong to the functional region, otherwise, belonging to the non-functional region, and obtaining two categories of the functional region and the non-functional region;

and finally, matching the clustered clusters with the attributes of the functional areas by adopting an ECoG intraoperative brain function positioning integration algorithm: after obtaining two classes of the boundary cluster and two classes of the attribute classification, respectively calculating the similarity of the two classes of the boundary cluster and the two classes of the attribute classification, wherein the attribute of the class with higher similarity is regarded as the attribute of the class cluster, and the result of the cluster with the attribute is regarded as the final brain function positioning result;

further, the method for positioning brain function in an ECoG operation based on clustering and classification algorithm is characterized in that the functional area boundary clustering algorithm process specifically comprises:

s11, preprocessing each lead ECoG data of the patient;

s12, performing 6-layer wavelet decomposition on the preprocessed ECoG data of each lead by using a db3 wavelet, and extracting the energy ratio of each single sub-band as a feature;

s13, performing two-class clustering on all lead data of the patient by using a hierarchical clustering method to obtain two cluster classes.

Further, the method for positioning brain function in an ECoG operation based on clustering and classification algorithms is characterized in that the functional area attribute classification algorithm process specifically comprises:

s21, preprocessing each lead data of the patient;

s22, dividing each lead data into a sample every 2 seconds;

s23, extracting time domain statistics and rhythm energy as features for each sample, and using PCA

Reducing the dimension;

s24, judging the attributes of the leads by using the pre-trained SVM classification model (when the proportion of the samples of a certain lead to the functional region is more than or equal to 0.5, the lead is considered to belong to the functional region, otherwise, the lead belongs to the non-functional region), and obtaining two classification categories.

Further, in step S11, the ECoG data sampling specifically includes: implanting a platinum electrode array (Ad-Tech, Racine, Wis., USA) having a diameter of 4mm and an inter-electrode distance of 10mm under the dura of the patient; the sampling frequency of the ECoG data is 500 Hz. After craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; after anesthesia, the patient is awakened, and an ECoG signal acquired by an intraoperative neuro-electrophysiological detector (Endevor Bravo, Nicolet, Inc., USA) in the resting state of the patient is acquired to obtain the ECoG in the resting state of the patient. The ECoG data preprocessing specifically comprises the following steps: firstly, carrying out 1Hz high-pass filtering processing, secondly, carrying out 49-51Hz notch filtering processing, and finally, carrying out ICA filtering processing.

Further, the step S12 is specifically: selecting db3 wavelet base, performing 6-layer wavelet decomposition on the EEG signal, extracting detail coefficients and approximation coefficients of each layer, reconstructing single subband signals corresponding to the coefficients of each layer, and calculating the energy ratio of each layer signal as an original characteristic.

Further, the step S13 is specifically: the clustering method used is hierarchical clustering, and the optimal parameters of the clusters found by grid search are as follows: definition and linkage are "euclidean" and "ward".

Further, in step S21, the ECoG data sampling specifically includes: implanting a platinum electrode array (Ad-Tech, Racine, Wis., USA) having a diameter of 4mm and an inter-electrode distance of 10mm under the dura of the patient; the sampling frequency of the ECoG data is 500 Hz. After craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; after anesthesia, the patient is awakened, and an ECoG signal acquired by an intraoperative neuro-electrophysiological detector (Endevor Bravo, Nicolet, Inc., USA) in the resting state of the patient is acquired to obtain the ECoG in the resting state of the patient. The ECoG data preprocessing specifically comprises the following steps: firstly, carrying out 1Hz high-pass filtering processing, secondly, carrying out 49-51Hz notch filtering processing, and finally, carrying out ICA filtering processing.

Further, the pre-training of the SVM classification model in step S24 specifically includes: the ECoG data collected in advance was used as data for training the model. First, the data is divided into training sets and test sets by patient. Second, the training set and test set data are preprocessed and the samples are partitioned. Again, time domain statistics (sum, mean, variance, standard deviation, maximum, minimum) and rhythm energy (α, β, γ, θ, δ) features were extracted for all samples, and the PCA was used to reduce the dimensions, which retained 85% of the variance information of the original features. And finally, optimizing the key parameters of the SVM by using grid search, so that the SVM parameter with the highest accuracy of the test set is the optimal parameter, finding the optimal parameter of the SVM, (C is 1, gamma is auto ', Kernel is rbf'), and finishing the training of the classification model.

The working process of the invention is as follows:

firstly, a functional area boundary clustering algorithm is adopted for conducting lead category clustering: firstly, intercepting resting state ECoG data of a patient and preprocessing the data, wherein the preprocessing process comprises the following steps: the method mainly comprises the following steps of 1HZ high-pass filtering, 49-51Hz notch filtering, ICA filtering and pretreatment flow, wherein the pretreatment flow is mainly used for removing power frequency interference existing on 50Hz and frequency multiplication thereof. Secondly, extracting the clustering characteristics, wherein the characteristics used in the clustering in the research are energy ratios of seven sub-bands obtained after six-layer wavelet decomposition, decomposing the original signal into the seven sub-bands by using a db3 wavelet basis, and calculating the energy ratio of each sub-band as the characteristics of the clustering. And finally, using a hierarchical clustering method and using the optimal clustering parameters (affinity: "iterative" and linkage: "ward") found by grid search to perform two types of clustering of a functional area and a non-functional area on all lead data of the patient, and completing a functional area boundary clustering process to obtain two clustering type clusters.

Secondly, classifying the lead categories by adopting a functional region attribute classification algorithm: firstly, intercepting resting state ECoG data of a patient, performing a preprocessing flow which is the same as a clustering algorithm, and dividing each lead data into a sample every 2 seconds. Second, the time domain statistics and the rhythm energy are extracted as features for each sample and the dimensionality reduction is performed using PCA. And finally, judging whether each sample is a functional region or a non-functional region by using a pre-trained SVM classification model, adopting a threshold value method when determining the lead attribute, and when the ratio of the number of the samples of a certain lead which are distributed to the functional region is more than or equal to 0.5, considering that the lead belongs to the functional region, otherwise, judging that the lead belongs to the non-functional region. Through the functional region classification algorithm process, all leads of a patient are divided into two classification categories of a functional region and a non-functional region.

And finally, matching the clustered clusters with the attributes of the functional areas by adopting an ECoG intraoperative brain function positioning integration algorithm: after the functional area boundary clustering result and the functional area attribute classification result of the patient are obtained, the similarity between the two clustered clusters and the two classified classes is respectively calculated, the attribute of the class with higher similarity is regarded as the attribute of the clusters, and the clustered cluster result with the attribute is regarded as the final brain function positioning result.

The ECoG intraoperative brain function positioning method based on the clustering and classifying algorithm is quick and effective in positioning, has good generalization capability, can realize accurate and quick clinical application of intraoperative brain function positioning, and can be widely applied to neuroscience researches such as intraoperative brain function positioning based on electroencephalogram analysis.

Compared with the prior art, the invention has the following advantages and beneficial effects:

1. the invention provides an ECoG intraoperative brain function positioning method based on clustering and classification algorithms, which comprises the steps of firstly, conducting lead category clustering by adopting a functional area boundary clustering algorithm to obtain a functional area boundary clustering result; secondly, conducting lead classification by adopting a functional region attribute classification algorithm to obtain a functional region attribute classification result; and finally, matching the clustered clusters with the attributes of the functional areas by adopting an ECoG intraoperative brain function positioning integration algorithm to obtain two clusters of boundary clustering and two categories of attribute classification, respectively calculating the similarity of the two clusters and the two categories of attribute classification, regarding the attribute of the category with higher similarity as the attribute of the cluster, regarding the result of the clustered clusters with the attributes as a final brain function positioning result, and completing intraoperative brain function area positioning. The method has the advantages of quick and accurate positioning and great clinical application significance.

2. The ECoG intraoperative brain function positioning method based on the clustering and classifying algorithm uses a positioning mode combining the clustering and classifying algorithm, integrates the advantages of more accurate boundary positioning capability of the clustering algorithm and attribute identification of the classifying algorithm, determines the boundary of a functional area by using the clustering algorithm, and assists in determining the attribute of a cluster by using the classifying algorithm.

3. The ECoG intraoperative brain function positioning method based on the clustering and classifying algorithm, disclosed by the invention, has the advantages that the clustering and classifying algorithm is combined to identify the specific electroencephalograms of the functional area and the non-functional area, the identification effect is better, the overall identification accuracy of the algorithm is 90.9%, the specificity is 88.9% and the sensitivity is 100% when the algorithm is measured in a laboratory. Therefore, the method has potential clinical application prospect in neurosurgery.

Drawings

Fig. 1 is an algorithm flow chart of a method for locating a brain functional region in an ECoG operation based on a clustering and classifying algorithm according to the present invention.

Fig. 2 is a flow chart of a functional area boundary clustering algorithm in the ECoG intraoperative brain functional area positioning method based on a clustering and classification algorithm according to the present invention.

Fig. 3 shows PSD during data preprocessing of patient a in the method for locating brain functional areas in ECoG surgery based on clustering and classification algorithms according to the present invention.

Fig. 4 is a flow chart of a functional region attribute classification algorithm in the ECoG intraoperative brain functional region localization method based on clustering and classification algorithms according to the present invention.

Fig. 5 is a flow chart of an ECoG intraoperative brain function localization integration algorithm in the ECoG intraoperative brain function region localization method based on clustering and classification algorithms according to the present invention.

Detailed Description

The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.

Examples

The method can scientifically detect specific electroencephalogram signals of a functional area and a non-functional area, and position the functional area and the non-functional area of cerebral cortex in real time in an operation to help a doctor to perform a neurosurgical operation. The invention is further described in the following with reference to the accompanying drawings and examples. As shown in fig. 1, the specific processing is as follows:

1. firstly, a functional area boundary clustering algorithm is adopted to perform lead category clustering, and the algorithm flow is shown in fig. 2 and comprises the following steps:

(1) signal acquisition:

after craniotomy, a platinum electrode array (Ad-Tech, Racine, Wis., USA) with a diameter of 4mm and an inter-electrode distance of 10mm was implanted under the dura mater of the patient; the sampling frequency of the ECoG data is 500 Hz. After craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; after anesthesia, the patient is awakened, and an ECoG signal acquired by an intraoperative neuro-electrophysiological detector (Endevor Bravo, Nicolet, Inc., USA) in the resting state of the patient is acquired to obtain the ECoG in the resting state of the patient. . The data samples collected for patient a are shown in table 1 below (all subsequent processing and calculations are for patient a).

Table 1 ECoG data samples for patient a

(2) Data preprocessing:

intercepting the data in the resting state, sequentially performing processing steps of 1Hz high-pass filtering, 49-51Hz notch filtering, ICA filtering and the like, and removing power frequency interference existing on 50Hz and frequency multiplication thereof. The change of the signal PSD during the preprocessing of each lead signal of the patient A is shown in figure 3.

(3) Feature extraction:

the original signal is decomposed into seven sub-bands by using a db3 wavelet base, the seven sub-bands are respectively represented by d1, d2, d3, d4, d5, d6 and a6, the energy ratio of each sub-band is calculated as a characteristic, and the calculation formula of the energy ratio is as follows:

in the formula E_jThe energy of the signal is reconstructed for the jth sub-band. J is the six-layer decomposition depth. ER_jThe energy fraction of the jth sub-band. The frequency ranges of the seven sub-bands are shown in table 2 below.

TABLE 2 Single sub-band frequency Range of six-layer wavelet decomposition

(4) Clustering analysis:

after the features are extracted, a hierarchical clustering method is used, and clustering of two types, namely a functional area and a non-functional area, is performed on all lead data of the patient A by using an optimal clustering parameter (affinity ═ eruydean "and linkage ═ ward") found by grid search, so that a functional area boundary clustering process is completed, and two clustering clusters are obtained. The clustering results are shown in table 3 below.

TABLE 3 functional zone boundary clustering results for patient A

2. Secondly, a functional area attribute classification algorithm is adopted to classify the lead attributes, and the algorithm flow is shown in fig. 4 and comprises the following steps in sequence:

(1) signal acquisition:

after craniotomy, a platinum electrode array (Ad-Tech, Racine, Wis., USA) with a diameter of 4mm and an inter-electrode distance of 10mm was implanted under the dura mater of the patient; the sampling frequency of the ECoG data is 500 Hz. After craniotomy, subdural electrode arrays were placed under the cerebral cortex of each patient; after anesthesia, the patient is awakened, and an ECoG signal acquired by an intraoperative neuro-electrophysiological detector (Endevor Bravo, Nicolet, Inc., USA) in the resting state of the patient is acquired to obtain the ECoG in the resting state of the patient. The data samples collected from patient a are shown in table 1 above (all subsequent processing and calculations are for patient a).

(2) Data preprocessing:

intercepting the data in the resting state, sequentially performing processing steps of 1Hz high-pass filtering, 49-51Hz notch filtering, ICA filtering and the like, and removing power frequency interference existing on 50Hz and frequency multiplication thereof. The PSD of each lead signal of the patient A changes in the preprocessing process as shown in figure 2.

(3) Feature extraction:

the ECoG data of each lead is divided into one sample every 2 seconds, two groups of characteristics of time domain statistics (sum, mean, variance, standard deviation, maximum value and minimum value) and rhythm (delta (0.5-3Hz), theta (4-7Hz), alpha (8-13Hz), beta (14-30Hz) and gamma (more than 30Hz) energy are extracted from 1000 sampling points of each sample, the specific calculation process is that the sum, mean, variance, standard deviation, maximum value and minimum value of 1000 sampling points of each sample are respectively calculated as the time domain statistics characteristics of the sample, band-pass filtering processing of corresponding rhythm frequency bands is respectively carried out on each sample, and the square sum of the filtered sample points is calculated as the rhythm energy characteristics of the sample.

(4) PCA dimension reduction:

for each sample, the dimensionality of the extracted feature quantity is reduced by using PCA, and the dimensionality-reduced feature retains 85% of variance information of the original feature.

(5) SVM classification:

inputting each sample into a pre-trained SVM classification model for classification, adopting a threshold value method when calculating the lead attribute of a patient, and when the ratio of the number of the samples of a certain lead distributed into a functional region is more than or equal to 0.5, considering that the lead belongs to the functional region, otherwise, considering that the lead belongs to a non-functional region. The results of the lead classification for patient a are shown in table 4 below.

TABLE 4 functional zone Attribute Classification of patient A

3. Finally, matching the clustered clusters with the functional area attributes by using an ECoG intraoperative brain function positioning integration algorithm, wherein the algorithm flow is shown in FIG. 5 and comprises the following steps in sequence:

(1) obtaining functional area boundary clustering clusters:

two clusters of the patient cluster are obtained by a functional area boundary clustering algorithm.

(2) Obtaining a functional area attribute classification category:

two categories of patient classification are derived by the functional zone attribute classification algorithm.

(3) Calculating similarity between cluster and classification category

After the functional area boundary clustering result and the functional area attribute classification result of the patient are obtained, the similarity between the two classes of the boundary clustering and the two classes of the attribute classification is respectively calculated, and the calculation formula of the similarity is as follows:

wherein a represents a cluster of classes a, b represents a class b, S_abRepresenting the similarity of class cluster a and class b.

After the similarity between the class cluster and the two classes is calculated, the attribute of the class with higher similarity is regarded as the attribute of the class cluster, and the result of the clustered class cluster with the attribute is regarded as the final brain function positioning result.

The following table 5 is an example of the cluster and classification of patient a to calculate the final functional localization result of patient a.

TABLE 5 calculate attributes of two clusters of patient A

It can be seen that according to the calculation result of the similarity calculation formula (2), the similarity between the first cluster of the cluster and the first classified class is 0.8, the similarity between the first cluster of the cluster and the second classified class is 0, the similarity between the first cluster of the cluster and the second classified class is higher, and meanwhile, the attribute of the first classified class is a functional area, so that the attribute of the first cluster of the cluster is considered to be a functional area. The brain function localization results for patient a are shown in table 6 below:

TABLE 6 ECoG intraoperative brain function region localization integration algorithm results

The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.