A METHOD AND SYSTEM FOR GENERATING A CORTEX MAPPING OF SUBSTANCE ADDICTION EXPLAINABLE FROM EEG SIGNALS.

Abstract

The invention relates to a computer-aided method and system for generating an explainable substance dependence cortex map from EEG signals. Within the scope of the invention, multiple feature selection operations are applied to the features obtained from EEG signals, the selected features are evaluated by multiple classifiers, and features found to be effective in classification are backmapped to EEG channels or channel relationships. The backmapped channels or channel relationships are converted into a regional symbol space independent of the number of channels, and regional distributions, directional regional transition matrices, and hemispheric transition matrices are generated from these symbols. Successive regional transitions are separated into Texture, Edge, Corner, and Breakpoint classes, and a directional XAI transition matrix is ​​obtained for these classes.Using the collected data, an explainable substance addiction cortex map is created, and the neurotopographic evaluation of the cortical regions associated with the decision is provided.

Description

1 TARIFF Substance addiction that can be explained by EEG signals from the cortex. MAP GENERATION METHOD AND SYSTEM TECHNICAL AREA 5 The invention is based on biomedical signal processing and electroencephalography (EEG) brain technology. This relates to the technical field of activity analysis systems; more specifically, EEG. processing features obtained from signals, and interpreting those features in EEG association with channels and brain regions, at the channel and region level Determining neurotopographic distributions, hemispheric and regional transition patterns 10 the formation and explanation of substance addiction in the cortex using this data It relates to a method and system for producing maps. The invention also involves the analysis of physiological signals obtained from EEG measurement systems. processing results in the emergence of cortical distribution patterns of brain regions. extraction, determination of regional and hemispheric interactions and the 15 obtained Techniques for visualizing results for biomedical evaluation purposes. It includes solutions. STATE OF THE ART Electroencephalography (EEG) is used to measure brain activity and various 20 commonly used in the assessment of neurological or behavioral conditions It is one of the biomedical signal processing methods. In the known technique, EEG signals various preprocessing, feature extraction, feature selection and classification operations are performed on it. systems for determining individual circumstances through implementation It is used. 25 For example, in document CN120105048A, characteristics from EEG signals are discussed. Feature selection is being performed using the adaptive sparse group Lasso model. And emotion recognition is performed based on the selected features. Word The subject of this system is the classification of features obtained from EEG signals. The evaluation is carried out, and the characteristics found to be effective in the classification are reviewed again. 30 mapping back to EEG channel space, associating it with brain regions, or Any solution for generating neurotopographic explainability It is not available. 2 Similarly, document CN116671936A describes multi-channel EEG signals. The data is processed, feature vectors are obtained, and the training and verification stages are completed. Then epilepsy is diagnosed. This document also describes the analysis of EEG signals. The obtained characteristics are used for classification purposes. Together, the features that contribute to the decision-making process are channels or pairs of channels. 5 backmap at the level, conversion to regional symbol space, inter-regional Identifying transition patterns or creating an explainable cortex map It is not taught. Therefore, current solutions are mainly based on classification from EEG signals. It focuses on producing a result or prediction output, and contributes to the decision-making process. The association of these characteristics with anatomical brain regions, different EEG Creating a common neurotopographic representation independent of its montages, regional and elucidation of hemispheric transit behaviors and the results obtained It falls short in presenting the information in the form of explainable cortex maps. Therefore, the distinguishing features obtained from EEG signals are channel and 15 symbolic neurotopographic transformation that can be mapped back at the regional level This allows for the identification of regional and hemispheric transition patterns, and Substance abuse cortex maps can be explained using the data in question. New technical solutions are needed that can be produced. THE TECHNICAL PROBLEM THAT IS INTENDED TO BE SOLVED The technical problem that the invention aims to solve is the processing of EEG signals. Which EEG channels and which brain regions produce the distinctive features obtained as a result? The inability to determine whether it is related to the regions, classification decisions The inability to explain it neurotopographically, based on 25 different EEG montages. the inability to evaluate the data within a common regional representation space and regional differences The inability to elucidate hemispheric interaction patterns. The invention also addresses existing systems that only produce a classification result. The inability to identify the brain regions that contributed to the decision-making process, The inability to analyze transition behaviors between regions, hemispheric 30 The inability to evaluate the distributions and the explainable cortical results obtained It aims to solve the problems of not being able to present information in map form. In this context, the invention relates the features found to be effective in classification to the EEG channel space. backmapping, backmapped features are regional, independent of the number of channels. 3 conversion to symbol space, regional and hemispheric transition patterns Identifying and explaining substance addiction using the data in question. By enabling the generation of a cortex map, solutions can be provided to the specified technical problems. It brings. A BRIEF DESCRIPTION OF THE INVENTION The results obtained with current EEG-based analysis systems are mostly The classification output is presented in the form of a risk score or decision label, and the word The subject is which brain regions are associated with the decision, and which factors play a role in the decision-making process. regional patterns are influential and how interactions between different regions affect the situation. It cannot be adequately demonstrated that this occurred. Furthermore, different EEGs... data obtained from their assemblies under a common neurotopographic representation Solutions for its evaluation also remain limited. The invention aims to solve these technical problems by obtaining signals from EEG. performs multiple feature selection operations on the selected features, 15 selected features evaluated with multiple classifiers, found to be effective in classification features that are mapped back to the EEG channel space or channel relationships and word its subject matter is a regional symbol space independent of the number of channels. It offers a transformative method and system. The backmapped channel or channel relationships within the scope of the invention are regional symbols 20 These symbols are used to create regional histograms and directional regional histograms. Transition matrices, hemispheric distributions, hemisphere transition matrices, and entropy measures. It is created. In addition, sequential regional transitions are created using Texture, Edge, Corner and The decision is made by separating the explanations into classes defined as breakpoints. Contributing regional behavioral patterns are revealed. Thus, only 25 Neurotopographic information regarding not only the decision result but also the mechanism by which the decision was made. is obtained. Common features are selected using multiple feature selection methods in an application of the invention. Subsets of features are created, and these subsets can be multiple. It is evaluated with a classifier. Thanks to this structure, 30 effective classifications are made. Identifying the characteristics found and relating these characteristics to brain regions It becomes possible to establish a connection. 4 Another aspect of the invention is the regional functionality that operates independently of EEG mounting. By using a symbol dictionary, different numbers of channels or different electrodes can be used. Data obtained from EEG systems with specific locations form a common neurotopographic They can be evaluated under representation. Thus, different data sets and different EEGs Comparable regional distribution patterns among registration systems 5 can be created. The regional frequency information and transit densities obtained as a result of the invention, hemispheric distributions, entropy measures, and explainability classes are combined An explainable substance addiction cortex map is being created. The issue in question... Cortical mapping; identification of cortical regions associated with the decision, regions 10 uncovering the interaction patterns between them, hemispheric dominance visual assessment and evaluation of cortical distribution structures associated with substance abuse. and enables its numerical presentation. Thus, data obtained from EEG signals The biomedical interpretability of the information obtained is increased, and the decision... The anatomical equivalent of its mechanism can be revealed. 15 LIST OF FIGURES AND BRIEF DESCRIPTION Figure 1 shows the system diagram illustrating the general workflow of the method described in the invention. It shows. Figure 2 shows the creation of a feature matrix from EEG signals and the normalization of the features. It shows the process flow related to the execution. Figure 3 shows the application of multiple feature selection methods and the selected feature sub-categories. It shows the process for forming clusters. Figure 4 shows the evaluation of selected feature subsets with multiple classifiers and This shows the process flow for carrying out decision fusion. 25 Figure 5 shows the features found to be effective in classification based on EEG channel space or channel. It shows the process of rematching their relationships. Figure 6 shows that backmapped channel or channel relationships are independent of the number of channels. It demonstrates the structure related to its transformation into a regional symbol space. Figure 7 shows the regional histogram and directional regional 30 through symbolic regional representation. This shows the process for creating a transition matrix. Figure 8 shows the determination of hemispheric distribution through symbolic regional representation and It shows the structure for creating the hemisphere transition matrix. Figure 9 shows the sequential regional transitions to the Texture, Edge, Corner, and Breakpoint classes. This shows the explainability analysis of the separation. Figure 10 shows regional distribution, transition density, hemispheric analysis, and explainability. Explainable substance addiction cortex obtained by combining its outputs This shows an example of a map representation. 5 DETAILED DESCRIPTION OF THE INVENTION This discovery can be explained by processing data obtained from EEG signals. a method and system for creating a substance addiction cortex map It is related. 10 Within the scope of the invention, EEG signals belonging to one or more individuals were entered into the system. EEG data is obtained directly from measurement systems as raw data. These could be signals or pre-processed EEG data. In practice, EEG data includes artifact removal, resampling, and band-pass filtering. windowing, segmentation, normalization, or any of these 15 It undergoes pre-processing stages that include a combination of these. However, The invention is not limited to a specific pre-processing method. A multichannel analysis is performed using pre-processed EEG data. A feature matrix is ​​being created. The feature matrix includes time-domain features and frequencies. domain properties, time-frequency domain properties, statistical properties, channel pairs 20 properties based on relationships between them, transition table-based properties, pattern-based properties It can be created from signatures or any combination thereof. The feature matrix created in the application is scaled using the min-max scaling method. normalization is performed. However, the use of different normalization methods is also possible. It is possible. 25 Method of selecting multiple features on the resulting feature matrix Feature selection methods such as NCA and chi-square are being implemented in an application. It includes mRMR and ReliefF methods. However, different feature selection is possible. It is also possible to use these methods. For each feature selection method, the feature must be... Importance scores are calculated, these scores are ranked, and the previously 30 Candidate feature subsets are generated based on predetermined threshold values. One aspect of the invention is determined by a method of selecting only a single feature. Feature subsets are not used. Instead, different feature selection methods are employed. Features jointly selected by the parties are determined, and intersection-based feature subcategories are created. 6 Sets are being formed. In this context, binary intersection sets, ternary intersection sets are used. features that are determined in common by sets or all feature selection methods Sets can be formed. Thus, they can be repeatedly used by different methods. This ensures the identification of selected and highly distinctive features. Each subset of features generated will be classified by one or more classifiers. 5 It is evaluated using classifiers such as kNN and tuned kNN. It includes SVM and tuned SVM classifiers. However, it differs. It is also possible to use classification methods. For each subset of features... accuracy, error rate, sensitivity, specificity, F1 score, AUC value, or similar. Performance metrics are being calculated. 10 The classifier with the highest performance in an application. It is used. In another application, the classifier results are evaluated by majority vote. using weighted vote or performance-based decision fusion methods They are combined. Thus, a more stable system is achieved without relying on a single classifier. Decision outcomes are obtained. 15 Distinguishing factors that contribute to decision-making following the classification process. The characteristics are identified. Then, these distinguishing characteristics are repeated on the EEG. It is mapped back to the channel space or channel relations. The backmapping process Depending on the feature extraction structure used, directly at the channel level Channel pairs, channel groups, or derived channel relationships can be implemented as well. 20 It can also be carried out at this level. In the backmap process, the selected feature indexes are those indexes of the respective indexes. It is analyzed according to the feature inference structure from which it was generated. This analysis As a result, each selected feature includes: the corresponding EEG channel, EEG channel pair, and channel group. or is associated with a derived channel relationship. Thus, 25 are effective in classification. the numerical properties found remain only as abstract property values These features are prevented and their equivalents in the EEG measurement setup are determined. The mapped channel or channel relationships are then used for regional neurotopography. It is transformed into a regional symbolic space in order to create representation. This In the conversion, a channel-region glossary specific to EEG montage is used. The 30 in question... The channel-region glossary corresponds to the channel placements in the EEG system used. It is a mapping structure that associates them with the cortical regions from which they originate. Thus, it has 16 channels. Obtained from 32-channel, 64-channel, or other different number of-channel EEG assemblies. The acquired channel information can be transferred to a common regional symbol space. 7 Regional symbol space, neurotopographic representation independent of channel number. It is created to provide this. In an application, this symbol space; frontal left, frontal right, temporal left, temporal right, parietal left, parietal right, occipital left, symbols representing the occipital right, central left, central right, and midline regions It includes, for example, the FL frontal left region, the FR frontal right region, 5 TL represents the left temporal region, TR represents the right temporal region, PL represents the left parietal region, PR parietal right region, OL occipital left region, OR occipital right region, CL central CR represents the left region, CR represents the right region (central), and Fz, Cz, Pz, and Oz represent the midline regions. is able to. Thanks to this structure, 10 different numbers of channels can be obtained from EEG recordings. The results can be expressed in the same regional symbol space, and different EEG montages can be used. Comparable neurotopographic representations can be created between them. In one application of the invention, individuals diagnosed with substance addiction and EEG recordings obtained from individuals in the control group were entered into the system as input. It has been received. 15 Preprocessing operations were performed on the acquired EEG data, and each A multi-channel feature matrix was created for the individual. Then, the feature in question... By applying a method of selecting multiple features on the matrix, a distinctive feature can be identified. Subsets have been obtained. The resulting feature subsets were evaluated with multiple classifiers. and the characteristics that were found to be effective in classification were determined. The determined characteristics are further They were then mapped back to their associated EEG channels and channel pairs. The channel information obtained as a result of the backmatch is predefined. transformed into a regional symbol space and regional symbol sequences for each individual. regional histograms were created using these symbol sequences, 25 Directional regional transition matrices and hemispheric transition matrices have been calculated. Sequential regional transitions are managed under the Texture, Edge, Corner, and Breakpoint classes. The data was evaluated and distributions for each transition class were obtained. Regional information on distribution data, transition densities, hemispheric distributions, and transition classes. By evaluating the results together, substance dependence can be explained for each individual. 30 A map of the cortex has been created. Using the created cortex maps, the substance abuse group was identified. cortical distribution patterns of the control group and cortical distribution patterns of the control group. 8 visual and comparative regional behaviors that contribute to decision-making. This has been presented numerically. Using channels or channel relationships converted into a regional symbol space Symbolic regional sequences are being created. Through these symbolic sequences... The frequency of occurrence of each regional symbol is calculated, and a regional histogram 5 is created. is being created. A regional histogram shows the features found to be effective in classification. It is a numerical distribution structure that shows in which cortical regions it is concentrated. Among the regional symbols that are placed consecutively in the symbolic sequence A directional regional transition matrix is ​​created by counting the transitions. This matrix is ​​a the direction of transitions from one regional symbol to another regional symbol and 10 It shows the intensity. Thus, it shows not only which regions are affected, The transition behavior between the affected regions is also determined. In one application of the invention, regional symbols also represent the left hemisphere, right hemisphere, and left hemisphere. It is converted into hemisphere and optionally midline classes. This conversion As a result, a sequence of hemispheric symbols is created, and consecutive hemispheric symbols 15 A hemispheric transition matrix is ​​obtained from this. The hemispheric transition matrix, The features found to be effective in classification were the left hemisphere, right hemisphere, and midline. It reveals the distribution and transition behavior at this level. Shannon entropy or normalization over regional symbol distributions. Entropy can be calculated. The entropy value is the regional distribution of 20 in a single region. whether it is concentrated or not, or a widespread pattern among different regions. It is used to show whether or not it has been created. Explanatory analysis of sequential regional symbol transitions within the scope of the invention. For this purpose, it is divided into Texture, Edge, Corner, and Breakpoint classes. The same area The repeating transitions within it are defined as Texture class. The same anatomical 25 Transitions that change direction within the family are defined as Edge class. Similarly Transitions that occur between different regions within a hemisphere are called the Corner class. It is determined. It is either interhemispheric or of a discontinuous regional nature. Transitions are defined as Breakpoint class. The transition frequency for Texture, Edge, Corner, and Breakpoint classes is 30. a directional XAI transition matrix is ​​being calculated between the classes in question. The XAI transition matrix is ​​being created, representing regional factors that contribute to decision-making. not only the spatial distribution of transitions, but also according to transition types. It also shows an explainable behavioral pattern. 9 Regional histogram, directional regional transition matrix, hemisphere transition matrix, The substance can be explained by considering entropy measures and the XAI transition matrix together. A cortical map of addiction is being created. This cortical map shows the channel. regional contribution map independent of number, survey pattern representation or cortical It can be presented in the form of a report output. Cortical report output; dominant regions, 5 It can include hemispheric asymmetry, entropy score, and XAI distribution. This application example is intended solely to facilitate understanding of the invention. This information is provided and does not limit the scope of protection of the invention. In one application of the invention, the system uses features found to be effective in classification. A back-map that performs the mapping back to EEG channels or channel relationships. The mapping unit maps the backmapped channel or channel relationships to the regional symbol space. a regional symbol transformation unit that transforms, a directional regional transition matrix and A transition analysis unit that creates a hemisphere transition matrix, Texture, Edge, Corner and An explainability analysis unit and XAI transition matrix that define breakpoint classes. This can be explained by cortical mapping that creates a cortex map of substance addiction. 15 It includes the unit. 25

Claims

SYSTEMS 1. Substance that can be explained by computer-assisted processing of EEG signals. This method focuses on generating a cortex map of addiction; its characteristic feature is: - Creating a feature matrix from multi-channel EEG signals, 5 - NCA, chi-square, mRMR and ReliefF are included in the feature matrix in question. Application of feature selection methods, - characteristics determined by the relevant feature selection methods forming a common intersection set among them, - the generated common intersection set is kNN, tuned kNN, SVM and tuned 10 Evaluation using SVM classifiers, - Identifying the characteristics that are effective in classification, - specified features to EEG channels, channel pairs or derived mapping back to channel relationships, - back-mapped channel or channel relationships frontal left, frontal right, 15 temporal left, temporal right, parietal left, parietal right, occipital left, including the occipital right, central left, central right, and midline regions. transformation into a regional symbol space, - directional regional transition through the symbolic regional representation created. creation of the matrix, 20 - the regional symbols in question also represent the left hemisphere, the right hemisphere, and optional conversion to midline classes and hemispheres creation of the transition matrix, - sequential regional transitions in Texture, Edge, Corner and Breakpoint dividing into classes, 25 - a directional guide for the Texture, Edge, Corner, and Breakpoint classes in question. Generation of an XAI transition matrix and - regional transition matrix, hemisphere transition matrix and XAI transition matrix substance addiction cortex map can be explained using It is characterized by including the steps involved in its creation. 30 2. It is a method according to Claim 1, and its characteristic is that at least two commonly determined characteristics... from the features selected together by the feature selection method It is characterized by its creation. 11 3. This method, according to claim 1, is characterized by its backmapped channel or channels. their relationships, a predefined EEG independent of the number of channels by converting it to a regional symbol space using an assembly dictionary It is characteristic.

4. The method according to claim 4, its characteristic is; the frontal left of the regional symbol space, 5 frontal right, temporal left, temporal right, parietal left, parietal right, occipital left, including the occipital right, central left, central right, and midline regions. It is characteristic.

5. The method according to Claim 1, its characteristic is that regional symbols also include the left hemisphere, conversion to right hemisphere and optionally midline classes 10 It is characterized by the formation of a hemispheric transition matrix.

6. The method according to Claim 1, its characteristic is; successive regional transitions within the same region. Texture refers to the transitions that occur within it, within the same anatomical family. Edge refers to the transitions that occur in different regions within the same hemisphere. For transitions occurring between corners and hemispheres, or discontinuous 15 It is characterized by being classified as a breakpoint for regional transitions.

7. This method, according to claim 7, has the following features: Texture, Edge, Corner, and Breakpoint. It is characterized by the creation of a directional XAI transition matrix for the classes.

8. Generating a substance abuse cortex map that can be explained from EEG signals. It is a computer-aided system structured to be; its feature is; 20 - Features found to be effective in classification are related to EEG channels and channel pairs. or a backmap that enables it to be mapped back to derived channel relationships unit, - back-mapped channel or channel relationships specific to EEG montages - A regional symbol, independent of channel number, is created using a regional dictionary. 25 a regional symbol conversion unit that enables its conversion into space, - via a directional regional transition matrix through the relevant regional symbols a transition analysis that enables the creation of a hemisphere transition matrix unit, - Consecutive regional transitions in Texture, Edge, Corner and Breakpoint 30 the division into classes and the transition frequencies for those classes an explainability analysis unit that enables its determination and 12 - A directional XAI transition for Texture, Edge, Corner, and Breakpoint classes. creation of the matrix and regional transition matrix, hemisphere transition Substance dependence can be explained using the matrix and the XAI transition matrix. a cortical mapping that enables the creation of a cortex map It is characterized by containing the unit 5. 15 25

Images