Consciousness level recognition device based on electroencephalogram data and computer readable storage medium

Abstract

The application discloses a consciousness level recognition device based on electroencephalogram data and a computer readable storage medium. The device can perform the following operations: obtaining multi-lead scalp electroencephalogram data of a patient to be recognized; inputting the multi-lead scalp electroencephalogram data into a trained neural network model for consciousness level recognition operation, to output a response vector composed of response intensities of each lead, the response intensity representing the correlation degree between the electroencephalogram feature of the lead and the target consciousness level; calculating the distance between the response vector and the average response vector of the target consciousness level group; and determining the consciousness level recognition result of the patient to be recognized according to the distance. The consciousness level recognition scheme based on electroencephalogram data can improve the accuracy of distinguishing between the vegetative state and the weak consciousness state, and realize objective and accurate evaluation of the consciousness level.

Description

Technical Field

[0001] This application generally relates to the interdisciplinary field of biomedical engineering and artificial intelligence. More specifically, this application relates to a consciousness level recognition device based on electroencephalogram (EEG) data and a computer-readable storage medium. Background Technology

[0002] Assessment of consciousness level is a crucial step in the clinical diagnosis and prognosis of comatose patients, especially in distinguishing between vegetative state (VS) and minimally conscious state (MCS). Accurate assessment results directly impact treatment options and rehabilitation intervention strategies. Currently, two types of assessment methods exist in clinical practice: one based on behavioral responses and the other on objective physiological signals. However, both have limitations in accuracy, objectivity, and applicability, making it difficult to meet the needs of refined diagnosis.

[0003] In existing technologies, behavioral response-based rating scales (such as CRS-R) are the mainstream means of assessing consciousness levels. This method involves clinicians conducting multiple rounds of stimulation and observation on patients, scoring them based on behavioral responses such as eye opening, command compliance, and limb movement. However, the assessment results are highly dependent on the physician's subjective judgment and the patient's level of cooperation. Significant diagnostic blind spots exist for uncooperative patients and those with sensory impairments, leading to a substantial decrease in sensitivity and accuracy. In particular, patients with VS and MCS exhibit highly similar behavioral behaviors, making misjudgment highly likely. Regarding objective indicators, while electroencephalography (EEG) has been widely studied for consciousness assessment, traditional EEG analysis relies heavily on expert interpretation of graphic features, lacking unified standards and exhibiting strong subjectivity. More importantly, traditional methods treat each lead as an independent entity, ignoring the spatial synergy between multiple leads, making it difficult to comprehensively characterize the functional state of the brain network and limiting the accuracy and reliability of the assessment.

[0004] In view of this, there is an urgent need to provide a consciousness level recognition scheme based on EEG data in order to achieve an objective and accurate assessment of consciousness level. Summary of the Invention

[0005] In order to at least address one or more of the technical problems mentioned above, this application proposes a consciousness level recognition scheme based on EEG data in several aspects.

[0006] In a first aspect, this application provides a consciousness level recognition device based on electroencephalogram (EEG) data, characterized in that it includes: a processor; and a memory storing consciousness level recognition program instructions based on EEG data. When the program instructions are executed by the processor, the device performs the following operations: acquiring multi-lead scalp EEG data of a patient to be identified; inputting the multi-lead scalp EEG data into a trained neural network model for consciousness level recognition, so as to output a response vector composed of the response intensities of each lead, wherein the response intensity characterizes the correlation between the EEG features of the lead and the target consciousness level; calculating the distance between the response vector and the average response vector of the target consciousness level group; and determining the consciousness level recognition result of the patient to be identified based on the distance.

[0007] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations to calculate the distance between the response vector and the average response vector of the target consciousness level group: mapping each lead to multiple regions based on the spatial location of each lead; determining the spatial weight of each lead based on the distance between each lead and the center lead of its region, using the region as the unit of calculation; and calculating the distance based on the spatial weight of each lead, the response vector, and the average response vector of the target consciousness level group.

[0008] In some embodiments, when the program instructions are executed by the processor, the device uses the following formula to determine the spatial weights of each lead. :

[0009]

[0010] in, Let be the distance between the i-th lead and the central lead of its region. This is the spatial sensitivity parameter.

[0011] In some embodiments, when the program instructions are executed by the processor, the device uses the following formula to calculate the distance. :

[0012]

[0013] in, For the response vector, The average response vector of the target awareness group. Spatial weights for each lead, Let i be the response intensity of the i-th lead. Let be the average response intensity of the i-th lead.

[0014] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations before calculating the distance based on the spatial weights of each lead, the response vector, and the average response vector of the target consciousness level group: for each region, determining a focus score for each region based on the response intensity of the central lead of each region, the average response intensity of the central lead, and the standard deviation of the response intensity; determining high-cluster regions among the plurality of regions based on the focus score of each region and a preset score; using a dynamic weighting factor to weight and amplify the response intensity of each lead in the high-cluster regions to obtain a weighted response vector, and subsequently using the weighted response vector to calculate the distance.

[0015] In some embodiments, when the program instructions are executed by the processor, the device uses the following formula to determine the clustering score for each region. :

[0016]

[0017] in, For the region Clustering score, For the region The response intensity of the central lead ck, The average response intensity of the target awareness group on the central lead ck. The mean standard deviation of the response intensity of the target awareness level group on the central lead ck.

[0018] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations: mapping the focus score of each region to the spatial location of each link within that region to obtain discrete lead points with color gradients; and interpolating the discrete lead points with color gradients to generate an EEG topology map.

[0019] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations to train the neural network model: acquiring multiple multi-lead scalp EEG data from different comatose patients and from different stages of the same patient; labeling the multiple multi-lead scalp EEG data according to the level of consciousness of each patient and the level of consciousness of the same patient at different stages, generating multiple multi-lead scalp EEG data with labels; and training the neural network model using five-fold cross-validation based on the multiple multi-lead scalp EEG data with labels.

[0020] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations to train the neural network model using five-fold cross-validation: during the training phase, inputting a first number of labeled multi-lead scalp EEG data as training data into the neural network model for training; during the validation phase, inputting a second number of labeled multi-lead scalp EEG data as validation data into the neural network model to output the response vector of the second number of multi-lead scalp EEG data; and calculating the average of the response vectors of each multi-lead scalp EEG data labeled as the target level of consciousness to obtain the average response vector of the target level of consciousness group.

[0021] In a second aspect, this application provides a computer-readable storage medium storing consciousness level recognition program instructions based on electroencephalogram (EEG) data, which, when executed by a processor, cause the operation performed by the EEG-based consciousness level recognition device according to the foregoing first aspect and several embodiments to be implemented.

[0022] By using the above-described EEG-based consciousness level identification scheme, this embodiment of the application acquires multi-lead scalp EEG data and utilizes the response vectors output by the trained neural network to reflect the correlation between the leads and the target consciousness level. The identification result is determined by calculating the distance to the average response vector of the target consciousness level group. This approach can eliminate the reliance on behavioral response-based rating scales and reduce diagnostic blind spots and subjective judgment errors caused by patient non-cooperation and sensory impairment. At the same time, by constructing multi-lead response vectors and analyzing spatial distance, the spatial synergy between leads is effectively explored, overcoming the shortcomings of traditional EEG analysis that ignore spatial organization, rely on manual interpretation, and lack unified standards. This can improve the accuracy of distinguishing between vegetative state (VS) and minimally conscious state (MCS) and achieve an objective and accurate assessment of consciousness level. Attached Figure Description

[0023] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:

[0024] Figure 1 An exemplary architecture diagram of a consciousness level recognition device based on electroencephalogram (EEG) data according to an embodiment of this application is shown;

[0025] Figure 2 An exemplary flowchart of the consciousness level recognition operation based on EEG data according to an embodiment of this application is shown;

[0026] Figure 3An exemplary flowchart of training a neural network model according to an embodiment of this application is shown;

[0027] Figure 4 An exemplary flowchart for calculating the Gaussian-weighted Euclidean distance according to an embodiment of this application is shown;

[0028] Figure 5 An exemplary flowchart illustrating the determination of highly clustered regions according to an embodiment of this application is shown. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0031] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0032] As used in this specification and claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0033] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0034] Figure 1 This is an exemplary structural block diagram illustrating a consciousness level recognition device 100 based on electroencephalogram (EEG) data according to an embodiment of this application. Figure 1As shown, the device 100 may include a processor 101 and a memory 102. The processor 101 may include, for example, a general-purpose processor (“CPU”) or a dedicated graphics processor (“GPU”), and the memory 102 stores program instructions executable on the processor. In some embodiments, the memory 102 may include, but is not limited to, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), and enhanced dynamic random access memory (EDRAM).

[0035] Furthermore, the aforementioned memory 102 may store consciousness level recognition program instructions based on EEG data. When the program instructions are executed by the processor, the device 100 performs actions such as... Figure 2 The consciousness level recognition operation based on EEG data is shown below:

[0036] In step S201, multi-lead scalp EEG data of the patient to be identified is acquired; in step S202, the multi-lead scalp EEG data is input into the trained neural network model to perform consciousness level recognition operation, so as to output a response vector composed of the response intensity of each lead; in step S203, the distance between the response vector and the average response vector of the target consciousness level group is calculated; in step S204, the consciousness level recognition result of the patient to be identified is determined based on the distance.

[0037] The aforementioned multi-lead scalp EEG data is potential change data of brain neuron electrical activity collected through multiple scalp electrodes. Its core is the EEG signal information corresponding to multiple leads (electrodes), which serves as the foundational data for recognizing level of consciousness. In practice, the GSN256 lead system can be used to collect multi-lead scalp EEG data. The GSN256 lead system contains 256 scalp electrodes, which have a standardized spatial arrangement and positional coordinates.

[0038] The aforementioned neural network model can be a one-dimensional convolutional neural network (1D-CNN). This model can learn the distribution characteristics of different levels of consciousness in the spatial structure of EEG leads, output the response intensity of each lead, and thus accurately indicate the influence of key leads on the judgment of the level of consciousness, ultimately outputting results with clear clinical interpretability. Furthermore, this one-dimensional convolutional neural network can run on computing devices equipped with a Python execution environment, the PyTorch deep learning framework, and the MNE EEG signal processing library.

[0039] In some implementation scenarios, the aforementioned neural network model can also be a graph neural network (GNN). Specifically, a GNN can treat each EEG lead as a node in a graph, and the spatial adjacency between leads as edges. By automatically learning the information transmission patterns and spatial correlations between nodes, it completes spatial modeling of EEG signals and assessment of consciousness states. However, it should be noted that the internal reasoning process of a GNN has difficulty accurately identifying leads that play a key role in the model's judgment, thus lacking intuitive interpretability in clinical applications.

[0040] like Figure 3 As shown in the embodiments of this application, the device 100 may further perform the following operations to train the neural network model:

[0041] In step S301, multiple multi-lead scalp EEG data are acquired from different comatose patients, as well as from the same patient at different stages. It is important to note that multi-lead scalp EEG data collected from the same patient at different stages are processed as independent samples.

[0042] Furthermore, abnormal lead removal and bandpass filtering can be performed on multiple multi-lead scalp EEG data to preprocess them, obtaining preprocessed multi-lead scalp EEG data for use in step S302 below. Specifically, when performing abnormal lead removal, low-quality leads (such as leads located at the temporal edge or leads significantly affected by electromyography contamination) can be screened out based on lead quality and noise level; when performing bandpass filtering, 0.5–40Hz bandpass spatial filtering can be performed on all multi-lead scalp EEG data, combined with noise processing and multiple linear regression to suppress interference and improve the signal-to-noise ratio.

[0043] In step S302, based on the level of consciousness of each patient and the level of consciousness of the same patient at different stages, multiple multi-lead scalp EEG data are labeled to generate multiple multi-lead scalp EEG data with labels.

[0044] Here, the patient's level of consciousness is the clinical diagnosis result, which can be either a vegetative state (VS) or a minimally conscious state (MCS). When labeling, 0 or 1 can be used to label the two different levels of consciousness. For example, VS is labeled 1 and MCS is labeled 0, or MCS is labeled 1 and VS is labeled 0, and the level of consciousness labeled 1 is taken as the target level of consciousness.

[0045] In step S303, the neural network model is trained using five-fold cross-validation based on multiple multi-lead scalp EEG data with labels.

[0046] Five-fold cross-validation is a model training and evaluation method. Specifically, the training sample set is randomly divided into 5 non-overlapping subsets (i.e., "folds"). During each training session, 4 subsets are selected as the training set, and the remaining subset is selected as the validation set. This process is repeated 5 times (with a different validation set each time) to comprehensively evaluate the model's performance on different data subsets.

[0047] In practice, it is important to ensure that all samples from the same patient belong to only one subset (to avoid bias caused by the same patient's data appearing in both the training and validation sets), and that the number of samples from patients in a vegetative state (VS) and a minimally conscious state (MCS) is roughly balanced within each subset to ensure the objectivity of the validation and the generalization ability of the model.

[0048] Based on this, during the training phase of the neural network model, a first batch of labeled multi-lead scalp EEG data is used as training data input into the neural network model for training. Furthermore, during the training phase, training strategies such as the Adam optimizer and L2 regularization (with a regularization coefficient of 1e-4) can be employed to prevent overfitting, ultimately obtaining the trained neural network model.

[0049] During the validation phase of the neural network model, the second number of multi-lead scalp EEG data with labels can be input into the neural network model as validation data to output the response vector of the second number of multi-lead scalp EEG data.

[0050] Furthermore, the average response vector of each multi-lead scalp EEG data labeled with the target level of consciousness can be calculated to obtain the average response vector of the target level of consciousness group; and the standard deviation of the response vector of each multi-lead scalp EEG data labeled with the target level of consciousness can be calculated to obtain the standard deviation response vector of the target level of consciousness group.

[0051] In other words, the average response vector is composed of the average response intensity of each lead, and the standard deviation response vector is composed of the standard deviation response intensity of each lead. The number of average response intensities in the average response vector and the number of standard deviation response intensities in the standard deviation response vector are both equal to the number of leads.

[0052] As an example, assuming there are m multi-channel scalp EEG data points labeled as target awareness level, the model will output m response vectors V, with the average response intensity of lead i being... The following formula can be used for calculation:

[0053] (1)

[0054] Standard deviation response intensity of lead i The following formula can be used for calculation:

[0055] (2)

[0056] in, For the j-th response vector The response intensity of the i-th lead.

[0057] The aforementioned response intensity characterizes the correlation between the EEG characteristics of the lead and the target level of consciousness, with a numerical range of [0,1]. The closer the value is to 1, the more the EEG characteristics of the lead are inclined towards the target level of consciousness; the closer the value is to 0, the more the EEG characteristics of the lead are inclined towards another target level of consciousness.

[0058] As an example, if the target's level of consciousness is a weak state of consciousness, the closer the response intensity value of the lead is to 1, the more the EEG characteristics of the lead tend to be a weak state of consciousness; the closer the response intensity value of the lead is to 0, the more the EEG characteristics of the lead tend to be a vegetative state.

[0059] In step S203 above, the distance between the response vector and the average response vector of the target consciousness level group is calculated, which can be done by calculating the Euclidean distance between the two. This distance identifies the spatial distribution difference between the patient to be identified and the target consciousness level group.

[0060] As described above, the target consciousness level group can be a group in a state of weak consciousness or a group in a vegetative state, and the Euclidean distance between the response vector and the average response vector of the target consciousness level group can be calculated using the following formula:

[0061] (3)

[0062] in, For the response vector, The average response vector of the target awareness group. Let i be the response intensity of the i-th lead. Let be the average response intensity of the i-th lead.

[0063] In step S204 above, the recognition result of the consciousness level of the patient to be identified is determined based on the distance. Specifically, the smaller the distance, the closer the EEG characteristics of the patient to be identified are to the average characteristics of the target consciousness level group; conversely, the larger the distance, the more the EEG characteristics of the patient to be identified deviate from the average characteristics of the target consciousness level group and are closer to another consciousness level.

[0064] To more intuitively express this degree of proximity, a similarity index can be introduced using the following formula, which combines distance and similarity. Convert to proximity scores within the interval [0,1]:

[0065] (4)

[0066] The closer the proximity value is to 1, the higher the similarity between the patient to be identified and the target group at the level of consciousness. Conversely, the closer it is to 0, the lower the similarity between the patient to be identified and the target group at the level of consciousness.

[0067] As described above, the consciousness level recognition device based on EEG data provided in this application acquires multi-lead scalp EEG data and uses a trained neural network to output a response vector reflecting the correlation between the lead and the target consciousness level. The recognition result is determined by calculating the distance to the average response vector of the target consciousness level group. This eliminates the reliance on behavioral response-based rating scales and reduces diagnostic blind spots and subjective judgment errors caused by patient non-cooperation and sensory impairment. At the same time, through the construction of multi-lead response vectors and spatial distance analysis, the spatial synergy between leads is effectively explored, overcoming the shortcomings of traditional EEG analysis that ignore spatial organization, rely on manual interpretation, and lack unified standards. This improves the accuracy of distinguishing between vegetative state (VS) and minimally conscious state (MCS), and achieves an objective and accurate assessment of consciousness level.

[0068] It is important to understand that the above combination Figure 1 The described structure or architecture of the consciousness level recognition device based on EEG data is merely exemplary, and the implementation method and implementation entity of this application are not limited thereto, but can be changed without departing from the spirit of this application.

[0069] In step S203, besides calculating the Euclidean distance between the response vector and the average response vector of the target consciousness level group, a Gaussian weighting mechanism can be introduced in some implementation scenarios to calculate the distance between these two vectors. The introduction of the Gaussian weighting mechanism is based on a consensus in neuroscience that spatially adjacent brain regions are more likely to produce co-activation. The characteristics of the Gaussian weighting function perfectly match this pattern of cortical activity, enabling a more realistic simulation of the brain's physiological activity.

[0070] Based on this, such as Figure 4 As shown, the aforementioned device 100 can further perform the following operations to calculate the Gaussian-weighted Euclidean distance between the response vector and the average response vector of the target consciousness level group:

[0071] In step S401, each lead is mapped to multiple regions according to its spatial location.

[0072] Specifically, the leads in the GSN256 lead system can be divided into k=5 region modules. These are the frontal lobe, central lobe, parietal lobe, occipital lobe, and temporal lobe, respectively denoted as... , , , and This division helps to avoid interference between spatially adjacent but functionally heterogeneous regions during analysis.

[0073] In step S402, the spatial weight of each lead is determined based on the distance between each lead and the center lead of its respective region, using the region as the calculation unit.

[0074] Specifically, the spatial weight of the central lead can be set to 1, and the spatial weight of each lead can be determined using the following formula. :

[0075] (5)

[0076] in, The distance between the i-th lead and the central lead of its region, such as the Euclidean distance. This is the spatial sensitivity parameter.

[0077] Understandably, when determining the central lead for a region, the following selection principles can be adopted to ensure its representativeness: prioritize referring to the recognized standard leads in the region, such as using the Cz lead by default in the central region, the Fz lead by default in the frontal region, and the Pz lead by default in the parietal region; if further accuracy is required, the spatial center (centroid) of the region can be calculated, and the lead that is closest to the spatial center and has a higher response intensity can be selected as the central lead, based on the response activity of each lead.

[0078] In addition, those skilled in the art can set the spatial sensitivity parameters according to actual needs. The specific value of is related to the lead weight distribution: when When the value is small, only leads that are close to the central reference point will be given a higher weight; conversely, when the value is large, only leads that are close to the central reference point will be given a higher weight. When the value is large, leads farther from the central reference point can still retain a certain weight. Through this parameterization strategy, the model can automatically adjust the weight allocation of each lead based on spatial distance, enhance the contribution of key leads that are spatially close to the target brain region to the classification decision, and suppress the interference of distant noisy leads, thereby improving the model's ability to selectively capture spatial features.

[0079] In step S403, the distance, i.e. Gaussian weighted Euclidean distance, is calculated based on the spatial weights of each lead, the response vector, and the average response vector of the target awareness level group.

[0080] Specifically, the Gaussian weighted Euclidean distance can be calculated using the following formula. :

[0081] (6)

[0082] in, For the response vector, The average response vector of the target awareness group. Spatial weights for each lead, Let i be the response intensity of the i-th lead. Let be the average response intensity of the i-th lead.

[0083] The above combination Figure 4 This paper describes a method for calculating the Gaussian-weighted Euclidean distance between the response vector and the average response vector of the target awareness level group. This is achieved by defining a method based on Euclidean distance and spatial sensitivity parameters. The spatial weighting method (also known as Gaussian weighting factor) automatically adjusts the spatial weight allocation based on the distance of each lead to the central lead of its region. Specifically, leads closer to the central lead are assigned higher spatial weights, while leads farther away receive lower weights. This mechanism enhances the contribution of key leads spatially adjacent to the target brain region to the classification task, while suppressing interference from distant noisy leads, thereby improving the model's accuracy in capturing EEG spatial features and optimizing classification performance.

[0084] As previously shown, each lead can be mapped to multiple regions. When examining the contribution of each region to consciousness level recognition, the inventors found that the contribution of each region differs. In order to improve the sensitivity to key regions (also known as highly clustered regions) and achieve selective attention at the structural level, in the embodiments of this application, these regions can be screened out, and the response intensity of the leads in these regions can be weighted and amplified, thereby improving the consciousness level recognition capability.

[0085] Based on this, such as Figure 5 As shown, the aforementioned device 100 can further perform the following operations to identify high-concentration areas among multiple regions:

[0086] In step S501, for each region, a focus score is determined based on the response intensity of the central lead in each region, and the average response intensity and standard deviation of the central lead response intensity. .

[0087] The focus score obtained here Used to characterize the similarity between the current region and the group response pattern at the target awareness level: when The closer the value is to 1, the higher the match between the response characteristics of this region and the standard template of the target awareness level group; when... When the value is close to 0, it means that the response characteristics of this area may not belong to the typical pattern of the target awareness level group.

[0088] Specifically, the following formula can be used to determine the clustering score for each region. :

[0089] (7)

[0090] in, For the region Clustering score, For the region The response intensity of the central lead ck, The average response intensity of the target awareness group on the central lead ck. The mean standard deviation of the response intensity of the target awareness level group on the central lead ck.

[0091] In step S502, based on the focus score and preset score of each region, highly clustered regions among multiple regions are identified. These highly clustered regions can be marked as key analysis regions for the current sample, serving as an important basis for subsequent weighted analysis. It should be understood that those skilled in the art can select specific values ​​for the preset score according to actual needs; for example, a preset score of 0.8 may be selected. This application does not limit this selection.

[0092] In step S503, a dynamic weighting factor is used to weight and amplify the response intensity of each lead in the highly clustered region to obtain a weighted response vector, which can be used to calculate the distance in step S203.

[0093] Understandably, the specific values ​​of the dynamic weighting factors can be set according to actual needs, for example, they can be combined with clustering scores. It exhibits a monotonically increasing relationship, that is... Regions with higher values ​​receive greater weight to ensure that regions with response patterns more similar to those of target-awareness groups are given priority in response enhancement.

[0094] It is also understandable that the focus rating While it is an abstract indicator of regional dimensions, the spatial distribution characteristics of EEG signals are crucial for recognizing levels of consciousness. Therefore, focusing on each region can be used to determine its level of awareness. By mapping to specific lead locations, abstract scores can be transformed into intuitive spatial distribution visualizations.

[0095] Specifically, tools such as MNE or Matplotlib can be used, combined with the spatial coordinates of the GSN256 lead system, to map the focus score of each region to the spatial location of each lead within that region, thus obtaining discrete lead points with color gradients. Therefore, the current numerical range [0,1] can be converted into a color gradient, which can visually display the focus scores of different regions. For example, the redder the color, the closer the response characteristics of that region are to the standard template of the target awareness level group.

[0096] Furthermore, due to the limited number of EEG leads actually collected, directly displaying discrete lead points cannot present a continuous distribution of EEG activity. Interpolation can be performed on discrete lead points with color gradients to generate an EEG topology map. This topology map visually presents the contribution of each brain region to the recognition of consciousness using color gradients, enabling doctors to quickly identify key brain regions (e.g., highly clustered areas correspond to darker colors), thereby enhancing the transparency and reliability of the model's decision-making.

[0097] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a computer-readable storage medium storing consciousness level recognition program instructions based on electroencephalogram (EEG) data. When these program instructions are executed by one or more processors, they can be used to implement the embodiments of this application. Figures 1 to 5 The described operation performed by the consciousness level recognition device based on EEG data.

[0098] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0099] Although the embodiments of this application are described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

[0100] Furthermore, the collection and acquisition of various data in this application comply with relevant laws and regulations and are authorized by the data providers. Any organization or individual that needs to obtain external data shall obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor shall it illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A consciousness level recognition device based on electroencephalogram (EEG) data, characterized in that, Includes: processor; A memory storing program instructions for recognizing levels of consciousness based on electroencephalogram (EEG) data, which, when executed by the processor, cause the device to perform the following operations: Acquire multi-lead scalp EEG data from the patient to be identified; The multi-lead scalp EEG data is input into a trained neural network model to perform a consciousness level recognition operation, so as to output a response vector composed of the response intensity of each lead, wherein the response intensity characterizes the degree of correlation between the EEG characteristics of the lead and the target consciousness level. Calculate the distance between the response vector and the average response vector of the target consciousness level group; Based on the distance, the recognition result of the level of consciousness of the patient to be identified is determined; The device further performs the following operations: calculating the distance between the response vector and the average response vector of the target consciousness level group. Based on the spatial location of each lead, each lead is mapped to multiple regions; Using regions as the unit of calculation, the spatial weight of each lead is determined based on its distance from the central lead of its region. The distance is calculated based on the spatial weights of each lead, the response vector, and the average response vector of the target awareness level group. The device uses the following formula to determine the spatial weight of each lead. : in, Let be the distance between the i-th lead and the central lead of its region. This refers to the spatial sensitivity parameter. The device calculates the distance using the following formula. : in, For the response vector, The average response vector of the target awareness group. Spatial weights for each lead, Let i be the response intensity of the i-th lead. Let be the average response intensity of the i-th lead.

2. The device according to claim 1, characterized in that, When the program instructions are executed by the processor, the device further performs the following operations before calculating the distance based on the spatial weights of each lead, the response vector, and the average response vector of the target awareness level group: For each region, a focus score is determined based on the response intensity of the central lead in each region, and the average response intensity and standard deviation response intensity of the central lead. Based on the focus score and preset score of each region, high-aggregation regions among the multiple regions are identified; A dynamic weighting factor is used to weight and amplify the response intensity of each lead in the highly clustered region to obtain a weighted response vector, which is then used to calculate the distance.

3. The device according to claim 2, characterized in that, When the program instructions are executed by the processor, the device uses the following formula to determine the clustering score for each region. : in, For the region Clustering score, For the region The response intensity of the central lead ck is The average response intensity of the target awareness group on the central lead ck. The mean standard deviation of the response intensity of the target awareness level group on the central lead ck.

4. The device according to claim 3, characterized in that, When the program instructions are executed by the processor, the device further performs the following operations: The focus score of each region is mapped to the spatial location of each lead within that region to obtain discrete lead points with color gradients. Interpolation is performed on the discrete lead points with color gradients to generate an EEG topology map.

5. The device according to claim 1, characterized in that, When the program instructions are executed by the processor, the device further performs the following operations to train the neural network model: Acquire multiple multi-lead scalp EEG data from different comatose patients, as well as from the same patient at different stages; Based on the level of consciousness of each patient, as well as the level of consciousness of the same patient at different stages, multiple multi-lead scalp EEG data were labeled to generate multiple multi-lead scalp EEG data with tags. The neural network model was trained using 5-fold cross-validation based on multiple multi-lead scalp EEG data with tags.

6. The device according to claim 5, characterized in that, When the program instructions are executed by the processor, the device further performs the following operations to train the neural network model using five-fold cross-validation: During the training phase, a first number of multi-lead scalp EEG data with labels are input into the neural network model as training data for training. During the verification phase, the second number of multi-lead scalp EEG data with labels is input into the neural network model as verification data to output the response vector of the second number of multi-lead scalp EEG data. The average response vector of each multi-lead scalp EEG data labeled with the target consciousness level is calculated to obtain the average response vector of the target consciousness level group.

7. A computer-readable storage medium storing program instructions for recognizing level of consciousness based on electroencephalogram (EEG) data, wherein when the program instructions are executed by a processor, the operation performed by the EEG-based level of consciousness recognition device according to any one of claims 1-6 is implemented.

Images