Multimodal fusion assisted evaluation information generation method and system and storage medium

Abstract

This invention discloses a method, system, and storage medium for generating multimodal fusion-assisted assessment information. The method acquires behavioral scale scores, EEG data, and clinical status information of the target subject, generating valid EEG signals and channel validity masks. Based on channel spatial locations, representative signals for multiple brain functional regions are generated, region EEG features are extracted, and a reference parameter library is used to complete standardization, feature direction unification, and nonlinear mapping. Based on the mapping relationship, EEG assessment scores for each functional dimension are calculated. Behavioral output limitation parameters are calculated based on clinical status information, and adaptive fusion weights are determined by combining EEG data quality parameters and behavioral assessment quality parameters. Finally, the EEG assessment scores are fused with the normalized behavioral scores to output the consciousness level-assisted classification result. This approach improves the functional interpretability of EEG results and reduces the risk of underestimation due to limited overt responses.

Description

Technical Field

[0001] This invention relates to the field of medical data processing and auxiliary assessment of disorders of consciousness, specifically to a method, system, and storage medium for generating multimodal fusion auxiliary assessment information. Background Technology

[0002] Ancillary assessments of patients with altered consciousness typically rely on behavioral scales, electroencephalography (EEG), imaging studies, and clinical observation. Among these, behavioral scales offer advantages such as clear procedures, high clinical acceptance, and ease of recording overt responses. However, their results are highly dependent on the patient's ability to express responses through eye movements, limb movements, vocalizations, or communication. Patients with severe limb paralysis, locked-in states, dystonia, aphasia, tracheotomy, eyelid closure disorders, the effects of sedative medications, or fluctuating arousal may not fully demonstrate auditory comprehension, visual processing, motor imagery, language processing, or communication-related brain functions in their behavioral scale scores, even if they retain some. This can lead to lower behavioral scores and affect the accuracy and stability of the ancillary assessment information.

[0003] Electroencephalography (EEG) signals, with their advantages of bedside acquisition, high temporal resolution, and ability to reflect cortical electrophysiological activity, have been used for auxiliary assessment of patients with impaired consciousness. However, current EEG analyses mostly focus on single-category indicators such as overall spectral power, complexity, event-related potentials, or functional connectivity. The output results typically remain at the level of overall abnormality or overall level of consciousness probability, making it difficult to establish a stable mapping with functional dimensions such as hearing, vision, motor, speech, communication, and arousal / cognition in behavioral scales. Furthermore, EEG acquisition is also plagued by problems such as bad leads, artifacts, low-quality brain regions, equipment differences, individual differences, and inconsistencies in feature orientation. Without quality constraints, reference parameter standardization, functional dimension mapping, and behavioral output limitation correction mechanisms, unexplained biases can easily arise between EEG results and behavioral results, limiting the reproducibility and clinical reference value of multimodal auxiliary assessment systems. Summary of the Invention

[0004] The purpose of this invention is to address the problems of existing auxiliary assessments of consciousness disorders, which struggle to stably map EEG results to behavioral functional dimensions, and which tend to underestimate residual consciousness levels in patients with limited behavioral output due to insufficient overt responses, as well as the lack of quality constraints, dynamic fusion, and mechanisms for identifying hidden consciousness risks.

[0005] The first aspect of this invention provides a method for generating multimodal fusion-assisted evaluation information, comprising the following steps:

[0006] Acquire multimodal assessment data of the target subject, including behavioral scale scores, electroencephalogram data, and clinical status information; The EEG data is preprocessed and abnormal channel classification is performed to generate valid EEG signals, channel validity masks, brain region quality factors, and EEG data quality parameters. Representative signals of multiple brain functional regions are generated based on the spatial location of the EEG channel. The EEG features of each brain functional region are extracted, and the reference parameter library is used to standardize, unify the direction, and perform nonlinear mapping on the EEG features of the regions to obtain normalized EEG features and generate feature quality factors. Based on the normalized EEG characteristics, feature quality factors, and intra-brain region feature weights, the brain region function retention value for each brain functional region is calculated. Based on the mapping matrix between brain functional regions and behavioral functional dimensions, the brain region quality factor, and the brain region function retention value, candidate EEG assessment scores for each behavioral functional dimension are calculated, and the effective mapping quality is used to determine whether to output the EEG assessment score for that dimension. Based on the clinical status information, behavioral output limitation parameters for each behavioral functional dimension are calculated, and the original behavioral quality parameters are corrected using the behavioral output limitation parameters. Based on the EEG data quality parameters and the corrected behavioral quality parameters, the EEG modality fusion weights are determined using a single-channel compensation method, and the EEG assessment scores are fused with the normalized behavioral scores to generate a functional dimension fusion index. Based on the EEG assessment score, behavioral normalized score, EEG data quality parameters, and functional dimension fusion index, the system outputs concealed consciousness risk warning information, overall auxiliary assessment results, confidence information, and standardized reports in parallel.

[0007] In a second aspect, the present invention provides a multimodal fusion-assisted evaluation system, comprising: The data acquisition module is used to acquire behavioral scale scores, electroencephalogram (EEG) data, and clinical status information of the target subjects. The EEG preprocessing module is used to preprocess the EEG data and perform abnormal channel classification processing to generate valid EEG signals, channel validity masks, brain region quality factors, and EEG data quality parameters. The brain functional area feature extraction module is used to generate representative signals of multiple brain functional areas based on the spatial location of the brain electrical channels, extract the brain electrical features of each brain functional area, and generate feature quality factors. A reference parameter library is used for hierarchical storage of standardized parameters, orientation factors, threshold parameters, slope parameters, brain region feature weights, mapping matrices, clinical state influence weights, effective mapping quality thresholds, and auxiliary assessment threshold parameters. The functional dimension EEG scoring module is used to calculate the EEG assessment score corresponding to each behavioral functional dimension based on normalized EEG features, feature quality factors, intra-brain region feature weights, brain region quality factors, effective mapping quality, and mapping matrix. The multimodal fusion module is used to calculate behavioral output limitation parameters based on clinical status information, correct the original behavioral quality parameters, determine the EEG modal fusion weights using a single-channel compensation method, and generate a functional dimension fusion index. The concealed consciousness risk identification module is used to generate concealed consciousness risk warning information based on EEG assessment scores, behavioral normalization scores, and EEG data quality parameters; The overall auxiliary assessment module is used to generate overall auxiliary assessment results based on the functional dimension fusion index, whole brain network indicators, and data quality admission criteria.

[0008] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the multimodal fusion-assisted assessment information generation method for objects with consciousness disorders as described above.

[0009] This invention performs abnormal channel classification processing on EEG data to generate brain region quality parameters and feature quality parameters. It then standardizes, unifies the direction of, and nonlinearly maps the regional EEG features using a reference parameter library. Next, it generates brain region function retention values ​​using feature weights within each brain region. Finally, it calculates the EEG assessment scores for each functional dimension using a mapping matrix between brain functional regions and behavioral functional dimensions. Therefore, the EEG results no longer merely represent the overall degree of abnormality but can correspond to the output of functional dimensions such as hearing, vision, motor, speech, communication, and arousal / cognition. This reduces the impact of low-quality brain regions and features on EEG scores and improves the dimensional interpretability, computational stability, and reproducibility of the auxiliary assessment results.

[0010] This invention quantifies behavioral output limitation parameters for each behavioral functional dimension based on clinical status information, and uses these parameters to correct behavioral assessment quality parameters. Furthermore, it combines these parameters with EEG data quality parameters to determine EEG modality fusion weights, enabling dynamic fusion of EEG assessment scores and normalized behavioral scores based on the patient's actual behavioral output conditions and data quality. When the target subject exhibits paralysis, aphasia, eye movement disorders, sedation effects, or arousal fluctuations, this invention can reduce underestimation caused by solely relying on low behavioral scores. It also outputs a hidden consciousness risk warning when the EEG assessment score is high, the behavioral score is low, and the EEG quality meets requirements, providing more stable reference information for subsequent review and auxiliary assessment.

[0011] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art are briefly described below. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0013] Figure 1 The overall flowchart of the multimodal fusion-assisted evaluation information generation method provided in the embodiments of the present invention is shown.

[0014] Figure 2 This is a block diagram of the multimodal fusion-assisted evaluation system provided in an embodiment of the present invention.

[0015] Figure 3 The flowchart illustrates the EEG preprocessing, abnormal channel grading processing, and generation of representative signals for brain functional regions provided in this embodiment of the invention.

[0016] Figure 4 A flowchart illustrating the construction of the reference parameter library, threshold calibration, and model verification provided for embodiments of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present invention, and not all embodiments. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention.

[0018] It should be noted that the "auxiliary assessment results," "auxiliary assessment information on level of consciousness," "risk warning information on concealed consciousness," and "functional dimension retention information" mentioned in this invention are all auxiliary assessment information generated by computer equipment based on acquired data. They do not replace the final medical diagnosis made by clinicians, nor do they include treatment steps or treatment plan selection steps.

[0019] This embodiment provides a method and system for generating multimodal fusion-based auxiliary assessment information. The method processes behavioral scale scoring data, electroencephalogram (EEG) data, and clinical status information of a target subject to generate fusion indices for each functional dimension, auxiliary assessment results, concealed consciousness risk warnings, and standardized reports. The main processing steps of this method are as follows: Figure 1 As shown.

[0020] Example 1: Multimodal Fusion-Assisted Evaluation Information Generation Method S1: Multimodal data acquisition.

[0021] The system acquires multimodal assessment data of the target object through the data acquisition module 110. The multimodal assessment data includes behavioral scale scoring data, electroencephalogram data and clinical status information.

[0022] The multimodal data acquisition process includes: acquiring behavioral scale scoring data, acquiring EEG data, acquiring clinical status information, normalizing behavioral scores, calculating behavioral assessment quality parameters 510, importing EEG data and unifying sampling rates, performing EEG segment quality control, and outputting standardized input data.

[0023] The behavioral scale scoring data can be CRS-R scale scoring data or scoring data from other consciousness disorder behavioral scales that correspond to CRS-R. The behavioral scale scoring data includes sub-item scores of multiple behavioral functional dimensions, including auditory function, visual function, motor function, verbal function, communication function, and arousal / cognitive function.

[0024] The system converts the original sub-item scores into normalized behavioral scores between 0 and 1 based on the highest score for each behavioral function dimension. The higher the normalized value, the more pronounced the overt behavioral response in that function dimension.

[0025] Clinical status information includes whether sedative drugs are used, whether there are arousal fluctuations, epileptiform discharges, significant limb paralysis, locked-in state, dystonia, eyelid closure disorder, aphasia, auditory pathway impairment, or visual pathway impairment.

[0026] To enable clinical status information to be quantified and processed by computer devices, this embodiment converts the clinical status information into several status variables, including a motor output impairment variable. Language output barrier variables Variables of eye movement or eyelid disorders Auditory input impairment variables Variables of visual input barriers Variables affected by sedative drugs Awakening fluctuation variables and variables affecting epileptic discharge All the above state variables are normalized to between 0 and 1.

[0027] EEG data can be either resting-state or task-oriented. Task-oriented EEG data can include auditory stimulation tasks, visual stimulation tasks, language comprehension tasks, and motor imagery tasks. EEG acquisition preferably uses a 10-20 or 10-10 international standard lead system, with sampling rates of 250Hz, 500Hz, or 1000Hz. For data from different sampling rates, the system can uniformly resample to the preset sampling rate.

[0028] In one specific implementation, the aforementioned state variables can be quantified as 0, 0.3, 0.6, and 1.0, respectively, indicating no impact, mild impact, moderate impact, and severe impact. This quantification rule can be used as the system's default initialization rule or updated based on historical samples in the reference parameter library 140. In another specific implementation, the grading rules for each normalized state variable can be pre-configured. Taking the motor output impairment variable as an example, the value is 0 when the target object can complete the main motor output corresponding to this functional dimension; 0.3 when there is bradykinesia, decreased muscle strength, or insufficient range of motion but a stable overt response can still be observed; 0.6 when there is significant limb paralysis, dystonia, or unstable motor output, resulting in some assessment items being unreliable; and 1.0 when there is a locked-in state, severe paralysis, or inability to express a response through limb movements.

[0029] Taking the language output disorder variable as an example, the value is 0 when the target can complete the vocalization or language expression related items; 0.3 when there is mild vocalization difficulty but the identifiable response can still be recorded; 0.6 when there is tracheotomy, aphasia or severe articulation disorder that significantly restricts language output; and 1.0 when the target cannot respond through vocalization or language expression.

[0030] Taking eye movement or eyelid disorder variables as an example, the value is 0 when the target subject can stably complete eye movement, fixation or eyelid-related responses; 0.3 when there is a slight delay in eye movement or eyelid response but it can still be observed repeatedly; 0.6 when eye movement, fixation or eyelid response is significantly unstable and affects the judgment of some behavioral items; and 1.0 when the subject cannot express a response through eye movement, fixation or eyelid movement.

[0031] For variables affecting sedative medication, arousal fluctuations, and epileptiform discharges, the system categorizes them based on sedative medication use, arousal stability, the proportion of effective assessment windows, and the proportion of epileptiform discharge fragments during the assessment period. These categorization rules are stored as default rules in the reference parameter library and can be updated based on historical samples, institutional assessment standards, or expert review results.

[0032] S2: EEG preprocessing and abnormal channel classification processing.

[0033] The system uses the EEG preprocessing module 120 to preprocess EEG data, identify abnormal channels, perform region-level quality control, and generate region-representative signals, producing valid EEG signals, channel validity masks, brain region quality factors, and EEG data quality parameters. For details, please refer to [link to detailed process documentation]. Figure 3 as follows.

[0034] S210: Filtering, Notch Filtering, and Artifact Removal The system performs basic preprocessing on the acquired EEG data. Specifically, the system can perform bandpass filtering on the EEG data from 0.5Hz to 45Hz, and select 50Hz or 60Hz power frequency notch filtering depending on the acquisition area to remove low-frequency drift, high-frequency noise, and power frequency interference.

[0035] For electrooculogram (EOG), electromyogram (EMG), and motor artifacts, the system can use independent component analysis, artifact subspace reconstruction, wavelet thresholding, or a combination thereof to process the EEG signals and obtain pre-cleaned signals.

[0036] S220: Bad Lead Identification The system identifies bad leads in the initially cleaned EEG signals. If any EEG channel continuously exhibits abnormal amplitude, abnormal variance, low correlation with neighboring channels, or a large proportion of saturated, broken, or flat signals within a preset time window, the system marks that EEG channel as an abnormal channel.

[0037] For example, if the signal amplitude of a channel exceeds a preset amplitude range within multiple consecutive time windows, or its signal variance is lower than a preset lower limit, or its correlation coefficient with neighboring channels is lower than a preset threshold, then the channel can be identified as an abnormal channel.

[0038] S230: Generate channel validity mask The system generates a channel validity mask based on the bad lead identification results. If a channel is a valid channel, the corresponding mask value is 1; if a channel is an abnormal channel, the corresponding mask value is 0.

[0039] The channel validity mask is used for subsequent channel interpolation, channel weight setting, brain region quality factor calculation, and generation of regional representative signals.

[0040] S240: Establish the correspondence between brain functional regions and channel sets. The system divides the EEG channels into different brain function zones based on their spatial location and preset partitioning rules, forming a channel set corresponding to each brain function zone.

[0041] In one embodiment, the brain functional regions include a prefrontal cortex region 310, a motor cortex region 320, a temporal lobe region 330, an occipital lobe region 340, a posterior parietal lobe region 350, and a whole-brain network region 360.

[0042] For example, the prefrontal cortex region 310 may include channels such as Fp1, Fp2, F3, F4, F7, F8, and Fz; the motor cortex region 320 may include channels such as C3, C4, Cz, FC3, FC4, CP3, and CP4; the temporal lobe region 330 may include channels such as T7, T8, P7, and P8; the occipital lobe region 340 may include channels such as O1, O2, and Oz; the posterior parietal lobe region 350 may include channels such as P3, P4, Pz, PO3, and PO4; and the whole-brain network region 360 is used to characterize the cross-regional connectivity between multiple brain functional regions.

[0043] It is important to note that the brain functional partitioning in this step does not involve re-anatomically segmenting the EEG signals. Instead, it assigns the acquired EEG channels to corresponding brain functional partitions based on a pre-defined lead-brain region correspondence for partition-level quality control and subsequent feature extraction. The whole-brain network partition is not a set of channels corresponding to a single anatomical brain region, but rather a cross-regional network functional unit composed of representative signals from multiple brain functional partitions. For whole-brain network partitioning, the system does not use the weighted average signal of a single channel set as the sole representative signal. Instead, it calculates network metrics such as global efficiency, average connectivity strength, prefrontal-parietal connectivity strength, clustering coefficient, or node strength based on the connection matrix between the prefrontal cortex, motor cortex, temporal lobe, occipital lobe, and posterior parietal lobe regions.

[0044] When whole-brain network regions are used in EEG scoring for arousal / cognitive function dimensions, their quality factors are jointly determined by the quality factors of multiple brain functional regions involved in the network calculation, the proportion of effective connectivity edges, and the proportion of available EEG segments. This avoids misinterpreting whole-brain network regions as single spatial brain regions and ensures consistency in the calculation methods for cross-regional network indicators and ordinary region features in subsequent fusion.

[0045] S250: Statistically analyze the proportion of abnormal channels according to brain functional regions and perform graded processing. The system uses a channel validity mask to calculate the proportion of abnormal channels in each brain functional region to the total number of channels in that region, and then performs hierarchical processing based on the proportion of abnormal channels.

[0046] When the proportion of abnormal channels in a certain brain functional area does not exceed 30%, the system uses weighted interpolation of nearby effective channels or spherical spline interpolation to reconstruct the abnormal channels, so that the brain functional area can still participate in subsequent EEG feature extraction and EEG score calculation.

[0047] When the proportion of abnormal channels in a certain brain functional region is greater than 30% but not more than 60%, the system does not force the reconstruction of all abnormal channels. Instead, it resets the weight of abnormal channels to zero during the generation of representative signals for subsequent regions and reduces the quality parameters of the brain functional region, so that the region participates in subsequent calculations in a deweighted manner.

[0048] When the proportion of abnormal channels in a certain brain functional area exceeds 60%, the system marks that brain functional area as a low-quality brain region. For low-quality brain regions, the system sets its brain region quality factor to 0, or to a low-quality value not higher than 0.2. Preferably, when calculating high-confidence EEG scores, the low-quality brain region is not included in the calculation of the corresponding behavioral functional dimension's EEG score.

[0049] S260: Calculating brain region quality factors The system generates brain region quality factors for each brain functional region. The brain region quality factor is used to represent the first... The reliability of data collected in this study on individual brain functional regions.

[0050] In one specific implementation, brain region quality factor Calculate using the following formula:

[0051] in, Indicates the first The effective channel ratio of each brain functional region; This indicates the proportion of effective EEG segments in that region; This represents the normalized signal-to-noise ratio of the partition; This indicates the interpolation channel ratio for that partition; This indicates that the calculation results will be limited to between 0 and 1. The weights mentioned above are exemplary initial weights for the system. The system can optimize these initial weights based on historical samples. During the optimization process, the goal is to minimize the difference between the original behavioral quality parameters and expert quality annotations, consistency of repeated evaluations, or subsequent review results, constraining each weight to be no less than 0 and the sum of the weights to be 1. After independent validation with validation samples, the weights that meet the preset consistency and stability requirements are written into the reference parameter library.

[0052] If the first If the proportion of abnormal channels in a brain functional area exceeds 60%, the system will... Set it to 0, or a low quality value no higher than 0.2, and reduce or exclude the contribution of this brain functional area in subsequent EEG scoring.

[0053] S270: Determine the set of valid channels for a partition and the partition channel weights. After completing the abnormal channel classification and brain region quality factor calculation, the system determines the final effective channel set and corresponding channel weights for each brain functional region to participate in the generation of representative signals.

[0054] For normal channels, their channel weights are determined based on the channel signal-to-noise ratio, stability, and channel validity mask. For abnormal channels that have been reconstructed through interpolation, their channel weights can be lower than those of normal channels. For abnormal channels that have not been reconstructed, their channel weights are reset to 0. For regions marked as low-quality brain areas, the channels within these regions are not included in the calculation of high-confidence EEG scores, or are only retained as low-confidence reference information.

[0055] Through this step, the system clarifies which channels in each brain functional region participate in the calculation, with what weight each channel participates in the calculation, and the reliability of that brain functional region in subsequent scoring.

[0056] S3: Brain functional regions and feature extraction.

[0057] The brain functional region feature extraction module 130 divides the effective brain electrical signals into multiple brain functional regions based on the spatial location of the brain electrical channels, and extracts the brain electrical features of each region.

[0058] Brain functional regions include the prefrontal cortex region, motor cortex region, temporal lobe region, occipital lobe region, posterior parietal lobe region, and whole-brain network region.

[0059] In one embodiment, the prefrontal cortex region includes Fp1, Fp2, F3, F4, F7, F8, and Fz; the motor cortex region includes C3, C4, Cz, FC3, FC4, CP3, and CP4; the temporal lobe region includes T7, T8, P7, and P8; the occipital lobe region 340 includes O1, O2, and Oz; the posterior parietal lobe region 350 includes P3, P4, Pz, PO3, and PO4; and the whole-brain network region is used to characterize the cross-regional connectivity between multiple brain functional regions.

[0060] For any brain functional region, the system generates a region representative signal based on the channel validity mask and channel quality score. The region representative signal is calculated using the following weighted average method:

[0061] in, Indicates the first Individual brain functional areas in time The representative signal; Indicates the first A collection of channels within each functional brain region; Indicates the first The EEG signals from each channel after cleaning or reconstruction; Indicates the first The weights of each channel are determined based on the channel validity mask and the channel quality score; To prevent constants with a denominator of zero.

[0062] Regional EEG features are extracted based on representative signals from each brain functional region. These regional EEG features include at least two of the following: frequency domain features, complexity features, event-related potential (ERP) features, and functional connectivity features. Frequency domain features may include absolute power, relative power, and frequency band ratios for frequency bands such as δ, θ, α, β, and γ; complexity features may include sample entropy, permutation entropy, and Lempel-Ziv complexity; EEPROM features may include N100, P300, N400, visual evoked potentials, or motor imagery-related desynchronization features; and functional connectivity features may include coherence, phase lock value, weighted phase lag exponent, prefrontal-parietal connectivity strength, global efficiency, node strength, and clustering coefficient.

[0063] Generate a feature quality factor for each EEG feature. In one specific implementation, if a product-type quality assessment method is adopted, the characteristic quality factor... Calculate using the following formula:

[0064] in, Indicates the calculation of the first The first brain functional region When considering individual EEG features, the normalized value of the number of segments can be used; This represents the normalized signal-to-noise ratio of the segment corresponding to this feature; This represents the proportion of effective trials for task-state features; for resting-state features, it can be set to 1. Indicates the first Channel effectiveness indicators for individual brain functional regions.

[0065] In another specific implementation, the characteristic quality factor Alternatively, it can be calculated using the following weighted method:

[0066] in, Normalized value representing the number of available fragments This represents the normalized signal-to-noise ratio. This represents the proportion of effective trials for task-state features; for resting-state features, it can be set to 1. This represents the channel effectiveness index corresponding to the brain's functional regions. If a certain feature cannot be calculated, then... This feature is not included in the subsequent calculation of brain region functional retention values.

[0067] The frequency domain features, complexity features, event-related potential features, and functional connectivity features mentioned above can all be calculated using conventional algorithms in this field. This embodiment will not elaborate on each of the known calculation formulas.

[0068] S4: Reference parameter library standardization and orientation mapping.

[0069] The system calls reference parameter library 140 to standardize, unify the direction, and perform nonlinear mapping on the regional EEG features to obtain normalized EEG features.

[0070] S4 can further include calling the reference parameter library 140, selecting matching hierarchical parameters, performing standardization processing, unifying feature directions, performing Sigmoid nonlinear mapping, and outputting normalized EEG features.

[0071] Reference parameter library 140 stores parameters hierarchically according to device model, lead layout, acquisition paradigm, preprocessing procedure, and population attributes. For any brain functional region and any EEG feature, reference parameter library 140 stores at least the mean / standard deviation, median / MAD, orientation factor, threshold parameter, slope parameter, brain region intra-region feature weight, brain region-functional dimension mapping matrix, clinical state influence weight, quality factor generation rules, and threshold calibration parameters.

[0072] When a reference sample matching the target object's acquisition conditions exists, and the sample size meets preset requirements, the system uses the mean and standard deviation of the reference sample for Z-score standardization. When the reference sample size is insufficient, the feature distribution is significantly skewed, or the proportion of outliers exceeds a preset proportion, the system uses the median and median absolute deviation for robust standardization. When no external reference library is available, the system can use the median and median absolute deviation of the effective EEG segments of the current target object for intra-patient relative standardization, and label this result as an intra-patient relative assessment result in the report.

[0073] To avoid fusion errors caused by opposing directions of different EEG features, feature directions are unified before nonlinear mapping:

[0074] in, Indicates the first The first brain functional region Standardized values ​​of individual EEG features; This represents the orientation factor stored in reference parameter library 140; a larger EEG characteristic value indicates a higher degree of functional preservation. When the value of this EEG characteristic is higher, it indicates a greater degree of functional impairment. ; This represents the eigenvalues ​​after the direction is unified.

[0075] The system then inputs the directionally unified feature values ​​into a nonlinear mapping function to obtain normalized EEG features between 0 and 1:

[0076] in, Indicates normalized EEG characteristics; This refers to the slope parameter stored in reference parameter library 140; This represents the threshold parameter stored in reference parameter library 140.

[0077] S5: Functional Dimension EEG Score.

[0078] The system calculates the EEG assessment scores corresponding to each behavioral functional dimension through the partition-function mapping module 150 and the functional dimension EEG scoring module 160.

[0079] The system divides the EEG channels into prefrontal cortex, motor cortex, temporal lobe, occipital lobe, posterior parietal lobe, and whole brain network regions according to spatial location. It also breaks down the behavioral scale scores into auditory function, visual function, motor function, speech function, communication function, and arousal / cognitive function dimensions.

[0080] Among them, the prefrontal cortex region is mainly related to the dimensions of speech function, communication function, and arousal / cognitive function; the motor cortex region is mainly related to the dimension of motor function; the temporal lobe region is mainly related to the dimensions of auditory function and speech function; the occipital lobe region is mainly related to the dimension of visual function; the posterior parietal lobe region is mainly related to the dimensions of visual function, motor function, and communication function; and the whole brain network region is mainly related to the dimension of arousal / cognitive function.

[0081] The system uses a mapping matrix M to represent the correspondence between the aforementioned brain functional regions and behavioral functional dimensions. Each row in the mapping matrix M corresponds to a behavioral functional dimension, and each column corresponds to a brain functional region. The matrix elements represent the spatial mapping contribution weight of the corresponding brain functional region to the corresponding behavioral functional dimension. Through this mapping matrix M, the system can convert the brain region functional retention values ​​of each brain functional region into EEG assessment scores corresponding to each behavioral functional dimension, thereby establishing a correspondence between EEG analysis results and functional dimensions in the behavioral scale. The mapping matrix M is used to characterize the contribution relationship between different brain functional regions to different behavioral functional dimensions.

[0082] To avoid the overlapping of weights for features within brain regions and weights for the brain region-function dimension mapping, this embodiment defines the meanings of the two types of weights in a hierarchical manner: weights for features within brain regions. Used only to indicate the first The contribution relationships between different EEG features within a single brain functional region do not represent the spatial contribution of that brain functional region to the behavioral functional dimension; mapping matrix Used only to indicate the first The brain functional regions are related to the first The spatial mapping contribution of each behavioral functional dimension no longer participates in the fusion of features within brain regions. The two types of weights are normalized separately and do not substitute for each other.

[0083] The system first calculates the first Functional retention values ​​of individual brain functional regions :

[0084] in, Indicates the first Functional retention values ​​of individual brain functional regions; Indicates the first The first brain functional region Normalized EEG characteristics; Indicates the first Within the first brain functional region Weights of brain region features for each EEG feature; Indicates the first The first brain functional region Characteristic quality factors of individual EEG features; To prevent constants with a denominator of zero.

[0085] Subsequently, the system combines the mapping matrix M and brain region quality factors. Calculate the candidate EEG assessment scores for any behavioral functional dimension:

[0086] in, Indicates the first Candidate EEG assessment scores corresponding to each behavioral functional dimension; Indicates the first The brain functional regions are related to the first Mapping weights for each behavioral functional dimension; Indicates the first Brain region quality factors for each functional brain region; Indicates the first The brain region functional retention value of each brain functional area.

[0087] To avoid the unilateral amplification of the remaining few minor brain regions after the primary mapping brain regions have become ineffective, this embodiment further calculates the... The effective mapping quality of each behavioral functional dimension and:

[0088] when Not lower than the preset effective mapping threshold At that time, As the first EEG assessment scores for each behavioral functional dimension; when Instead of performing unilateral normalization amplification calculations on the remaining brain regions, the EEG assessment score for that dimension is marked as invalid, low confidence, or assigned a preset default low confidence value.

[0089] In one specific implementation, a preset effective mapping threshold is defined. It can be 0.5. When a whole-brain network region experiences an excessively high proportion of abnormal channels or insufficient effective segments... And make the corresponding arousal / cognitive function dimensions When the score is below 0.5, instead of using the remaining few brain regions such as the prefrontal cortex to forcibly amplify the EEG assessment score for this dimension, we mark the EEG score for this dimension as having low confidence in the report and suggest supplementing EEG collection or repeating the assessment.

[0090] In one specific implementation, the mapping matrix M can be initialized as follows: The columns of the matrix correspond to the prefrontal cortex region, motor cortex region, temporal lobe region, occipital lobe region, posterior parietal lobe region, and whole-brain network region, respectively; the rows of the matrix correspond to the auditory function dimension, visual function dimension, motor function dimension, speech function dimension, communication function dimension, and arousal / cognitive function dimension, respectively.

[0091] The sum of the weights of each row in the matrix above is 1. This matrix is ​​only an example initialization matrix, and the system can update it based on historical samples, expert consensus labels, and follow-up results from the reference parameter library 140.

[0092] In one specific implementation, the determination of the mapping matrix M includes three stages: initialization, constraint optimization, and verification.

[0093] During the initialization phase, the system generates an initial mapping matrix based on the preset correspondence between brain functional areas and behavioral functional dimensions, and normalizes each row in the matrix so that the sum of the mapping weights of each brain functional area corresponding to the same behavioral functional dimension is 1.

[0094] During the constraint optimization phase, the system optimizes the mapping matrix M based on the functional dimension reference scores from historical samples. These functional dimension reference scores can be generated by combining multiple behavioral review results, task-based EEG re-examination results, expert consensus records, and follow-up results. The optimization aims to minimize the difference between the fusion index of each functional dimension and the functional dimension reference score, and constrains the matrix elements in mapping matrix M to be non-negative and the sum of the weights in each row to be 1.

[0095] During the verification phase, the system uses independent test samples to verify the updated mapping matrix M. When the updated mapping matrix M outperforms the initial matrix in terms of macro-average F1 score, test-retest consistency, functional dimension-level score consistency, or EEG-behavioral inconsistency recognition rate, and its specificity is not lower than the preset requirements, the mapping matrix M is written into the reference parameter library; otherwise, the initial matrix or the previous version of the mapping matrix M continues to be used.

[0096] Through the above processing, the mapping matrix M is not simply determined by experience, but is determined by initialization rules, historical sample calibration and independent test verification.

[0097] S6: Generation of the index for the integration of behavioral output limitation correction and functional dimension fusion.

[0098] Using the multimodal fusion module 170, based on clinical status information, behavioral assessment quality parameters, EEG data quality parameters, EEG assessment scores, and normalized behavioral scores, the module calculates behavioral output limitation parameters, behavioral usability factors, corrected behavioral quality parameters, EEG modal fusion weights, and functional dimension fusion indices for each behavioral functional dimension. The processing results of S6 serve as input for subsequent hidden consciousness risk identification and overall auxiliary assessment result generation.

[0099] In this embodiment, S6 includes: calculating behavioral output constraint parameters, calculating behavioral usability factors, correcting behavioral quality parameters, determining EEG modality fusion weights, and generating functional dimension fusion indices. To avoid confusion between quality parameters at different levels, in this embodiment, the brain region quality factor is denoted as... , used to represent the reliability of the data in the b-th brain functional region; the overall EEG quality parameter is used to represent the overall data quality of this EEG acquisition; the functional dimension-level EEG data quality parameter is denoted as , used to represent the reliability of the EEG score corresponding to the f-th behavioral functional dimension. When determining the weights for EEG modality fusion, the system prioritizes functional dimension-level EEG data quality parameters. Instead of directly using overall EEG quality parameters, the system assigns an excessively high EEG fusion weight to a specific behavioral functional dimension if the quality of the main brain functional area corresponding to that dimension is insufficient, even if the overall EEG data quality is high.

[0100] First, the system calculates the first [number] based on the clinical status information. Constrained parameters for behavioral output in each behavioral functional dimension The behavioral output constraint parameter represents the degree of limitation on the target object's expression of its true response through overt behavior within this behavioral functional dimension. The behavioral output constraint parameter can be calculated as follows:

[0101] in, Represents the obstacle variable in motion output. Represents language output barrier variables. Indicates variables related to eye movement or eyelid disorders. Represents variables related to auditory input impairment. Represents variables related to visual input barriers. Indicates the variable affecting sedative drugs. Represents the arousal fluctuation variable, Indicates variables affecting epileptiform discharges; This indicates the weight of the influence of each clinical state variable on different behavioral function dimensions; This means that the calculation result will be limited to between 0 and 1.

[0102] Then, the system outputs restricted parameters based on the behavior. Computational behavioral usability factor :

[0103] in, Indicates the first Behavioral usability factors for each behavioral functional dimension; This represents the attenuation strength parameter. The larger the behavioral output limitation parameter, the smaller the behavioral usability factor, indicating that the behavioral score in this functional dimension is more significantly affected by the limitation of explicit responses.

[0104] The behavioral assessment quality parameter 510 can be denoted as the raw behavioral quality parameter. In one implementation, the original behavioral quality parameters are determined by item completion, repeatability, arousal stability, and assessment interference.

[0105] in, Indicates the first The completion rate of assessment items for each behavioral functional dimension. This indicates consistency across repeated tests. Indicates the stability of the waking state. This indicates the degree to which the assessment process was undisturbed. All indicators are normalized to between 0 and 1. Among them, project completion and repeatability directly affect the completeness and reproducibility of the behavioral assessment results, therefore they are assigned relatively high initial weights of 0.35 and 0.35 respectively; arousal stability affects whether the target subject can consistently demonstrate responsiveness within the assessment time window, and is assigned an initial weight of 0.20; the degree to which the assessment was undisturbed reflects the external influences of the environment, equipment, and operational processes, and is assigned an initial weight of 0.10. All weights are non-negative and sum to 1.

[0106] The system utilizes behavioral usability factors The original behavioral quality parameters were corrected to obtain the corrected behavioral quality parameters:

[0107] in, Indicates the first The modified behavioral quality parameter corresponds to each behavioral function dimension. This parameter indicates the usability of the behavioral score within that function dimension after considering behavioral output limitations.

[0108] EEG data quality parameters can be denoted as: In one implementation, the system uses a mapping matrix. Brain region quality factors And the effective mapping threshold determines the first EEG data quality parameters corresponding to each behavioral functional dimension:

[0109] Indicates the first The effective mapping quality of each behavioral functional dimension and This sets a preset effective mapping threshold. When... When the data quality is below the preset effective mapping threshold, it indicates that the quality of the main brain regions corresponding to the functional dimension is insufficient. The system sets the EEG data quality parameter of the dimension to 0 to avoid unilaterally amplifying the EEG score by using a few residual brain regions.

[0110] Subsequently, the system calculates the EEG modality fusion weights using a single-channel compensation method. This single-channel compensation method means that only the behavioral quality parameter is reduced through the behavioral usability factor to reflect the impact of behavioral output limitation on behavioral scores, without applying an additional amplification term based on the behavioral output limitation parameter to the EEG data quality parameters. The EEG modality fusion weights are calculated as follows:

[0111] in, Indicates the first The fusion weights of EEG modalities under each behavioral functional dimension are determined. In this way, when behavioral output is severely restricted but EEG quality meets the requirements, the behavioral quality parameter decreases and the EEG modal weight increases accordingly; when the EEG quality is insufficient or the effective mapping quality is below the threshold, the EEG modal weight decreases or is set to zero.

[0112] It should be noted that the single-channel compensation method does not directly improve the EEG data quality parameters, nor does it multiply the EEG modality weights by an additional behavioral output limitation parameter. Instead, it reduces the behavioral availability of the corresponding behavioral functional dimension by using the behavioral output limitation parameter, thereby decreasing the corrected behavioral quality parameters. This increases the relative proportion of EEG modalities in the fusion weights when the EEG quality meets the requirements.

[0113] When the target subject has severe motor output impairment, language output impairment, or eye movement impairment, this method can reduce the impact of insufficient overt behavioral responses on behavioral scores; when the quality of EEG data is insufficient or the quality of effective mapping is insufficient, this method will not forcibly increase the weight of EEG modalities due to limited behavioral output.

[0114] Finally, the system will... EEG assessment scores corresponding to each behavioral functional dimension Behavioral Normalized Scores The functions are integrated to generate a functional dimension integration index. :

[0115] in, Indicates the first Fusion index of behavioral functional dimensions; This indicates the EEG assessment score corresponding to this functional dimension; This represents the behavioral normalized score corresponding to this functional dimension; This represents the fusion weight of EEG modalities within this functional dimension.

[0116] After processing by S6, the system obtains the EEG data quality parameters corresponding to each behavioral functional dimension. Corrected behavioral quality parameters EEG modal fusion weights Integration index with functional dimensions .in, , , and It will serve as input for S7 concealed awareness risk identification.

[0117] S7: Risk identification of concealed consciousness based on EEG-behavioral inconsistency.

[0118] After generating the functional dimension fusion index of each behavioral functional dimension in S6, the concealed consciousness risk identification module 180 further judges whether there is a significant inconsistency between the EEG assessment results and the behavioral scores, and generates concealed consciousness risk warning information based on the judgment results.

[0119] It should be noted that S7 does not repeat the calculation of EEG assessment scores or functional dimension fusion, but instead utilizes the EEG assessment scores already generated in S5 and S6. Behavioral Normalization Scores EEG data quality parameters and functional dimension integration index Consistent assessments and risk warnings are made for each behavioral functional dimension.

[0120] Specifically, the system calculates the EEG-behavioral difference value for each behavioral functional dimension:

[0121] in, Indicates the first The difference between EEG assessment scores and normalized behavioral scores for each behavioral functional dimension. The larger the value, the more significantly the degree of functional preservation indicated by the EEG results in that dimension is higher than the behavioral score performance.

[0122] For any behavioral functional dimension, the system will trigger a risk warning regarding the concealment of awareness in that dimension when the following conditions are met simultaneously:

[0123]

[0124]

[0125]

[0126] in, This represents the EEG retention threshold, used to determine whether there is relatively preserved EEG activity in this dimension; This represents the threshold for low behavioral response, used to determine whether the explicit behavioral response in this dimension is too low. This represents the EEG-behavioral difference threshold, used to determine whether the difference between the EEG assessment score and the normalized behavioral score reaches a prompting level. This represents the EEG quality threshold, used to ensure that risk warnings are based on EEG data of sufficient quality.

[0127] The technical meaning of the above four conditions is as follows: Under the premise that the EEG quality meets the requirements, if the EEG assessment score of a certain functional dimension reaches the retention level, while the corresponding behavioral normalized score is at the low response level, and the difference between the two reaches the preset difference threshold, it indicates that there may be an inconsistency between "EEG prompt function retention and behavioral performance inadequate" in that functional dimension. Based on this, the system generates a hidden consciousness risk warning.

[0128] If any condition is not met, the system will not trigger a hidden awareness risk warning for that dimension. Specifically, when Even at that time If the value exceeds the EEG preservation threshold, the system does not trigger a risk warning, but instead marks that dimension as having insufficient EEG quality; when When, it indicates that the EEG did not reach the level required to retain the cue; when When, it indicates that the behavioral score is not a low response; when At this time, it indicates that the difference in EEG-behavioral data is insufficient to trigger a risk warning.

[0129] When generating a risk warning for concealed consciousness, the system can output the behavioral functional dimensions that trigger the risk, EEG assessment scores, behavioral normalization scores, EEG-behavioral variance values, EEG data quality parameters, and suggested review methods in the standardized report. For example, when the motor function dimension triggers a risk warning for concealed consciousness, the system can output a suggestion for EEG review in the motor imagery task state; when the verbal function dimension triggers a risk warning for concealed consciousness, the system can output a suggestion for EEG review in the language comprehension task state.

[0130] The information on the risk of concealed consciousness is output in parallel with the overall auxiliary assessment results. In other words, the risk of concealed consciousness is not a necessary prerequisite for generating auxiliary cues for low-response states, nor does it directly replace the overall auxiliary assessment results. Instead, it serves as a verification cue for EEG-behavioral inconsistencies and is written into the standardized report along with the functional dimension fusion index, the overall auxiliary assessment results, and confidence information.

[0131] S8: Auxiliary assessment results and report output.

[0132] The system generates overall auxiliary assessment results, confidence information, and standardized reports through the overall auxiliary assessment module 190, the confidence calculation module 200, and the report generation and visualization module 210.

[0133] After identifying the risk of concealed awareness, an overall auxiliary assessment result is generated, the confidence level is calculated, and a standardized report and visualization results are output.

[0134] To avoid a circular relationship between the overall auxiliary assessment results and the hidden consciousness risk warning, this embodiment sets them up as parallel outputs. The overall auxiliary assessment results are mainly determined by the fusion index of each functional dimension, whole-brain network indicators, and data quality admission criteria; the hidden consciousness risk warning serves as an overlay or verification warning, and is not a necessary prerequisite for the "low-reaction state auxiliary warning".

[0135] In one implementation, the system can use a rule-based model to generate auxiliary assessment results. For example, when the fusion index of the communication function dimension or the speech function dimension is not lower than the high retention threshold, the system outputs auxiliary prompts related to a higher level of consciousness; when the fusion index of at least one of the visual function dimension, auditory function dimension, or motor function dimension is not lower than the medium retention threshold, and the conditions for prompting a higher level of consciousness are not met, the system outputs auxiliary prompts related to partial function retention; when the fusion index of each function dimension is lower than the low retention threshold, and the network integration index corresponding to the whole-brain network partition is lower than the whole-brain network integration threshold, the system outputs auxiliary prompts related to a low-responsiveness state. If there is also a hidden consciousness risk prompt, an EEG-behavioral inconsistency review prompt is superimposed on the report.

[0136] The report generation and visualization module 210 generates a standardized report. The standardized report includes basic information of the target subject, assessment time, behavioral scale scores, EEG assessment scores for each functional dimension, fusion index for each functional dimension, EEG-behavioral difference results, risk warning of concealed consciousness, EEG quality description, overall auxiliary assessment results, confidence information, and follow-up recommendations.

[0137] Example 2: Multimodal Fusion-Assisted Evaluation System like Figure 2 As shown, this embodiment also provides a multimodal fusion-assisted assessment system. The system includes a data acquisition module 110, an EEG preprocessing module 120, a brain functional region feature extraction module 130, a reference parameter library 140, a region-function mapping module 150, a functional dimension EEG scoring module 160, a multimodal fusion module 170, a concealed consciousness risk identification module 180, an overall assisted assessment module 190, a confidence calculation module 200, and a report generation and visualization module 210.

[0138] The data acquisition module 110 is used to acquire behavioral scale scores, electroencephalogram (EEG) data, and clinical status information of the target subject.

[0139] The EEG preprocessing module 120 is used to filter, notch filter, identify bad leads, grade abnormal channels, remove artifacts, segment and perform quality control on EEG data, and generate valid EEG signals, channel validity masks, brain region quality factors and EEG data quality parameters.

[0140] The brain functional region feature extraction module 130 is used to generate representative signals of the prefrontal cortex region, motor cortex region, temporal lobe region, occipital lobe region, posterior parietal lobe region and whole brain network region according to the spatial location of the channel, and extract at least two of the following: frequency domain features, complexity features, event-related potential features and functional connectivity features.

[0141] Reference parameter library 140 is used to store standardized parameters, orientation factors, threshold parameters, slope parameters, brain region feature weights, mapping matrix parameters, clinical status influence weights, quality parameter generation rules, effective mapping quality thresholds, and model parameters in a hierarchical manner according to device model, lead layout, acquisition paradigm, preprocessing process, and population attributes.

[0142] The partition-function mapping module 150 is used to map the brain functional partitions to the auditory function dimension, visual function dimension, motor function dimension, speech function dimension, communication function dimension, and arousal / cognitive function dimension according to the mapping relationship between brain functional partitions and behavioral function dimensions.

[0143] The functional dimension EEG scoring module 160 is used to assess brain region functional retention values. Brain region quality factors Effective mapping quality and And the mapping matrix M calculates the EEG assessment scores corresponding to each behavioral functional dimension.

[0144] The multimodal fusion module 170 is used to calculate behavioral output limitation parameters based on clinical status information, and to determine adaptive fusion weights based on behavioral output limitation parameters, behavioral assessment quality parameters 510 and EEG data quality parameters, thereby generating fusion indices for each functional dimension.

[0145] The concealed consciousness risk identification module 180 is used to identify functional dimensions with significant inconsistencies between EEG and behavior based on EEG assessment scores, behavioral normalization scores, EEG-behavioral discrepancies, and EEG data quality parameters, and to generate concealed consciousness risk warning information.

[0146] The overall auxiliary assessment module 190 is used to generate overall auxiliary assessment results based on the fusion index of each functional dimension, whole brain network indicators, and data quality admission criteria.

[0147] The confidence calculation module 200 is used to generate auxiliary assessment confidence based on EEG data quality, behavioral assessment quality, modal consistency, and model output probability.

[0148] The report generation and visualization module 210 is used to generate standardized evaluation reports and visualize the results.

[0149] Example 3: Construction of Reference Parameter Library, Threshold Calibration, and Model Validation like Figure 4As shown, this embodiment also provides a method for constructing a reference parameter library, threshold calibration, and model verification.

[0150] First, the system performs S910 to collect historical samples. The historical samples include raw EEG data, preprocessed EEG features, behavioral scale scores, clinical status information, expert consensus labels, and follow-up results.

[0151] Subsequently, the system performs S920 sample stratification. Sample stratification factors include device model, lead layout, acquisition paradigm, preprocessing procedure, and population attributes. These population attributes may include age group, etiological type, disease stage, and auxiliary assessment labels related to level of consciousness.

[0152] Next, the system executes S930 to calculate statistical parameters. For each brain functional region and EEG characteristics under each layer, the system calculates the mean, standard deviation, median, and median absolute deviation, and forms the basic statistical parameters in the reference parameter library 140.

[0153] Subsequently, the system performs S940 calibration of the direction factor, threshold parameter, and slope parameter. The direction factor is used to unify the directional relationship between EEG features and the degree of functional preservation; the threshold parameter is used to determine the intermediate discriminant position in the feature nonlinear mapping; and the slope parameter is used to control the steepness of the nonlinear mapping curve.

[0154] Then, the system performs S950 optimization of the mapping matrix and feature weights. The system can optimize the feature weights within brain regions based on expert consensus results, follow-up results, and functional dimension reference results from historical samples. Mapping matrix M between brain functional regions and behavioral functional dimensions; weight matrix of clinical status influence. Effective mapping quality threshold , and rules for deweighting low-quality brain regions.

[0155] Next, the system executes S960 to divide the system into a construction set, a calibration set, and a test set. The construction set is used to form the basic statistical parameters in the reference parameter library 140; the calibration set is used to determine the threshold parameter, slope parameter, mapping matrix, fusion weight parameter, etc. , , , , , In addition to the overall auxiliary evaluation threshold; the test set is used to independently verify system performance.

[0156] To make the multi-parameter adaptive calibration process executable, this embodiment can adopt the following optimization objective function:

[0157] in, This represents the set of parameters to be calibrated, including brain region feature weights, mapping matrices, clinical state influence weights, orientation mapping parameters, quality threshold parameters, and threshold parameters. Indicates the first The historical sample in the first Integration index across functional dimensions; This represents the standardized form used to characterize the first... The historical sample in the first A continuous reference score for the actual retention of each functional dimension. The continuous reference score can be obtained by mapping the performance information corresponding to the functional dimension based on multiple behavioral review results, task-based EEG review results, expert consensus records, or follow-up. It is not directly equivalent to a disease diagnosis classification label. The labels represent the overall auxiliary assessment categories of historical samples. This indicates the overall auxiliary evaluation category of the system output; and This is a weighting factor.

[0158] In one implementation, the system may use a construction set to determine basic statistical parameters, a calibration set for grid search or constraint optimization, and a test set for independent verification. For You can search in 1, 1.5, 2, 2.5, 3; for , , , , The search can be performed in steps of 0.05 or 0.10 within the range of 0 to 1. The search objective can be to maximize the macro-average F1 value while meeting the minimum specificity requirement, or to maximize the weighted sum of sensitivity and specificity. If multiple parameter combinations yield similar results, the combination with smaller parameter variations and more stable model output should be selected first.

[0159] The system further performs independent testing and validation using the S970. Validation metrics may include accuracy, sensitivity, specificity, macro-average F1 score, EEG-behavioral inconsistency recognition rate, test-retest consistency, and confidence intervals.

[0160] Finally, the system executes S980 to update the reference parameter library 140. Statistical parameters, direction factors, threshold parameters, slope parameters, mapping matrix parameters, rules for generating behavioral output constraints, rules for generating quality parameters, effective mapping quality thresholds, fusion weight parameters, and model parameters, verified through independent testing, can be written into the reference parameter library 140 for use in generating auxiliary evaluation information for the target object.

[0161] In one specific embodiment, target subject A completes an auxiliary assessment of consciousness impairment. The system acquires behavioral scale scores, EEG data, and clinical status information of target subject A through data acquisition module 110. The behavioral scale scores include sub-item scores for functional dimensions such as hearing, vision, motor, speech, communication, and arousal / cognition; the EEG data includes resting-state EEG data and task-oriented EEG data; the clinical status information shows that target subject A has significant motor output impairment, some degree of language output limitation, and mild arousal fluctuations.

[0162] The system first normalizes the behavioral scale scores to obtain normalized behavioral scores for each behavioral function dimension. Because target subject A has motor output impairment and limited language output, his normalized behavioral scores for the motor function dimension and the verbal function dimension are low. However, this score only indicates a weak overt behavioral response and does not directly equate to a complete loss of the corresponding brain function.

[0163] Subsequently, the system uses the EEG preprocessing module 120 to filter, notch, remove artifacts, and identify abnormal channels in the EEG data. For brain functional regions with an abnormal channel ratio of no more than 30%, the system uses neighboring effective channel weighted interpolation or spherical spline interpolation for reconstruction; for brain functional regions with an abnormal channel ratio greater than 30% but not exceeding 60%, the system resets the abnormal channel weights to zero and reduces the corresponding brain region quality factor; for brain functional regions with an abnormal channel ratio exceeding 60%, the system marks them as low-quality brain regions and prohibits them from participating in the calculation of high-confidence EEG scores.

[0164] Subsequently, the system generates representative signals for prefrontal cortex, motor cortex, temporal lobe, occipital lobe, posterior parietal lobe, and whole-brain network regions based on the spatial location of the EEG channels. Frequency domain features, complexity features, event-related potential features, and functional connectivity features are extracted from each functional brain region. The system calls reference parameter library 140 to standardize, unify the direction of, and perform nonlinear mapping on the EEG features of the aforementioned regions, obtaining normalized EEG features. Simultaneously, the system generates feature quality factors based on the number of available segments, signal-to-noise ratio, number of effective trial runs, and channel effectiveness index. If a certain EEG feature cannot be calculated, its feature quality factor is set to 0.

[0165] The system further calculates the brain region function retention value for each brain functional region based on normalized EEG features, feature quality factors, and intra-region feature weights. The intra-region feature weights are only used for the fusion of different EEG features within the same brain functional region and are not used to represent the spatial contribution of the brain functional region to the behavioral function dimensions. Subsequently, based on the mapping matrix between brain functional regions and behavioral function dimensions, the system maps the brain region function retention value of each brain functional region to EEG assessment scores corresponding to behavioral function dimensions such as hearing, vision, motor, speech, communication, and arousal / cognition.

[0166] When calculating EEG assessment scores, the system simultaneously calculates the sum of effective mapping quality for each behavioral functional dimension. If the sum of effective mapping quality for a certain behavioral functional dimension is lower than a preset effective mapping threshold, such as 0.5, the system does not use the remaining few brain regions to perform unilateral amplification calculations on the EEG score for that dimension. Instead, it marks the EEG assessment score for that dimension as invalid or low-confidence. If the sum of effective mapping quality is not lower than the preset effective mapping threshold, the system outputs the EEG assessment score for that dimension.

[0167] Next, the system calculates the behavioral output limitation parameters for each behavioral function dimension based on clinical status information. For example, for target subject A, due to significant motor output impairment, the system increases the behavioral output limitation parameters for the motor function dimension; due to limited language output, the system increases the behavioral output limitation parameters for the verbal function dimension. The system then calculates a behavioral usability factor based on the behavioral output limitation parameters and uses this behavioral usability factor to correct the original behavioral quality parameters, obtaining the corrected behavioral assessment quality parameters.

[0168] The system employs a single-channel compensation approach to determine EEG modality fusion weights. Specifically, the system reflects the impact of behavioral output limitation only by reducing the modified behavioral assessment quality parameter, without introducing an additional product amplification term based on the behavioral output limitation parameter into the EEG data quality parameter. Therefore, although the normalized behavioral score is low in the motor and verbal function dimensions of target subject A, due to the strong behavioral output limitation, the system reduces the influence of the behavioral score on the fusion result and enhances the reference value of the EEG assessment score when the EEG data quality meets the requirements.

[0169] The system then integrates the EEG assessment score with the behavioral normalized score to generate a functional dimension fusion index for each behavioral functional dimension. If a certain functional dimension shows a high EEG assessment score, a low behavioral normalized score, and a difference exceeding a preset difference threshold, while the EEG data quality meets the requirements, the system generates a hidden consciousness risk warning for that dimension.

[0170] In this embodiment, due to low behavioral scores but relatively preserved EEG scores in the motor and verbal function dimensions of target subject A, the system outputs an EEG-behavioral inconsistency warning and suggests task-based EEG re-examination or repeated behavioral assessment. Simultaneously, the system generates an overall auxiliary assessment result in parallel based on the fusion index of each functional dimension, whole-brain network indicators, and data quality admission criteria. The standardized report outputs the fusion index of each functional dimension, EEG quality description, concealed consciousness risk warning, confidence information, and review recommendations.

Claims

1. A method for generating multimodal fusion-assisted evaluation information, characterized in that, Includes the following steps: Acquire multimodal assessment data of the target subject, including behavioral scale scores, electroencephalogram data, and clinical status information; The EEG data is preprocessed and abnormal channel classification is performed to generate valid EEG signals, channel validity masks, brain region quality factors, and EEG data quality parameters. Representative signals of multiple brain functional regions are generated based on the spatial location of the EEG channel. The EEG features of each brain functional region are extracted, and the reference parameter library is used to process the EEG features of the regions to obtain normalized EEG features and generate feature quality factors. Based on the normalized EEG characteristics, feature quality factors, and intra-brain region feature weights, the brain region function retention value for each brain functional region is calculated. Based on the mapping matrix between brain functional regions and behavioral functional dimensions, the brain region quality factor, and the brain region function retention value, candidate EEG assessment scores for each behavioral functional dimension are calculated, and the effective mapping quality is used to determine whether to output EEG assessment scores. Based on the clinical status information, behavioral output limitation parameters for each behavioral functional dimension are calculated, and the original behavioral quality parameters are corrected using the behavioral output limitation parameters. Based on the EEG data quality parameters and the corrected behavioral quality parameters, the EEG modality fusion weights are determined using a single-channel compensation method, and the EEG assessment scores are fused with the normalized behavioral scores to generate a functional dimension fusion index. Based on the EEG assessment score, behavioral normalized score, EEG data quality parameters, and functional dimension fusion index, the system outputs concealed consciousness risk warning information, overall auxiliary assessment results, confidence information, and standardized reports in parallel.

2. The method according to claim 1, characterized in that, The behavioral functional dimensions include one or more of the following: auditory functional dimension, visual functional dimension, motor functional dimension, verbal functional dimension, communication functional dimension, and arousal / cognitive functional dimension; The brain functional regions include the prefrontal cortex region, the motor cortex region, the temporal lobe region, the occipital lobe region, the posterior parietal lobe region, and the whole brain network region.

3. The method according to claim 1, characterized in that, The preprocessing and abnormal channel classification of the EEG data includes: When the proportion of abnormal channels in any brain functional area does not exceed the first proportion threshold, the abnormal channels are reconstructed by weighted interpolation of adjacent effective channels or spherical spline interpolation. When the proportion of abnormal channels is greater than the first proportion threshold but not more than the second proportion threshold, the weight of the abnormal channels will be reset to zero during the subsequent generation of the representative signal of the partition, and the quality parameters of the corresponding brain functional partition will be reduced. When the proportion of abnormal channels exceeds the second proportion threshold, the corresponding brain functional area is marked as a low-quality brain region, and the brain functional area is reduced or prohibited from participating in the EEG score calculation of the corresponding functional dimension.

4. The method according to claim 1, characterized in that, The brain region quality factor is calculated as follows: in, Indicates the first The effective channel ratio of each brain functional region; Indicates the proportion of effective EEG segments in a region; This represents the normalized signal-to-noise ratio of the partition; Indicates the interpolation channel ratio for the partition; , , , These are the weighting coefficients; This means that the calculation result will be limited to between 0 and 1.

5. The method according to claim 4, characterized in that, The characteristic quality factor is calculated as follows: in, Indicates the calculation of the first The first brain functional region When considering individual EEG features, the normalized value of the number of segments can be used; This represents the normalized signal-to-noise ratio of the segment corresponding to the feature. This represents the proportion of effective trials for task-state features; for resting-state features, it can be set to 1. Indicates the first Channel effectiveness indicators for individual brain functional regions; To prevent constants with a denominator of zero.

6. The method according to claim 5, characterized in that, Processing the EEG features of the region by calling the reference parameter library includes: Z-score standardization or robust standardization is performed on the regional EEG features to obtain standardized values; The standardized EEG features were oriented in the following manner: in, Represents the standardized value. Indicates the direction factor. The eigenvalues ​​represent the eigenvalues ​​after the orientation is unified; Then, perform nonlinear mapping as follows to obtain normalized EEG features: in, Indicates normalized EEG characteristics, Represents the slope parameter. This represents the threshold parameter.

7. The method according to claim 6, characterized in that, The brain region function retention values ​​were calculated as follows: in, Indicates the first Functional retention values ​​of individual brain functional regions Indicates normalized EEG characteristics, Indicates the weight of features within a brain region. Represents the characteristic quality factor. To prevent constants with a denominator of zero.

8. The method according to claim 7, characterized in that, The candidate EEG assessment scores were calculated as follows: in, Indicates the first Candidate EEG assessment scores for each behavioral functional dimension, Indicates the first The brain functional regions are related to the first Spatial mapping weights of behavioral functional dimensions Indicates brain region quality factors, This indicates the value of brain region function retention.

9. The method according to claim 8, characterized in that, The effective mapping quality is calculated as follows: when When the candidate EEG assessment score is not lower than a preset effective mapping threshold, the score will be... EEG scores as a dimension of behavioral function ; when When the score is below the preset effective mapping threshold, the EEG assessment score of the behavioral function dimension is marked as invalid or given a low confidence default value, and the corresponding EEG data quality parameter is reduced.

10. The method according to claim 1, characterized in that, The method of determining the EEG modality fusion weights using single-channel compensation includes: Calculate the first EEG data quality parameters corresponding to each behavioral functional dimension , The EEG modality fusion weights are calculated using the following formula: in, This represents the corrected behavioral quality parameters; The EEG modality fusion weights only indirectly reflect the impact of behavioral output limitation through the modified behavioral quality parameters, without introducing additional product amplification terms based on behavioral output limitation parameters into the EEG data quality parameters; The functional dimension fusion index is calculated as follows: in, Indicates the integration index of functional dimensions. Indicates the EEG assessment score, This represents the normalized behavioral score.

11. A multimodal fusion-assisted evaluation system, characterized in that, include: The data acquisition module is used to acquire behavioral scale scores, electroencephalogram (EEG) data, and clinical status information of the target subjects. The EEG preprocessing module is used to preprocess the EEG data and perform abnormal channel classification processing to generate valid EEG signals, channel validity masks, brain region quality factors, and EEG data quality parameters. The brain functional area feature extraction module is used to generate representative signals of multiple brain functional areas based on the spatial location of the brain electrical channels, extract the brain electrical features of each brain functional area, and generate feature quality factors. A reference parameter library is used for hierarchical storage of standardized parameters, orientation factors, threshold parameters, slope parameters, brain region feature weights, mapping matrices, clinical state influence weights, effective mapping quality thresholds, and auxiliary assessment threshold parameters. The functional dimension EEG scoring module is used to calculate the EEG assessment score corresponding to each behavioral functional dimension based on normalized EEG features, feature quality factors, intra-brain region feature weights, brain region quality factors, effective mapping quality, and mapping matrix. The multimodal fusion module is used to calculate behavioral output limitation parameters based on clinical status information, correct the original behavioral quality parameters, determine the EEG modal fusion weights using a single-channel compensation method, and generate a functional dimension fusion index. The concealed consciousness risk identification module is used to generate concealed consciousness risk warning information based on EEG assessment scores, behavioral normalization scores, and EEG data quality parameters; The overall auxiliary assessment module is used to generate overall auxiliary assessment results based on functional dimension fusion index, whole brain network indicators and data quality admission criteria. The report generation and visualization module is used to generate and output standardized reports that include hidden awareness risk warnings, overall auxiliary assessment results, confidence level information, and review prompts.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multimodal fusion-assisted evaluation information generation method according to any one of claims 1 to 10.

Images