Anesthesia consciousness determination method, system, device and medium

Abstract

The application provides a method, system, device and medium for determining anesthesia consciousness, which comprises: obtaining a target electroencephalogram signal of a target object; determining consciousness state information of the target object based on the target electroencephalogram signal through a consciousness determination model, wherein the consciousness state includes a conscious state, a light anesthesia state and an unconscious state; wherein the consciousness determination model is a machine learning model, and the consciousness determination model is obtained based on training samples, wherein the training samples include sample electroencephalogram signals, and the sample electroencephalogram signals include sample conscious state data, sample light anesthesia state data and sample unconscious state data. Through the trained consciousness determination model, the change in anesthesia depth of the target object during the operation and the smooth transition thereof can be accurately tracked on the basis of accurately determining the unconscious state or the non-unconscious state of the target object, so that the regulation of the amount of anesthetic drugs for the target patient during the operation is more accurate.

Description

Technical Field

[0001] This invention relates to the medical field, and in particular to a method, system, device, and medium for determining anesthesia consciousness. Background Technology

[0002] During induction of anesthesia, the anesthesiologist assesses the consciousness of the target subject (e.g., the patient) using a pre-set anesthesia depth assessment scale (such as the MOAAS score). Subsequent surgical procedures will only proceed after confirming the patient is completely unconscious. During the surgery, the doctor will adjust the dosage of anesthetic drugs in real time based on the patient's various physiological indicators to maintain the patient in a state of unconsciousness (i.e., general anesthesia).

[0003] Identifying a patient's state of consciousness can typically be achieved by analyzing their electroencephalogram (EEG) signals. However, due to objective limitations in clinical data collection, the amount of EEG data varies significantly across different depths of anesthesia. For example, there is limited EEG data for patients in a state of light anesthesia or moderate sedation (e.g., MOAAS2 and MOAAS3 scores). Excluding EEG data from patients in a state of light anesthesia when assessing anesthesia depth leads to inaccurate assessments, making it difficult to capture subtle changes in the patient's state of consciousness. This, in turn, results in inaccurate assessment of anesthetic dosage, potentially affecting the surgical procedure or causing negative impacts on the patient.

[0004] Therefore, a method, system, device, and medium for determining anesthesia consciousness are provided, which makes the determination of consciousness state more accurate and the tracking of changes in the patient's anesthesia state more precise. Summary of the Invention

[0005] This invention provides a method for determining consciousness under anesthesia, comprising acquiring target EEG signals of a target subject; and determining the consciousness state information of the target subject based on the target EEG signals using a consciousness determination model, wherein the consciousness state includes a conscious state, a lightly anesthetized state, and an unconscious state; wherein the consciousness determination model is a machine learning model, and the consciousness determination model is obtained based on training samples, wherein the training samples include sample EEG signals, and the sample EEG signals include sample conscious state data, sample lightly anesthetized state data, and sample unconscious state data.

[0006] This invention provides an anesthesia consciousness determination system, comprising: an acquisition module configured to acquire target EEG signals of a target object; and a determination module configured to determine the consciousness state information of the target object based on the target EEG signals and through a consciousness determination model, wherein the consciousness state information includes a conscious state, a lightly anesthetized state, and an unconscious state; wherein the consciousness determination model is a machine learning model, and the consciousness determination model is obtained by training on training samples, wherein the training samples include conscious state data, lightly anesthetized state data, and unconscious state data.

[0007] The present invention provides an anesthesia consciousness determination device, comprising at least one storage medium and at least one processor; the at least one storage medium is used to store computer instructions; the at least one processor is used to execute the computer instructions to implement the above-described anesthesia consciousness determination method.

[0008] The present invention provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the above-described method for determining anesthesia consciousness.

[0009] The beneficial effects of this invention include, but are not limited to: (1) the trained consciousness determination model can accurately determine the unconscious or non-unconscious state of the target object, ensuring the normal progress of the surgery; (2) it can accurately track the changes in the target object's different states of consciousness and / or depth of anesthesia during the operation and their smooth transition, enabling medical staff to more accurately assess and adjust the dosage of anesthetic drugs for the target patient during the operation; (3) in the process of training the consciousness determination model, it makes full use of the sample awake state data, sample light anesthesia state data and sample unconscious state data, so that the consciousness determination model can fully learn the relationship between the EEG signals of different depths of anesthesia and their corresponding states of consciousness and / or depths of anesthesia, thereby making the trained consciousness determination model more refined and accurate in determining the state of consciousness and / or depth of anesthesia of the target object. Attached Figure Description

[0010] The present invention will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same reference numerals denote the same structures, wherein:

[0011] Figure 1 This is a schematic diagram illustrating an application scenario of the anesthesia consciousness determination system according to some embodiments of the present invention;

[0012] Figure 2 This is a schematic diagram of a module of an anesthesia consciousness determination system according to some embodiments of the present invention;

[0013] Figure 3This is an exemplary flowchart of a method for determining anesthesia consciousness according to some embodiments of the present invention;

[0014] Figure 4 This is a schematic diagram of a consciousness determination model according to some embodiments of the present invention;

[0015] Figure 5 This is a schematic diagram of a training method for a consciousness determination model according to some embodiments of the present invention;

[0016] Figure 6 This is a schematic diagram of a training method for another consciousness determination model according to some embodiments of the present invention;

[0017] Figure 7 This is a schematic diagram of the structure of an anesthesia consciousness determination device according to some embodiments of the present invention;

[0018] Figure 8 This is a schematic diagram of the structure of a computer-readable storage medium according to some embodiments of the present invention. Detailed Implementation

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of the present invention. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0020] It should be understood that the terms “system,” “device,” “unit,” and / or “module” used herein are one way to distinguish different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0021] Unless the context clearly indicates an exception, words such as "a," "an," "a kind," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0022] This invention uses flowcharts to illustrate the operations performed by the system according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0023] Figure 1 This is a schematic diagram illustrating an application scenario of an anesthesia consciousness determination system according to some embodiments of the present invention.

[0024] like Figure 1 As shown, the application scenario 100 of the anesthesia consciousness determination system may include a processing device 110, a network 120, a terminal 130, a storage device 140, and an EEG information database 150.

[0025] Processing device 110 can process data and / or information obtained from terminal 130, storage device 140, and EEG information database 150. For example, processing device 110 can acquire EEG information from EEG information database 150. In some embodiments, processing device 110 can generate training samples based on EEG information (such as EEG signals) to train and obtain a consciousness determination model. More details about the consciousness determination model can be found elsewhere in this invention (e.g., Figure 3 ).

[0026] In some embodiments, processing device 110 may be a single server or a group of servers. In some embodiments, processing device 110 may be local or remote. Processing device 110 may be directly connected to terminal 130, storage device 140, and EEG information database 150 to access stored or retrieved information and / or data. In some embodiments, processing device 110 may be implemented on a cloud platform. By way of example only, cloud platforms may include private clouds, public clouds, hybrid clouds, community clouds, distributed clouds, internal clouds, multi-tiered clouds, etc., or any combination thereof. In some embodiments, processing device 110 may be a distributed group of servers, which may include multiple server nodes.

[0027] Network 120 may include any suitable network that facilitates information and / or data exchange within the application scenario 100 of the anesthesia consciousness assessment system. In some embodiments, one or more components of the application scenario 100 of the anesthesia consciousness assessment system (e.g., processing device 110, terminal 130, storage device 140, or EEG information database 150) may transmit information and / or data with one or more other components of the application scenario 100 of the anesthesia consciousness assessment system via network 120. For example, processing device 110 may retrieve EEG-related information from storage device 140 and / or EEG information database 150 via network 120.

[0028] In some embodiments, network 120 can be any one or more of wired or wireless networks. In some embodiments, the network can be various topologies such as point-to-point, shared, centralized, or a combination of multiple topologies.

[0029] Terminal 130 may include mobile device 130-1, tablet computer 130-2, laptop computer 130-3, etc., or any combination thereof. In some embodiments, terminal 130 may interact with other components in application scenario 100 of the anesthesia consciousness assessment system via network 120. In some embodiments, terminal 130 may receive information and / or instructions input by a user and send the received information and / or instructions to processing device 110 via network 120. For example, terminal 130 may receive instructions from a user (such as a doctor or nurse) and retrieve EEG information from storage device 140 and / or EEG information database 150 via network 120. In some embodiments, terminal 130 may present consciousness state information (such as awake state, subconscious state, unconscious state, etc.) corresponding to the EEG information (such as target EEG signal) of a target object (such as an intraoperative patient).

[0030] Storage device 140 can store data and / or instructions. Storage device 140 may include mass storage devices, removable storage devices, volatile read-write memory, read-only memory (ROM), etc., or any combination thereof. In some embodiments, storage device 140 may be implemented on a cloud platform. In some embodiments, storage device 140 may be part of processing device 110.

[0031] In some embodiments, storage device 140 may store data and / or instructions for processing device 110 to perform the exemplary methods described in this invention. For example, storage device 140 may store instructions for processing device 110 to perform the methods shown in the flowcharts. In some embodiments, storage device 140 may store data acquired from processing device 110, terminal 130, and / or EEG information database 150. For example, storage device 140 may store data acquired from EEG information database 150 (such as EEG signals of a target object). In some embodiments, storage device 140 may store consciousness determination models (such as an initial consciousness determination model or a trained consciousness determination model).

[0032] The EEG information database 150 refers to the data source used to provide EEG information. For example, it may be a database of an operating entity (such as a hospital) and / or a third-party service platform (such as a research institution or healthcare provider). In some embodiments, the EEG information database 150 may be part of the storage device 140, for example, it may be a database deployed on the storage device 140.

[0033] Electroencephalogram (EEG) information includes various types of data related to brain signals, which may include, but are not limited to, EEG data, metadata, and feature data.

[0034] Electroencephalogram (EEG) signal data includes EEG signals (such as raw voltage time-series data) and EEG signal types (such as alpha waves, theta waves, delta waves, etc.). Metadata includes user information (such as age, gender, health or disease information, etc.) and acquisition device information (such as device type, device model, acquisition parameters, etc.) of the object corresponding to the EEG signal data (such as patients, subjects, etc.). Feature data includes the time-domain features, frequency-domain features, and nonlinear features of the EEG signals.

[0035] In some embodiments, the EEG information database 150 includes EEG information for specific medical scenarios. These specific medical scenarios include, but are not limited to, induction of anesthesia, intraoperative monitoring, disease diagnosis (such as epilepsy diagnosis), and scientific research. The EEG information can be determined according to actual needs (such as medical scenarios). For example, for the application scenario of induction of anesthesia, the EEG information also includes information on the state of consciousness corresponding to the EEG signals (such as conscious state, light anesthesia state, unconscious state, etc.).

[0036] In some embodiments, the EEG information database 150 can be used to provide training samples for training the consciousness determination model, and the training samples can be generated based on EEG information. For example, EEG signals corresponding to different states of consciousness of one or more objects can be used as sample EEG signals. Exemplarily, sample EEG signals include sample awake state data, sample light anesthesia state data, and sample unconscious state data. More information regarding training samples and the training of the consciousness determination model can be found elsewhere in this invention (e.g., Figure 3 ).

[0037] In some embodiments, the EEG information database 150 can interact with other components in the application scenario 100 of the anesthesia consciousness determination system via the network 120. For example, the EEG information database 150 can send training samples to the storage device 140 and / or processing device 110 via the network 120.

[0038] The anesthesia consciousness assessment system also includes an EEG signal acquisition device (not shown in the figure). This device includes an electroencephalogram (EEG) machine, intraoperative EEG monitoring equipment (such as a bispectral index monitor), portable / wearable EEG devices (such as wireless head-mounted EEG devices, flexible electrode patches), etc. In some embodiments, the EEG signal acquisition device can acquire the target subject's EEG signals at various stages of the procedure (such as the induction of anesthesia) to determine the target subject's state of consciousness.

[0039] The above description is for illustrative purposes only, and actual application scenarios may vary.

[0040] It should be noted that the application scenario 100 of the anesthesia consciousness assessment system is provided for illustrative purposes only and is not intended to limit the scope of the invention. Those skilled in the art can make various modifications or variations based on the description of this invention. However, these modifications and variations will not depart from the scope of the invention.

[0041] Figure 2 This is a schematic diagram of a module of an anesthesia consciousness determination system according to some embodiments of the present invention.

[0042] like Figure 2 As shown, the anesthesia consciousness determination system 200 (hereinafter referred to as the consciousness determination system) may include an acquisition module 210, a determination module 220 and a training module 230.

[0043] The acquisition module 210 is configured to acquire the target EEG signal of the target object.

[0044] The determination module 220 is configured to determine the consciousness state information of the target object based on the target EEG signal and through a consciousness determination model. The consciousness state includes a conscious state, a lightly anesthetized state, and an unconscious state.

[0045] In some embodiments, the consciousness determination model is a machine learning model, which is obtained by training on training samples. The training samples include sample EEG signals, which include sample awake state data, sample light anesthesia state data, and sample unconscious state data.

[0046] In some embodiments, the consciousness determination model includes a classification model, where consciousness state information represents the probability that the target object is in a non-unconscious state. The consciousness determination model is trained based on positive sample EEG signals and negative sample EEG signals, wherein the positive sample EEG signals include sample awake state data and sample light anesthesia state data, and the negative sample EEG signals include sample unconscious state data.

[0047] In some embodiments, the consciousness determination model includes a data preprocessing layer, a feature extraction layer, and an exponent calculation layer. The determination module 220 is further configured to: determine the consciousness state information of the target object based on the target EEG signal using the consciousness determination model, including: preprocessing the target EEG signal using the data preprocessing layer to obtain a preprocessed target EEG signal, wherein the preprocessing includes at least one of segmentation, filtering, noise reduction, and downsampling; extracting features from the preprocessed target EEG signal using the feature extraction layer to obtain at least one feature; and processing the at least one feature using the exponent calculation layer to determine the consciousness state information of the target object.

[0048] The training module 230 is configured to construct a loss function based on sample conscious state data, sample light anesthesia state data, and sample unconscious state data. The loss function includes a first loss term and a second loss term. The first loss term is used to indicate the first difference in the output of the consciousness determination model, which represents the difference between the predicted consciousness state corresponding to the training sample and the actual consciousness state corresponding to the training sample. The second loss term is used to indicate the second difference in the output of the consciousness determination model, which represents the difference between the probabilities of the consciousness state of positive sample EEG signals corresponding to multiple first state levels being a non-unconscious state. Based on the loss function, the model parameters of the consciousness determination model are adjusted.

[0049] In some embodiments, the first sedation level includes a fifth sedation level, a fourth sedation level, a third sedation level, and a second sedation level.

[0050] In some embodiments, the second difference satisfies a first preset deviation condition, which is related to the difference between each first state level.

[0051] In some embodiments, the loss function further includes a third loss term, which indicates a third difference in the output of the consciousness determination model. The third difference characterizes the difference between the probabilities of the consciousness state being unconscious among the negative sample EEG signals corresponding to multiple second state levels.

[0052] In some embodiments, the second state level includes a first sedation level and a complete sedation level.

[0053] In some embodiments, the third difference satisfies a second preset deviation condition, which is related to the difference between each of the second state levels.

[0054] In some embodiments, the training module 230 is further configured to: divide the negative sample EEG signals into multiple negative sample subsets, wherein the number of samples in each negative sample subset is the same as the number of samples in the positive sample EEG signals; combine each negative sample subset with the positive sample EEG signals respectively to obtain multiple training sample subsets; and train a consciousness determination model based on the multiple training sample subsets.

[0055] In some embodiments, the training module 230 is further configured to: divide each training sample subset into multiple training sample groups, wherein the number of positive sample EEG signals and the number of negative sample EEG signals in each training sample group meet a first preset condition, and the number of positive sample EEG signals corresponding to each first state level meets a second preset condition; and train a consciousness determination model based on the multiple training sample groups.

[0056] In some embodiments, the first preset condition includes: the ratio of the number of positive sample EEG signals to the number of negative sample EEG signals is within a first ratio threshold range.

[0057] In some embodiments, the ratio between the number of positive sample EEG signals for each first state level is within a second ratio threshold range.

[0058] In some embodiments, the training module 230 is further configured to: perform a first-stage training on the initial judgment model based on a first training sample to obtain a first judgment model, wherein the first training sample includes sample conscious state data and sample unconscious state data; perform a second-stage training on the first-stage consciousness judgment model based on a second training sample to obtain a second judgment model, wherein the second training sample includes sample conscious state data and sample light anesthesia state data; perform a third-stage training on the first-stage consciousness judgment model based on a third training sample to obtain a third judgment model, wherein the third training sample includes sample light anesthesia state data and sample unconscious state data; and determine a consciousness judgment model based on the second judgment model and / or the third judgment model.

[0059] It should be noted that the above description of the anesthesia consciousness determination system and its modules is for ease of description only and should not limit the invention to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principle of the system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from this principle. For example, the acquisition module 210, the determination module 220, and the training module 230 may be different modules in the system, or a single module may implement the functions of two or more of the above-mentioned modules. For example, the modules may share a storage module, or each module may have its own storage module. Such modifications are all within the protection scope of this invention.

[0060] Figure 3 This is an exemplary flowchart of a method for determining anesthesia consciousness according to some embodiments of the present invention.

[0061] In some embodiments, the procedure 300 of the anesthesia consciousness determination method can be executed by a consciousness determination system. For example... Figure 3 As shown, the procedure 300 for determining anesthesia consciousness includes the following steps.

[0062] Step S310: Obtain the target EEG signal of the target object.

[0063] The target object refers to the object whose state of consciousness needs to be determined. For example, it could be a patient or subject who needs anesthesia for surgery.

[0064] Target EEG signals refer to EEG signals used to determine the consciousness status of the target subject. Target EEG signals include the intraoperative EEG signals of the target subject; for example, they can be EEG signals from the patient during induction of anesthesia, maintenance of anesthesia (such as general anesthesia), and / or recovery. Target EEG signals can be used to determine the target subject's consciousness status during surgery to assess whether the target subject requires intraoperative adjustment of anesthetic drug dosage (e.g., adjusting the concentration of inhaled anesthetics or the infusion rate of intravenous anesthetics).

[0065] In some embodiments, the consciousness determination system can acquire target EEG signals through an EEG signal acquisition device and store them in an EEG signal database (such as EEG information database 150) and / or a storage device (such as storage device 140). For more information on EEG signal acquisition devices, EEG signal databases, and storage devices, please refer to [link to relevant documentation]. Figure 1 And its description.

[0066] Step S320: Based on the target EEG signal, determine the consciousness state information of the target object through the consciousness determination model. The consciousness state includes a conscious state, a lightly anesthetized state, and an unconscious state.

[0067] A state of consciousness is used to reflect the ability of a target object to perceive and / or react to itself and / or the external environment. For example, if a target object can perceive and / or respond to different types or degrees of external stimuli (e.g., auditory stimuli such as calling, tactile stimuli such as patting), it indicates that the target object is in a conscious state.

[0068] Different levels of perceptual behavior (e.g., clear, vague) and different levels of stress response behavior (e.g., dull, sensitive) of a target object can reflect the target object's different levels of consciousness. For example, normal awake state, sleep state, general anesthesia state, etc.

[0069] Level of consciousness can be used to characterize the level of alertness or sedation of a target subject, and it can be determined based on medical experience (such as a pre-defined consciousness rating scale). In some embodiments, the level of consciousness of the target subject and its corresponding state of consciousness can be assessed using the Modified Observer's Assessment of Alertness / Sedation (MOAAS).

[0070] The MOAAS (Moral Awareness Assessment System) includes six sedation levels (or ratings), each corresponding to a different level of consciousness. A higher sedation level indicates a higher level of alertness. For example, MOAAS includes MOAAS5, MOAAS4, MOAAS3, MOAAS2, MOAAS1, and MOAAS0. MOAAS5 indicates that the target subject responds to commands in a normal tone (such as being called by name); MOAAS4 indicates that the target subject responds to commands in a normal tone but with a somewhat sluggish response; MOAAS3 indicates that the target subject responds to loud or repetitive commands; MOAAS2 indicates that the target subject responds to mild stimuli (such as shaking or tapping the shoulder); MOAAS1 indicates that the target subject does not respond to mild stimuli; and MOAAS0 indicates that the target subject does not respond to strong painful stimuli (such as squeezing the trapezius muscle).

[0071] In some embodiments, the state of consciousness includes a conscious state, a state of light anesthesia, and a state of unconsciousness.

[0072] A conscious state refers to the state of consciousness of a target when they are awake. In some embodiments, the consciousness determination system can use the consciousness state corresponding to the sedation levels of MOAAS5 and MOAAS4 as the conscious state.

[0073] Unconscious state refers to the state of consciousness of a target subject while under anesthesia. In some embodiments, the consciousness determination system can use the state of consciousness corresponding to the sedation levels MOAAS1 and MOAAS0 as the unconscious state.

[0074] A state of light anesthesia is a state of consciousness between wakefulness and unconsciousness. In some embodiments, the consciousness determination system can use the consciousness state corresponding to the sedation levels of MOAAS3 and MOAAS2 as a state of light anesthesia.

[0075] In some embodiments, the consciousness determination system can analyze and / or process the target EEG signal using brain signal analysis methods such as frequency domain analysis, time domain analysis, and time-frequency analysis to determine the consciousness state of the target object.

[0076] In some embodiments, the consciousness determination system can analyze and / or process the target EEG signal using a consciousness determination model to determine the consciousness state of the target object.

[0077] A consciousness determination model is a model used to determine the consciousness state of a target object. It can be a model generated based on various algorithms (such as mathematical models). In some embodiments, the consciousness determination model is a trained machine learning model, which can be a neural network model built based on various machine learning algorithms. For example, a deep neural network (DNN) model.

[0078] The input to the consciousness determination model includes the target EEG signal, and the output includes the consciousness state information of the target object. In some embodiments, the consciousness state information output by the consciousness determination model is one of a conscious state, a lightly anesthetized state, and an unconscious state. In some embodiments, the input to the consciousness determination model may be features corresponding to the target EEG signal.

[0079] For more information on the consciousness determination model, please refer to [link / reference]. Figure 4 And its description.

[0080] In some embodiments, the consciousness determination model is obtained based on training samples, which include sample EEG signals, including sample awake state data, sample light anesthesia state data, and sample unconscious state data.

[0081] Sample EEG signals can be obtained from historical data. For example, a consciousness assessment system can acquire EEG signals from multiple sample subjects (such as sample patients) during induction of anesthesia over a historical period (such as the past year or month) as sample EEG signals.

[0082] As an example only, for a given sample subject, the consciousness assessment system can acquire the EEG signals collected when the subject is awake before anesthesia (e.g., sedation level MOAAS5 and / or MOAAS4), as the sample awake state data; acquire the EEG signals collected when the subject is under light anesthesia after injection of anesthetic drugs (e.g., sedation level MOAAS3 and / or MOAAS2), as the sample light anesthesia state data; and acquire the EEG signals collected when the subject is unconscious after anesthesia (e.g., sedation level MOAAS1 and / or MOAAS0), as the sample unconscious state data. The sample awake state data, sample light anesthesia state data, and sample unconscious state data are each used as a set of training samples.

[0083] The training label for each set of training samples can be determined based on the actual state of consciousness of the sample subjects. For example, the training labels for samples in a conscious state, samples in a state of light anesthesia, and samples in a state of unconsciousness are conscious, lightly anesthetized, and unconscious, respectively. Training labels can be based on manual annotation or other methods.

[0084] The consciousness assessment system can construct multiple training samples based on data from multiple patients in conscious, lightly anesthetized, and unconscious states. During training, the loss function value can be determined based on the difference between the output of the initial model and the training labels. The parameters of the initial model can be iteratively updated based on the loss function value until the training termination condition is met (e.g., loss function convergence, a specific number of iterations, etc.). The updated initial model can then be used as the trained consciousness assessment model.

[0085] In some embodiments, the consciousness determination model includes a classification model, where consciousness state information represents the probability that the target object is in a non-unconscious state.

[0086] The classification model can be a classification model used to determine the consciousness category of the target EEG signal. The consciousness category can be multiple categories corresponding to several different preset states of consciousness (such as conscious state, light anesthesia state, unconscious state).

[0087] In some embodiments, the classification model can be a binary classification model, used to determine whether the target object's conscious state is non-unconscious or unconscious. A non-unconscious state can be represented by 1, and an unconscious state by 0. When the trained consciousness determination model, based on the processing of the target EEG signal, outputs a predicted value greater than a preset threshold (e.g., 0.5), it indicates that the target EEG signal is in a non-unconscious state; otherwise, it is in an unconscious state. The closer the predicted value is to 1, the greater the probability that the target EEG signal is in a non-unconscious state.

[0088] In some embodiments, the binary classification model may be one of the following: a trained Adaptive-Network-Based Fuzzy Inference System (ANFIS), a support vector machine, a random forest, a gradient boosting tree, or other deep neural networks.

[0089] In some embodiments, the consciousness determination model is trained based on positive and negative EEG signals.

[0090] Positive sample EEG signals refer to sample EEG signals corresponding to non-unconscious states. These include data from conscious and lightly anesthetized states. The training label for positive sample EEG signals can be set to 1.

[0091] Negative sample EEG signals refer to sample EEG signals corresponding to unconscious states. They include sample unconscious state data, and the training label for negative sample EEG signals can be set to 0.

[0092] In some embodiments, the consciousness determination system can construct a loss function based on sample conscious state data, sample light anesthesia state data, and sample unconscious state data. The loss function includes a first loss term and a second loss term.

[0093] The first loss term is used to indicate the first difference in the output of the consciousness determination model. The first difference represents the difference between the predicted consciousness state corresponding to the training sample and the real consciousness state corresponding to the training sample.

[0094] Predicted state of consciousness refers to the predicted state of consciousness corresponding to the training samples output by the consciousness determination model during training. It includes one of the two consciousness categories: non-unconscious state and unconscious state. The predicted state of consciousness can be represented by a numerical value in the interval [0,1]. The larger the value (e.g., 0.9), the greater the probability that the predicted state of consciousness is non-unconscious; the smaller the value (e.g., 0.1), the greater the probability that the predicted state of consciousness is unconscious.

[0095] The true state of consciousness refers to the actual or expected value of the state of consciousness corresponding to the training sample, which can be determined based on the training label (such as 1 or 0) of the training sample.

[0096] The first difference is determined based on the difference between the predicted and actual values ​​(e.g., the absolute value of the difference), which can be represented by a value in the range [0,1]. The smaller the first difference, the smaller the deviation (loss) between the predicted and expected values ​​during the training process of the consciousness determination model. This indicates that the consciousness determination model can more accurately distinguish between positive and negative EEG signals, thus accurately determining the consciousness state corresponding to each training sample (e.g., positive or negative EEG signal).

[0097] In some embodiments, the first loss term is generated based on the cross-entropy loss function, which can be represented based on equation (1) as shown below:

[0098] In equation (1), L ce The value of the first loss term represents the first difference (value); N represents the number of training samples; i represents the i-th training sample, y i p represents the training label (e.g., 1 or 0) of the i-th training sample. i This represents the predicted value corresponding to the predicted consciousness state output by the consciousness determination model during training.

[0099] When the input is a positive sample EEG signal, its corresponding training label is 1, and the above equation (1) becomes the form of the following equation (2).

[0100] The variables in equation (2) have the same meaning as the variables in equation (1).

[0101] When the input is a negative sample EEG signal, its corresponding training label is 0, and the above equation (1) becomes the form of equation (3) below.

[0102] The variables in equation (3) have the same meaning as the variables in equation (1).

[0103] The second loss term is used to indicate the second difference in the output of the consciousness determination model. The second difference represents the difference between the probabilities of the conscious state being a non-unconscious state among the positive sample EEG signals corresponding to multiple first state levels.

[0104] First state levels are used to distinguish multiple different states of consciousness within a non-unconscious state. Multiple first state levels can be determined based on information such as the number and classification of non-unconscious states. For example, multiple first state levels may include the first state level corresponding to a conscious state, the first state level corresponding to a state of light anesthesia, etc.

[0105] In some embodiments, the first sedation level includes a fifth sedation level, a fourth sedation level, a third sedation level, and a second sedation level. The fifth sedation level, the fourth sedation level, the third sedation level, and the second sedation level correspond to MOAAS5, MOAAS4, MOAAS3, and MOAAS2, respectively.

[0106] The positive sample EEG signals corresponding to the fifth sedation level, the fourth sedation level, the third sedation level, and the second sedation level refer to the sample EEG signals of the sample subjects when they were under sedation levels of MOAAS5, MOAAS4, MOAAS3, and MOAAS2, respectively, and are referred to as the fifth sample EEG signal, the fourth sample EEG signal, the third sample EEG signal, and the second sample EEG signal, respectively, below.

[0107] The states of consciousness corresponding to multiple positive EEG samples at the first state level include the states of consciousness corresponding to the fifth, fourth, third, and second EEG samples. These will be referred to as the fifth, fourth, third, and second states of consciousness, respectively, below.

[0108] It can be understood that the fifth, fourth, third, and second states of consciousness are non-unconscious states. Among them, the fifth and fourth states of consciousness are conscious states, while the third and second states of consciousness are states of light anesthesia. Different non-unconscious states can be used to reflect changes in the non-unconscious state of the target subject in actual medical scenarios (induction of anesthesia, maintenance of anesthesia during surgery, recovery, etc.).

[0109] In some embodiments, predicting states of consciousness includes predicting a fifth state of consciousness, a fourth state of consciousness, a third state of consciousness, and a second state of consciousness.

[0110] In some embodiments, the second difference includes the difference (e.g., the absolute value of the difference) between the predicted values ​​corresponding to the predicted fifth state of consciousness, the predicted fourth state of consciousness, the predicted third state of consciousness, and the predicted second state of consciousness. For example, the second difference includes the difference between any two of the predicted fifth state of consciousness, the predicted fourth state of consciousness, the predicted third state of consciousness, and the predicted second state of consciousness.

[0111] In some embodiments, the second difference satisfies a first preset deviation condition, which is related to the difference between each first state level.

[0112] The first preset deviation condition refers to the condition that the second difference needs to meet. For example, the second difference is that the difference between the probability that the conscious state of any two positive sample EEG signals corresponding to the first state level is a non-unconscious state (such as the difference between predicting the fifth conscious state and predicting the fourth conscious state) is within a preset difference threshold (such as 0.05, 0.1, etc.).

[0113] In some embodiments, the differences between the various first state levels (such as fifth sedation level, fourth sedation level, third sedation level, and second sedation level) are used to reflect differences in the target subject's level of consciousness (e.g., decreasing levels of consciousness sequentially). A first preset deviation condition may be that the predicted values ​​(probability values) corresponding to the predicted fifth, fourth, third, and second states of consciousness decrease sequentially. For example, the first preset deviation condition may be that the difference between the probability values ​​of two adjacent predicted states of consciousness is within the range of 0.05.

[0114] In some embodiments, the consciousness determination system can set probability intervals for predicting the fifth, fourth, third, and second states of consciousness, respectively, so that the predicted values ​​of the predicted states of consciousness corresponding to each first state level satisfy a first preset deviation condition. For example, the probability interval for predicting the fifth state of consciousness is [0.95, 1], the probability interval for predicting the fourth state of consciousness is [0.9, 0.95), the probability interval for predicting the third state of consciousness is [0.85, 0.9), and the probability interval for predicting the second state of consciousness is [0.8, 0.85]. That is, the predicted values ​​of the predicted states of consciousness corresponding to each first state level are different and have a preset sequential nature.

[0115] In some embodiments of the present invention, by setting a first preset deviation condition to take into account the differences between different non-unconscious states, the second loss term can constrain the differences and orderliness of the predicted values ​​corresponding to different non-unconscious states, so that the subsequent output predicted values ​​are more in line with the actual situation (such as differences in the degree of consciousness).

[0116] In some embodiments, the second loss term is constructed based on the marginal ranking loss function, which can be represented based on the following equation (4):

[0117] In equation (4), L rank1 This represents the value of the second loss term. P represents the set of positive EEG signals, and C1 represents the size of set P (the number of positive EEG signals).

[0118] p i p represents the predicted value output of the i-th positive sample EEG signal (e.g., a positive sample EEG signal at the fifth level of sedation). j p represents the predicted value of the output of the j-th positive sample EEG signal (such as the positive sample EEG signal of the fourth sedation level). i and p j The corresponding positive sample EEG signals constitute a sample pair (p i ,p j For example, a sample pair (positive EEG signal of level 5 sedation, positive EEG signal of level 4 sedation).

[0119] s i p i The corresponding first status level (such as MOAAS level or rating), s j p j The corresponding first state level. For example, s i The value is the numerical value corresponding to the fifth level of sedation (MOAAS5) (e.g., 5), s j The value is the value corresponding to the fourth level of sedation (MOAAS4) (e.g., 4).

[0120] w ij Indicates sample pairs (p) i ,p j The sample pair weights can be preset values. In some embodiments, the sample pair weights can be determined based on the sample pair (p). i ,p j p) i and p j The difference in the corresponding first state level is set. For example, for the sample pair corresponding to MOAAS5 and MOAAS2, w ijIt can be set to 1; for the sample pairs corresponding to MOAAS5 and MOAAS4, w ij It can be set to 0.5.

[0121] L margin (p i ,p j ,s i ,s i ) represents a sample pair (p) i ,p j The second difference corresponding to ) is p i With p j The difference between the corresponding predicted values. The margin represents the boundary parameter, which can be a preset boundary threshold (e.g., 0.1, 0.05, etc.). The preset boundary threshold is used to constrain p. i With p j The difference between the corresponding predicted values ​​satisfies the first preset deviation condition.

[0122] λ represents the penalty factor, which can be a preset value used to adjust the impact of the second loss term on the total loss value of the loss function (total loss function) of the consciousness determination model. For example, when λ is set to 0, the value of the second loss term L... rank1 When the value is 0, the total loss function is transformed into the value of the first loss term.

[0123] In some embodiments of the present invention, a second loss term is constructed by introducing a marginal ranking loss function. During training, this can constrain the probability or ranking relationship of the positive EEG signals of the two first state levels in the positive sample pair as non-unconscious states. This allows the model to learn the differences in EEG signals between different non-unconscious states, thereby enabling the trained consciousness determination model to accurately distinguish non-unconscious states (such as the fifth consciousness state, the fourth consciousness state, etc.) corresponding to different first state levels, in addition to distinguishing between non-unconscious states and unconscious states.

[0124] In some embodiments, the consciousness determination system can construct a loss function for the consciousness determination model based on a first loss term and a second loss term, which can be represented by the following equation (5). L total =L ce +L rank1 (5)

[0125] Wherein, L in equation (5) ce L represents the value of the first loss term, which can be determined based on equation (1). rank1 L represents the value of the second loss term, which can be determined based on equation (4). total This represents the total loss value, which is determined based on the sum of the values ​​of the first loss term and the second loss term.

[0126] In some embodiments, the consciousness determination system can train the initial consciousness determination model based on multiple sets of training samples according to the loss function shown in equation (5). These multiple sets of training samples include a set of single samples and a set of positive sample pairs.

[0127] A single-sample set refers to a set of samples that uses a single positive or negative EEG signal as training input for binary classification tasks. The loss value for binary classification training is determined based on the value of the first loss term. The consciousness determination system can input either positive or negative EEG signals from the single-sample set into the initial consciousness determination model and determine the value of the first loss term based on the difference between the output of the initial consciousness determination model and its training label (e.g., 1 or 0). For more information on constructing training samples using positive and negative EEG signals, see [link to relevant documentation]. Figure 5 And its description.

[0128] A positive sample pair set refers to a sample set that uses positive sample EEG signals corresponding to two first-level states as training input for a ranking task. The loss value for the ranking task training is determined based on the value of the second loss term. The consciousness determination system can input each positive sample pair from the positive sample pair set into the initial consciousness determination model and determine the value of the second loss term based on the difference between the output of the initial consciousness determination model and its training label (e.g., 1 or -1). Here, label 1 or -1 indicates the magnitude or arrangement relationship of the predicted values ​​corresponding to the two samples in the positive sample pair. For example, for the sample pair corresponding to MOAAS5 and MOAAS4, the level of consciousness in the non-unconscious state (fifth state of consciousness) corresponding to MOAAS5 is greater than the level of consciousness in the non-unconscious state (fourth state of consciousness) corresponding to MOAAS4. The training label for this sample pair is set to 1, indicating that the predicted value corresponding to the fifth state of consciousness needs to be greater than the predicted value corresponding to the fourth state of consciousness.

[0129] The consciousness determination system determines the value of the total loss function based on the values ​​of the first and second loss terms, and iteratively updates the model parameters of the initial consciousness determination model based on the value of the total loss function. The updated initial consciousness determination model can be used as the trained consciousness determination model once the training termination condition is met (e.g., loss function convergence, a specific number of iterations, etc.).

[0130] In some embodiments, the consciousness determination system can train the initial consciousness determination model in stages. For more details, see [link to documentation]. Figure 6 And its description.

[0131] In some embodiments, the loss function further includes a third loss term, which indicates a third difference in the output of the consciousness determination model. The third difference characterizes the difference between the probabilities of the consciousness state being unconscious among the negative sample EEG signals corresponding to multiple second state levels.

[0132] The second state level is used to distinguish multiple different conscious states within the unconscious state. Multiple second state levels can be determined based on information such as the number and classification of unconscious states.

[0133] In some embodiments, the second sedation level includes a first sedation level and a complete sedation level. The first sedation level and the complete sedation level correspond to MOAAS1 and MOAAS0, respectively.

[0134] The negative sample EEG signals corresponding to the first sedation level and the complete sedation level refer to the sample EEG signals of the sample subjects when they were in a sedation level of MOAAS1 and MOAAS0, respectively, and are referred to as the first negative sample EEG signal and the second negative sample EEG signal below.

[0135] The states of consciousness corresponding to multiple negative sample EEG signals at multiple second-state levels include the states of consciousness corresponding to the first negative sample EEG signal and the second negative sample EEG signal. These are hereinafter referred to as the first unconscious state and the complete unconscious state, respectively. The first unconscious state and the complete unconscious state can be used to reflect changes in the unconscious state of a target subject in practical medical scenarios (such as the intraoperative maintenance phase, the resuscitation phase, etc.).

[0136] In some embodiments, the third difference satisfies a second preset deviation condition, which is related to the difference between each of the second state levels.

[0137] The second preset deviation condition refers to the condition that the third difference needs to meet. The content of the second preset deviation condition and the third difference is similar to that of the first preset deviation condition and the second difference, and will not be repeated here. The difference is that the third difference satisfies the requirement that the difference in the probability of the conscious state of any two negative sample EEG signals corresponding to the second state level being an unconscious state is within a preset threshold (such as 0.05, 0.1, etc.). The probability intervals corresponding to the completely unconscious state and the first unconscious state can be set to [0,0.1], [0.1,0.2], or other preset threshold ranges (such as [0,0.5)).

[0138] In some embodiments, the third loss term is constructed based on the marginal ranking loss function, which can be represented based on the following equation (6):

[0139] The variables in equation (6) are similar to those in equation (4), L rank2This represents the value of the third loss term. Q represents the set of negative sample EEG signals, and C2 represents the size of set Q (the number of negative sample EEG signals). p i p represents the predicted value output of the i-th negative sample EEG signal (such as the negative sample EEG signal of the first sedation level). j p represents the predicted value of the output of the j-th negative sample EEG signal (such as a negative sample EEG signal at the level of complete sedation). i and p j The corresponding negative sample EEG signals constitute a sample pair (p i ,p j For example, sample pairs (negative sample EEG signals of the first sedation level, negative sample EEG signals of the complete sedation level).

[0140] L in equation (6) margin (p i ,p j ,s i ,s i ),w ij λ can be found in the description of equation (5).

[0141] In some embodiments, the consciousness determination system can construct a loss function for the consciousness determination model based on a first loss term, a second loss term, and a third loss term, which can be represented by the following equation (7). L total =L cs +L rank1 +L rank2 (7)

[0142] Wherein, L in equation (7) ce L represents the value of the first loss term, which can be determined based on equation (1). rank1 L represents the value of the second loss term, which can be determined based on equation (4). rank2 L represents the value of the third loss term, which can be determined based on equation (6). total This represents the total loss value, which is determined based on the sum of the values ​​of the first loss term, the second loss term, and the third loss term.

[0143] In some embodiments, the consciousness determination system can train the initial consciousness determination model based on multiple sets of training samples according to the loss function shown in Equation (7). The training method is similar to the method described above, which uses the loss function shown in Equation (5) and trains the set by constructing a single sample set and positive samples.

[0144] For example, training can be conducted on a binary classification task targeting the first loss term by constructing a set of single samples (such as positive sample EEG signals and negative sample EEG signals), on a ranking task targeting the second loss term by constructing a set of positive sample pairs (such as sample pairs of MOAAS5 and MOAAS4, sample pairs of MOAAS4 and MOAAS3, etc.), and on a ranking task targeting the third loss term by constructing a set of negative sample pairs (such as sample pairs of MOAAS1 and MOAAS0). Further details are omitted here.

[0145] Compared to using MOAAS5 and MOAAS4 sample EEG signals as positive samples and MOAAS1 and MOAAS0 sample EEG signals as negative samples to train the binary classification model, the intermediate MOAAS2 and MOAAS3 sample EEG signals are not included in the model training due to their small data volume. In some embodiments of the present invention, by incorporating the data of the light anesthesia state into the training of the consciousness determination model, the trained consciousness determination model can more accurately determine the consciousness state of the target object under light anesthesia. Simultaneously, it can track the smooth transition between the target object's conscious state, light anesthesia state, and unconscious state, thus making the intraoperative anesthesia plan more accurate and ensuring the normal progress of the surgery.

[0146] Figure 4 This is a schematic diagram of a consciousness determination model according to some embodiments of the present invention.

[0147] In some embodiments, such as Figure 4 As shown, the consciousness determination model 410 includes a data preprocessing layer 411, a feature extraction layer 412, and an index calculation layer 413. The consciousness determination model 410 can process the input target EEG signal 401 and output consciousness state information 402. For details regarding the target EEG signal and consciousness state information, please refer to... Figure 3 And its description.

[0148] The data preprocessing layer 411 refers to a data processing module or data processing unit built based on a data preprocessing algorithm. The preprocessing algorithm can be determined according to actual needs (such as environmental noise, motion artifacts, etc.). The consciousness determination system constructs the data preprocessing layer 411 based on one or more of the preset data preprocessing algorithms (such as denoising algorithms, motion artifact correction algorithms, etc.).

[0149] In some embodiments, preprocessing includes at least one of slice processing, filtering processing, noise reduction processing, and downsampling processing.

[0150] Slicing is used to divide the acquired EEG signals into multiple target EEG signals within a preset time window. The preset time window can be 2.5 seconds, etc. As an example, sliceing can be a sliding window process based on a 2.5-second preset time window to obtain multiple EEG signal segments, each of which can serve as a target EEG signal. For example, an EEG signal segment from 0 seconds to 2.5 seconds is one target EEG signal, an EEG signal segment from 2.5 seconds to 5 seconds is another target EEG signal, and so on. Slicing can also take other forms, such as overlapping sliding windows.

[0151] Filtering is used to filter specific frequencies in the target EEG signal, retaining frequency bands that reflect the target subject's state of consciousness. Filtering includes, but is not limited to, high-pass filtering and low-pass filtering, which can be determined based on the specific circumstances. For example, high-pass filtering removes low-frequency components such as breathing artifacts; low-pass filtering suppresses high-frequency noise.

[0152] Noise reduction processing is used to separate and remove non-brain-derived noise or artifacts. Noise includes, but is not limited to, physiological artifacts (such as eye movement artifacts, electromyography artifacts, etc.) and environmental artifacts (such as power supply noise, electrosurgical interference, etc.).

[0153] Downsampling is used to downsample the target EEG signal to a preset frequency (e.g., 200 Hz) to reduce the amount of data and improve processing efficiency.

[0154] The preprocessed target EEG signal refers to the EEG signal that meets preset requirements (such as noise reduction and motion artifact correction). The consciousness determination model can perform subsequent analysis and / or processing based on the preprocessed target EEG signal.

[0155] It should be noted that the data preprocessing layer 411 in the consciousness determination model 410 is optional. In some embodiments, the data preprocessing layer 411 is set outside the consciousness determination model 410 as an independent data processing module or processing unit. In this case, the consciousness determination model preprocesses the collected EEG signals of the target object based on the data preprocessing layer 411, and uses the preprocessed EEG signals as the target EEG signals as input to the consciousness determination model 410.

[0156] Feature extraction layer 412 refers to a feature extraction module or feature extraction unit based on manual feature extraction or automatic feature extraction by neural networks.

[0157] In some embodiments, the feature extraction layer 412 is used to extract features from the target EEG signal to obtain at least one EEG signal feature (not shown in the figure).

[0158] Electroencephalogram (EEG) signal features are used to reflect the characteristics of a target EEG signal, which include one or more combinations of time-domain features, frequency-domain features, and nonlinear features.

[0159] Temporal characteristics reflect the amplitude changes and waveform characteristics of the target EEG signal, including but not limited to kurtosis (the sharpness of signal peaks), skewness (the asymmetry of signal distribution), mean (such as the average voltage level of the signal), variance (the degree of signal fluctuation), and peak count (the number of peaks / troughs per unit time).

[0160] Frequency domain characteristics reflect the energy distribution characteristics of different frequency bands of the target EEG signal, including but not limited to the energy distribution of signals in frequency bands such as delta waves, theta waves, alpha waves, beta waves, and gamma waves.

[0161] Nonlinear features reflect the complex and chaotic characteristics of the target EEG signal, including but not limited to entropy features (such as spectral entropy, state entropy, response entropy, permutation entropy, etc.) and chaotic features (such as fractal dimension, etc.).

[0162] Different target EEG signals require different EEG signal features, which are pre-set based on actual conditions (such as subject information (age, gender, etc.) and anesthetic drug information (type, dosage, etc.). The consciousness determination system extracts the EEG signal features of the target EEG signal based on feature extraction layer 412.

[0163] The exponential calculation layer 413 is used to process at least one EEG signal feature of the target EEG signal to determine the consciousness state corresponding to the target EEG signal. The exponential calculation layer 413 can be one of support vector machines, random forests, gradient boosting trees, or other deep neural networks.

[0164] In some embodiments, the exponent calculation layer 413 is constructed and generated based on an adaptive fuzzy logic neural network. In some embodiments of the present invention, considering that the EEG signal characteristics of the target EEG signal have nonlinear relationships, the exponent calculation layer constructed using an adaptive fuzzy logic neural network can better process the nonlinear characteristics of the target EEG signal.

[0165] In some embodiments, the consciousness determination system can input the EEG signal features corresponding to the target EEG signal into the exponential calculation layer 413, and after processing by the exponential calculation layer 413, output consciousness state information 402.

[0166] Consciousness state information 402 includes a predicted value of the target object's current consciousness state, which can be determined during training based on actual needs. For example, in a scenario involving induction of anesthesia, users (such as doctors and nurses) are concerned with whether the target object is already unconscious to determine whether surgery can be performed. In this case, consciousness state information 402 could be the probability of being unconscious or not unconscious. For instance, a predicted value less than or equal to 0.5 indicates that the target object is unconscious; otherwise, it is not unconscious.

[0167] For example, in addition to focusing on whether the target object is unconscious or not, users also need to pay attention to changes in the target object's state of consciousness (such as the transition from a conscious to an unconscious state during the induction of anesthesia, or the gradual awakening from an unconscious state during the maintenance or recovery phases). Consciousness state information 402 can be the probability of consciousness states corresponding to different first and second state levels. For example, the probability of a conscious state (e.g., the consciousness states corresponding to MOAAS5 and MOAAS4), a light anesthesia state (e.g., the consciousness states corresponding to MOAAS3 and MOAAS2), and an unconscious state (e.g., the consciousness states corresponding to MOAAS1 and MOAAS0). For instance, a predicted value in the probability interval [0.9, 1] indicates the target object is conscious; a predicted value in the probability interval [0.8, 0.9) indicates the target object is in a light anesthesia state; and a predicted value in the probability interval [0, 0.2] indicates the target object is unconscious. The content regarding the training of the first state level, the second state level, and the consciousness determination model can be found elsewhere in this invention (e.g., Figure 3 , Figure 6 ).

[0168] Figure 5 This is a schematic diagram of a training method for a consciousness determination model according to some embodiments of the present invention.

[0169] In some embodiments, the training process 500 of the consciousness determination model can be executed by the consciousness determination system. For example... Figure 5 As shown, the training method 500 for the consciousness determination model includes the following steps.

[0170] Step S510: Divide the negative sample EEG signals into multiple negative sample subsets, wherein the number of samples in each negative sample subset is the same as the number of samples in the positive sample EEG signals.

[0171] A negative sample subset refers to a collection of multiple negative EEG signals selected from a pool of negative sample EEG signals. For example, a negative sample subset could be a collection of 50 (or other) negative sample EEG signals.

[0172] In some embodiments, the consciousness determination system divides the EEG signals into a predetermined number of negative sample subsets based on the number of negative sample EEG signals and the number of positive sample EEG signals. For example, in response to the fact that the number of negative sample EEG signals (e.g., n) is larger than the number of positive sample EEG signals (e.g., m), the consciousness determination system divides the negative sample EEG signals into k negative sample subsets based on the number of positive sample EEG signals, where the number of negative sample EEG signals in each negative sample subset is m, and k = n / m, where k is an integer.

[0173] Step S520: Combine each negative sample subset with the positive sample EEG signal to obtain multiple training sample subsets.

[0174] The training sample subset refers to the set of samples used to train the consciousness determination model, which can be constructed based on positive sample EEG signals and / or negative sample EEG signals.

[0175] In some embodiments, the consciousness determination system can add positive sample EEG signals (e.g., m positive sample EEG signals) to each of multiple (e.g., k) negative sample subsets to obtain multiple (e.g., k) training sample subsets. That is, each training sample subset includes m positive sample EEG signals and m negative sample EEG signals.

[0176] Step S530: Train a consciousness determination model based on multiple training sample subsets.

[0177] In some embodiments, the consciousness determination system can use multiple subsets of training samples or a portion thereof as training samples to train an initial consciousness determination model to obtain a trained consciousness determination model. More information about the consciousness determination model and training process can be found elsewhere in this invention (e.g., Figure 3 , Figure 6 ).

[0178] In some embodiments, for each subset of training samples, the consciousness determination system divides the subset of training samples into multiple training sample groups, and trains a consciousness determination model based on the multiple training sample groups.

[0179] A training sample set refers to a collection of positive and / or negative EEG signals extracted from a subset of training samples. For example, a training sample set may be a collection of positive and negative EEG signals selected from a subset of training samples, representing a predetermined proportion (e.g., 10%).

[0180] In some embodiments, the number of positive sample EEG signals and the number of negative sample EEG signals in each training sample group meet a first preset condition, and the number of positive sample EEG signals corresponding to each first state level in the positive sample EEG signals meets a second preset condition.

[0181] The first preset condition refers to the condition that the number of positive and negative EEG signals in each training sample group must meet. For example, the first preset condition could be that the number of positive and negative EEG signals is within a preset threshold range (such as 40, 50, etc.).

[0182] In some embodiments, the first preset condition includes the ratio of the number of positive sample EEG signals to the number of negative sample EEG signals within a first ratio threshold range. The first ratio threshold range may be a preset interval, for example, [1.2, 0.8].

[0183] In some embodiments, the ratio of the number of positive sample EEG signals to the number of negative sample EEG signals is 1, that is, the number of positive samples is equal to the number of negative samples. It should be noted that the first preset condition can be set according to actual needs. For example, if it is desired that the consciousness determination model be more sensitive to positive sample EEG signals, the proportion of positive sample EEG signals can be adjusted towards a larger range of the first ratio threshold, so that the proportion of positive sample EEG signals in the training sample group is greater.

[0184] The second precondition refers to the requirement that the number of positive EEG samples corresponding to each first state level in the training sample group must meet. Specifically, the positive EEG samples corresponding to the first state level include the fifth, fourth, third, and second sample EEG signals. For more information on the first state level and its corresponding positive EEG samples, please refer to [link to relevant documentation]. Figure 3 And its description.

[0185] In some embodiments, the second preset condition can be used to balance the number of positive EEG samples corresponding to different first state levels. Considering that in actual medical scenarios, the number of positive EEG samples corresponding to different first state levels can vary significantly—for example, the number of third and second EEG samples may be smaller than that of the fifth and fourth sample EEG signals—the consciousness determination system can divide the training sample groups based on the actual number of positive EEG samples corresponding to each first state level using the second preset condition.

[0186] In some embodiments, the second preset condition includes a ratio between the number of positive sample EEG signals at each first state level within a second ratio threshold range. The second ratio threshold range can be a preset interval, which can be the same as the first ratio threshold range. As an example only, in each training sample group, the number of fifth sample EEG signals, fourth sample EEG signals, third sample EEG signals, and second sample EEG signals are all equal.

[0187] It should be noted that, for each subset of training samples, when dividing into multiple training sample groups based on the first and / or second preset conditions, considering that the number of fifth, fourth, third, and second sample EEG signals within the subset of training samples may differ, the consciousness determination system can further determine multiple training sample groups based on methods such as resampling. For example, for the third sample EEG signal, when its quantity is minimal, when dividing into multiple training sample groups, the third sample EEG signal or a portion thereof can be repeatedly assigned to one or more training sample groups to ensure that the first and / or second preset conditions are met.

[0188] In some embodiments, the consciousness determination system can also set up a test sample set for verifying the consciousness determination model obtained during the training phase. The training samples and test samples in the test sample set can be sample EEG signals (e.g., positive and / or negative sample EEG signals) from the same data source (e.g., EEG information database 150). As an example only, for sample EEG signals (e.g., positive and negative sample EEG signals) from multiple sample objects, the consciousness determination system can generate multiple training samples and multiple test samples from the sample EEG signals in a preset ratio (e.g., 7:3). Multiple training samples are used to generate a training sample set, and multiple test samples are used to generate a test sample set.

[0189] In some embodiments, the consciousness assessment system can evaluate the performance of the trained consciousness assessment model using a test sample set and preset evaluation metrics. In some embodiments, the evaluation metrics include classification evaluation metrics to assess the accuracy of the trained consciousness assessment model in predicting the state of consciousness corresponding to different test samples. Classification evaluation metrics include, but are not limited to, true positive rate, true negative rate, F1 score, and Matthews correlation coefficient (MCC). In some embodiments, the evaluation metrics also include volatility metrics to reflect the stability of the model output. The smaller the value of the volatility metric, the higher the stability of the trained consciousness assessment model in predicting the state of consciousness corresponding to different test samples. In some embodiments, the evaluation metrics include the Pearson correlation coefficient (PCC) between the model output and the MOAAS score, used to assess whether the output of the trained consciousness assessment model (such as predicting the state of consciousness corresponding to the EEG signal) has a good correlation with clinical standards (such as the MOAAS score).

[0190] In some embodiments of the present invention, on the one hand, dividing the negative sample EEG signals into multiple negative sample subsets based on the number of positive sample EEG signals and generating training sample subsets can avoid the problem of an imbalance in the number of positive and negative sample EEG signals. On the other hand, by balancing the amount of positive sample EEG signals corresponding to each first state level, the problem of insufficient number of positive sample EEG signals (such as third sample EEG signals, second sample EEG signals) corresponding to the actual light anesthesia state is avoided, enabling the training samples to be generated normally. At the same time, it allows the consciousness determination model to fully learn the characteristics of EEG signals in different non-unconscious states, thereby making the model's determination of consciousness state more accurate.

[0191] Figure 6 This is a schematic diagram of a training method for another consciousness determination model according to some embodiments of the present invention.

[0192] In some embodiments, the consciousness determination model can be obtained through phased training based on training samples.

[0193] like Figure 6 As shown, the consciousness determination system can perform a first-stage training on the initial determination model 610 based on the first training sample 601 to obtain a first determination model 620; perform a second-stage training on the first determination model 620 based on the second training sample 602 to obtain a second determination model 630; perform a third-stage training on the first determination model 620 based on the third training sample 603 to obtain a third determination model 640; and determine the consciousness determination model 410 based on the second determination model 630 and / or the third determination model 640.

[0194] The first stage of training refers to the training phase in which the first decision model 620 is obtained. In some embodiments, the first decision model 620 is a pre-trained binary classification model, and the initial decision model is the initial binary classification model. That is, the first stage of training corresponds to binary classification task training, so that the pre-trained first decision model 620 can determine whether the consciousness state corresponding to the target EEG signal is a non-unconscious state or an unconscious state. For more information on binary classification models, target EEG signals, non-unconscious states, and unconscious states, please refer to... Figure 3 And its description.

[0195] In some embodiments, the first training sample 601 includes sample conscious state data and sample unconscious state data. In some embodiments, the sample conscious state data includes a fifth sample EEG signal and a fourth sample EEG signal; the sample unconscious state data includes a first negative sample EEG signal and a second negative sample EEG signal.

[0196] During the first phase of training, the consciousness determination system can construct multiple sets of positive EEG signals based on the data of the conscious state samples, with the training label set to 1; and construct multiple sets of negative EEG signals based on the data of the unconscious state samples, with the training label set to 0. Thus, the first training sample 601 is generated based on the multiple sets of positive and negative EEG signals.

[0197] During training, the value of the loss function is determined based on the difference between the output of the initial decision model and the training labels. The parameters of the initial decision model can be iteratively updated based on the value of the loss function until the training termination condition is met (e.g., the loss function converges, a certain number of iterations are performed, etc.). The updated initial decision model can be used as the trained first decision model 620. The loss function can be the cross-entropy loss function, which can be constructed according to the equation (1) corresponding to the first loss term.

[0198] For more information on the data of the conscious state of the samples, the data of the unconscious state of the samples, the EEG signals of the fifth and fourth samples, the EEG signals of the first and second negative samples, and the first loss term, please refer to [link to relevant documentation]. Figure 3 And its description.

[0199] The second stage of training refers to the training phase in which the second decision model 630 is obtained, which can be performed after the first stage of training has obtained the first decision model 620. In some embodiments, the consciousness determination system uses the trained first decision model 620 as the initial model for the second stage of training (hereinafter referred to as the second initial model) based on the transfer learning method, so as to retain the parameters of the first decision model 620 after the first stage of training.

[0200] In some embodiments, the second training sample 602 includes sample awake state data and sample light anesthesia state data. In some embodiments, the sample light anesthesia state data includes third sample EEG signal and second sample EEG signal. The consciousness determination system constructs a first loss term based on equation (1), constructs a second loss term based on equation (4), and then constructs a loss function corresponding to the second initial model based on the first loss term and the second loss term, such as equation (5).

[0201] For more information on data from samples in a state of light anesthesia, data from samples in a state of unconsciousness, and EEG signals from the third and second samples, as well as the second loss term, please refer to [link to relevant documentation]. Figure 3 And its description.

[0202] In the second phase of training, the consciousness determination system constructs a single-sample set and a positive-sample pair set based on sample conscious state data and sample light anesthesia state data. The single-sample set is used for training a binary classification task based on the second initial model, with its loss value determined based on the value of the first loss term. The positive-sample pair set is used for training a ranking task based on the second initial model, with its loss value determined based on the value of the second loss term, to constrain the probabilities of the non-unconscious states corresponding to the conscious state data and the light anesthesia state data in the positive sample pairs to satisfy a second difference (such as satisfying a first preset bias condition). For more information on single-sample sets, positive-sample pair sets, the second difference, the first preset bias condition, and training, please refer to [link to relevant documentation]. Figure 3 And its description.

[0203] The third stage of training refers to the training phase in which the third decision model 640 is obtained. It can be performed after the first stage of training has obtained the first decision model 620. In some embodiments, the second and third stages of training can be performed simultaneously. The consciousness decision system can use the trained first decision model 620 as the initial model for the third stage of training (hereinafter referred to as the third initial model) based on transfer learning. The second and third initial models have the same parameters, thereby enabling the second and third stages of training to be conducted based on the binary classification task completed in the first stage of training, improving training efficiency.

[0204] In some embodiments, the third training sample 603 includes sample data of light anesthesia and sample data of unconsciousness. In some embodiments, the sample data of unconsciousness includes the EEG signal of the first negative sample and the EEG signal of the second negative sample. The loss function corresponding to the third initial model includes a first loss term and a third loss term. The consciousness determination system constructs the first loss term based on equation (1), constructs the third loss term based on equation (6), and then constructs the loss function corresponding to the third initial model based on the first loss term and the third loss term. For more information on the first negative sample EEG signal, the second negative sample EEG signal, and the third loss term, please refer to [link to relevant documentation]. Figure 3 And its description.

[0205] In the third phase of training, the consciousness determination system constructs a single-sample set based on sample data of light anesthesia and sample data of unconsciousness, and a set of negative sample pairs (such as a negative sample pair combining the EEG signals of the first and second negative samples) based on sample data of unconsciousness. The single-sample set is used for training a binary classification task based on the third initial model, with its loss value determined based on the value of the first loss term. The negative sample pair set is used for training a ranking task based on the third initial model, with its loss value determined based on the value of the third loss term, to constrain the probability of unconsciousness corresponding to the first and second negative sample EEG signals in the negative sample pair to satisfy the third difference (such as satisfying the second preset bias condition). For more information on the negative sample pair set, the third difference, the second preset bias condition, and training, please refer to [link to relevant documentation]. Figure 3 And its description.

[0206] In some embodiments, the consciousness determination system may be based on either a trained second determination model 630 or a third determination model 640 as the consciousness determination model 410.

[0207] The above descriptions of the first, second, and third stages of training are merely examples. Actual training stages can vary depending on the number and difficulty of obtaining training samples. For instance, the consciousness determination system can also construct training samples based on the second determination model 630 or the third determination model 640, using data on conscious state, light anesthesia, and unconscious states. Furthermore, it can train the system by constructing a loss function based on the first, second, and third loss terms to obtain a trained consciousness determination model 410.

[0208] In some embodiments of the present invention, a trained consciousness determination model is obtained through phased training, which reduces the difficulty of training. Furthermore, considering that there is a large amount of readily available data on conscious and unconscious states, but a relatively small amount of data on lightly anesthetized states, phased model training avoids the impact of imbalanced training sample data on training, improving training speed while also making training more targeted.

[0209] Figure 7 This is a schematic diagram of the structure of an anesthesia consciousness determination device according to some embodiments of the present invention.

[0210] like Figure 7 As shown, the anesthesia consciousness determination device 700 includes a processor 710 and a storage medium 720 coupled to the processor 710.

[0211] The storage medium 720 stores program instructions for implementing the methods of any of the above embodiments. The processor 710 executes the program instructions stored in the storage medium 720 to implement the steps of the above method embodiments. The processor 710 may also be referred to as a CPU (Central Processing Unit). The processor 710 may be an integrated circuit chip with signal processing capabilities. The processor 710 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor may be a microprocessor or any conventional processor.

[0212] In some embodiments, the processor 710 is configured to execute an anesthesia consciousness determination method, the method comprising: acquiring target EEG signals of a target object; and determining the consciousness state information of the target object based on the target EEG signals using a consciousness determination model, wherein the consciousness state includes a conscious state, a lightly anesthetized state, and an unconscious state; wherein the consciousness determination model is a machine learning model, the consciousness determination model is obtained by training on training samples, the training samples include sample EEG signals, and the sample EEG signals include sample conscious state data, sample lightly anesthetized state data, and sample unconscious state data.

[0213] In some embodiments, the consciousness determination model includes a classification model, where consciousness state information represents the probability that the target object is in a non-unconscious state. The consciousness determination model is trained based on positive sample EEG signals and negative sample EEG signals, wherein the positive sample EEG signals include sample awake state data and sample light anesthesia state data, and the negative sample EEG signals include sample unconscious state data.

[0214] In some embodiments, the consciousness determination model includes a data preprocessing layer, a feature extraction layer, and an exponent calculation layer. Based on the target EEG signal, the consciousness determination model determines the consciousness state information of the target object by: preprocessing the target EEG signal based on the data preprocessing layer to obtain a preprocessed target EEG signal, wherein the preprocessing includes at least one of segmentation, filtering, noise reduction, and downsampling; extracting features from the preprocessed target EEG signal based on the feature extraction layer to obtain at least one feature; and processing the at least one feature based on the exponent calculation layer to determine the consciousness state information of the target object.

[0215] In some embodiments, the processor 710 is further configured to perform: constructing a loss function based on sample conscious state data, sample light anesthesia state data, and sample unconscious state data, the loss function including a first loss term and a second loss term, wherein the first loss term is used to indicate a first difference in the output of the consciousness determination model, the first difference characterizing the difference between the predicted consciousness state corresponding to the training sample and the true consciousness state corresponding to the training sample, and the second loss term is used to indicate a second difference in the output of the consciousness determination model, the second difference characterizing the difference between the probabilities of the consciousness state of positive sample EEG signals corresponding to multiple first state levels being a non-unconscious state; and adjusting the model parameters of the consciousness determination model based on the loss function.

[0216] In some embodiments, the loss function further includes a third loss term, which indicates a third difference in the output of the consciousness determination model. The third difference characterizes the difference between the probabilities of the consciousness state of the negative sample EEG signals corresponding to multiple second state levels being a non-unconscious state.

[0217] In some embodiments, the processor 710 is further configured to perform: dividing the negative sample EEG signals into multiple negative sample subsets, wherein the number of samples in each negative sample subset is the same as the number of samples in the positive sample EEG signals; combining each negative sample subset with the positive sample EEG signals respectively to obtain multiple training sample subsets; and training a consciousness determination model based on the multiple training sample subsets.

[0218] In some embodiments, the processor 710 is further configured to perform: training a consciousness determination model based on multiple training sample subsets, including: for each training sample subset, dividing the training sample subset into multiple training sample groups, wherein the number of positive sample EEG signals and the number of negative sample EEG signals in each training sample group satisfy a first preset condition, and the number of positive sample EEG signals corresponding to each first state level satisfies a second preset condition; and training a consciousness determination model based on the multiple training sample groups.

[0219] In some embodiments, the processor 710 is further configured to perform: a first stage of training on an initial determination model based on a first training sample to obtain a first determination model, wherein the first training sample includes sample conscious state data and sample unconscious state data; a second stage of training on the first stage of the consciousness determination model based on a second training sample to obtain a second determination model, wherein the second training sample includes sample conscious state data and sample light anesthesia state data; a third stage of training on the first stage of the consciousness determination model based on a third training sample to obtain a third determination model, wherein the third training sample includes sample light anesthesia state data and sample unconscious state data; and determining a consciousness determination model based on the second determination model and / or the third determination model.

[0220] Figure 8This is a schematic diagram of the structure of a computer-readable storage medium according to some embodiments of the present invention.

[0221] like Figure 8 As shown, the computer-readable storage medium 800 stores program instructions 810, which, when executed, implement the methods provided in the above embodiments of the present invention. The program instructions 810 can form a program file and be stored in the computer-readable storage medium 800 in the form of a software product, so that a computer device (which may be a personal computer, server, or network device, etc.) or processor can execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned computer-readable storage medium 800 includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or terminal devices such as computers, servers, mobile phones, and tablets.

[0222] It should be noted that the above descriptions of the processes and diagrams are merely illustrative and do not limit the scope of the invention. Those skilled in the art can make various modifications and changes to the processes under the guidance of this invention. However, these modifications and changes are still within the scope of this invention.

[0223] The basic concepts have been described above. It is clear that the detailed disclosure above is merely illustrative and does not constitute a limitation of the present invention. Although not explicitly stated herein, various modifications, improvements, and corrections may be made to the present invention by those skilled in the art. Such modifications, improvements, and corrections are suggested in this invention and therefore remain within the spirit and scope of the exemplary embodiments of the present invention.

[0224] Meanwhile, specific terms are used to describe embodiments of the invention. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the invention. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this invention do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics in one or more embodiments of the invention can be appropriately combined.

[0225] For each patent, patent application, patent application publication, and other material, such as articles, books, specifications, publications, and documents, referenced in this invention, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this invention, as well as documents that limit the broadest scope of the claims of this invention (currently or subsequently appended to this invention). It should be noted that if there are any inconsistencies or conflicts between the descriptions, definitions, and / or terminology used in the appended materials of this invention and the content of this invention, the descriptions, definitions, and / or terminology used in this invention shall prevail.

[0226] Finally, it should be understood that the embodiments described in this invention are merely illustrative of the principles of the invention. Other modifications may also fall within the scope of this invention. Therefore, alternative configurations of the embodiments of this invention are considered as examples and not limitations, and are regarded as consistent with the teachings of this invention. Accordingly, the embodiments of this invention are not limited to those explicitly described and illustrated herein.

Claims

1. A method for determining anesthesia consciousness, characterized in that, include: Acquire the target brainwave signals of the target object; Based on the target EEG signal, the consciousness state information of the target object is determined by a consciousness determination model. The consciousness state includes a conscious state, a lightly anesthetized state, and an unconscious state. The consciousness determination model is a machine learning model, trained on training samples, which include sample EEG signals. These sample EEG signals include data on conscious state, data on light anesthesia, and data on unconscious state. The consciousness determination model also includes a classification model, where the consciousness state information represents the probability that the target object is in a non-unconscious state. This model is trained on positive and negative sample EEG signals, where the positive sample EEG signals include data on conscious state and data on light anesthesia, and the negative sample EEG signals include data on unconscious state. The consciousness determination model is obtained based on training samples and includes: A loss function is constructed based on the sample conscious state data, the sample light anesthesia state data, and the sample unconscious state data. The loss function includes a first loss term and a second loss term. The first loss term is used to indicate the first difference output by the consciousness determination model. The first difference represents the difference between the predicted consciousness state corresponding to the training sample and the actual consciousness state corresponding to the training sample. The second loss term is used to indicate the second difference output by the consciousness determination model. The second difference represents the difference between the probabilities of the consciousness state of the positive sample EEG signals corresponding to multiple first state levels being the non-unconscious state. Based on the loss function, adjust the model parameters of the consciousness determination model; The consciousness determination model, trained based on training samples, also includes: The negative sample EEG signals are divided into multiple negative sample subsets, wherein the number of samples in each negative sample subset is the same as the number of samples in the positive sample EEG signals; Each negative sample subset is combined with the positive sample EEG signal to obtain multiple training sample subsets; For each training sample subset, the training sample subset is divided into multiple training sample groups, wherein the number of positive sample EEG signals and the number of negative sample EEG signals in each training sample group satisfy a first preset condition, and the number of positive sample EEG signals corresponding to each first state level satisfies a second preset condition. The consciousness determination model is obtained by training based on the multiple training sample groups.

2. The method according to claim 1, characterized in that, The loss function further includes a third loss term, which is used to indicate the third difference output by the consciousness determination model. The third difference represents the difference between the probabilities of the conscious state being the unconscious state of the negative sample EEG signals corresponding to multiple second state levels.

3. The method according to claim 1, characterized in that, The consciousness determination model includes a data preprocessing layer, a feature extraction layer, and an index calculation layer. The process of determining the consciousness state information of the target object based on the target EEG signal through the consciousness determination model includes: Based on the data preprocessing layer, the target EEG signal is preprocessed to obtain the preprocessed target EEG signal. The preprocessing includes at least one of the following: slice processing, filtering processing, noise reduction processing, and downsampling processing. Based on the feature extraction layer, feature extraction is performed on the preprocessed target EEG signal to obtain at least one feature; Based on the index calculation layer, the at least one feature is processed to determine the consciousness state information of the target object.

4. A system for determining anesthesia consciousness, characterized in that, include: The acquisition module is configured to acquire the target EEG signal of the target object. The determination module is configured to determine the consciousness state information of the target object based on the target EEG signal and through a consciousness determination model. The consciousness state information includes a conscious state, a lightly anesthetized state, and an unconscious state. The consciousness determination model is a machine learning model, which is trained based on training samples, including conscious state data, lightly anesthetized state data, and unconscious state data. The determining module is further configured to: A loss function is constructed based on the sample conscious state data, the sample light anesthesia state data, and the sample unconscious state data. The loss function includes a first loss term and a second loss term. The first loss term is used to indicate the first difference output by the consciousness determination model. The first difference represents the difference between the predicted consciousness state corresponding to the training sample and the actual consciousness state corresponding to the training sample. The second loss term is used to indicate the second difference output by the consciousness determination model. The second difference represents the difference between the probabilities of the consciousness state being a non-unconscious state among the positive sample EEG signals corresponding to multiple first state levels. Based on the loss function, adjust the model parameters of the consciousness determination model; The determining module is further configured to: The negative sample EEG signals are divided into multiple negative sample subsets, wherein the number of samples in each negative sample subset is the same as the number of samples in the positive sample EEG signals; Each negative sample subset is combined with the positive sample EEG signal to obtain multiple training sample subsets; For each training sample subset, the training sample subset is divided into multiple training sample groups, wherein the number of positive sample EEG signals and the number of negative sample EEG signals in each training sample group satisfy a first preset condition, and the number of positive sample EEG signals corresponding to each first state level satisfies a second preset condition. The consciousness determination model is obtained by training based on the multiple training sample groups.

5. A device for determining anesthesia consciousness, characterized in that, Includes at least one storage medium and at least one processor; The at least one storage medium is used to store computer instructions; The at least one processor is used to execute the computer instructions to implement the anesthesia consciousness determination method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions, which, when executed by a processor, implement the anesthesia consciousness determination method as described in any one of claims 1 to 3.

Images