Severe patient sedation state evaluation method based on multi-modal network model

Abstract

The application relates to the technical field of patient state evaluation, and provides a critical patient sedation state evaluation method based on a multi-modal network model, which comprises the following steps: S1, acquiring original physiological data of a critical patient in real time; S2, converting EEG data into an electroencephalogram spectrum diagram; S3, filling the generated missing EEG data into complete EEG data and updating the complete EEG data; S4, cutting the updated complete EEG data through a sliding window to obtain complete EEG data segments; and matching RASS scores as labels with each complete EEG data segment on a time axis to form labeled EEG data; S5, training based on the multi-modal network model and using the labeled EEG data; and S6, RASS score prediction on unlabeled data. The application realizes objective evaluation of the sedation depth of a patient in an intensive care unit by monitoring multi-modal physiological parameters of the patient.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of patient state assessment, and particularly relates to a method for assessing the sedation state of a critical patient based on a multi-modal network model. BACKGROUND

[0002] Critical patients in an intensive care unit (ICU) usually use sedative drugs, which can relieve the pain of patients and promote clinical care. Sedative treatment can reduce stress and protect organ function, but it can also inhibit the important physiological functions of certain organs or increase the metabolic burden of certain organs and cause damage to organ function. However, for patients in the ICU who use sedative drugs, monitoring the sedation state is a challenge. Inappropriate use of sedatives can lead to over-sedation and under-sedation. In clinical practice, long-term or excessive use of sedatives can cause various damage to the human body. Over-sedation can cause prolonged mechanical ventilation, increased medical costs, prolonged ICU hospitalization, mental damage, and other side effects. On the contrary, under-sedation can cause pain, anxiety, and agitation, which can lead to the removal of invasive devices such as intubation and catheterization. Therefore, achieving the optimal sedation state is crucial for improving patient outcomes.

[0003] In clinical practice, the sedation level of critical patients is usually monitored by periodically assessing the behavioral response of patients to stimulation, such as the Richmond Agitation-Sedation Scale (RASS) scale for monitoring sedation level and the Confusion Assessment Method-ICU assessment (CAM-ICU) for detecting delirium. A large amount of evidence shows that the use of these sedation assessment scores and sedation regimens based thereon can effectively reduce the use of mechanical ventilation, shorten the time of tracheal intubation, and shorten the hospitalization time. As an alternative, electroencephalogram monitoring methods have been proposed and initially applied in clinical trials. However, due to the severity of the patient's condition, individual differences in patients, and different types of surgery, the required sedation time will also vary, which makes it challenging to apply electroencephalogram (EEG) signals in the assessment of the sedation level of patients in the ICU.

[0004] At the same time, when using sedative drugs, the ICU also closely monitors the basic vital signs of patients (consciousness, heart rate, respiration, blood pressure, urine output, and body temperature) to select appropriate drugs and their dosages, determine the efficacy of observation and monitoring, develop the best individualized treatment plan, and achieve the minimum adverse reactions and the best efficacy.

[0005] At present, in the clinic, the evaluation of the sedation state of patients is still determined according to the clinical experience of doctors and the corresponding scale, and there is no intelligent method for feature extraction based on electroencephalogram signals and other peripheral physiological signals and identification of the depth of sedation of patients. In fact, patent CN112006658A describes a method for evaluating the depth of sedation using EEG, heartbeat signals and respiratory signals. However, this method is limited because it does not effectively combine the patient's blood pressure, illness and other information collected in the clinic, and does not extract and utilize the features of the EEG signals more deeply. SUMMARY

[0006] The present application mainly solves the technical problem that the evaluation of the sedation state of patients is still determined according to the clinical experience of doctors and the corresponding scale, and there is no intelligent method for feature extraction based on electroencephalogram signals and other peripheral physiological signals and identification of the depth of sedation of patients, and proposes a method for evaluating the sedation state of critically ill patients based on a multi-modal network model, which realizes the objective evaluation of the depth of sedation of patients in the intensive care unit by monitoring the multi-modal physiological parameters of patients, and allows real-time tracking of the sedation state of patients by automatically preprocessing and analyzing the collected physiological parameters to form and verify the sedation level of patients.

[0007] The present application provides a method for evaluating the sedation state of critically ill patients based on a multi-modal network model, comprising:

[0008] Step S1: real-time acquisition of original physiological data of critically ill patients; the physiological data includes EEG data and peripheral data, and the peripheral data includes HR data, SpO2 data and BP data;

[0009] Step S2: preprocessing the original EEG data, and converting the EEG data into an electroencephalogram spectrum graph;

[0010] Step S3: generating missing EEG data, filling the generated missing EEG data into complete EEG data, and updating the complete EEG data;

[0011] Step S4: cutting the updated complete EEG data through a sliding window to obtain complete EEG data segments; and matching the RASS score as a label with each complete EEG data segment on the time axis to form labeled EEG data;

[0012] Step S5: training based on a multi-modal network model using labeled EEG data;

[0013] Step S6: predicting the RASS score of unlabeled data through the trained sedation state evaluation model.

[0014] Further, step S2 includes steps S201 to S203 as follows:

[0015] Step S201: rejecting abnormal points in the collected original EEG data;

[0016] Step S202: denoising the EEG data;

[0017] Step S203: converting the denoised EEG data into an electroencephalogram spectrum diagram.

[0018] Further, step S202 includes: decomposing the EEG data into subbands of different scales, applying threshold processing on the high-frequency subbands, reconstructing the signal after removing the noise, then analyzing the wavelet coefficients on each scale, extracting specific patterns, and finally performing lossy compression on the signal by retaining the low-frequency subbands and discarding the high-frequency subbands exceeding the threshold; the mathematical expression is:

[0019]

[0020] Wherein, x(t) represents an input signal, ψ(t) is a mother wavelet, a is a scaling parameter, and b is a translation parameter.

[0021] Further, step S3 specifically includes steps S301 to S304:

[0022] Step S301: determining the integrity of the EEG data and the peripheral data to obtain complete EEG data, incomplete EEG data, complete peripheral data, and incomplete peripheral data;

[0023] Step S302: training an RBM model using the complete EEG data and the complete peripheral data;

[0024] Step S303: using the trained RBM model to predict the missing data in the incomplete EEG data through the characteristics of the hidden nodes;

[0025] Step S304: randomly initializing the missing data in the incomplete EEG data and generating missing EEG data using the RBM model, and filling the generated missing EEG data into the incomplete EEG data to form complete EEG data.

[0026] Further, step S4 includes steps S401 to S403:

[0027] Step S401: cutting the complete EEG data into segments using a sliding window to obtain complete EEG data segments;

[0028] Step S402: aligning the discrete peripheral data with the complete EEG data segments through data interpolation;

[0029] Step S403, match the RASS score evaluated by the medical staff with each interpolated EEG data segment to form labeled EEG data.

[0030] Further, step S402 aligns the discrete peripheral data with the complete EEG data segment through data interpolation, denoted as:

[0031]

[0032] Wherein, y is the value to be interpolated, x is the position to be inserted, (x1, y1) and (x2, y2) are valid data points before and after invalid or noise data.

[0033] Further, step S5 includes steps S501 to S503 as follows:

[0034] Step S501: using WaveNet model to train labeled EEG signal, labeled HR signal, labeled SpO2 signal and labeled BP signal;

[0035] Step S502: using convolutional neural network model to train the electroencephalogram spectrogram generated in step S2;

[0036] Step S503: based on the multi-modal network model, weighting and fusing the WaveNet model trained in step S501 and the convolutional neural network model trained in step S502 to obtain the final sedation state evaluation model.

[0037] Further, the step S503 includes the following process:

[0038] Let X1 be the instance matrix of the first modality, X2 be the instance matrix of the second modality, Xi be the instance matrix of the i-th modality; N is the number of instances, d i is the dimension of the features extracted by each modality respectively;

[0039] In order to nonlinearly convert the original features of the modalities, deep neural networks are constructed for the two modalities as follows:

[0040] O1=f1(X1;W1)

[0041] O2=f2(X2;W2)

[0042] O i =f i (X i ;W i )

[0043] Wherein, Wi Represents all parameters of the nonlinear transformation. It is the output of the neural network, where d represents the output dimension;

[0044]

[0045] For weighted and fused operations, initialize the hyperparameter α. i Find the optimal values ​​for the weights and fuse the different modalities, as shown below:

[0046]

[0047] This invention provides a method for assessing the sedation status of critically ill patients based on a multimodal network model. It integrates data collected from multiple bedside monitoring devices in the ICU, and performs special weighted fusion of features from each modality. This method monitors the patient's sedation status, completes missing data, and predicts the patient's sedation level, enabling safe monitoring and early warning of sedation in critically ill patients. It boasts advantages such as short response time and high accuracy. This invention can simultaneously monitor sedation depth and identify changes in sedation level, meeting the needs of ICU medical staff to understand the patient's sedation level and providing a reference for subsequent infusion rates. This allows for providing the most suitable sedation state for critically ill patients, avoiding over-sedation or under-sedation. The method according to this invention makes it possible, particularly based on automated analysis of patient physiological parameters such as EEG, HR, and BP, to form and verify the patient's sedation level. Through this invention, ICU patients can be monitored in three dimensions, analyzed in real time, and personalized treatment plans can be developed, quantifying the RASS score to help patients better adapt to the intensive care unit environment. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating the implementation of the sedation status assessment method for critically ill patients based on a multimodal network model provided by the present invention.

[0049] Figure 2 This is a logic diagram of data segmentation and tagging in this invention;

[0050] Figure 3 This is a schematic diagram illustrating the multimodal feature fusion method in this invention.

[0051] Figure 4 This is a graph showing the EEG signal denoising and filtering in this invention;

[0052] Figure 5 The spectrum diagram of the EEG signal generated by this invention;

[0053] Figure 6 This represents the true value of the patient's RASS score during the training process;

[0054] Figure 7 The RASS prediction value generated by the present application in combination with the patient multi-modal data. DETAILED DESCRIPTION

[0055] In order to make the technical problems solved by the present application, the technical solutions adopted and the technical effects achieved more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not limit the present application. In addition, it should be noted that, in order to facilitate the description, only the parts related to the present application are shown in the drawings, not all.

[0056] The present application provides a method for evaluating the sedation state of a critically ill patient based on a multi-modal network model, which can be used to automatically and continuously measure the sedation state of a patient in an intensive care unit (hereinafter referred to as a critically ill patient). Sedation is a medical technology and management strategy, which is usually used to treat patients with acute life-threatening diseases to alleviate symptoms, improve survival conditions or create conditions for further treatment. Its core goal is to adjust the patient's state of consciousness, physical response or psychological burden through drugs or other means to ensure that they remain stable during treatment and reduce pain and risk.

[0057] Upon arrival at the intensive care unit, the patient is connected to multiple parameter monitoring devices. These parameter monitoring devices collect measurements of various physiological variables of the patient. Among these signals extracted from the monitoring devices, the present application relies on EEG measurements, ECG (electrocardiogram) monitoring, NIBP (non-invasive blood pressure), SpO2 measurements. For example, EEG is a graphical representation of the electrical activity of the patient's brain over time. EEG can be obtained by using electrodes placed on the patient's scalp.

[0058] As shown in Figure 1 The method for evaluating the sedation state of a critically ill patient based on a multi-modal network model provided by the embodiment of the present application includes the following processes:

[0059] Step S1: Real-time acquisition of original physiological data of a critically ill patient; the physiological data includes EEG data and peripheral data, and the peripheral data includes HR data, SpO2 data and BP data.

[0060] In this step, EEG (brain wave) data, HR (heart rate) data, SpO2 (blood oxygen saturation) data, BP (blood pressure) data and other physiological data are collected through the rich bedside parameter monitoring devices in the ICU, wherein the HR data, SpO2 data and BP data are peripheral data. Specifically, real-time EEG data, HR data and SpO2 data are collected using electrode patches or electrode caps; a blood pressure measuring device is set to measure BP data every 30 minutes.

[0061] Step S2: Preprocessing the original EEG data and converting the EEG data into an electroencephalogram spectrum diagram.

[0062] This step preprocesses the original EEG data to remove artifacts and low-quality signals in various original EEG data. Specifically, the purpose is to ensure that the collected signals are as complete and acceptable as possible. The direct current offset and low-frequency drift of the signal are eliminated, and the true dynamic characteristics are restored. Random noise and device noise are eliminated, signal features are highlighted, and baseline drift of the signal is removed, power frequency noise is filtered, wavelet denoising is applied, and artifacts caused by non-physiological or external interference are identified and removed. Step S2 specifically includes the following steps S201 to S203:

[0063] Step S201: Rejecting abnormal points in the collected original EEG data.

[0064] After detecting the amplitude value, low standard deviation, and amplitude change of the original EEG data, the threshold method is used to reject abnormal data. The specific process of this step is to detect EEG data with an abnormal high amplitude value exceeding 500 μV, observe EEG data with a low standard deviation <0.2 μV, identify data with a rapid amplitude change exceeding 900 μV within 0.1 s and other abnormal data, and reject them. This step can be realized by automatic combination with manual method.

[0065] Step S202: Denoising the EEG data.

[0066] This step is performed by the wavelet transform method, which is used to filter interference signals in the acquisition process. The specific process is as shown in Figure 2 : The EEG data is decomposed into sub-bands of different scales, threshold processing is applied on the high-frequency sub-band to remove noise, the signal is reconstructed after noise removal, then the wavelet coefficients on each scale are analyzed, specific patterns are extracted, and finally the signal is lossy compressed by retaining the low-frequency sub-band and discarding the high-frequency sub-band exceeding the threshold. The mathematical expression is:

[0067]

[0068] Where x(t) represents the input signal, ψ(t) is the mother wavelet, a is the scaling parameter, and b is the translation parameter.

[0069] Step S203: Converting the denoised EEG data into an electroencephalogram spectrum diagram.

[0070] This step converts the denoised EEG data into an electroencephalogram spectrum diagram (as shown in Figure 3The spectrogram is trained as one of the modalities, so as to perform time-frequency analysis on the EEG data. In this step, the absolute value or square value of time b, frequency f and wavelet coefficient W(a, b) is plotted as an electroencephalogram spectrogram, which can be expressed as:

[0071] Spectrogram(b, f) = |W(a, b)| 2

[0072] Step S3: generating missing EEG data, filling the generated missing EEG data into the complete EEG data, and updating the complete EEG data.

[0073] The complex environment in the ICU and the interference of the patient also cause signal artifacts and missing data. Therefore, the missing data is supplemented in this step. Specifically, the relationship between the EEG data and each peripheral data is learned through a multi-modal learning method, the peripheral data is used as supplementary information, and the missing part of the EEG data is inferred and filled, and vice versa. Step S3 specifically includes the following steps S301 to S304:

[0074] Step S301: determining the completeness of the EEG data and the peripheral data to obtain complete EEG data, incomplete EEG data, complete peripheral data and incomplete peripheral data.

[0075] Incomplete EEG data means that the EEG data is missing, and incomplete peripheral data means that the peripheral data is missing. The method for determining the completeness of the EEG data and the peripheral data is: monitoring the packet loss rate of the EEG data and the peripheral data in real time, and if the continuous loss time is > 2 seconds or the cumulative loss rate is > 5%, it is determined as an incomplete data segment.

[0076] Step S302: training an RBM model using complete EEG data and complete peripheral data.

[0077] The RBM model has a hidden layer and a visible layer, and is a random neural network model connected symmetrically and without self-feedback, fully connected between layers and no connection within layers. The data nodes in the hidden layer are hidden nodes. The data nodes in the visible layer are visible nodes. In the present application, each modality corresponds to a group of visible nodes, and then the data of the visible nodes is input into the RBM model, and the state of the hidden nodes is calculated through the weight and the activation function.

[0078] For a given state vector and The current energy function of the RBM (Restricted Boltzmann Machine) model can be expressed as:

[0079]

[0080] where n v is the number of visible layer neurons, n h is the number of hidden layer neurons, a i is the bias of the i-th neuron in the visible layer, b j is the bias of the j-th neuron in the hidden layer, and W ij is the connection weight between the i-th neuron in the visible layer and the j-th neuron in the hidden layer.

[0081] Based on the energy function, the joint probability distribution of the visible and hidden layers can be expressed as:

[0082]

[0083] where Z is the partition function, which is used to ensure the normalization of the probability distribution.

[0084] Given the visible layer v, the activation probability of the j-th neuron in the hidden layer is:

[0085]

[0086] where σ is the sigmoid function.

[0087] Given the hidden layer h, the activation probability of the i-th neuron in the visible layer is:

[0088]

[0089] The update formulas for the weights W, the visible layer bias a, and the hidden layer bias b are as follows:

[0090] ΔW ij = ε(<v i h j - data <v i h j ) recon

[0091] Δa i = ε(<v i - data <v i ) recon

[0092] Δb i = ε(<h i - data <h i ) recon

[0093] where ε is the learning rate, <·> data ​​​denotes the expectation on the training data, recon denotes the expectation on the reconstructed data.

[0094] The RBM model is trained using the complete EEG data and the complete peripheral data.

[0095] Step S303: Using the trained RBM model, the missing data in the incomplete EEG data is predicted through the features of the hidden nodes.

[0096] The complete EEG data and the complete peripheral data of the visible nodes are input into the RBM model, the state of the hidden nodes is calculated through the weights and the activation function, and the data of the visible nodes is reconstructed through the weights and the activation function according to the state of the hidden nodes, wherein the state of the hidden nodes is the extracted latent features.

[0097] Step S304: The missing data in the incomplete EEG data is randomly initialized, and the RBM model is used to generate the missing EEG data, the generated missing EEG data is filled into the incomplete EEG data to form the complete EEG data. The random initialization can be to fill the missing part with the mean value of the known data.

[0098] Then the multi-modal data is updated according to the following equation:

[0099]

[0100] Where t is the iteration number, l represents the learning rate, v i is the i-th missing value of the sample, denotes the conditional probability of the missing value given the hidden node. In addition, F is an energy function satisfying the following equation:

[0101]

[0102] Where V E denotes the EEG data, V P denotes the peripheral data. θ is the parameter of the model, Σ h denotes all possible hidden nodes.

[0103] Step S4: The updated complete EEG data is cut through a sliding window to obtain complete EEG data segments; and the RASS score is matched with each complete EEG data segment on the time axis to form labeled EEG data;

[0104] This step performs data segmentation and labeling. Specifically, as shown in the data segmentation and labeling process shown in Figure 4 Step S4 specifically includes the following steps S401 to S403:

[0105] Step S401, using a sliding window to cut the complete EEG data into segments to obtain complete EEG data segments.

[0106] The window length is set as w l , and the window sliding step is set as w s . The complete EEG data is time series data, and each complete EEG data segment after cutting is also time series data.

[0107] Step S402, aligning the discrete peripheral data with the complete EEG data segment through a data interpolation method, and the alignment is represented as:

[0108]

[0109] Wherein, y is the value to be interpolated, x is the position of the interpolated value, (x1, y1) and (x2, y2) are valid data points before and after invalid or noise data.

[0110] The peripheral data is discrete data, and the interpolated EEG data formed is also a data segment.

[0111] Step S403, matching the RASS score evaluated by medical staff with each data segment of the interpolated EEG data to form labeled EEG data.

[0112] Specifically, the time point of the RASS score is matched with the time point of the peripheral data, and each data segment has a one-to-one corresponding RASS score (label value).

[0113] Step S5: training based on a multi-modal network model using the labeled EEG data.

[0114] Specifically, the complementarity of multi-modal data is used to fuse different types of information through a multi-modal learning framework to improve the performance and generalization ability of the model. Step S5 includes the following steps S501 to S503:

[0115] Step S501, using a WaveNet model to train the labeled EEG signal, the labeled HR signal, the labeled SpO2 signal and the labeled BP signal.

[0116] The present application uses the basic components of the WaveNet model (time series prediction model) to process the complete EEG data, the input of the model is the complete EEG data, and then rectifies it into a time channel tensor The signal X C of each channel can be represented as:

[0117] X C ={x c,1 ,xc,2 ,…,x c,T},c=1,2,…,C

[0118] Input signals are processed through multiple layers of dilated convolution and residual concatenation. Each layer of dilated convolution enables the network to capture features at different scales, thus improving the expressiveness of the model. Then, the model processes input data from different channels separately. For a channel of input X C , the calculation formula of convolution is:

[0119]

[0120] where K is the size of the convolution kernel, ω k is the weight of the convolution kernel, and x c,t-k is the input value in X C .

[0121] WaveNet is a Neural Vocoder architecture proposed by Google DeepMind in 2016. The model is a probabilistic model based on dilated causal convolution.

[0122] Next, the outputs of each channel are combined through a concatenation layer to form a composite feature vector. Then, the composite vector is passed through a dense layer with ReLU activation function to output multiple features. The mathematical expression of the gated activation unit is as follows:

[0123] z c (t)=tanh(W f *y c (t)+b f )⊙σ(W g *y c (t)+b g )

[0124] where tanh is the hyperbolic tangent activation function. σ is the Sigmoid activation function. W f and W g are the weights of the convolution kernel, b f and b g are the bias terms. ⊙ represents element-wise multiplication.

[0125] The composite feature vector through the concatenation layer is:

[0126] Z=Flatten([Concat(Z1,Z2,…,Z C )])

[0127] where Concat is the concatenation operation in the channel dimension, and Flatten flattens the concatenated tensor into a one-dimensional vector.

[0128] The output z of the gating activation unit c (t) is used to generate the skip connection and the residual connection. The expression of the residual connection is:

[0129] r c (t) = y c (t) + W r *z c (t)

[0130] where W r is the weight of the residual convolution kernel.

[0131] Finally, the fully connected layer with the softmax activation function outputs the probability distribution of different categories.

[0132]

[0133] where K is the number of categories. W k and b k are the weights and bias terms of the softmax layer.

[0134] Furthermore, the Rectified Linear Unit (ReLU) is used as the activation function, and its expression is:

[0135] f(x) = max(0, x)

[0136] This formula means that if the input x is positive, the output is x, and if x is negative, the output is 0.

[0137] The Kullback-Leibler Divergence Loss (KLDivLoss) is used as an indicator to measure the performance of the model, and its expression is:

[0138]

[0139] represents the difference between a probability distribution Q and a reference probability distribution P.

[0140] Step S502, using a convolutional neural network model, training the electroencephalogram spectrum generated in step S2.

[0141] Specifically, the data is first divided into training set, validation set and test set, and then the time-frequency features of the spectrogram are extracted using the convolutional neural network and classified. The cross-entropy loss function is used to calculate the error between the predicted class probability and the true class, and then the model is optimized. The loss is calculated and the weight is updated batch by batch, and after each training round, the model performance is evaluated with the validation set to monitor overfitting.

[0142] Step S503: Based on the multi-modal network model, the WaveNet model trained in step S501 and the convolutional neural network model trained in step S502 are weighted and fused to obtain a final sedation state evaluation model.

[0143] Specifically, as shown in the idea diagram of multi-modal feature fusion. Figure 5

[0144] Let X1be the instance matrix of the first modality, X2be the instance matrix of the second modality, Xibe the instance matrix of the i-th modality. Here, N is the number of instances, d i is the dimension of the features extracted by each modality. In order to nonlinearly convert the original features of the modalities, the present application constructs a deep neural network for the two modalities as follows:

[0145] O1=f1(X1;W1)

[0146] O2=f2(X2;W2)

[0147] O i =f i (X i ;W i )

[0148] where W i represents all parameters of the nonlinear transformation, is the output of the neural network, and d represents the output dimension.

[0149]

[0150] For weighted and fused, the present application initializes the hyperparameter a i , finds the best value of the weight, and fuses different modalities as follows:

[0151]

[0152] Step S6: Through the trained sedation state evaluation model, the RASS score prediction of the unlabeled data is performed.

[0153] ​The present step is a result evaluation step, and the unlabeled data is the original EEG data, HR data, SpO2 data and BP data collected, which are to be subjected to RASS scoring.

[0154] Specifically, the newly collected patient data is used to output a probability distribution of the trained sedation state evaluation model, the class corresponding to the maximum probability is selected as the prediction result, and the prediction result of the model is compared with the RASS score given by the expert to evaluate the prediction accuracy. As follows:

[0155]

[0156] Under the action of the patient sedation state evaluation method based on multi-modal physiological signals in the intensive care unit, as shown in Figure 6 and Figure 7 the present application can effectively and accurately evaluate the RASS score of the critical patient.

[0157] The present application provides a kind of based on multi-modal network model's critical patient sedation state evaluation method, can realize monitoring sedation depth simultaneously, and realize the identification of sedation grade change, satisfy the need of ICU medical staff to master patient sedation grade, and can provide reference for subsequent infusion rate, further provide the most suitable sedation state for critical patient, avoid sedation too deep or too shallow.According to the method of the present application, it is possible to form and verify the sedation level of the patient based on the automatic analysis of the patient physiological activity parameters such as EEG, HR, BP, etc.More particularly, a single modality signal is difficult to comprehensively describe the patient's consciousness state, and the real-time evolution of such multi-parameter makes it possible to comprehensively describe the consciousness state.Through the invention, ICU patients are stereoscopic monitored, real-time analyzed and personalized treatment plans are developed, and RASS score is quantified, so that patients can better adapt to the environment of intensive care unit.

[0158] The present application can fuse the data collected by various bedside monitoring devices in ICU, and simultaneously perform special weighting fusion on the features of each modality, has the function of monitoring the sedation state of the patient, can complete the missing data, and can predict the sedation level of the patient, realizes the safe monitoring and early warning of the sedation state of the critical patient, has the advantages of short response time, high precision, etc.

[0159] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions described in the foregoing embodiments can be modified, or some or all of the technical features can be replaced by equivalents, without changing the essence of the corresponding technical solutions out of the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for assessing the sedation status of critically ill patients based on a multimodal network model, characterized in that, include: Step S1: Acquire raw physiological data of critically ill patients in real time; The physiological data includes EEG data and peripheral data, which includes HR data, SpO2 data, and BP data. Step S2: Preprocess the raw EEG data and convert it into an electroencephalogram (EEG) spectrogram; Step S3: Generate missing EEG data. Fill the missing EEG data generated using the RBM model into the complete EEG data and update the complete EEG data. Step S4: Segment the updated complete EEG data using a sliding window to obtain complete EEG data segments; align the discrete peripheral data with the complete EEG data segments using data interpolation; and match the RASS score as a label with each complete EEG data segment along the time axis to form labeled EEG data. Step S5: Based on the multimodal network model, train it using labeled EEG data; Step S5 includes the following steps S501 to S503: Step S501: Use the WaveNet model to train the labeled EEG signal, labeled HR signal, labeled SpO2 signal, and labeled BP signal; Step S502: Use a convolutional neural network model to train the EEG spectrogram generated in step S2; Step S503: Based on the multimodal network model, the WaveNet model trained in step S501 and the convolutional neural network model trained in step S502 are weighted and fused to obtain the final sedation state assessment model. Step S6: Predict RASS scores on unlabeled data using the trained sedation assessment model.

2. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 1, characterized in that, Step S2 includes the following steps S201 to S203: Step S201: Remove outliers from the collected raw EEG data; Step S202: Denoise the EEG data; Step S203: Convert the denoised EEG data into an electroencephalogram (EEG) spectrogram.

3. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 2, characterized in that, Step S202 includes: decomposing the EEG data into subbands of different scales, applying threshold processing to the high-frequency subbands to remove noise and reconstruct the signal, then analyzing the wavelet coefficients at each scale to extract specific patterns, and finally performing lossy compression on the signal by retaining the low-frequency subbands and discarding the high-frequency subbands that exceed the threshold; the mathematical expression is: ； in, Indicates the input signal. It is the mother wavelet. For scaling parameters, These are the translation parameters.

4. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 2, characterized in that, Step S3 specifically includes the following steps S301 to S304: Step S301: Determine the integrity of EEG data and peripheral data to obtain complete EEG data, incomplete EEG data, complete peripheral data, and incomplete peripheral data; Step S302: Train the RBM model using complete EEG data and complete peripheral data; Step S303: Using the trained RBM model, predict missing data in incomplete EEG data by using the features of hidden nodes; Step S304: Randomly initialize the missing data in the incomplete EEG data, generate missing EEG data using the RBM model, and fill the incomplete EEG data with the generated missing EEG data to form complete EEG data.

5. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 1, characterized in that, Step S4 includes the following steps S401 to S403: Step S401: Use a sliding window to cut the complete EEG data into segments to obtain complete EEG data segments; Step S402: Align the discrete peripheral data with the complete EEG data segment using data interpolation. Step S403: Match the RASS score assessed by medical staff with the data segment of each interpolated EEG data to form labeled EEG data.

6. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 5, characterized in that, Step S402: Align the discrete peripheral data with the complete EEG data segment using data interpolation, as shown below: ； in, It is the value to be interpolated. This is the position where the interpolation value will be inserted. and These are the valid data points before and after invalid or noisy data, respectively.

7. The method for assessing the sedation status of critically ill patients based on a multimodal network model according to claim 5, characterized in that, Step S503 includes the following process: make This is the instance matrix of the first mode. This is the instance matrix of the second mode. For the first The instance matrix of modalities; It is the number of instances. These are the dimensions of the features extracted from each modality; To non-linearly transform the original features of the modalities, a deep neural network was constructed for both modalities, as shown below: ； in, Represents all parameters of the nonlinear transformation. It is the output of the neural network. Indicates the output dimension; ； For weighted and fused operations, initialize hyperparameters. Find the optimal values ​​for the weights and fuse the different modalities, as shown below: 。

Images