A method and system for prognosis prediction of chronic doc patients

Abstract

The application provides a prognosis prediction method and system for chronic Doc patients, relates to the technical field of clinical prognosis evaluation, and collects multi-modal data such as neuroimaging, neuroelectrophysiological data, biochemical indexes and clinical data, carries out standardized processing and feature extraction, and fuses spatial features, time sequence features and key features.Combining a trained consciousness state grading evaluation model and a prognosis prediction model, a feature level data set is generated, and the consciousness state grading and prognosis development trend of the patient are accurately predicted.The application effectively integrates the complementary information of multi-modal data, improves the accuracy and reliability of prognosis prediction, and provides scientific support for clinical decision-making, individualized treatment and medical resource optimization.

Description

Technical Field

[0001] This invention relates to the field of clinical prognostic assessment technology, and in particular to a method and system for predicting the prognosis of patients with chronic Doc. Background Technology

[0002] Assessment of consciousness and prognosis prediction in patients with chronic disorder of consciousness (Doc) are core issues in clinical treatment. However, existing single-modal data analysis methods are insufficient to meet complex clinical needs, resulting in low accuracy in diagnosis and prognostic assessment, high rates of missed diagnoses, and a lack of personalized treatment strategies.

[0003] In recent years, multimodal brain data fusion technology has gradually become a research hotspot. By integrating multi-dimensional indicators such as brain neural circuits, metabolism, connectivity, electrophysiology, and biochemistry, it is possible to deeply mine hidden information and reveal the diversity of brain network connectivity and neural activity. However, research on multimodal brain data fusion models at home and abroad is still limited, especially in the grading assessment and prognostic prediction of chronic dysplasia (CD) patients, where standardized and intelligent solutions have not yet been formed. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for predicting the prognosis of patients with chronic Doc (Chronic Disease) that can accurately predict the patient's level of consciousness and prognostic trends, provide scientific decision support for clinicians, optimize personalized treatment plans, improve the accuracy and reliability of prognostic prediction, and thus improve the patient's rehabilitation effect and the efficiency of medical resource utilization.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A prognostic prediction method for patients with chronic Doc includes:

[0007] Multimodal data of patients with chronic Doc were collected, and each type of multimodal data was standardized to obtain a standardized dataset; the multimodal data included: neuroimaging data, neuroelectrophysiological data, biochemical indicators, and clinical data;

[0008] Feature extraction is performed on each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data.

[0009] The spatial features, the time series features, and the key features are fused to obtain a fused feature vector.

[0010] The fused feature vector is input into the trained consciousness state hierarchical evaluation model to obtain the classification results and intermediate layer features of the model;

[0011] The classification results are used as labels, and combined with the features of the intermediate layer of the model to generate a feature-level dataset.

[0012] The feature-level dataset is input into the trained patient prognosis prediction model to obtain the prediction results.

[0013] Preferably, the standardization process includes: noise reduction, normalization, and format conversion.

[0014] Preferably, feature extraction is performed on each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time-series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data, including:

[0015] For neuroimaging data, convolutional neural networks are used to extract spatial features;

[0016] Time-frequency analysis was used to extract time-series features from neuroelectrophysiological data.

[0017] Key features were extracted from biochemical indicators and clinical data using a random forest algorithm.

[0018] Preferably, time-series features are extracted from neuroelectrophysiological data using time-frequency analysis methods, including:

[0019] Time-frequency features are extracted using the short-time Fourier transform method; the expression for the time-frequency features is: Where x(τ) represents the original neurophysiological data, w(τ-t,σ) is the time window function, t is the time point representing the local spectrum of the signal at different time intervals, f is the frequency representing the frequency components of the signal, X(t,f) is the time-frequency characteristic representing the frequency distribution of the signal at each time point, and σ is the width of the window function, which is dynamically adjusted as follows: α is an adjustment parameter used to control the sensitivity of the window width to the signal amplitude;

[0020] The power spectral density is extracted based on the time-frequency features, and then normalized to obtain the normalized power spectral density; the expression for the power spectral density is: P(t,f)=|X(t,f)| 2 Where P(t,f) is the power spectral density, representing the energy intensity of the signal at time t and frequency f; the expression for the normalized power spectral density is: Among them, f min and f max P represents the minimum and maximum values ​​of the analysis frequency range, respectively. norm(t,f) represents the normalized power spectral density;

[0021] Nonlinear feature analysis is performed based on short-time multi-scale entropy to obtain frequency-weighted multi-scale entropy; wherein, the expression for the short-time multi-scale entropy is as follows: Where E(t,τ) is the short-time multiscale entropy at time t and time scale τ, used to quantify the complexity of the signal and reflect the dynamic change pattern of the neurophysiological signal, τ is the time scale, representing the window size, and p i (τ) represents the probability distribution of the signal at time scale τ; the expression for the frequency-weighted multi-scale entropy is: E f (t,τ)=∑ f w(f)E(t,τ,f); where E(t,τ,f) is the short-time multi-scale entropy at frequency f, w(f) is the frequency weighting function, which is adaptively adjusted according to the importance of a specific frequency band, and E f (t,τ) represents the frequency-weighted multi-scale entropy;

[0022] The normalized power spectral density is converted into a time-frequency weighted feature; the expression for the time-frequency weighted feature is: Among them, f β For frequency weighting, when β>0, high-frequency components are emphasized, and when β<0, low-frequency components are emphasized. -F(t) is the time-frequency weighted feature, which represents the projection of the time-frequency feature onto the time dimension. σ'(t) is the local standard deviation of the signal in the neurophysiological data, and μ(t) is the local mean of the signal in the neurophysiological data.

[0023] The time-frequency features, the normalized power spectral density, the frequency-weighted multi-scale entropy, and the time-frequency weighted features are concatenated into multi-dimensional features to obtain the fused time series features.

[0024] Preferably, the key features include: blood glucose, lactate dehydrogenase, C-reactive protein, medical history, behavioral assessment scales, and medication records.

[0025] Preferably, the fused feature vector is input into the trained consciousness state hierarchical assessment model to obtain classification results and intermediate layer features of the model, including:

[0026] Construct an initial depth model based on a multilayer perceptron;

[0027] The preset sample fusion feature vector and sample consciousness state level are input into the initial deep model to obtain the trained consciousness state hierarchical evaluation model; the loss function of the consciousness state hierarchical evaluation model is the cross-entropy loss function; the optimizer of the cross-entropy loss function is the Adam optimizer; the network structure of the consciousness state hierarchical evaluation model includes: an input layer, a first fully connected layer, a second fully connected layer and an output layer connected in sequence;

[0028] The fused feature vector is input into the trained consciousness state hierarchical evaluation model to obtain the classification result and the intermediate layer features of the model; the intermediate layer features of the model are the output of the second fully connected layer to represent the deep features of the fused feature vector.

[0029] Preferably, the feature-level dataset is input into a trained patient prognosis prediction model to obtain prediction results, including:

[0030] Obtain a preset feature-level data sample set;

[0031] Construct the initial convolutional neural network;

[0032] The initial convolutional neural network is trained based on the feature-level data sample set to obtain a trained classifier;

[0033] A trained LSTM neural network is connected after the classifier to obtain the patient prognosis prediction model.

[0034] The feature-level dataset is input into the patient prognosis prediction model to obtain the prediction results.

[0035] Preferably, the prediction results include: the patient's potential for recovery of consciousness, the prediction of recovery time, the possibility of disease deterioration, and the specific prognostic level.

[0036] Preferably, the prognostic levels include: complete recovery, partial recovery, and no recovery.

[0037] A prognostic prediction system for patients with chronic Doc includes:

[0038] The data standardization module is used to collect multimodal data from patients with chronic Doc and to standardize each type of multimodal data to obtain a standardized dataset. The multimodal data includes: neuroimaging data, neuroelectrophysiological data, biochemical indicators, and clinical data.

[0039] The feature extraction module is used to extract features from various types of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data.

[0040] The feature fusion module is used to fuse the spatial features, the time series features, and the key features to obtain a fused feature vector.

[0041] The classification module is used to input the fused feature vector into the trained consciousness state hierarchical evaluation model to obtain the classification result and the intermediate layer features of the model;

[0042] The feature-level generation module is used to generate a feature-level dataset by combining the classification results as labels with the features of the intermediate layer of the model.

[0043] The prediction module is used to input the feature-level dataset into the trained patient prognosis prediction model to obtain the prediction results.

[0044] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0045] This invention provides a method and system for predicting the prognosis of patients with chronic respiratory disease (DRD). The method includes: collecting multimodal data of DRD patients and standardizing each type of multimodal data to obtain a standardized dataset; the multimodal data includes: neuroimaging data, neuroelectrophysiological data, biochemical indicators, and clinical data; extracting features from each type of data in the standardized dataset to obtain spatial features corresponding to the neuroimaging data, time-series features corresponding to the neuroelectrophysiological data, and key features corresponding to the biochemical indicators and clinical data; fusing the spatial features, time-series features, and key features to obtain a fused feature vector; inputting the fused feature vector into a trained consciousness state grading assessment model to obtain classification results and intermediate layer features of the model; using the classification results as labels and combining them with the intermediate layer features of the model to generate a feature-level dataset; and inputting the feature-level dataset into a trained patient prognosis prediction model to obtain prediction results. This invention integrates multimodal data (including neuroimaging, neuroelectrophysiological data, biochemical indicators, and clinical data) to achieve comprehensive feature extraction and fusion analysis of DRD patients, fully utilizing the complementary information of spatial features, time-series features, and key features. This invention combines a trained consciousness level assessment model and a prognostic prediction model to accurately predict a patient's consciousness level and prognostic trend, providing clinicians with scientific decision support, optimizing personalized treatment plans, and improving the accuracy and reliability of prognostic prediction, thereby enhancing patient rehabilitation and the efficiency of medical resource utilization. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The purpose of this invention is to provide a method and system for predicting the prognosis of patients with chronic Doc (Chronic Disease) that can accurately predict the patient's level of consciousness and prognostic trends, provide scientific decision support for clinicians, optimize personalized treatment plans, improve the accuracy and reliability of prognostic prediction, and thus improve the patient's rehabilitation effect and the efficiency of medical resource utilization.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, the present invention provides a method for predicting the prognosis of patients with chronic Doc, comprising:

[0053] Step 100: Collect multimodal data of patients with chronic Doc and standardize each type of multimodal data to obtain a standardized dataset; the multimodal data includes: neuroimaging data, neuroelectrophysiological data, biochemical indicators and clinical data;

[0054] Step 200: Extract features from each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data;

[0055] Step 300: Perform feature fusion on spatial features, time series features, and key features to obtain a fused feature vector;

[0056] Step 400: Input the fused feature vector into the trained consciousness state hierarchical assessment model to obtain the classification results and intermediate layer features of the model;

[0057] Step 500: Use the classification results as labels and combine them with the features from the intermediate layers of the model to generate a feature-level dataset;

[0058] Step 600: Input the feature-level dataset into the trained patient prognosis prediction model to obtain the prediction results.

[0059] Preferably, the standardization process includes: noise reduction, normalization, and format conversion.

[0060] Specifically, step 100 in this embodiment includes:

[0061] Step 101: First, for patients with chronic Doc (Chronic Disease) using multimodal data, collect neuroimaging data (e.g., MRI, fMRI, DTI), neuroelectrophysiological data (e.g., EEG, ERP), biochemical indicators (e.g., blood glucose, lactate dehydrogenase, C-reactive protein), and clinical data (e.g., medical history, behavioral assessment scales, medication records). During data collection, ensure the reliability and consistency of data sources, and conduct preliminary checks on data from different modalities to remove obvious outliers and incomplete records, laying the foundation for subsequent processing.

[0062] Step 102: Next, denoising is performed on each type of data. For neuroimaging data, filtering algorithms (such as Gaussian filtering or wavelet transform) can be used to remove image noise; for neuroelectrophysiological data, bandpass filters can be used to remove power frequency interference and low-frequency drift; for biochemical indicators and clinical data, outliers can be detected and removed using statistical methods. After denoising, the data is normalized. For example, the pixel values ​​of the image data are normalized to the [0,1] interval, the amplitudes of the neuroelectrophysiological data are standardized to zero mean and unit variance, and the biochemical indicators are scaled proportionally to their normal range to ensure the comparability of data from different modalities.

[0063] Step 103: Finally, perform format conversion to standardize the data format. Neuroimaging data can be converted to standardized NIfTI or DICOM formats, neuroelectrophysiological data can be saved as EDF or MAT formats, and biochemical indicators and clinical data can be organized into structured tables (such as CSV or JSON formats). Through denoising, normalization, and format conversion, a standardized multimodal dataset is obtained, providing high-quality data input for subsequent feature extraction and analysis.

[0064] Preferably, feature extraction is performed on each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time-series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data, including:

[0065] For neuroimaging data, convolutional neural networks are used to extract spatial features;

[0066] Time-frequency analysis was used to extract time-series features from neuroelectrophysiological data.

[0067] Key features were extracted from biochemical indicators and clinical data using a random forest algorithm.

[0068] Preferably, time-series features are extracted from neuroelectrophysiological data using time-frequency analysis methods, including:

[0069] Time-frequency features are extracted using the short-time Fourier transform method; the expression for the time-frequency features is: Where x(τ) represents the original neurophysiological data, w(τ-t,σ) is the time window function, t is the time point representing the local spectrum of the signal at different time intervals, f is the frequency representing the frequency components of the signal, X(t,f) is the time-frequency characteristic representing the frequency distribution of the signal at each time point, and σ is the width of the window function, which is dynamically adjusted as follows: α is an adjustment parameter used to control the sensitivity of the window width to the signal amplitude. Exemplarily, this adjustment parameter dynamically controls the width of the time window to balance time resolution and frequency resolution, thereby more accurately capturing the frequency distribution characteristics of the signal at different time points. The specific adjustment method in this embodiment is to dynamically change the window width according to the frequency components of the signal: for high-frequency signals, a narrower time window is used to obtain higher time resolution, thereby more accurately capturing rapidly changing signal characteristics; for low-frequency signals, a wider time window is used to improve frequency resolution, thereby more clearly describing the distribution of frequency components. This dynamic adjustment typically involves setting an adjustment parameter that adaptively optimizes the width of the time window based on the range of signal frequencies and the importance of the analysis task, ensuring a relatively balanced time-frequency characteristic representation across different frequency bands.

[0070] The power spectral density is extracted based on the time-frequency features, and then normalized to obtain the normalized power spectral density; the expression for the power spectral density is: P(t,f)=|X(t,f)| 2 Where P(t,f) is the power spectral density, representing the energy intensity of the signal at time t and frequency f; the expression for the normalized power spectral density is: Among them, f min and f max P represents the minimum and maximum values ​​of the analysis frequency range, respectively. norm(t,f) represents the normalized power spectral density;

[0071] Nonlinear feature analysis is performed based on short-time multi-scale entropy to obtain frequency-weighted multi-scale entropy; wherein, the expression for the short-time multi-scale entropy is as follows: Where E(t,τ) is the short-time multiscale entropy at time t and time scale τ, used to quantify the complexity of the signal and reflect the dynamic change pattern of the neurophysiological signal, τ is the time scale, representing the window size, and p i (τ) represents the probability distribution of the signal at time scale τ; the expression for the frequency-weighted multi-scale entropy is: E f (t,τ)=∑ f w(f)E(t,τ,f); where E(t,τ,f) is the short-time multi-scale entropy at frequency f, w(f) is the frequency weighting function, which is adaptively adjusted according to the importance of a specific frequency band, and E f (t,τ) represents the frequency-weighted multiscale entropy. Exemplarily, the frequency weighting function reflects the importance of specific frequency bands in signal analysis by assigning weights to different frequency components, thereby adjusting the multiscale entropy accordingly. Specifically, the weight distribution is customized or learned based on the needs of the target task and the characteristics of the signal. For example, in neurophysiological signal analysis, certain frequency bands (such as alpha, beta, and gamma waves) may be closely related to the patient's state of consciousness or pathological characteristics; therefore, these bands can be assigned higher weights, while other irrelevant or noisy bands can be assigned lower weights. This weight allocation can be preset through expert experience or adaptively adjusted through data-driven methods (such as optimizing weight parameters using training data) to ensure that the frequency-weighted multiscale entropy can more effectively capture the key dynamic features and complexity of the signal.

[0072] The normalized power spectral density is converted into a time-frequency weighted feature; the expression for the time-frequency weighted feature is: Among them, f β For frequency weighting, when β>0, high-frequency components are emphasized, and when β<0, low-frequency components are emphasized. F(t) is the time-frequency weighted feature, which represents the projection of the time-frequency feature onto the time dimension. σ'(t) is the local standard deviation of the signal in the neurophysiological data, and μ(t) is the local mean of the signal in the neurophysiological data.

[0073] The time-frequency features, the normalized power spectral density, the frequency-weighted multi-scale entropy, and the time-frequency weighted features are concatenated into multi-dimensional features to obtain the fused time series features.

[0074] Preferably, the key features include: blood glucose, lactate dehydrogenase, C-reactive protein, medical history, behavioral assessment scales, and medication records.

[0075] Specifically, step 200 in this embodiment includes:

[0076] Step 201: Spatial feature extraction from neuroimaging data

[0077] For neuroimaging data (such as MRI, fMRI, DTI, etc.), preprocessing is first performed, including format conversion (such as NIfTI or DICOM format), normalization, and denoising. Then, convolutional neural networks (CNNs) are used to extract spatial features, capturing local texture, shape, and structural features in the image data through multi-layer convolution and pooling operations. Finally, high-dimensional spatial features are compressed into low-dimensional feature vectors using fully connected layers or global average pooling layers, serving as the spatial features of the neuroimaging data. These features can reflect the spatial differences in the structure and function of the patient's brain.

[0078] Step 202: Extraction of time-series features from neuroelectrophysiological data

[0079] For neurophysiological data (such as EEG or ERP), preprocessing is first performed, including filtering and denoising (e.g., removing power line interference and low-frequency drift) and normalization. Then, short-time Fourier transform (STFT) is used to perform time-frequency analysis on the signal, extracting the frequency distribution characteristics of the signal at different time points. By dynamically adjusting the time window width, the time resolution and frequency resolution are balanced to capture the time-frequency characteristics of the signal, providing a basis for subsequent analysis.

[0080] Step 203: Extraction of normalized power spectral density

[0081] Based on the time-frequency analysis results, the energy distribution (power spectral density) of the signal in time and frequency is calculated and normalized to reflect the relative energy distribution of the signal within the frequency range. The normalized power spectral density can more intuitively represent the energy intensity of the signal at different frequencies, providing a standardized input for subsequent feature fusion.

[0082] Step 204: Nonlinear Feature Analysis and Frequency-Weighted Multiscale Entropy Extraction

[0083] Short-time multiscale entropy analysis is used to analyze signal complexity and quantify the dynamic patterns of neurophysiological signals. By combining a frequency-weighted function, frequency-weighted multiscale entropy is further extracted to capture the importance and complexity of the signal in different frequency bands. These nonlinear features reflect the dynamic changes and frequency characteristics of neurophysiological signals, providing support for a comprehensive description of time-series features.

[0084] Step 205: Calculation of time-frequency weighted features

[0085] By combining normalized power spectral density with frequency weights, time-frequency weighted features are calculated to adjust the weights of high-frequency or low-frequency components. Furthermore, incorporating the signal's local mean and standard deviation further enhances the descriptive power of the time-frequency features. These time-frequency weighted features reflect the signal's frequency projection over time, providing crucial information for the fusion of time-series features.

[0086] Step 206: Multidimensional feature splicing and time series feature fusion

[0087] Time-frequency features, normalized power spectral density, frequency-weighted multi-scale entropy, and time-frequency weighted features are multidimensionally concatenated to form fused time-series features. This concatenation operation preserves the unique information of each feature while enhancing the overall descriptive ability of neurophysiological signals. These fused features can capture the dynamic change patterns, frequency distribution, and nonlinear complexity of signals, providing high-quality time-series feature input for subsequent prognostic prediction.

[0088] Step 207: Extraction of key features from biochemical indicators and clinical data

[0089] Biochemical indicators (such as blood glucose, lactate dehydrogenase, and C-reactive protein) and clinical data (such as medical history, behavioral assessment scales, and medication records) were preprocessed, including missing value imputation and standardization. Then, a random forest algorithm was used to extract key features. By evaluating the importance of these features, those contributing most to the target task, such as blood glucose levels, inflammatory markers (C-reactive protein), patient history, and behavioral scores, were selected. These key features, combined with neuroimaging and neurophysiological characteristics, provide important references for multimodal data fusion.

[0090] Furthermore, in step 300 of this embodiment, various features are preprocessed to ensure consistency in their dimensions, range, and scale (e.g., through normalization or standardization). Then, spatial features (such as features extracted from neuroimaging), time-series features (such as time-frequency features of neurophysiological signals), and key features (such as features extracted from biochemical indicators and clinical data) are concatenated in a certain order to form a high-dimensional feature vector. If the spatial features are of higher importance, they can be weighted and then fused. Finally, the fused feature vector integrates the key information of multimodal data, providing a unified input for subsequent modeling.

[0091] As an example, this embodiment utilizes a random forest model to assign an importance score to each feature. This score represents the average contribution of that feature to the prediction of the target variable across all decision trees. A higher score indicates that the feature is more important to the model's prediction. The feature importance score can be directly extracted from the model's attributes. Based on the extracted feature importance scores, all features are sorted in descending order to obtain a feature importance ranking. Features ranked higher are those that contribute the most to the prediction of the target variable, while features ranked lower may be redundant or irrelevant. To verify the rationality of the feature ranking, low-importance features can be gradually removed, and the model can be retrained to observe changes in model performance. If the model performance does not significantly decrease after removing certain features, it indicates that these features contribute little to the task, and the feature set can be further optimized. In this way, this embodiment not only determines the importance of features but also provides a basis for subsequent feature selection and model optimization, thereby improving model efficiency and predictive performance.

[0092] Preferably, the fused feature vector is input into the trained consciousness state hierarchical assessment model to obtain classification results and intermediate layer features of the model, including:

[0093] Construct an initial depth model based on a multilayer perceptron;

[0094] The preset sample fusion feature vector and sample consciousness state level are input into the initial deep model to obtain the trained consciousness state hierarchical evaluation model; the loss function of the consciousness state hierarchical evaluation model is the cross-entropy loss function; the optimizer of the cross-entropy loss function is the Adam optimizer; the network structure of the consciousness state hierarchical evaluation model includes: an input layer, a first fully connected layer, a second fully connected layer and an output layer connected in sequence;

[0095] The fused feature vector is input into the trained consciousness state hierarchical evaluation model to obtain the classification result and the intermediate layer features of the model; the intermediate layer features of the model are the output of the second fully connected layer to represent the deep features of the fused feature vector.

[0096] Specifically, step 400 in this embodiment includes:

[0097] Step 401: Construct an initial depth model based on a multilayer perceptron

[0098] First, a Multilayer Perceptron (MLP) is designed as the initial deep model for assessing the classification of states of consciousness. The network structure of this model includes an input layer, two fully connected layers (a first fully connected layer and a second fully connected layer), and an output layer. The dimension of the input layer is the same as the size of the fused feature vector, used to receive the feature vector after the fusion of multimodal data. The first and second fully connected layers are used to extract higher-order features, respectively, and the output layer is used to generate the classification result. Activation functions (such as ReLU) can be added after each layer to enhance the model's non-linear expressive power, and regularization (such as Dropout) can be added to prevent overfitting.

[0099] Step 402: Training the Consciousness State Grading Assessment Model

[0100] The pre-defined sample fusion feature vector and corresponding sample consciousness state level (label) are input into the initial deep model for training. During training, the cross-entropy loss function is used as the objective function to measure the difference between the model's prediction and the true label. To optimize the model parameters, the Adam optimizer is selected, which can dynamically adjust the learning rate and has a fast convergence speed and good robustness. Through multiple rounds of iterative training, the model gradually learns the mapping relationship between the fused feature vector and the consciousness state level, ultimately obtaining a trained consciousness state classification and evaluation model.

[0101] Step 403: Input the fused feature vector for inference

[0102] The new fused feature vector is input into the trained consciousness level assessment model, which generates classification results through forward propagation. The classification results represent the patient's level of consciousness (e.g., fully conscious, partially conscious, or unconscious), providing clinicians with an assessment basis for the patient's current state of consciousness. Simultaneously, during inference, the output of the model's intermediate layer (i.e., the second fully connected layer) is extracted as a deep feature of the fused feature vector.

[0103] Step 404: Extract intermediate layer features of the model

[0104] The intermediate layer features are the output of the second fully connected layer, representing high-order features extracted from the fused feature vector by the deep model. These features capture deep-seated correlations and complex patterns in multimodal data, exhibiting higher expressive power. These intermediate layer features can serve as input for subsequent analyses, such as training patient prognosis prediction models or further feature analysis. This approach not only provides classification results for states of consciousness but also offers rich feature information for subsequent tasks through the intermediate layer features.

[0105] Further, in step 500 of this embodiment, the classification result of the trained consciousness state grading assessment model is used as the label to represent the consciousness state level of each sample. Then, the intermediate layer features of the model (i.e., the output of the second fully connected layer) are extracted. These features are high-order features extracted from the fused feature vectors by the deep model, which can reflect the deep-level correlations and complex patterns of multimodal data. The intermediate layer features of each sample are combined with the corresponding classification result (label) to form a new dataset, namely the feature-level dataset. The samples in this dataset are composed of intermediate layer feature vectors, and the labels are the classification results of consciousness state grading.

[0106] The feature-level dataset mainly consists of two parts: first, deep feature vectors extracted from the intermediate layers of the model, serving as the input features of the dataset; and second, the classification results from the consciousness state grading model, serving as the dataset labels. This dataset can be used for subsequent training of patient prognosis prediction models because the intermediate layer feature vectors contain high-order information from multimodal data, and the classification results, as labels, can guide the prognosis prediction model to learn the relationship between patient consciousness state and prognosis. This method of constructing feature-level datasets can effectively improve the performance and generalization ability of prognosis prediction models.

[0107] Preferably, the feature-level dataset is input into a trained patient prognosis prediction model to obtain prediction results, including:

[0108] Obtain a preset feature-level data sample set;

[0109] Construct the initial convolutional neural network;

[0110] The initial convolutional neural network is trained based on the feature-level data sample set to obtain a trained classifier;

[0111] A trained LSTM neural network is connected after the classifier to obtain the patient prognosis prediction model.

[0112] The feature-level dataset is input into the patient prognosis prediction model to obtain the prediction results.

[0113] Preferably, the prediction results include: the patient's potential for recovery of consciousness, the prediction of recovery time, the possibility of disease deterioration, and the specific prognostic level.

[0114] Preferably, the prognostic levels include: complete recovery, partial recovery, and no recovery.

[0115] Specifically, step 600 in this embodiment includes:

[0116] Step 601: Obtain the feature-level data sample set

[0117] First, a feature-level data sample set is collected and organized. This dataset consists of deep feature vectors extracted from the intermediate layers of the model and classification result labels. These feature vectors contain high-order information of multimodal fusion features, reflecting the patient's state of consciousness and related complex patterns. Preprocessing (such as normalization and standardization) ensures the quality and consistency of the feature-level data sample set. Then, the dataset is divided into training, validation, and test sets for subsequent training and evaluation of the patient prognosis prediction model.

[0118] Step 602: Construct the initial convolutional neural network and train the classifier

[0119] An initial convolutional neural network (CNN) is constructed to further extract local patterns and higher-order features from the feature-level data sample set. The input layer of the CNN has the same dimension as the feature vector. Convolutional layers extract local correlations of features through filters, while pooling layers are used for dimensionality reduction and to enhance the robustness of the model. The feature-level data sample set is input into the CNN for training, using cross-entropy as the loss function, and the model parameters are optimized using the Adam optimizer. After training, a trained classifier is obtained, which can perform preliminary classification of the feature-level data and extract deep features related to patient prognosis.

[0120] Step 603: Connect LSTM neural network to build prognostic prediction model

[0121] A LSTM neural network is connected after the trained CNN classifier to capture the temporal dynamic patterns and sequence dependencies of feature-level data. LSTM can handle long-term dependencies in sequence data, making it ideal for analyzing the temporal evolution of features during patient prognosis. The high-order features extracted by the CNN are used as input to the LSTM network. By training the LSTM network, the dynamic changes in the patient's state of consciousness and prognostic-related time-series features are learned. Ultimately, the combination of CNN and LSTM constitutes a complete patient prognostic prediction model.

[0122] Step 604: Input feature-level dataset and generate prediction results

[0123] The feature-level dataset is input into a pre-trained patient prognostic prediction model, which generates predictions through forward propagation. These predictions include the patient's potential for recovery of consciousness (e.g., the probability of complete or partial recovery), the predicted recovery time (e.g., the expected timeframe for recovery), the likelihood of disease deterioration (e.g., the probability of deterioration or risk level), and the specific prognostic grade (e.g., grading score or classification result). These predictions provide clinicians with comprehensive prognostic information, aiding in the development of personalized treatment plans and interventions, and also provide families with a scientific basis for decision-making.

[0124] Corresponding to the above method, this embodiment also provides a prognostic prediction system for patients with chronic Doc, including:

[0125] The data standardization module is used to collect multimodal data from patients with chronic Doc and to standardize each type of multimodal data to obtain a standardized dataset. The multimodal data includes: neuroimaging data, neuroelectrophysiological data, biochemical indicators, and clinical data.

[0126] The feature extraction module is used to extract features from various types of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data.

[0127] The feature fusion module is used to fuse the spatial features, the time series features, and the key features to obtain a fused feature vector.

[0128] The classification module is used to input the fused feature vector into the trained consciousness state hierarchical evaluation model to obtain the classification result and the intermediate layer features of the model;

[0129] The feature-level generation module is used to generate a feature-level dataset by combining the classification results as labels with the features of the intermediate layer of the model.

[0130] The prediction module is used to input the feature-level dataset into the trained patient prognosis prediction model to obtain the prediction results.

[0131] The beneficial effects of this invention are as follows:

[0132] This invention combines a trained consciousness level assessment model and a prognostic prediction model to accurately predict a patient's consciousness level and prognostic trend, providing clinicians with scientific decision support, optimizing personalized treatment plans, and improving the accuracy and reliability of prognostic prediction, thereby enhancing patient rehabilitation and the efficiency of medical resource utilization.

[0133] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0134] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for prognosis prediction of a chronic Doc patient, characterized by, The application relates to a method for predicting the prognosis of a patient with chronic doc, and a device thereof. The method comprises the following steps: Collecting multi-modal data of a chronic doc patient, and standardizing each type of the multi-modal data to obtain a standardized data set; The multi-modal data comprises neuroimaging data, neuroelectrophysiological data, biochemical indicators and clinical data; Extracting features from each type of data in the standardized data set to obtain spatial features corresponding to the neuroimaging data, time sequence features corresponding to the neuroelectrophysiological data and key features corresponding to the biochemical indicators and the clinical data; Fusing the spatial features, the time sequence features and the key features to obtain a fusion feature vector; Inputting the fusion feature vector into a trained consciousness state grading evaluation model to obtain a classification result and an intermediate layer feature of the model; Taking the classification result as a label and combining the intermediate layer feature to generate a feature level data set; Inputting the feature level data set into a trained patient prognosis prediction model to obtain a prediction result; The feature extraction of each type of data in the standardized data set comprises the following steps: For the neuroimaging data, a convolutional neural network is used to extract spatial features; For the neuroelectrophysiological data, time-frequency analysis is used to extract time sequence features; For the biochemical indicators and the clinical data, a random forest algorithm is used to extract key features; extracting time-frequency features by using short-time Fourier transform; the expression of the time-frequency features is: ; wherein, is the original neurophysiological data, is a time window function, is a time point, representing the local spectrum of the signal in different time periods, is a frequency, representing the frequency component of the signal, is a time-frequency feature, representing the frequency distribution of the signal at each time point, is the width of the window function, dynamically adjusted as: , is an adjustment parameter, used to control the sensitivity of the window width to the signal amplitude; According to the time-frequency feature, a power spectrum density is extracted, and the power spectrum density is normalized to obtain a normalized power spectrum density; an expression of the power spectrum density is: ; wherein, is a power spectrum density, and represents an energy intensity of a signal at time and frequency ; an expression of the normalized power spectrum density is: ; wherein, are respectively a minimum value and a maximum value of an analysis frequency range, is a normalized power spectrum density. According to short-time multiscale entropy for nonlinear feature analysis, frequency weighted multiscale entropy is obtained; wherein, the expression of the short-time multiscale entropy is ; wherein, is the short-time multiscale entropy on the time and time scale , used for quantifying the complexity of the signal and reflecting the dynamic change mode of the neural electrophysiological signal, is the time scale, representing the size of the window, is the probability distribution of the signal on the time scale ; the expression of the frequency weighted multiscale entropy is: ; wherein, is the short-time multiscale entropy on the frequency , is the frequency weight function, which is adaptively adjusted according to the importance of a specific frequency band, is the frequency weighted multiscale entropy; transforming the normalized power spectral density into a time-frequency weighted feature; the expression of the time-frequency weighted feature is: ; wherein, is a frequency weighting term, emphasizing high frequency components when is a frequency weighting term, emphasizing low frequency components when is a frequency weighting term, emphasizing low frequency components when is a time-frequency weighted feature, representing the projection of the time-frequency feature on the time dimension; , is a local standard deviation of the signal of the neuroelectrophysiological data, is a local mean of the signal of the neuroelectrophysiological data; For the neuroelectrophysiological data, a time-frequency analysis method is used to extract time sequence features, which comprises the following steps:

2. The method of prognosis prediction of chronic Doc patients according to claim 1, characterized in that, The time-frequency features, the normalized power spectral density, the frequency weighted multi-scale entropy and the time-frequency weighted features are multi-dimensionally spliced to obtain fused time sequence features.

3. The prognosis prediction method for a chronic Doc patient according to claim 1, characterized by, The standardization process comprises denoising, normalization and format conversion.

4. The prognosis prediction method for a chronic Doc patient according to claim 1, characterized by, The key features comprise blood glucose, lactate dehydrogenase, C-reactive protein, medical history, behavior evaluation scale and medication record. The method comprises the following steps: An initial deep model based on a multilayer perception is constructed; A preset sample fusion feature vector and a sample consciousness state grade are input into the initial deep model to obtain a trained consciousness state grading evaluation model; a loss function of the consciousness state grading evaluation model is a cross-entropy loss function; an optimizer of the cross-entropy loss function is an Adam optimizer; a network structure of the consciousness state grading evaluation model comprises an input layer, a first full connection layer, a second full connection layer and an output layer which are connected in sequence; 5. The method of prognosis prediction of chronic Doc patients according to claim 1, characterized in that, The fusion feature vector is input into the trained consciousness state grading evaluation model to obtain a classification result and an intermediate layer feature of the model; the intermediate layer feature of the model is the output of the second full connection layer to represent the deep features of the fusion feature vector. The method comprises the following steps: A preset feature level data sample set is obtained; An initial convolutional neural network is constructed; The initial convolutional neural network is trained according to the feature level data sample set to obtain a trained classifier; A trained LSTM neural network is connected behind the classifier to obtain the patient prognosis prediction model; The feature-level dataset is input into the patient prognosis prediction model to obtain a prediction result.

6. The method of prognosis prediction of chronic Doc patients according to claim 1, characterized in that, The prediction result includes the recovery potential of the patient's consciousness state, the prediction of the recovery time, the possibility of disease deterioration, and the specific prognosis level.

7. The method of prognosis prediction of chronic Doc patients according to claim 6, characterized in that, The prognosis level includes complete recovery, partial recovery, and no recovery.

8. A prognosis prediction system for chronic Doc patients, characterized by, Comprise: A data standardization module is configured to collect multi-modal data of chronic Doc patients, and standardize each type of multi-modal data to obtain a standardized dataset; the multi-modal data includes neuroimaging data, neuroelectrophysiological data, biochemical indicators, and clinical data; A feature extraction module is configured to extract features from each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data; A feature fusion module is configured to fuse the spatial features, the time series features, and the key features to obtain a fusion feature vector; A classification module is configured to input the fusion feature vector into a trained consciousness state grading evaluation model to obtain a classification result and an intermediate layer feature of the model; A feature-level generation module is configured to generate a feature-level dataset by combining the classification result as a label and the intermediate layer feature of the model; A prediction module is configured to input the feature-level dataset into a trained patient prognosis prediction model to obtain a prediction result; Feature extraction is performed on each type of data in the standardized dataset to obtain spatial features corresponding to neuroimaging data, time series features corresponding to neuroelectrophysiological data, and key features corresponding to biochemical indicators and clinical data, including: For neuroimaging data, a convolutional neural network is used to extract spatial features; For neuroelectrophysiological data, time-frequency analysis is used to extract time series features; For biochemical indicators and clinical data, a random forest algorithm is used to extract key features; For neuroelectrophysiological data, time-frequency analysis is used to extract time series features, including: extracting time-frequency features using short-time Fourier transform; the expression of the time-frequency features is: ; wherein, is the original neurophysiological data, is a time window function, is a time point, representing the local spectrum of the signal in different time periods, is a frequency, representing the frequency component of the signal, is a time-frequency feature, representing the frequency distribution of the signal at each time point, is the width of the window function, dynamically adjusted as: , is an adjustment parameter, used to control the sensitivity of the window width to the signal amplitude; According to the time-frequency feature, a power spectrum density is extracted, and the power spectrum density is normalized to obtain a normalized power spectrum density; an expression of the power spectrum density is: ; wherein, is a power spectrum density, indicating an energy intensity of a signal at a time and a frequency ; an expression of the normalized power spectrum density is: ; wherein, are respectively a minimum value and a maximum value of an analysis frequency range, is a normalized power spectrum density. According to short-time multiscale entropy for nonlinear feature analysis, frequency-weighted multiscale entropy is obtained; wherein, the expression of the short-time multiscale entropy is ; wherein, is a short-time multiscale entropy on a time and a time scale , used for quantifying complexity of a signal and reflecting a dynamic change mode of a neural electrical physiological signal, is a time scale, representing a size of a window, is a probability distribution of a signal on a time scale ; the expression of the frequency-weighted multiscale entropy is: ; wherein, is a short-time multiscale entropy on a frequency , is a frequency weight function, which is adaptively adjusted according to importance of a specific frequency band, is the frequency-weighted multiscale entropy; transforming the normalized power spectral density into a time-frequency weighted feature; the expression of the time-frequency weighted feature is: ; wherein, is a frequency weighting term, emphasizing high frequency components when is a frequency weighting term, emphasizing low frequency components when is a frequency weighting term, emphasizing low frequency components when is a time-frequency weighted feature, representing the projection of the time-frequency feature on the time dimension; , is a local standard deviation of the signal of the neuroelectrophysiological data, is a local mean of the signal of the neuroelectrophysiological data; The time-frequency features, the normalized power spectral density, the frequency-weighted multi-scale entropy, and the time-frequency weighted features are multi-dimensionally spliced to obtain fused time series features.

Images