A method for detecting intracranial pressure

Abstract

The application discloses an intracranial pressure detection method, which comprises the following steps: acquiring regional signal envelope data corresponding to a preset acquisition region, the acquisition region being set as at least two key brain regions, each key brain region corresponding to one regional signal envelope data; inputting the regional signal envelope data into a corresponding intracranial pressure estimation model which has been pre-trained, and outputting initial intracranial pressure data of each region; extracting an intracranial pressure correlation feature vector based on the initial intracranial pressure data of each region, inputting the intracranial pressure correlation feature vector into a contribution index prediction model which has been pre-trained, and outputting a contribution index corresponding to the initial intracranial pressure data of the region; and calculating target intracranial pressure data based on the initial intracranial pressure data of each region and the contribution index. Thus, the dynamic change difference of intracranial pressure in different brain regions is captured, and the accuracy and reliability of intracranial pressure detection are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data detection, and in particular to an intracranial pressure detection method. BACKGROUND

[0002] Intracranial pressure monitoring is the pressure of the intracranial content on the skull wall, a probe of an intracranial pressure detection instrument needs to be placed in the intracranial cavity, the probe is placed on the forehead and the occiput, and a waveform of the intracranial pressure is transmitted to a workstation through a sensor, so that the change condition of the intracranial pressure can be completely understood. By analyzing the change of the intracranial pressure of a patient, the condition of the patient's injury and brain edema can be judged, so that treatment and prognosis can be known. With people paying more and more attention to their own physical health, detection of various physical indexes is very important. At present, the intracranial pressure detection is inaccurate, and the intracranial pressure information of a patient cannot be accurately and timely obtained.

[0003] A kind of intracranial pressure detection method and system are disclosed in Chinese patent application No. 202310987825.6, which extracts signal envelope data from head PPG signal, uses neural network to infer signal envelope, generates initial intracranial pressure estimation data, obtains local coefficient representing local details and global coefficient representing global correlation through range projection and global binary, generates key coefficient combining local and global coefficient, enhances the expression of correlation between data, optimizes data representation, and improves the reliability of results.

[0004] However, due to the differences in signal characteristics at different regions of the human head, the intracranial pressure detection data of the human head is affected to varying degrees, and a single region or limited region may not accurately detect intracranial pressure abnormalities caused by unilateral brain edema and other local lesions, or the overall intracranial pressure detection may be abnormal due to intracranial pressure abnormalities in a certain local area, which will have some impact on the final intracranial pressure data. Due to differences in physiological structure, illness, and other factors, the characteristics of intracranial pressure change are different, the signal characteristics are different, and the intracranial physiological state is complex and variable. Single monitoring may not fully and accurately reflect this complexity. SUMMARY

[0005] The application provides an intracranial pressure detection method that captures the dynamic changes and differences of intracranial pressure in different brain regions, improving the accuracy and reliability of intracranial pressure detection.

[0006] The application provides an intracranial pressure detection method, comprising:

[0007] S101, acquiring regional signal envelope data corresponding to a preset acquisition region, the acquisition region is set to at least two key brain regions, and each key brain region corresponds to a regional signal envelope data;

[0008] S102, input the regional signal envelope data into the pre-trained intracranial pressure estimation model to output the initial intracranial pressure data of each region;

[0009] S103, based on the initial intracranial pressure data of each region, extract the intracranial pressure correlation feature vector and input it into the pre-trained contribution index prediction model to output the contribution index corresponding to the initial intracranial pressure data of the region;

[0010] S104, based on the initial intracranial pressure data of each region and its contribution index, calculate the target intracranial pressure data.

[0011] Preferably, the extraction of the intracranial pressure correlation feature vector comprises:

[0012] S201, obtain the regional monitoring position information;

[0013] S202, obtain the mean, variance, dynamic range and trend value of the initial intracranial pressure data of the region;

[0014] S203, the monitoring position information, mean, variance, dynamic range and trend slope value are combined to form the intracranial pressure correlation feature vector.

[0015] Preferably, the dynamic range value is set as: the initial intracranial pressure data is curve-fitted to generate an intracranial pressure fitting curve; all turning points of the fitting curve are obtained to obtain the sum of all triangular area surrounded by all turning points under the fitting curve as the dynamic range value; the trend value is set as: the absolute value of the slope value determined by all adjacent turning points is averaged to obtain the trend value.

[0016] Preferably, the pre-trained contribution index prediction model is obtained in the following manner:

[0017] A1, a large amount of historical initial intracranial pressure data corresponding to the intracranial pressure correlation feature vector is collected, and each intracranial pressure correlation feature vector is labeled with a label, and the label content is set as the contribution index; wherein the contribution index is determined in the following manner: based on each historical initial intracranial pressure data, the actual intracranial pressure data recorded thereby is obtained, and the similarity between the historical initial intracranial pressure data and the actual intracranial pressure data is calculated, and the similarity is determined as the contribution index;

[0018] A2, the labeled intracranial pressure correlation feature vector is used as a training sample set, and a pre-selected neural network structure is trained using the training sample set to obtain a final contribution index prediction model.

[0019] Preferably, the target intracranial pressure data is calculated according to the following formula:

[0020]

[0021] wherein, is target intracranial pressure data, is initial intracranial pressure data of the i-th region, is the contribution index corresponding to the initial intracranial pressure data of the i-th region, and n is the total number of key brain regions.

[0022] Preferably, the S102 further comprises:

[0023] S301, based on the regional signal envelope data of each region, extracting a brain region correlation feature vector of the regional signal envelope data of the region;

[0024] S302, for the i-th acquisition region, splicing the regional signal envelope data and the corresponding brain region correlation feature vector to obtain an enhanced input feature: , r is the monitoring position information of the acquisition region;

[0025] S303, inputting the enhanced input feature into a pre-trained intracranial pressure estimation model to output initial intracranial pressure data corresponding to the acquisition region;

[0026] The pre-trained intracranial pressure estimation model is obtained in the following manner:

[0027] B1, collecting a large number of regional signal envelope data corresponding to the enhanced input features of different acquisition regions in history and corresponding real intracranial pressure data, and labeling each enhanced input feature with real intracranial pressure data;

[0028] B2, using the labeled enhanced input features as a training sample set, training a preselected neural network structure using the training sample set, and generating a final intracranial pressure estimation model.

[0029] Preferably, in the S301, the brain region correlation feature vector of the regional signal envelope data of the region is extracted, specifically including:

[0030] S401, based on all acquisition regions, performing time domain cross-correlation analysis on the regional signal envelope data of any two acquisition regions to generate cross-correlation feature information of any two acquisition regions;

[0031] S402, based on each acquisition region, obtaining cross-correlation feature information of the regional signal envelope data of the region and the regional signal envelope data of other acquisition regions, respectively, and according to the distance values between the three-dimensional coordinates of the acquisition region and other acquisition regions, arranging the cross-correlation feature information in ascending order to generate a cross-correlation feature information sequence of the acquisition region;

[0032] S403, determine the cross-correlation feature information sequence of the acquisition region as the brain region correlation feature vector.

[0033] Preferably, the time domain cross-correlation analysis specifically comprises:

[0034] C1, traverse the two region signal envelope data by using a preset sliding window and a preset sliding step, calculate the covariance of the amplitude of the region signal envelope data of the i-th acquisition region and the j-th acquisition region in each window; and .

[0035] C2, based on each sliding window, obtain the strong correlation information between and .

[0036] C3, based on the covariance and the strong correlation information of each sliding window, calculate the cross-correlation feature information:

[0037]

[0038] wherein, is the cross-correlation feature information, m is the total number of sliding windows, is the covariance of the k-th sliding window, is the strong correlation information of the k-th sliding window, and are weight factors of the covariance and the strong correlation information respectively, used to reflect the influence degree on the cross-correlation feature information.

[0039] Preferably, the S303 further comprises:

[0040] S501, according to the brain region correlation feature vector in the enhanced input feature , obtain the cross-brain region correlation heterogeneity index for comprehensively reflecting the fluctuation intensity and the complexity of the correlation mode between brain regions;

[0041] S502, according to the cross-brain region correlation heterogeneity index of the enhanced input feature, determine the route input to the intracranial pressure estimation model, and output the initial intracranial pressure data corresponding to the acquisition region;

[0042] Wherein, the intracranial pressure estimation model comprises a plurality of transmission routing modules, each transmission routing module is used for learning according to the input enhanced input feature, and outputs the initial intracranial pressure data.

[0043] Preferably, the acquisition method of the plurality of transmission routing modules comprises:

[0044] E1, obtain a training sample set of an intracranial pressure estimation model, based on the brain region correlation feature vector of each training sample, use a preset clustering algorithm to classify the training sample set, obtain a plurality of clusters, each cluster includes at least one brain region correlation feature vector of the training sample, and average all brain region correlation feature vectors in the cluster to obtain the center feature label of the cluster; E2, based on the training sample corresponding to each cluster, train the preselected neural network structure, continuously optimize the model parameters, and generate the corresponding transmission routing module, each transmission routing module is attached with the corresponding center feature label;

[0045] The determination of the route input into the intracranial pressure estimation model specifically comprises:

[0046] The brain region correlation feature vector of the enhanced input feature is obtained, the Euclidean distance value between each transmission routing module and the center feature label is calculated respectively, and the transmission routing module corresponding to the minimum Euclidean distance value is determined as the route of the enhanced input feature.

[0047] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0048] The multi-acquisition region setting can comprehensively reflect the intracranial physiological state, avoid the limitations of single-point measurement, help capture the dynamic change difference of intracranial pressure in different brain regions, and improve the accuracy of detection; the multi-region data covers different brain regions, and the dynamic weighting mechanism is combined to more comprehensively reflect the overall state of intracranial pressure, especially for complex pathologies; the refined preprocessing steps (noise reduction, envelope extraction) and contribution index model effectively suppress motion artifacts, high-frequency noise and abnormal data interference, and the contribution index prediction model can evaluate the contribution of the initial intracranial pressure data in each region, further enhancing the reliability of the data; the multi-dimensional and multi-feature analysis method is more in line with the actual clinical needs, which can provide more comprehensive and accurate intracranial pressure evaluation results for doctors, and help to develop more reasonable treatment plans. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The flowchart of the intracranial pressure detection method of the embodiment of the present application is shown. DETAILED DESCRIPTION

[0050] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the related drawings; the preferred embodiments of the present application are shown in the drawings, but the present application can be realized in many different forms, and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0051] It should be noted that the terms "vertical", "horizontal", "upper", "lower", "left", "right", and similar terms used herein are for illustrative purposes only and are not intended to be limiting.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application; the use herein of the terms "and / or" includes a set of one or more associated listed items.

[0053] Embodiment one: Figure 1 is a flowchart of an intracranial pressure detection method according to an embodiment of the application.

[0054] As shown in Figure 1 , an intracranial pressure detection method comprises the following steps:

[0055] S101, obtaining a region signal envelope data corresponding to a preset acquisition region, the acquisition region being set as at least two key brain regions, each key brain region corresponding to a region signal envelope data.

[0056] Specifically, the key brain regions are selected on the human head, and PPG sensors are respectively arranged to form a multi-channel signal acquisition system, so as to obtain the original physiological signal data of each acquisition region, and the original physiological signal data is preprocessed to obtain the region signal envelope data of each key brain region.

[0057] It can be understood that the multiple acquisition regions can comprehensively reflect the intracranial physiological state, avoid the limitation of single-point measurement, capture the dynamic change difference of intracranial pressure in different brain regions, help to find the intracranial pressure abnormalities caused by local lesions, and provide a multi-dimensional data basis for subsequent analysis; solve the problem that a single sensor cannot accurately reflect the overall intracranial pressure condition due to uneven distribution of intracranial pressure caused by local brain region lesions (such as unilateral brain edema).

[0058] Preprocessing can be set as:

[0059] Noise reduction: wavelet threshold denoising method is used, and the reserved frequency band is set as 0.5-5Hz to effectively suppress high-frequency noise and motion artifacts;

[0060] Envelope extraction: Hilbert transform technology is used to extract the region signal envelope data from the denoised signal (i is set as the number of the acquisition region, for example, if there are four acquisition regions, i takes values of 1, 2, 3, and 4, which respectively correspond to frontal lobe, occipital lobe, temporal lobe, and parietal lobe channels);

[0061] Finally, the regional signal envelope data of each collection area is obtained, such as the frontal lobe envelope data (5-second time domain waveform), the noise reduction processing improves the signal quality, the envelope extraction provides key signal characteristics for subsequent feature analysis, and the influence of noise interference on the analysis result is reduced.

[0062] S102, input the regional signal envelope data into the pre-trained intracranial pressure estimation model to output the initial intracranial pressure data of each region.

[0063] S103, based on the initial intracranial pressure data of each region, an intracranial pressure correlation feature vector is extracted and input into the pre-trained contribution index prediction model to output the contribution index corresponding to the initial intracranial pressure data of the region.

[0064] Specifically, the intracranial pressure correlation feature vector is extracted, including:

[0065] S201, obtaining the regional monitoring position information, represented as a three-dimensional coordinate, simulating and establishing a coordinate system with the center position of the human head as the origin, and obtaining the three-dimensional coordinate corresponding to the regional monitoring position information;

[0066] In other embodiments, the method of obtaining the regional monitoring position information can also be set as medical image assisted positioning (CT or MRI image of the patient, three-dimensional model of the head), which is not limited by the application.

[0067] S202, obtaining the mean, variance, dynamic range and trend value of the initial intracranial pressure data of the region;

[0068] Wherein, the dynamic range value is set as: the initial intracranial pressure data is curve-fitted to generate an intracranial pressure fitting curve; all turning points (determined as points where the curve trend changes, the trend is rising or falling) of the fitting curve are obtained, and the sum of all triangular area areas (triangular area is the area formed by three adjacent turning points under the fitting curve) surrounded by all turning points under the fitting curve is obtained as the dynamic range value; the trend value is set as: the absolute value of the slope value determined by all adjacent turning points is averaged to obtain the trend value;

[0069] S203, the monitoring position information, mean, variance, dynamic range and trend slope value are combined to form an intracranial pressure correlation feature vector.

[0070] In some embodiments, the pre-trained contribution index prediction model is obtained in the following manner:

[0071] A1, a large amount of historical initial intracranial pressure data corresponding to the intracranial pressure correlation feature vector is collected, and each intracranial pressure correlation feature vector is labeled with a label, and the label content is set as the contribution index.

[0072] wherein the contribution index is determined in the following manner:

[0073] Based on each historical initial intracranial pressure data, the recorded actual intracranial pressure data is obtained, the similarity of the historical initial intracranial pressure data and the actual intracranial pressure data is calculated, and the similarity is determined as the contribution index.

[0074] It should be noted that the actual intracranial pressure data can be determined by expert doctors according to the initial intracranial pressure data of each region of the patient, or can be determined according to existing invasive monitoring data. If there is no invasive monitoring data, a virtual data set can be constructed, and different brain region pressure distributions are simulated through fluid mechanics simulation, so as to determine the global actual intracranial pressure data. The present application does not repeat and limit this.

[0075] A2, the labeled intracranial pressure correlation feature vector is used as a training sample set, and the training sample set is used to train a preselected neural network structure, and model parameter optimization is continuously performed to obtain a final contribution index prediction model.

[0076] S104, based on the initial intracranial pressure data of each region and the contribution index thereof, target intracranial pressure data is calculated and transmitted to a medical care system.

[0077] Specifically, the target intracranial pressure data is calculated according to the following formula:

[0078]

[0079] wherein, is the target intracranial pressure data, is the initial intracranial pressure data of the i-th region, is the contribution index corresponding to the initial intracranial pressure data of the i-th region, and n is the total number of key brain regions.

[0080] Therefore, the problem of inaccurate intracranial pressure data caused by single region or limited region detection is solved. At least two key brain regions are selected on the human head, and PPG sensors are respectively arranged to form a multi-channel signal acquisition system to obtain regional signal envelope data of each key brain region. After these data are preprocessed, they are input into a corresponding intracranial pressure estimation model which has been pre-trained, and initial intracranial pressure data of each region is output. Further, based on the initial intracranial pressure data of each region, an intracranial pressure correlation feature vector is extracted and input into a contribution index prediction model, and the contribution index corresponding to the initial intracranial pressure data of the region is output. Finally, according to the initial intracranial pressure data of each region and the contribution index thereof, target intracranial pressure data is calculated and transmitted to a medical care system. The target intracranial pressure data obtained finally comprehensively considers the intracranial pressure conditions of multiple key brain regions, and weights each regional data through the contribution index, so that the result is more accurate and reliable.

[0081] The technical solutions in the embodiments of the application have at least the following technical effects or advantages:

[0082] The multi-acquisition region setting can comprehensively reflect the intracranial physiological state, avoid the limitations of single-point measurement, help capture the dynamic change differences of intracranial pressure in different brain regions, and improve the accuracy of detection; the multi-region data covers different brain regions, and the dynamic weighting mechanism is combined to more comprehensively reflect the overall state of intracranial pressure, especially for complex pathologies; the refined preprocessing steps (noise reduction, envelope extraction) and contribution index model effectively suppress motion artifacts, high-frequency noise and abnormal data interference, and the contribution index prediction model can evaluate the contribution of the initial intracranial pressure data in each region, further enhancing the reliability of the data; the multi-dimensional and multi-feature analysis method is more in line with the actual clinical needs, can provide doctors with more comprehensive and accurate intracranial pressure evaluation results, and helps to develop more reasonable treatment plans.

[0083] Embodiment two: The intracranial pressure estimation model of embodiment one is further limited.

[0084] Therefore, the embodiments of the application are optimized on the basis of the above embodiments.

[0085] In some embodiments, step S102 further includes:

[0086] S301, based on the signal envelope data of each region, extracting the brain region correlation feature vector of the signal envelope data of the region.

[0087] S302, for the i-th acquisition region, the signal envelope data of the region is spliced with the corresponding brain region correlation feature vector to obtain an enhanced input feature: , r is the monitoring position information of the acquisition region.

[0088] S303, inputting the enhanced input feature into the pre-trained intracranial pressure estimation model to output the initial intracranial pressure data corresponding to the acquisition region.

[0089] In some embodiments, the pre-trained intracranial pressure estimation model is obtained in the following manner:

[0090] B1, collect a large number of enhanced input features corresponding to the signal envelope data of the region of the different acquisition regions in history and the corresponding real intracranial pressure data, and label each enhanced input feature with the real intracranial pressure data.

[0091] It should be noted that the real intracranial pressure data for each enhanced input feature may come from invasive monitoring equipment in the corresponding acquisition area, or be obtained through expert annotation or experimental simulation, to ensure that the data covers different physiological and pathological conditions, such as normal intracranial pressure, local edema, tumor compression, etc. The input for each sample is the enhanced input feature of the signal envelope data of each region, and the output is the real intracranial pressure value of each region, i.e., the initial intracranial pressure data.

[0092] B2. The labeled enhanced input features are used as the training sample set. The pre-selected neural network structure is trained using the training sample set, and the model parameters are continuously optimized to generate the final intracranial pressure estimation model.

[0093] Therefore, brain region association feature vectors are introduced. In step S102, based on the signal envelope data of each region, the brain region association feature vector of that region signal envelope data is extracted and concatenated with the region signal envelope data to obtain enhanced input features. The enhanced input features are then input into the pre-trained intracranial pressure estimation model, which outputs the initial intracranial pressure data corresponding to the acquisition region, thus obtaining more accurate initial intracranial pressure data. This is because the enhanced input features contain brain region association information, which helps the model better understand the mutual influence between different brain regions.

[0094] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0095] By introducing brain region association feature vectors, the accuracy of the intracranial pressure estimation model is improved, enabling the model to better adapt to the physiological and pathological conditions of different individuals.

[0096] Example 3: Further defining the brain region association feature vector in Example 2.

[0097] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0098] In some embodiments, step S301, extracting the brain region association feature vector of the signal envelope data of the region, specifically includes:

[0099] S401, based on all acquisition areas, performs time-domain cross-correlation analysis on the regional signal envelope data of any two acquisition areas to generate cross-correlation feature information of any two acquisition areas.

[0100] The time-domain cross-correlation analysis specifically includes:

[0101] C1. Using a preset sliding window and a preset sliding step size, traverse the signal envelope data of the two regions, and calculate the amplitude of the regional signal envelope data of the i-th acquisition region and the j-th acquisition region within each window. and covariance of

[0102]

[0103] wherein, is the covariance of and in the window, N is the number of sampling points in the window, is the amplitude of the regional signal envelope data of the i-th acquisition region at the t-th sampling point in the window, is the amplitude of the regional signal envelope data of the j-th acquisition region at the t-th sampling point in the window, is the mean square error value of the signal envelope data amplitude of the i-th acquisition region in the window, is the mean square error value of the signal envelope data amplitude of the j-th acquisition region in the window.

[0104] wherein, the length and sliding step of the preset sliding window are set according to actual conditions, for example, the sliding window length is 10 seconds and the sliding step is 2 seconds. Specifically, a fixed length window is slid on the signal sequence, which is used to divide the continuous signal into multiple discrete parts, so as to calculate the probability distribution of strong correlation information in each part for subsequent use.

[0105] C2, based on each sliding window, strong correlation information (which can be understood as the calculation result of mutual information) between and is obtained:

[0106]

[0107] wherein, is the strong correlation information between the two in the sliding window, is the joint probability distribution of and signal amplitudes, and are the marginal probability distributions of and signal amplitudes, respectively.

[0108] Specifically, the acquisition method of the joint probability distribution and the marginal probability distribution includes:

[0109] D1, based on each sliding window, the occurrence frequency of each signal amplitude is counted and normalized to a probability distribution.

[0110] D2, the number of times that and two signal amplitudes appear simultaneously in each sliding window is counted, and is normalized to , that is, under the given condition, the signal the probability of the signal taking value x, the probability of the signal taking value y.

[0111] D3, respectively counting and the frequency of the signal amplitude in each sliding window, normalized as and , representing the signal the probability of the signal taking value x, representing the signal the probability of the signal taking value y.

[0112] Specifically, for each sliding window, the frequency of occurrence of each amplitude of and is counted, the frequency of occurrence of and amplitudes appearing at the same time is counted, and a joint frequency table is constructed (a joint frequency table is a statistical tool for recording the frequency of each variable (or signal) taking a specific value when appearing at the same time. In signal processing or data analysis, a joint frequency table can help understand the co-occurrence relationship of two signals under different values); the frequency of each signal amplitude is divided by the total frequency to obtain the marginal probability distribution P(x) and P(y); each frequency in the joint frequency table is divided by the total frequency (i.e. the number of all windows) to obtain the joint probability distribution P(x, y).

[0113] Exemplarily, P(x=1, y=2) represents the proportion of the window in which x takes value 1 and y takes value 2 among all windows; P(x=1) represents the proportion of the window in which x takes value 1 among all windows; similarly, P(y=2) represents the proportion of the window in which y takes value 2 among all windows. Specific application concepts related to mutual information can be referred to related prior art, which will not be described herein.

[0114] C3, based on the covariance and strong correlation information of each sliding window, the cross-correlation feature information is calculated:

[0115]

[0116] wherein, is the cross-correlation feature information, m is the total number of sliding windows, is the covariance of the kth sliding window, is the strong correlation information of the kth sliding window, and are respectively the weight factors of the covariance and the strong correlation information, used to reflect the degree of influence on the cross-correlation feature information, for example, is 0.6, is 0.4, the covariance is used to quantify the linear synchronization of signal fluctuations, and the strong correlation information is used to supplement the nonlinear correlation, so as to make up for the deficiency of the covariance reflecting only the linear relationship, capture more complex physiological coordination mechanisms, reduce the interference of random noise on the correlation calculation through probability distribution estimation, and prompt the doctor to pay attention to the pathological linkage between specific brain regions (such as local edema diffusion path).

[0117] S402, based on each acquisition region, obtain the cross-correlation feature information of the region signal envelope data of the region and the region signal envelope data of other acquisition regions respectively, and arrange the cross-correlation feature information in ascending order according to the distance value between the three-dimensional coordinates of the acquisition region and other acquisition regions, to generate a cross-correlation feature information sequence of the acquisition region.

[0118] S403, determine the cross-correlation feature information sequence of the acquisition region as the brain region correlation feature vector of the region signal envelope data of the region .

[0119] Therefore, when extracting the brain region correlation feature vector of the region signal envelope data of the region, the time domain cross-correlation analysis method is adopted, the cross-correlation feature information of any two acquisition regions is generated by performing time domain cross-correlation analysis on the region signal envelope data of the two acquisition regions, and the cross-correlation feature information is arranged according to the distance value between the three-dimensional coordinates of the acquisition region and other acquisition regions, to generate a cross-correlation feature information sequence of the acquisition region as a brain region correlation feature vector, so that a more accurate brain region correlation feature vector is obtained, which can more accurately reflect the mutual influence relationship feature distribution between different brain regions.

[0120] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages:

[0121] Through time domain cross-correlation analysis, the accuracy of the brain region correlation feature vector is improved, and more complex physiological coordination mechanisms can be captured to provide more comprehensive diagnostic information for doctors. The intracranial physiological state not only contains linear relationships, but also complex nonlinear relationships. Through cross-correlation analysis, the correlation feature distribution between different brain regions can be comprehensively reflected, so that when the enhanced input features of a certain acquisition region are extracted, the signal connection features between the distribution of the acquisition region and other regions can be accurately expressed. Based on the complexity of the intracranial physiological state, the brain region correlation feature vector can reflect a certain specific intracranial physiological state function to some extent when different body functions of different patients are targeted. The initial intracranial pressure data conforming to the current state of each region can be accurately predicted based on the trained model, and the homogeneity and limitations of traditional monitoring methods can be avoided.

[0122] Embodiment four: in the case of intracranial physiological state complex and changeable, there are significant differences in intracranial pressure data in different regions of different individuals and different pathological conditions. Traditional monitoring methods often uniformly process all data, resulting in decreased model performance and inaccurate diagnosis. The intracranial pressure estimation model in embodiment two is further limited.

[0123] Therefore, the embodiments of the present application optimize the above embodiments to a certain extent.

[0124] In some embodiments, step S303 further comprises:

[0125] S501, according to the brain region correlation feature vector in the enhanced input feature , obtain the cross-brain region correlation heterogeneity index, which is used to comprehensively reflect the fluctuation intensity and complexity of the correlation mode between brain regions.

[0126] Specifically, the cross-brain region correlation heterogeneity index is calculated according to the following formula:

[0127]

[0128] Wherein, is the cross-brain region correlation heterogeneity index, is the brain region correlation feature vector, is the variance value of the brain region correlation feature vector, reflecting the fluctuation intensity of the correlation between brain regions, is the Shannon entropy of the brain region correlation feature vector, indicating the complexity of the correlation mode. It should be noted that the calculation method of the variance value and the Shannon entropy is not described in detail herein, and can refer to the related prior art; and respectively represent the weight values of the variance value and the Shannon entropy, = 0.7, = 0.3.

[0129] Therefore, by introducing the cross-brain region correlation heterogeneity index, the different physiological states are indirectly reflected, which serves as a subsequent physiological state clustering index. The clustering index needs to be able to effectively distinguish different intracranial physiological states.

[0130] S502, according to the cross-brain region correlation heterogeneity index of the enhanced input feature, determine the route input to the intracranial pressure estimation model, and output the initial intracranial pressure data corresponding to the collection region.

[0131] Specifically, the intracranial pressure estimation model comprises a plurality of transmission routing modules, each transmission routing module being configured to learn according to the input enhanced input feature and output initial intracranial pressure data.

[0132] Wherein, the acquisition method of the plurality of transmission routing modules comprises:

[0133] E1, obtain a training sample set of an intracranial pressure estimation model, based on the brain region correlation feature vector of each training sample, use a preset clustering algorithm to classify the training sample set, obtain a plurality of clusters, each cluster includes at least one brain region correlation feature vector of the training sample, and average all brain region correlation feature vectors in the cluster to obtain the center feature label of the cluster.

[0134] E2, based on the training sample corresponding to each cluster, train the preselected neural network structure, constantly optimize the model parameters, and generate the corresponding transmission routing module, each transmission routing module is attached with the corresponding center feature label.

[0135] In some embodiments, determining the route input to the intracranial pressure estimation model specifically includes:

[0136] Obtaining the brain region correlation feature vector of the enhanced input feature, respectively calculating the Euclidean distance value of each transmission routing module center feature label, and determining the transmission routing module corresponding to the minimum Euclidean distance value as the route of the enhanced input feature.

[0137] Thus, the initial intracranial pressure data optimized for different intracranial physiological states is obtained, which improves the accuracy and reliability of detection. Through clustering analysis, the data of different intracranial physiological states are trained separately, which avoids the performance decline of the model caused by training multiple types of data in one model; each transmission routing module is optimized for a specific intracranial physiological state, so that the model can better adapt to different physiological and pathological conditions of different individuals. Through the cross-brain region correlation heterogeneity index and clustering analysis, different intracranial physiological states can be differentiated, providing more accurate diagnostic basis for doctors.

[0138] Avoid the problem of model performance decline and inaccurate diagnosis caused by mixing different intracranial physiological state data in traditional monitoring method. Through clustering analysis, the data of different intracranial physiological states are trained separately, which improves the generalization ability of the model under different individuals and different pathological conditions.

[0139] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An intracranial pressure detecting method characterized by comprising: Comprise: S101, obtain the region signal envelope data corresponding to the preset acquisition region, the acquisition region is set to at least two key brain regions, each key brain region corresponds to a region signal envelope data; S102, input each region signal envelope data into the pre-trained intracranial pressure estimation model to output the initial intracranial pressure data of each region; S103, based on the initial intracranial pressure data of each region, the intracranial pressure correlation feature vector is extracted and input into the pre-trained contribution index prediction model to output the contribution index corresponding to the initial intracranial pressure data of the region; The pre-trained contribution index prediction model is obtained in the following manner: A1, collect a large number of historical initial intracranial pressure data corresponding to the intracranial pressure correlation feature vector, and label each intracranial pressure correlation feature vector; The annotation content is set to the contribution index; Wherein, the determination method of the contribution index is: based on each historical initial intracranial pressure data, the actual intracranial pressure data recorded is obtained, the similarity between the historical initial intracranial pressure data and the actual intracranial pressure data is calculated, and the similarity is determined as the contribution index; A2, the labeled intracranial pressure correlation feature vector is used as a training sample set, and the preselected neural network structure is trained by using the training sample set to obtain the final contribution index prediction model; S104, based on the initial intracranial pressure data of each region and its contribution index, the target intracranial pressure data is calculated; The target intracranial pressure data is calculated according to the following formula: wherein, is the target intracranial pressure data, is the initial intracranial pressure data of the ith region, is the contribution index corresponding to the initial intracranial pressure data of the ith region, and n is the total number of key brain regions.

2. The intracranial pressure detecting method according to claim 1, wherein The extraction of the intracranial pressure correlation feature vector comprises: S201, obtain the region monitoring position information; S202, obtain the mean, variance, dynamic range and trend value of the initial intracranial pressure data of the region; S203, the monitoring position information, mean, variance, dynamic range and trend slope value are combined to form the intracranial pressure correlation feature vector.

3. The intracranial pressure detecting method according to claim 2, wherein The dynamic range value is set as: the initial intracranial pressure data is curve fitted to generate an intracranial pressure fitting curve; All the turning points of the fitting curve are obtained, and the sum of all the triangular area surrounded by all the turning points under the fitting curve is taken as the dynamic range value; The trend value is set as: the absolute value of the slope value determined by all adjacent turning points is averaged to obtain the trend value.

4. The intracranial pressure detecting method according to claim 1, wherein The S102 further comprises: S301, based on each region signal envelope data, the brain region correlation feature vector of the region signal envelope data is extracted; S302, for the i-th acquisition area, the region signal envelope data The corresponding brain region associated feature vector Splicing is carried out to obtain enhanced input features: r is the monitoring position information of the acquisition area; S303, input the enhanced input feature into the pre-trained intracranial pressure estimation model to output the initial intracranial pressure data corresponding to the acquisition region; The pre-trained intracranial pressure estimation model is obtained in the following manner: B1, collect a large number of region signal envelope data corresponding to the enhanced input feature of different acquisition regions in history and the corresponding real intracranial pressure data, and label each enhanced input feature with the real intracranial pressure data; B2, the labeled enhanced input feature is used as a training sample set, and the preselected neural network structure is trained by using the training sample set to generate the final intracranial pressure estimation model.

5. The intracranial pressure detecting method according to claim 4, wherein In the S301, the brain region correlation feature vector of the region signal envelope data is extracted, which specifically comprises: S401, based on all acquisition regions, time domain cross-correlation analysis is performed on the region signal envelope data of any two acquisition regions to generate cross-correlation feature information of any two acquisition regions; S402, based on each acquisition region, cross-correlation feature information of the region signal envelope data of the region and the region signal envelope data of other acquisition regions is obtained, and according to the distance value between the three-dimensional coordinates of the acquisition region and other acquisition regions, the cross-correlation feature information is arranged in ascending order to generate a cross-correlation feature information sequence of the acquisition region; S403, the cross-correlation feature information sequence of the acquisition region is determined as the brain region correlation feature vector.

6. The intracranial pressure detecting method according to claim 5, wherein The time domain cross-correlation analysis specifically includes: C1, traverse the two region signal envelope data by using a preset sliding window and a preset sliding step, calculate the covariance of the amplitude of the region signal envelope data of the i-th acquisition region and the j-th acquisition region in each window and ​ C2. Based on each sliding window, obtain strong association information between and ​ C3, based on the covariance and strong correlation information of each sliding window, the cross-correlation feature information is calculated: wherein, is the cross-correlation feature information, m is the total number of sliding windows, is the covariance of the kth sliding window, is the strong correlation information of the kth sliding window, and are the weight factors of the covariance and the strong correlation information, respectively, for reflecting the degree of influence on the cross-correlation feature information.

7. The intracranial pressure detecting method according to claim 4, wherein The S303 further includes: S501, obtaining a brain region correlation feature vector in the enhanced input features , obtaining a cross-brain region correlation heterogeneity index for comprehensively reflecting fluctuation intensity and complexity of correlation patterns between brain regions; S502, according to the cross-brain region correlation heterogeneity index of the enhanced input feature, determining the route input to the intracranial pressure estimation model, and outputting the initial intracranial pressure data corresponding to the acquisition region; Wherein, the intracranial pressure estimation model includes a plurality of transmission routing modules, each transmission routing module is used for learning according to the input enhanced input feature, and outputs the initial intracranial pressure data.

8. The intracranial pressure detecting method according to claim 7, wherein The acquisition method of the plurality of transmission routing modules includes: E1, obtaining a training sample set of the intracranial pressure estimation model, based on the brain region correlation feature vector of each training sample, using a preset clustering algorithm to classify the training sample set, obtaining a plurality of clusters, each cluster including at least one brain region correlation feature vector of the training sample, and averaging all brain region correlation feature vectors in the cluster to obtain the center feature label of the cluster; E2, based on the training sample corresponding to each cluster, training the preselected neural network structure, constantly optimizing the model parameters, and generating the corresponding transmission routing module, each transmission routing module is attached with the corresponding center feature label; The determination of the route input to the intracranial pressure estimation model specifically includes: The brain region correlation feature vector of the enhanced input feature is obtained, and the Euclidean distance value between each transmission routing module and the center feature label is calculated respectively, and the transmission routing module corresponding to the minimum Euclidean distance value is determined as the route of the enhanced input feature.

Images