Intracranial pressure noninvasive detection method and system based on infrared light density analysis

Abstract

The invention relates to the technical field of data detection. The intracranial pressure noninvasive detection method and system based on infrared light density analysis are provided, and the method comprises the steps that multiple sets of near-infrared light signals are emitted to a head monitoring area through a self-adaptive wavelength switching module, and light density signals of superficial tissue and deep tissue are separated; performing dynamic baseline drift correction processing on the optical density signal of the superficial tissue to generate a dynamic calibration signal, and performing time-space synchronous acquisition processing on the optical density signal of the deep tissue through a double-optical-path sensor to generate a dynamic optical density variation of a cerebrospinal fluid layer; and performing real-time correction processing on the intracranial pressure dynamic estimation value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and outputting a space weighted intracranial pressure curve and a multi-stage early warning signal based on the trained cerebrospinal fluid flow distribution model. The problem of conflict between optical signal interference separation and real-time correction in an optical signal processing technology is solved, so that the precision and adaptability of optical signal processing are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data detection, and particularly to a non-invasive intracranial pressure detection method and system based on infrared optical density analysis. Background Art

[0002] With the continuous development of medical detection technologies, non-invasive intracranial pressure monitoring has become increasingly important in fields such as neurosurgery and intensive care. Traditional intracranial pressure detection mainly relies on invasive means (such as ventricular puncture or lumbar puncture for pressure measurement), which have limitations such as a high risk of infection and the inability to continuously monitor. In recent years, non-invasive detection technologies based on near-infrared spectroscopy (NIRS) have gradually become a research hotspot.

[0003] However, there are conflicts in the relevant optical signal processing technologies in the separation and real-time correction of optical signal interference. Although the optical path separation technology can separate the optical density signals of superficial tissues and deep tissues, its adaptive wavelength switching module and optical path adaptive compensation mechanism are vulnerable to tissue blood flow fluctuations and ambient light transients when dealing with complex optical path differences, resulting in insufficient signal separation accuracy. At the same time, although the ambient light transient suppression algorithm can identify transient interference events when real-time correcting the dynamic estimated value of intracranial pressure, the mechanism of dynamically adjusting the cut-off frequency of the sliding window filter is difficult to fully adapt to the rapid change of interference intensity, causing signal correction lag. Summary of the Invention

[0004] Based on this, it is necessary to provide a non-invasive intracranial pressure detection method and system based on infrared optical density analysis for the above technical problems, so as to solve the conflict problems existing in the separation and real-time correction of optical signal interference in optical signal processing technologies, thereby improving the accuracy and adaptability of optical signal processing.

[0005] In a first aspect, the present application provides a non-invasive intracranial pressure detection method based on infrared optical density analysis, and the method includes:

[0006] Emitting multiple groups of near-infrared optical signals to the head monitoring area through an adaptive wavelength switching module, and separating the optical density signals of superficial tissues and deep tissues based on an optical path adaptive compensation mechanism;

[0007] Performing dynamic baseline drift correction processing on the optical density signals of superficial tissues to generate a dynamic calibration signal, and performing spatio-temporal synchronous acquisition processing on the optical density signals of deep tissues through a double optical path sensor to generate a dynamic optical density change amount of the cerebrospinal fluid layer;

[0008] Perform shallow interference subtraction on the dynamic optical density change of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronous calibration optical density signal; perform motion artifact suppression processing on the synchronous calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and strip the cross-absorption interference of melanin and hemoglobin through a non-linear absorption model to generate the optical density change related to the thickness of the cerebrospinal fluid layer;

[0009] Input the optical density change related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into the trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimate;

[0010] Perform real-time correction processing on the dynamic intracranial pressure estimate through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and output a spatially weighted intracranial pressure curve and multi-level warning signals based on the trained cerebrospinal fluid flow distribution model.

[0011] Further, input the optical density change related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into the trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimate, including:

[0012] Perform dynamic weight allocation on the optical density change related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters through a cross-modal attention mechanism to generate a spatio-temporal enhanced feature vector;

[0013] Perform time series analysis on the spatio-temporal enhanced feature vector based on a long short-term memory network to capture the dynamic evolution law of intracranial pressure and generate a dynamic pressure feature;

[0014] Construct a multi-modal training dataset based on animal experiment data and clinical invasive detection data, enhance sample diversity through a generative adversarial network, and optimize the initial weights of the model in combination with a transfer learning algorithm to generate a trained multi-modal neural network model;

[0015] Input the dynamic pressure feature into the fully connected layer of the trained multi-modal neural network model for non-linear transformation to generate a dynamic intracranial pressure estimate.

[0016] Further, perform time series analysis on the spatio-temporal enhanced feature vector based on a long short-term memory network to capture the dynamic evolution law of intracranial pressure and generate a dynamic pressure feature, including:

[0017] Extract local temporal correlation of the spatio-temporal enhanced feature vector based on a gated recurrent unit to generate preliminary temporal features;

[0018] Perform weighted fusion of the preliminary temporal features and the original spatio-temporal enhanced feature vector through cross-layer residual connections to generate enhanced temporal features;

[0019] Based on the intracranial pressure phase synchronization characteristics, perform multi-scale time window attention weight allocation on the enhanced temporal features, capture the cross-cycle correlation of sudden pressure fluctuations, and generate context-aware temporal features;

[0020] Perform tensor splicing on the context-aware temporal features and the cerebrospinal fluid flow direction parameters, and perform dimensional compression through a bottleneck layer to generate dynamic pressure features.

[0021] Furthermore, perform real-time correction processing on the intracranial pressure dynamic estimation value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and output a spatially weighted intracranial pressure curve and multi-level warning signals based on the trained cerebrospinal fluid flow distribution model, including:

[0022] Perform transient interference event recognition on the intracranial pressure dynamic estimation value through the ambient light transient suppression algorithm to generate an interference marker signal, and dynamically adjust the cut-off frequency of the sliding window filter based on the interference marker signal to generate an anti-interference intracranial pressure value;

[0023] Perform multi-brain region spatial mapping on the anti-interference intracranial pressure value based on the trained cerebrospinal fluid flow distribution model, and combine cerebral hemodynamic parameters to generate a cerebrospinal fluid pressure spatial distribution heat map;

[0024] Perform threshold matching and weight allocation on the pressure levels of each region in the cerebrospinal fluid pressure spatial distribution heat map. When the pressure level exceeds the preset pressure threshold and conforms to the brain hernia spatial diffusion pattern, trigger multi-level warning signals corresponding to the pressure level.

[0025] Furthermore, perform multi-brain region spatial mapping on the anti-interference intracranial pressure value based on the trained cerebrospinal fluid flow distribution model, and combine cerebral hemodynamic parameters to generate a cerebrospinal fluid pressure spatial distribution heat map, including:

[0026] Perform anatomical structure segmentation on the monitoring area based on a standard brain atlas to generate a multi-brain region spatial coordinate mapping relationship;

[0027] Based on the multi-brain region spatial coordinate mapping relationship, input the anti-interference intracranial pressure value and cerebral hemodynamic parameters into the cerebrospinal fluid flow distribution model, and eliminate the physiological parameter heterogeneity through a cross-modal feature alignment algorithm to generate a fused pressure-blood flow feature vector;

[0028] Based on the fused pressure-blood flow feature vector, calculate the pressure gradient distribution of each brain region through the hydrodynamic simulation layer of the cerebrospinal fluid flow distribution model, and combine the spatial interpolation algorithm to generate a continuous cerebrospinal fluid pressure spatial distribution heat map.

[0029] Further, perform dynamic baseline drift correction processing on the optical density signals of the shallow tissue to generate dynamic calibration signals, and perform spatio-temporal synchronous acquisition processing on the optical density signals of the deep tissue through a dual optical path sensor to generate the dynamic optical density change of the cerebrospinal fluid layer, including:

[0030] Extract the low-frequency baseline drift characteristics of the optical density signals of the shallow tissue based on the sliding time window statistical analysis method to generate drift compensation coefficients, and perform dynamic compensation on the original optical density signals through an adaptive filter to generate dynamic calibration signals;

[0031] Take the dynamic calibration signal as the shallow interference reference, and based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, perform optical path delay calibration and shallow interference deduction on the optical density signals of the deep tissue to generate synchronous calibration optical density signals;

[0032] Perform timestamp alignment and spatial registration on the synchronous calibration optical density signals, and combine the stability parameters of the dynamic calibration signals to strip the head micro-motion interference through the motion artifact suppression algorithm to generate the dynamic optical density change of the cerebrospinal fluid layer.

[0033] Further, perform shallow interference deduction on the dynamic optical density change of the cerebrospinal fluid layer based on the dynamic calibration signal to generate synchronous calibration optical density signals; perform motion artifact suppression processing on the synchronous calibration optical density signals based on the cerebral blood flow velocity monitoring signals to generate anti-interference optical density features, and strip the cross-absorption interference of melanin and hemoglobin through the non-linear absorption model to generate the optical density change related to the thickness of the cerebrospinal fluid layer, including:

[0034] Based on the time-domain fluctuation characteristics of the cerebral blood flow velocity monitoring signals, identify the artifact interference intervals caused by head movement through the adaptive threshold segmentation algorithm to generate motion artifact marking signals;

[0035] According to the optical density stability indexes of the motion artifact marking signals and the dynamic calibration signals, dynamically adjust the cut-off frequency and attenuation slope of the time-frequency domain joint filter to generate anti-interference optical density features;

[0036] Input the anti-interference optical density features into the non-linear absorption model, and generate the optical density change related to the thickness of the cerebrospinal fluid layer through the spectral feature extraction of the melanin-dominated absorption region and the residual compensation calculation of the hemoglobin absorption spectrum.

[0037] In a second aspect, the present application also provides a non-invasive intracranial pressure detection system based on infrared optical density analysis. The system includes:

[0038] A light source modulation and optical path separation module, configured to emit multiple groups of near-infrared light signals to the head monitoring area through an adaptive wavelength switching module, and separate the optical density signals of the shallow tissue and the deep tissue based on the optical path adaptive compensation mechanism;

[0039] A baseline correction and signal synchronization module is used to perform dynamic baseline drift correction processing on the optical density signal of the shallow tissue, generate a dynamic calibration signal, and perform spatio-temporal synchronization acquisition processing on the optical density signal of the deep tissue through a double optical path sensor to generate the dynamic optical density change of the cerebrospinal fluid layer;

[0040] An artifact suppression and absorption stripping module is used to perform shallow interference deduction on the dynamic optical density change of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronized calibration optical density signal; perform motion artifact suppression processing on the synchronized calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and strip the cross-absorption interference of melanin and hemoglobin through a non-linear absorption model to generate the optical density change related to the thickness of the cerebrospinal fluid layer;

[0041] A multi-modal neural network processing module is used to process the optical density change related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters by inputting them into a trained multi-modal neural network model to generate a dynamic intracranial pressure estimation value;

[0042] A transient suppression and flow warning module is used to perform real-time correction processing on the dynamic intracranial pressure estimation value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and output a spatially weighted intracranial pressure curve and multi-level warning signals based on a trained cerebrospinal fluid flow distribution model.

[0043] In a third aspect, the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and it is characterized in that when the processor executes the computer program, the steps of any method in the first aspect of the present application are implemented.

[0044] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and it is characterized in that when the computer program is executed by a processor, the steps of any method in the first aspect of the present application are implemented.

[0045] The technical solutions provided by this application include the following technical effects: By providing a non-invasive intracranial pressure detection method and system based on infrared optical density analysis, the method includes: emitting multiple groups of near-infrared light signals to the head monitoring area through an adaptive wavelength switching module, and separating the optical density signals of superficial tissues and deep tissues based on an optical path adaptive compensation mechanism; performing dynamic baseline drift correction processing on the optical density signals of superficial tissues to generate a dynamic calibration signal, and performing spatio-temporal synchronous acquisition processing on the optical density signals of deep tissues through a double optical path sensor to generate the dynamic optical density change of the cerebrospinal fluid layer; performing superficial interference deduction on the dynamic optical density change of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronous calibration optical density signal; performing motion artifact suppression processing on the synchronous calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and stripping the cross-absorption interference of melanin and hemoglobin through a non-linear absorption model to generate the optical density change related to the thickness of the cerebrospinal fluid layer; inputting the optical density change related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into a trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimation value; performing real-time correction processing on the dynamic intracranial pressure estimation value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and outputting a spatially weighted intracranial pressure curve and multi-level warning signals based on a trained cerebrospinal fluid flow distribution model, so as to solve the conflict problem existing in the optical signal interference separation and real-time correction of optical signal processing technology, thereby improving the accuracy and adaptability of optical signal processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a flowchart of a non-invasive intracranial pressure detection method based on infrared optical density analysis in an embodiment of the present invention;

[0048] Figure 2 It is a structural diagram of a non-invasive intracranial pressure detection system based on infrared optical density analysis in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To make the above objects, features, and advantages of the present application more apparent and understandable, the following detailed description of the specific implementation manners of the present application will be given in conjunction with the accompanying drawings. Many specific details are set forth in the following description to fully understand the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the application. Therefore, the present application is not limited by the specific embodiments disclosed below.

[0050] As Figure 1 shown, the present application provides a non-invasive intracranial pressure detection method based on infrared optical density analysis. The method includes:

[0051] S101: Transmit multiple groups of near-infrared light signals to the head monitoring area through an adaptive wavelength switching module, and separate the optical density signals of the shallow tissue and the deep tissue based on the optical path adaptive compensation mechanism.

[0052] Specifically, an adaptive wavelength switching module is used. This module can automatically adjust and transmit multiple groups of near-infrared light signals with different wavelengths to the head monitoring area according to the preset wavelength range and step size to cover tissues at different depths and levels, thus providing a basis for subsequent signal separation. At the same time, with the help of the optical path adaptive compensation mechanism, by real-time monitoring the transmission characteristics of the optical signals in the shallow tissue (such as the scalp and skull) and the deep tissue (such as the cerebrospinal fluid layer and brain tissue), such as light intensity attenuation and scattering, the optical path parameters, including the incident angle, reflection angle, and optical path of the light, are dynamically adjusted, so as to effectively separate the optical density signals of the shallow tissue and the deep tissue, providing an accurate data basis for subsequent signal processing and intracranial pressure estimation.

[0053] S102: Perform dynamic baseline drift correction processing on the optical density signal of the shallow tissue to generate a dynamic calibration signal, and perform spatio-temporal synchronous acquisition processing on the optical density signal of the deep tissue through a double optical path sensor to generate the dynamic optical density change amount of the cerebrospinal fluid layer.

[0054] Specifically, dynamic baseline drift correction processing is performed on the optical density signals of the shallow tissue to generate dynamic calibration signals. This step uses the sliding time window statistical analysis method to extract the low-frequency baseline drift characteristics of the optical density signals of the shallow tissue, generates drift compensation coefficients, and dynamically compensates the original optical density signals through an adaptive filter, thereby generating dynamic calibration signals. Subsequently, the optical density signals of the deep tissue are collected and processed in a spatio-temporal synchronous manner by a dual optical path sensor to generate the dynamic optical density change amount of the cerebrospinal fluid layer. Among them, the dynamic calibration signal is used as the shallow interference reference, and based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, optical path delay calibration and shallow interference subtraction are performed on the optical density signals of the deep tissue to generate synchronous calibration optical density signals. Subsequently, timestamp alignment and spatial registration are performed on the synchronous calibration optical density signals, and combined with the stability parameters of the dynamic calibration signals, the head micro-motion interference is removed through a motion artifact suppression algorithm, thereby generating the dynamic optical density change amount of the cerebrospinal fluid layer.

[0055] S103: Based on the dynamic calibration signals, perform shallow interference subtraction on the dynamic optical density change amount of the cerebrospinal fluid layer to generate synchronous calibration optical density signals; based on the cerebral blood flow velocity monitoring signals, perform motion artifact suppression processing on the synchronous calibration optical density signals to generate anti-interference optical density features, and strip the cross-absorption interference of melanin and hemoglobin through a non-linear absorption model to generate the optical density change amount related to the thickness of the cerebrospinal fluid layer.

[0056] Specifically, based on the dynamic calibration signals, perform shallow interference subtraction on the dynamic optical density change amount of the cerebrospinal fluid layer to generate synchronous calibration optical density signals. This step uses the dynamic calibration signals as the shallow interference reference, and based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, performs optical path delay calibration and shallow interference subtraction on the optical density signals of the deep tissue, thereby generating synchronous calibration optical density signals. Subsequently, based on the cerebral blood flow velocity monitoring signals, perform motion artifact suppression processing on the synchronous calibration optical density signals to generate anti-interference optical density features. Among them, the adaptive threshold segmentation algorithm is used to identify the artifact interference interval caused by head movement to generate a motion artifact marking signal, and according to the marking signal and the optical density stability index of the dynamic calibration signals, the cut-off frequency and attenuation slope of the time-frequency domain joint filter are dynamically adjusted, thereby generating anti-interference optical density features. Subsequently, the cross-absorption interference of melanin and hemoglobin is stripped through a non-linear absorption model to generate the optical density change amount related to the thickness of the cerebrospinal fluid layer. In this step, the anti-interference optical density features are input into the non-linear absorption model, and the spectral feature extraction of the melanin-dominated absorption region and the residual compensation calculation of the hemoglobin absorption spectrum are used, and then the optical density change amount related to the thickness of the cerebrospinal fluid layer is generated.

[0057] S104: Input the optical density change amount related to the cerebrospinal fluid layer thickness and the intracranial microcirculation parameters into the trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimate value.

[0058] Specifically, it is necessary to perform feature extraction and fusion on the optical density change amount related to the cerebrospinal fluid layer thickness and the intracranial microcirculation parameters to form a feature vector that can comprehensively represent the intracranial state. Then, input this feature vector into the multi-modal neural network model. This model usually consists of multiple neural network layers, including an input layer, hidden layers, and an output layer, and may also include convolutional layers, pooling layers, fully connected layers, etc., to perform complex non-linear transformations and learning on the input feature vector. During the model training stage, a training data set containing various modal data (such as optical density change amounts, microcirculation parameters, etc.) is used, and the weights and bias parameters in the network are adjusted through optimization algorithms (such as gradient descent) so that the output of the model is as close as possible to the actual intracranial pressure value. In actual applications, the trained model can automatically perform feature extraction, fusion, and transformation based on the input optical density change amount and microcirculation parameters to generate a dynamic intracranial pressure estimate value, providing important reference information for clinicians and assisting in the judgment of intracranial pressure status and the diagnosis of diseases.

[0059] S105: Perform real-time correction processing on the dynamic intracranial pressure estimate value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and output a spatially weighted intracranial pressure curve and multi-level warning signals based on the trained cerebrospinal fluid flow distribution model.

[0060] Specifically, perform real-time correction processing on the dynamic intracranial pressure estimate value through the ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value. In this step, the ambient light transient suppression algorithm can identify transient interference events in the dynamic intracranial pressure estimate value and generate an interference marking signal. According to this marking signal, dynamically adjust the cut-off frequency of the sliding window filter to generate an anti-interference intracranial pressure value. Then, based on the trained cerebrospinal fluid flow distribution model, perform multi-brain region spatial mapping on the anti-interference intracranial pressure value, and combine cerebral hemodynamic parameters to generate a cerebrospinal fluid pressure spatial distribution heat map. Then, perform threshold matching and weight assignment on the pressure levels of each region in the cerebrospinal fluid pressure spatial distribution heat map. When the pressure level exceeds the preset pressure threshold and conforms to the brain hernia spatial diffusion pattern, trigger multi-level warning signals corresponding to the pressure level.

[0061] The embodiment of the present application provides a non-invasive intracranial pressure detection method based on infrared optical density analysis. A multi-group of near-infrared light signals are emitted to the head monitoring area through an adaptive wavelength switching module, and the optical density signals of the shallow tissue and the deep tissue are separated based on an optical path adaptive compensation mechanism; the optical density signals of the shallow tissue are subjected to dynamic baseline drift correction processing to generate a dynamic calibration signal, and the optical density signals of the deep tissue are collected and processed in space-time synchronization through a double optical path sensor to generate a dynamic optical density change amount of the cerebrospinal fluid layer; the dynamic optical density change amount of the cerebrospinal fluid layer is subjected to shallow interference deduction based on the dynamic calibration signal to generate a synchronous calibration optical density signal; the synchronous calibration optical density signal is subjected to motion artifact suppression processing based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and the cross-absorption interference of melanin and hemoglobin is stripped through a non-linear absorption model to generate an optical density change amount related to the thickness of the cerebrospinal fluid layer; the optical density change amount related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters are input into a trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimation value; the dynamic intracranial pressure estimation value is subjected to real-time correction processing through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and a spatially weighted intracranial pressure curve and multi-level warning signals are output based on a trained cerebrospinal fluid flow distribution model, so as to solve the conflict problem existing in the separation and real-time correction of optical signal interference in optical signal processing technology, thereby improving the accuracy and adaptability of optical signal processing.

[0062] Further, inputting the optical density change amount related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into a trained multi-modal neural network model for processing to generate a dynamic intracranial pressure estimation value includes:

[0063] Performing dynamic weight allocation on the optical density change amount related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters through a cross-modal attention mechanism to generate a spatio-temporal enhanced feature vector;

[0064] Performing time series analysis on the spatio-temporal enhanced feature vector based on a long short-term memory network to capture the dynamic evolution law of intracranial pressure and generate a dynamic pressure feature;

[0065] Constructing a multi-modal training data set based on animal experiment data and clinical invasive detection data, enhancing sample diversity through a generative adversarial network, and optimizing the initial weights of the model in combination with a transfer learning algorithm to generate a trained multi-modal neural network model;

[0066] Inputting the dynamic pressure feature into the fully connected layer of the trained multi-modal neural network model for non-linear transformation to generate a dynamic intracranial pressure estimation value.

[0067] Specifically, a dynamic weight allocation is performed on the optical density change related to the cerebrospinal fluid layer thickness and the intracranial microcirculation parameters through a cross-modal attention mechanism to generate a spatio-temporal enhanced feature vector. The cross-modal attention mechanism enables the model to focus on the information related between different modalities, and assigns different weights by calculating the similarity or correlation between different modalities. Then, based on the long short-term memory network (LSTM), a time series analysis is performed on the spatio-temporal enhanced feature vector to capture the dynamic evolution law of intracranial pressure and generate dynamic pressure features. LSTM is a special recurrent neural network that can effectively solve the problems of gradient vanishing and gradient explosion in traditional RNN when dealing with long sequence data, and controls the information flow through a gating mechanism to capture long-term dependencies. In the model training stage, a multi-modal training dataset is constructed based on animal experiment data and clinical invasive detection data, the sample diversity is enhanced through a generative adversarial network, and the initial weights of the model are optimized by combining a transfer learning algorithm to generate a trained multi-modal neural network model. Then, the dynamic pressure features are input into the fully connected layer of the trained multi-modal neural network model for non-linear transformation to generate a dynamic intracranial pressure estimate value.

[0068] Furthermore, based on the long short-term memory network, a time series analysis is performed on the spatio-temporal enhanced feature vector to capture the dynamic evolution law of intracranial pressure and generate dynamic pressure features, including:

[0069] Based on the gated recurrent unit, local temporal correlation extraction is performed on the spatio-temporal enhanced feature vector to generate preliminary temporal features;

[0070] The preliminary temporal features and the original spatio-temporal enhanced feature vector are weighted and fused through cross-layer residual connections to generate enhanced temporal features;

[0071] Based on the intracranial pressure phase synchronization characteristics, multi-scale time window attention weight allocation is performed on the enhanced temporal features to capture the cross-cycle correlation of sudden pressure fluctuations and generate context-aware temporal features;

[0072] The context-aware temporal features and the cerebrospinal fluid flow direction parameters are tensor-concatenated and dimensionally compressed through a bottleneck layer to generate dynamic pressure features.

[0073] Specifically, a gated recurrent unit (GRU) is used to extract local temporal correlations from the spatio-temporal enhanced feature vector to generate preliminary temporal features. Through the control of the update gate and the reset gate, the GRU can effectively capture the local dependencies in the sequence data and reduce the vanishing gradient problem. Then, the preliminary temporal features are weighted and fused with the original spatio-temporal enhanced feature vector through cross-layer residual connections to generate enhanced temporal features. Cross-layer residual connections not only ensure the effective transmission of information, but also enhance the flow of gradients, avoid gradient vanishing, and improve information flow, enabling the input information to be transmitted across layers and avoiding information loss caused by the increase in the number of layers. Then, based on the intracranial pressure phase synchronization characteristics, multi-scale time window attention weight allocation is performed on the enhanced temporal features to capture the cross-cycle correlation of sudden pressure fluctuations and generate context-aware temporal features. Multi-scale time window attention weight allocation enables the model to focus on important information at different time scales, thus better capturing the dynamic changes of intracranial pressure. Then, the context-aware temporal features are tensor-concatenated with the cerebrospinal fluid flow direction parameters and dimensionally compressed through a bottleneck layer to generate dynamic pressure features. The design of the bottleneck layer helps reduce the feature dimension and improve the computational efficiency and expressive power of the model.

[0074] Furthermore, the dynamic estimated value of intracranial pressure is processed in real time through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and a spatially weighted intracranial pressure curve and multi-level warning signals are output based on the trained cerebrospinal fluid flow distribution model, including:

[0075] The ambient light transient suppression algorithm is used to identify transient interference events in the dynamic estimated value of intracranial pressure to generate interference marker signals, and the cut-off frequency of the sliding window filter is dynamically adjusted based on the interference marker signals to generate an anti-interference intracranial pressure value;

[0076] The anti-interference intracranial pressure value is subjected to multi-brain region spatial mapping based on the trained cerebrospinal fluid flow distribution model, and a cerebrospinal fluid pressure spatial distribution heat map is generated in combination with cerebral hemodynamic parameters;

[0077] Threshold matching and weight allocation are performed on the pressure levels of each region in the cerebrospinal fluid pressure spatial distribution heat map. When the pressure level exceeds the preset pressure threshold and conforms to the brain hernia spatial diffusion pattern, multi-level warning signals corresponding to the pressure level are triggered.

[0078] Specifically, the transient interference event recognition is performed on the dynamically estimated intracranial pressure value through the ambient light transient suppression algorithm to generate an interference marking signal, and the cut-off frequency of the sliding window filter is dynamically adjusted based on the interference marking signal to generate an anti-interference intracranial pressure value. Then, based on the trained cerebrospinal fluid flow distribution model, multi-brain region spatial mapping is performed on the anti-interference intracranial pressure value, and a cerebrospinal fluid pressure spatial distribution heat map is generated by combining cerebral hemodynamic parameters. Then, threshold matching and weight assignment are performed on the pressure levels of each region in the cerebrospinal fluid pressure spatial distribution heat map. When the pressure level exceeds the preset pressure threshold and conforms to the brain hernia spatial diffusion pattern, a multi-level warning signal corresponding to the pressure level is triggered.

[0079] Further, multi-brain region spatial mapping is performed on the anti-interference intracranial pressure value based on the trained cerebrospinal fluid flow distribution model, and a cerebrospinal fluid pressure spatial distribution heat map is generated by combining cerebral hemodynamic parameters, including:

[0080] Performing anatomical structure segmentation on the monitoring area based on a standard brain atlas to generate a multi-brain region spatial coordinate mapping relationship;

[0081] Based on the multi-brain region spatial coordinate mapping relationship, the anti-interference intracranial pressure value and cerebral hemodynamic parameters are input into the cerebrospinal fluid flow distribution model, and the physiological parameter heterogeneity is eliminated through a cross-modal feature alignment algorithm to generate a fused pressure-blood flow feature vector;

[0082] Based on the fused pressure-blood flow feature vector, the pressure gradient distribution of each brain region is calculated through the hydrodynamic simulation layer of the cerebrospinal fluid flow distribution model, and a continuous cerebrospinal fluid pressure spatial distribution heat map is generated by combining a spatial interpolation algorithm.

[0083] Specifically, anatomical structure segmentation is performed on the monitoring area based on a standard brain atlas to generate a multi-brain region spatial coordinate mapping relationship. Then, based on the multi-brain region spatial coordinate mapping relationship, the anti-interference intracranial pressure value and cerebral hemodynamic parameters are input into the cerebrospinal fluid flow distribution model, and the physiological parameter heterogeneity is eliminated through a cross-modal feature alignment algorithm to generate a fused pressure-blood flow feature vector. Then, based on the fused pressure-blood flow feature vector, the pressure gradient distribution of each brain region is calculated through the hydrodynamic simulation layer of the cerebrospinal fluid flow distribution model, and a continuous cerebrospinal fluid pressure spatial distribution heat map is generated by combining a spatial interpolation algorithm.

[0084] Further, dynamic baseline drift correction processing is performed on the optical density signal of the shallow tissue to generate a dynamic calibration signal, and spatio-temporal synchronous acquisition processing is performed on the optical density signal of the deep tissue through a dual optical path sensor to generate the dynamic optical density change amount of the cerebrospinal fluid layer, including:

[0085] Extract the low-frequency baseline drift characteristics of the optical density signal of the shallow tissue based on the sliding time window statistical analysis method, generate the drift compensation coefficient, and dynamically compensate the original optical density signal through an adaptive filter to generate a dynamic calibration signal;

[0086] Use the dynamic calibration signal as the shallow interference reference. Based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, perform optical path delay calibration and shallow interference subtraction on the optical density signal of the deep tissue to generate a synchronous calibration optical density signal;

[0087] Perform timestamp alignment and spatial registration on the synchronous calibration optical density signal. Combine the stability parameters of the dynamic calibration signal, and strip the head micro-motion interference through the motion artifact suppression algorithm to generate the dynamic optical density change of the cerebrospinal fluid layer.

[0088] Specifically, perform dynamic baseline drift correction on the optical density signal of the shallow tissue to generate a dynamic calibration signal. This step uses the sliding time window statistical analysis method to extract the low-frequency baseline drift characteristics of the optical density signal of the shallow tissue, generate the drift compensation coefficient, and dynamically compensate the original optical density signal through an adaptive filter to generate a dynamic calibration signal. Then, use the dynamic calibration signal as the shallow interference reference. Based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, perform optical path delay calibration and shallow interference subtraction on the optical density signal of the deep tissue to generate a synchronous calibration optical density signal. Then, perform timestamp alignment and spatial registration on the synchronous calibration optical density signal. Combine the stability parameters of the dynamic calibration signal, and strip the head micro-motion interference through the motion artifact suppression algorithm to generate the dynamic optical density change of the cerebrospinal fluid layer.

[0089] Furthermore, based on the time-domain fluctuation characteristics of the cerebral blood flow velocity monitoring signal, identify the artifact interference interval caused by head movement through the adaptive threshold segmentation algorithm to generate a motion artifact marking signal;

[0090] According to the optical density stability index of the motion artifact marking signal and the dynamic calibration signal, dynamically adjust the cut-off frequency and attenuation slope of the time-frequency domain joint filter to generate an anti-interference optical density feature;

[0091] Input the anti-interference optical density feature into the non-linear absorption model, and generate the optical density change related to the thickness of the cerebrospinal fluid layer through the spectral feature extraction of the melanin-dominated absorption region and the residual compensation calculation of the hemoglobin absorption spectrum.

[0092] Specifically, the dynamic optical density change of the cerebrospinal fluid layer is subtracted for shallow interference based on the dynamic calibration signal to generate a synchronous calibration optical density signal. This step uses the dynamic calibration signal as a shallow interference reference, and through the fixed spacing of the dual optical path sensor and the geometric relationship between the light transmission path, the optical density signal of the deep tissue is calibrated for optical path delay and subtracted for shallow interference, thereby generating a synchronous calibration optical density signal.

[0093] Afterwards, the synchronous calibration optical density signal is subjected to motion artifact suppression processing based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature. The artifact interference interval caused by head movement is identified by an adaptive threshold segmentation algorithm to generate a motion artifact marker signal. According to the optical density stability index of the marker signal and the dynamic calibration signal, the cutoff frequency and attenuation slope of the time-frequency domain joint filter are dynamically adjusted to generate an anti-interference optical density feature.

[0094] After that, the anti-interference optical density characteristics are input into the nonlinear absorption model, and the optical density variation related to the thickness of the cerebrospinal fluid layer is generated by extracting the spectrum characteristics of the melanin-dominated absorption area and calculating the residual compensation of the hemoglobin absorption spectrum. In this step, the nonlinear absorption model can effectively remove the cross-absorption interference between melanin and hemoglobin, thereby obtaining a more accurate optical density variation related to the thickness of the cerebrospinal fluid layer.

[0095] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0096] In one embodiment, the present application also provides a non-invasive intracranial pressure detection system 200 based on infrared light density analysis, the system comprising:

[0097] The light source modulation and optical path separation module 201 is used to transmit multiple groups of near-infrared light signals to the head monitoring area through the adaptive wavelength switching module, and separate the optical density signals of the shallow tissue and the deep tissue based on the optical path adaptive compensation mechanism;

[0098] The baseline correction and signal synchronization module 202 is used to perform dynamic baseline drift correction processing on the optical density signals of shallow tissues, generate dynamic calibration signals, and perform spatio-temporal synchronization acquisition processing on the optical density signals of deep tissues through a double optical path sensor to generate the dynamic optical density change amount of the cerebrospinal fluid layer;

[0099] The artifact suppression and absorption stripping module 203 is used to perform shallow interference deduction on the dynamic optical density change amount of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronized calibration optical density signal; perform motion artifact suppression processing on the synchronized calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate anti-interference optical density features, and strip the cross-absorption interference of melanin and hemoglobin through a non-linear absorption model to generate the optical density change amount related to the thickness of the cerebrospinal fluid layer;

[0100] The multi-modal neural network processing module 204 is used to process the optical density change amount related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters by inputting them into a trained multi-modal neural network model to generate a dynamic intracranial pressure estimation value;

[0101] The transient suppression and flow warning module 205 is used to perform real-time correction processing on the dynamic intracranial pressure estimation value through a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and output a spatially weighted intracranial pressure curve and multi-level warning signals based on a trained cerebrospinal fluid flow distribution model.

[0102] Specifically, the light source modulation and optical path separation module 201 uses an adaptive wavelength switching and optical path compensation mechanism to separate shallow and deep tissue signals; the baseline correction and signal synchronization module 202 corrects the shallow signals and synchronously acquires deep signals to generate the dynamic optical density change amount of the cerebrospinal fluid layer; the artifact suppression and absorption stripping module 203 deducts shallow interference and suppresses motion artifacts, and strips cross-absorption interference to obtain the optical density change amount related to the thickness of the cerebrospinal fluid layer; the multi-modal neural network processing module 204 fuses the optical density change amount and microcirculation parameters, and estimates the intracranial pressure through a trained model; the transient suppression and flow warning module 205 corrects the estimation value in real time and outputs an anti-interference intracranial pressure value and warning signals. The above steps effectively solve the problems of interference separation and real-time correction in optical signal processing, and improve the accuracy and reliability of detection.

[0103] The multi-modal neural network processing module 204 is further used for:

[0104] Performing dynamic weight allocation on the optical density change amount related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters through a cross-modal attention mechanism to generate a spatio-temporal enhanced feature vector;

[0105] Performing time series analysis on the spatio-temporal enhanced feature vector based on a long short-term memory network to capture the dynamic evolution law of intracranial pressure and generate dynamic pressure features;

[0106] Construct a multimodal training dataset based on animal experiment data and clinical invasive detection data, enhance sample diversity through a generative adversarial network, and optimize the initial weights of the model by combining a transfer learning algorithm to generate a trained multimodal neural network model;

[0107] Input the dynamic pressure features into the fully connected layer of the trained multimodal neural network model for non-linear transformation to generate an intracranial pressure dynamic estimate.

[0108] The multimodal neural network processing module 204 is also used for:

[0109] Extract local temporal correlations from the spatio-temporal enhanced feature vector based on a gated recurrent unit to generate preliminary temporal features;

[0110] Perform weighted fusion of the preliminary temporal features and the original spatio-temporal enhanced feature vector through cross-layer residual connections to generate enhanced temporal features;

[0111] Based on the intracranial pressure phase synchronization characteristics, perform multi-scale time window attention weight allocation on the enhanced temporal features to capture the cross-cycle correlation of sudden pressure fluctuations and generate context-aware temporal features;

[0112] Perform tensor concatenation on the context-aware temporal features and the cerebrospinal fluid flow direction parameters, and perform dimension compression through a bottleneck layer to generate dynamic pressure features.

[0113] The transient suppression and flow warning module 205 is also used for:

[0114] Identify transient interference events in the intracranial pressure dynamic estimate through an ambient light transient suppression algorithm to generate an interference marker signal, and dynamically adjust the cut-off frequency of a sliding window filter based on the interference marker signal to generate an interference-resistant intracranial pressure value;

[0115] Perform multi-brain region spatial mapping on the interference-resistant intracranial pressure value based on a trained cerebrospinal fluid flow distribution model, and combine cerebral hemodynamic parameters to generate a cerebrospinal fluid pressure spatial distribution heat map;

[0116] Perform threshold matching and weight allocation on the pressure levels of each region in the cerebrospinal fluid pressure spatial distribution heat map. When the pressure level exceeds a preset pressure threshold and conforms to the brain hernia spatial diffusion pattern, trigger a multi-level warning signal corresponding to the pressure level.

[0117] The transient suppression and flow warning module 205 is also used for:

[0118] Segment the anatomical structure of the monitoring area based on a standard brain atlas to generate a multi-brain region spatial coordinate mapping relationship;

[0119] Based on the spatial coordinate mapping relationship of multiple brain regions, the anti-interference intracranial pressure value and cerebral hemodynamic parameters are input into the cerebrospinal fluid flow distribution model. The physiological parameter heterogeneity is eliminated through the cross-modal feature alignment algorithm, and a fused pressure-blood flow feature vector is generated.

[0120] Based on the fused pressure-blood flow feature vector, the pressure gradient distribution of each brain region is calculated through the hydrodynamic simulation layer of the cerebrospinal fluid flow distribution model, and a continuous cerebrospinal fluid pressure spatial distribution heat map is generated by combining the spatial interpolation algorithm.

[0121] The baseline correction and signal synchronization module 202 is also used for:

[0122] Based on the sliding time window statistical analysis method, the low-frequency baseline drift characteristics of the optical density signal of the shallow tissue are extracted to generate a drift compensation coefficient, and the original optical density signal is dynamically compensated through an adaptive filter to generate a dynamically calibrated signal.

[0123] Taking the dynamically calibrated signal as the shallow interference reference, based on the fixed spacing of the dual optical path sensor and the geometric relationship of the optical transmission path, the optical path delay calibration and shallow interference deduction of the optical density signal of the deep tissue are performed to generate a synchronized calibrated optical density signal.

[0124] The time stamp alignment and spatial registration of the synchronized calibrated optical density signal are performed. Combining the stability parameters of the dynamically calibrated signal, the head micro-motion interference is stripped through the motion artifact suppression algorithm to generate the dynamic optical density change amount of the cerebrospinal fluid layer.

[0125] The artifact suppression and absorption stripping module 203 is also used for:

[0126] Based on the time-domain fluctuation characteristics of the cerebral blood flow velocity monitoring signal, the artifact interference interval caused by head movement is identified through the adaptive threshold segmentation algorithm to generate a motion artifact marking signal.

[0127] According to the optical density stability index of the motion artifact marking signal and the dynamically calibrated signal, the cut-off frequency and attenuation slope of the time-frequency domain joint filter are dynamically adjusted to generate anti-interference optical density characteristics.

[0128] The anti-interference optical density characteristics are input into the nonlinear absorption model. Through the spectral feature extraction of the melanin-dominated absorption region and the residual compensation calculation of the hemoglobin absorption spectrum, the optical density change amount related to the thickness of the cerebrospinal fluid layer is generated.

[0129] In one embodiment, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0130] In one embodiment, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0131] In one embodiment, photons are emitted from a light source and received by a detector. The path shape of the photons passing through the brain tissue is similar to a banana. The light emitted by the light source passes through the scalp and skull, then reaches the brain tissue, and after a series of absorption and scattering in the brain tissue, it is detected by the detector at the end of the optical path. According to the derivation of Lambert-Beer's law, the changes in hemoglobin, blood oxygen, etc. can be obtained. Lambert's law states that the proportion of photons absorbed by a transparent medium is independent of the intensity of the light entering the medium. Along the photon transmission path, the proportion of photons absorbed by the same thickness of the medium is the same. Beer's law states that the number of photons absorbed by the medium is proportional to the number of molecules capable of absorbing photons in the entire optical path. Lambert-Beer's law requires that the medium has no scattering, is evenly distributed, and has a single composition. It is defined as the ratio of the intensity of incident light to the intensity of transmitted light, and its expression is as follows:

[0132] I = I 0 e -∈CL

[0133] where I represents the intensity of the light emerging after the incident light passes through the medium, I 0 is the light emitted by the light source, that is, the intensity of the light entering the medium, ∈ represents the molecular extinction coefficient of this medium, C represents the concentration of the medium, and L represents the optical path that the incident light passes through the medium.

[0134] The absorbance is defined as the base-10 logarithm of the ratio of the intensity of the incident light before passing through a solution or a certain substance to the intensity of the transmitted light after passing through the medium:

[0135]

[0136] Optical density (OD) is a commonly used object in the current research of biological tissue spectroscopy. It is the reciprocal of the absorbance. The following formula describes the magnitude of the energy attenuation when light passes through biological tissue:

[0137]

[0138] Also, because it is found in the research that the intensity of the light emitted by the light source is not easy to detect, scientific researchers usually detect the change in the optical density (△OD) of biological tissue. First, the intensity of the transmitted light detected under a certain specific condition is selected as the reference state, and then the intensity of the light detected under other conditions is compared with the intensity of the light in the reference state. This change value is △OD:

[0139]

[0140] Among them, represents the intensity of the emitted light detected by the detector at time t 0 , and I t represents the intensity of the emitted light detected at time t. By this method, the detection of the incident light, which is a difficult object to accurately detect, is avoided.

[0141] For intracranial pressure detection, only the change in the thickness of the dura mater needs to be obtained. Therefore, the distances between the two photosensitive sensors and the light source are reasonably set to eliminate the loss of light energy by the remaining tissues, further improving the accuracy.

[0142] The absorption coefficient of melanin in the human brain monitoring area for 700 nm red light is much greater than that of hemoglobin. Therefore, when constructing a cerebral blood oxygen monitoring model, it is approximately considered that the change in the optical density of the 700 nm emitted light is caused by the absorption of melanin. Moreover, within the above-mentioned "optical window" band, as the wavelength increases, although the absorption coefficient of melanin for near-infrared light decreases, the trend is relatively stable. Therefore, the change in the optical density at 700 nm wavelength, that is, the absorption amount of melanin at 700 nm wavelength, is used to replace the influence of melanin on the change in optical density at 660 nm and 940 nm near-infrared light wavelengths to reduce the complexity during model construction. Based on the above principle, according to the framework of the hardware system, the model is constructed as follows:

[0143]

[0144] Among them, ΔOD λ represents the change in the optical density of the emitted light with a wavelength of λ, where the value of λ includes 700 nm, 640 nm, 805 nm, and 940 nm, DPF represents the differential path factor of a partial optical path, and Δρ represents the difference in the distances between the two photodetectors and the light, respectively represent the molar extinction coefficients of reduced hemoglobin for the wavelength light sources of 660 nm, 805 nm, and 940 nm, respectively represent the molar extinction coefficients of oxyhemoglobin for the wavelength light sources of 660 nm, 805 nm, 940 nm, and 700 nm, represents the protein concentration value in the human brain monitoring area, C Hb represents the concentration value of reduced hemoglobin in the human monitoring area, C M represents the melanin concentration in the human brain monitoring area. Solving the above equations gives the following:

[0145]

[0146] rSO 2 = kx 2 + bx + c

[0147] Among them, C tHb Indicates the total serum protein concentration value of the human brain monitoring area. 2 The absorption capacity of Hb to near-infrared wavelength is equal to that of Hb, so HbO is used in the above formula. 2 The extinction coefficient G for the 805nm source represents the absorption coefficient of the two hemoglobins in this environment, and x is the calculated rSO 2 .

[0148] The accurate k, b, and c are obtained by fitting clinical data. At this point, the cerebral blood oxygen value has been calculated.

[0149] Calculation of intracranial pressure values:

[0150] Intracranial pressure is mainly calculated by detecting the change in dura mater thickness using a light source of 805nm. The data received by the two photosensors are A1 and A2 respectively. The following formula is used to establish a mathematical model and calculate the accurate value of IPC:

[0151] ΔOD=lg(A1 / A2)

[0152] ICP=ΔOD*a+b

[0153] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0154] The above-mentioned embodiments only express several implementation methods of the embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the embodiments of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the embodiments of the present application, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A non-invasive detection method for intracranial pressure based on infrared optical density analysis, characterized in that: The method comprises: Multiple groups of near-infrared light signals are emitted to the head monitoring area through the adaptive wavelength switching module, and the optical density signals of shallow tissue and deep tissue are separated based on the optical path adaptive compensation mechanism; Performing dynamic baseline drift correction processing on the optical density signal of the shallow tissue to generate a dynamic calibration signal, and performing spatiotemporal synchronous acquisition processing on the optical density signal of the deep tissue through a dual optical path sensor to generate a dynamic optical density change of the cerebrospinal fluid layer; Based on the dynamic calibration signal, shallow interference is subtracted from the dynamic optical density change of the cerebrospinal fluid layer to generate a synchronous calibration optical density signal; based on the cerebral blood flow velocity monitoring signal, motion artifact suppression is performed on the synchronous calibration optical density signal to generate an anti-interference optical density feature, and the cross-absorption interference between melanin and hemoglobin is stripped off through a nonlinear absorption model to generate an optical density change related to the thickness of the cerebrospinal fluid layer; Inputting the optical density variation related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into the trained multimodal neural network model for processing to generate a dynamic estimation value of intracranial pressure; The dynamic estimated value of intracranial pressure is corrected in real time by a preset ambient light transient suppression algorithm to generate an interference-resistant intracranial pressure value, and a spatially weighted intracranial pressure curve and a multi-level warning signal are output based on a trained cerebrospinal fluid flow distribution model.

2. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 1 is characterized in that: The step of inputting the optical density variation related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into a trained multimodal neural network model for processing to generate a dynamic estimation value of intracranial pressure includes: Dynamically weighting the optical density variation related to the cerebrospinal fluid layer thickness and the intracranial microcirculation parameters through a cross-modal attention mechanism to generate a spatiotemporal enhancement feature vector; Performing time series analysis on the spatiotemporal enhancement feature vector based on a long short-term memory network to capture the dynamic evolution of intracranial pressure and generate dynamic pressure features; A multimodal training dataset was constructed based on animal experimental data and clinical invasive detection data. The sample diversity was enhanced through the adversarial generative network. The initial weights of the model were optimized by combining the transfer learning algorithm to generate a trained multimodal neural network model. The dynamic pressure feature is input into the fully connected layer of the trained multimodal neural network model for nonlinear transformation to generate the dynamic estimation value of intracranial pressure.

3. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 2 is characterized in that: The time series analysis of the spatiotemporal enhancement feature vector based on the long short-term memory network is performed to capture the dynamic evolution law of intracranial pressure and generate dynamic pressure characteristics, including: Extracting local temporal correlation of the spatiotemporal enhancement feature vector based on a gated recurrent unit to generate preliminary temporal features; The preliminary time series features are weightedly fused with the original spatiotemporal enhancement feature vector through cross-layer residual connections to generate enhanced time series features; Based on the phase synchronization characteristics of intracranial pressure, multi-scale time window attention weight allocation is performed on the enhanced time series features to capture the cross-cycle correlation of sudden pressure fluctuations and generate context-aware time series features; The context-aware time series features are tensor-concatenated with the cerebrospinal fluid flow direction parameters, and dimension compression is performed through a bottleneck layer to generate the dynamic pressure features.

4. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 1, characterized in that: The method performs real-time correction processing on the dynamic estimated intracranial pressure value by using a preset ambient light transient suppression algorithm to generate an anti-interference intracranial pressure value, and outputs a spatially weighted intracranial pressure curve and a multi-level warning signal based on a trained cerebrospinal fluid flow distribution model, including: Performing transient interference event identification on the dynamic estimation value of intracranial pressure by using the ambient light transient suppression algorithm to generate an interference marker signal, and dynamically adjusting the cutoff frequency of the sliding window filter based on the interference marker signal to generate the anti-interference intracranial pressure value; Based on the trained cerebrospinal fluid flow distribution model, the anti-interference intracranial pressure value is spatially mapped to multiple brain regions, and a cerebrospinal fluid pressure spatial distribution heat map is generated in combination with cerebral hemodynamic parameters; The pressure levels of each area in the cerebrospinal fluid pressure spatial distribution heat map are threshold matched and weighted, and when the pressure level exceeds a preset pressure threshold and meets the spatial diffusion pattern of brain herniation, the multi-level warning signal corresponding to the pressure level is triggered.

5. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 4 is characterized in that: The multi-brain region spatial mapping of the anti-interference intracranial pressure value based on the trained cerebrospinal fluid flow distribution model and the generation of a cerebrospinal fluid pressure spatial distribution heat map in combination with cerebral hemodynamic parameters include: The anatomical structure of the monitoring area is segmented based on the standard brain atlas to generate the spatial coordinate mapping relationship of multiple brain regions; Based on the multi-brain region spatial coordinate mapping relationship, the anti-interference intracranial pressure value and cerebral hemodynamic parameters are input into the cerebrospinal fluid flow distribution model, and the heterogeneity of physiological parameters is eliminated through a cross-modal feature alignment algorithm to generate a fused pressure-blood flow feature vector; Based on the fused pressure-blood flow feature vector, the pressure gradient distribution of each brain region is calculated through the fluid mechanics simulation layer of the cerebrospinal fluid flow distribution model, and a continuous cerebrospinal fluid pressure spatial distribution heat map is generated in combination with a spatial interpolation algorithm.

6. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 1, characterized in that: The method of performing dynamic baseline drift correction processing on the optical density signal of the shallow tissue to generate a dynamic calibration signal, and performing spatiotemporal synchronous acquisition processing on the optical density signal of the deep tissue by a dual optical path sensor to generate a dynamic optical density change of the cerebrospinal fluid layer includes: Extracting low-frequency baseline drift features of the optical density signal of the shallow tissue based on a sliding time window statistical analysis method to generate a drift compensation coefficient, and dynamically compensating the original optical density signal through an adaptive filter to generate the dynamic calibration signal; Using the dynamic calibration signal as a shallow interference reference, based on the fixed spacing of the dual optical path sensor and the geometric relationship between the optical transmission path, the optical density signal of the deep tissue is calibrated for optical path delay and subtracted for shallow interference to generate a synchronous calibration optical density signal; The synchronous calibration optical density signal is time-stamp aligned and spatially registered, and combined with the stability parameter of the dynamic calibration signal, the head micro-motion interference is stripped off through a motion artifact suppression algorithm to generate the dynamic optical density change of the cerebrospinal fluid layer.

7. The non-invasive detection method of intracranial pressure based on infrared optical density analysis according to claim 1, characterized in that: The method comprises: performing shallow interference subtraction on the dynamic optical density variation of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronous calibration optical density signal; performing motion artifact suppression processing on the synchronous calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and stripping the cross-absorption interference of melanin and hemoglobin through a nonlinear absorption model to generate an optical density variation related to the thickness of the cerebrospinal fluid layer, including: Based on the time domain fluctuation characteristics of the cerebral blood flow velocity monitoring signal, an adaptive threshold segmentation algorithm is used to identify the artifact interference interval caused by head movement, and a motion artifact marking signal is generated; According to the optical density stability index of the motion artifact marker signal and the dynamic calibration signal, dynamically adjusting the cutoff frequency and attenuation slope of the time-frequency domain joint filter to generate the anti-interference optical density feature; The anti-interference optical density characteristics are input into the nonlinear absorption model, and the optical density variation related to the thickness of the cerebrospinal fluid layer is generated by extracting the spectrum characteristics of the melanin-dominated absorption area and calculating the residual compensation of the hemoglobin absorption spectrum.

8. A non-invasive intracranial pressure detection system based on infrared optical density analysis, characterized in that: The system comprises: The light source modulation and optical path separation module is used to transmit multiple groups of near-infrared light signals to the head monitoring area through the adaptive wavelength switching module, and separate the optical density signals of shallow tissue and deep tissue based on the optical path adaptive compensation mechanism; A baseline correction and signal synchronization module is used to perform dynamic baseline drift correction processing on the optical density signal of the shallow tissue to generate a dynamic calibration signal, and to perform spatiotemporal synchronous acquisition processing on the optical density signal of the deep tissue through a dual optical path sensor to generate a dynamic optical density change of the cerebrospinal fluid layer; An artifact suppression and absorption stripping module is used to perform shallow interference subtraction on the dynamic optical density variation of the cerebrospinal fluid layer based on the dynamic calibration signal to generate a synchronous calibration optical density signal; perform motion artifact suppression processing on the synchronous calibration optical density signal based on the cerebral blood flow velocity monitoring signal to generate an anti-interference optical density feature, and strip the cross-absorption interference of melanin and hemoglobin through a nonlinear absorption model to generate an optical density variation related to the thickness of the cerebrospinal fluid layer; A multimodal neural network processing module, used for inputting the optical density variation related to the thickness of the cerebrospinal fluid layer and the intracranial microcirculation parameters into a trained multimodal neural network model for processing to generate a dynamic estimation value of intracranial pressure; The transient suppression and flow warning module is used to perform real-time correction processing on the dynamic estimation value of intracranial pressure through a preset ambient light transient suppression algorithm, generate an interference-resistant intracranial pressure value, and output a spatially weighted intracranial pressure curve and a multi-level warning signal based on a trained cerebrospinal fluid flow distribution model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the non-invasive intracranial pressure detection method based on infrared optical density analysis described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the non-invasive intracranial pressure detection method based on infrared optical density analysis described in any one of claims 1 to 7 are implemented.

Images