Real-time brain function evaluation intelligent nerve monitoring system for critical patient

Abstract

The invention provides a real-time brain function evaluation intelligent nerve monitoring system for critical patients, which efficiently monitors the brain function state and performs intelligent early warning by means of sound field accurate control, differential acquisition anti-interference, multi-dimensional feature fusion analysis and graded early warning optimization. The system adopts a double-layer piezoelectric transduction array structure, and is matched with self-adaptive sound field regulation and control and a multi-stage anti-interference strategy, so that the signal acquisition quality is improved; a brain functional region quantitative evaluation index is constructed through multi-dimensional feature extraction and fusion analysis, and a real-time risk early warning and dynamic optimization mechanism is provided based on a hierarchical risk evaluation model; according to the invention, intelligent and reliable brain function monitoring can be realized in a complex environment, and accurate and real-time decision support is provided for clinical management of critical patients.

Description

Technical Field

[0001] The present invention belongs to the field of medical health technology, and specifically relates to an intelligent neural monitoring system for real-time brain function evaluation of critically ill patients. Background Art

[0002] With the development of neurocritical care medicine, the assessment of brainstem function in critically ill patients has received increasing attention. The brainstem is an important center for maintaining life, and contains key functional areas such as consciousness, breathing, and circulation. In critically ill neurological patients, accurate assessment of brainstem function is of great significance for determining the severity of the disease, predicting prognosis, and guiding treatment. Currently, the commonly used clinical assessment methods mainly include:

[0003] A. Clinical scoring and physical examination. The most widely used ones are the Glasgow Coma Scale (GCS) and brainstem reflex examination.

[0004] (1) The main problems of GCS scoring are as follows:

[0005] The scoring process is highly subjective, with poor consistency between examiners; the assessment of sedated or intubated patients is limited; continuous monitoring and dynamic assessment are not possible; the relationship between the scoring results and anatomical location is unclear;

[0006] (2) Limitations of brainstem reflex examination:

[0007] It needs to be completed manually by professionals and cannot be automated; the inspection results rely on the operator's experience; the evaluation results cannot be quantified; and it is not suitable for frequent repeated inspections.

[0008] B. Electrophysiological examination methods

[0009] (1) Brainstem auditory evoked potential (BAEP):

[0010] It can only evaluate the auditory conduction pathway; the examination process is easily affected by external interference; the equipment is complex to operate and requires professional technicians; it is not suitable for continuous bedside monitoring.

[0011] (2) Electroencephalogram (EEG) monitoring:

[0012] Signal acquisition is easily affected by electromyography, environmental electromagnetic interference, etc.; it is sensitive to specific treatment measures for critically ill patients (such as ECMO, CRRT, etc.); data analysis requires interpretation by professionals; and the correlation with brainstem function is not direct enough.

[0013] C. Imaging examination methods

[0014] Although CT and MRI can provide information on brainstem morphology, they have obvious shortcomings:

[0015] It can only provide static anatomical information; it cannot evaluate real-time functional status; it is not suitable for frequent repeated examinations; it requires the transfer of patients, which poses a safety hazard;

[0016] D. Invasive monitoring methods;

[0017] Such as intracranial pressure monitoring: It requires surgical implantation, which carries the risk of trauma and infection; It can only reflect changes in intracranial pressure; The monitoring indicators are single and the functional evaluation is of limited significance; It is difficult to implement in some situations (such as abnormal coagulation function).

[0018] In summary, the prior art generally has the following problems:

[0019] A. The evaluation method is highly subjective and lacks objective quantitative indicators;

[0020] B. It is difficult to achieve continuous real-time monitoring;

[0021] C. Unable to accurately locate the function;

[0022] D. Limited assessment of sedated patients;

[0023] E. Lack of early warning mechanism;

[0024] F. The evaluation results are not sufficiently correlated with clinical significance.

[0025] These problems seriously restrict the precise assessment and treatment of critically ill neurological patients, and there is an urgent need to develop new assessment methods and monitoring means. Summary of the invention

[0026] The present invention provides an intelligent neural monitoring system for real-time brain function assessment of critically ill patients, comprising:

[0027] The sound wave induction unit adopts a double-layer piezoelectric transducer array structure and realizes precise control of the sound field through the coordinated work of the main array and the auxiliary array. The unit selectively stimulates different functional areas of the brainstem through sound field modulation technology, can monitor the sound field in real time, and integrates an adaptive sound field control algorithm, which can dynamically adjust the sound wave parameters according to feedback data. In order to ensure the safety of stimulation, energy density limitation and frequency modulation limitation mechanisms are designed to realize full-process monitoring of sound wave output.

[0028] The oscillation detection unit adopts a differential acquisition structure and effectively solves the complex interference problems in critical care environments through a three-level processing method of hardware-level differential amplification, adaptive filtering, and interference identification. The unit has a signal enhancement processing function and extracts weak neural oscillation characteristics through time-frequency analysis and nonlinear enhancement algorithms. Data preprocessing uses signal standardization and segmentation processing to ensure data quality.

[0029] The functional evaluation unit realizes the quantitative evaluation of different functional areas of the brainstem; this unit includes time-frequency feature extraction and nonlinear feature extraction, and establishes specific evaluation indicators for the midbrain, pons and medulla oblongata functional areas respectively; a comprehensive evaluation of the functional status is achieved through a multidimensional feature fusion method, and a signal quality control and anomaly detection mechanism is provided.

[0030] The early warning diagnosis unit adopts a multi-level risk assessment model, realizes risk identification through two mechanisms: immediate warning and trend warning, and has a risk classification and response strategy matching mechanism. The unit has false alarm control and early warning optimization functions, and can adaptively adjust the early warning strategy according to the system operation status.

[0031] The control display unit receives and processes the system's control instructions, uniformly dispatches various modules, and realizes basic system status monitoring and instruction distribution functions; displays system status, important data information and alarm information in real time, and provides users with a concise visual interface; provides standardized external interfaces, and supports data interaction with other medical systems.

[0032] The present invention is beneficial in that:

[0033] 1. It realizes accurate positioning and evaluation of brainstem function: through the double-layer piezoelectric transducer array structure and sound field modulation technology, it can selectively act on different functional areas of the brainstem, breaking through the limitation of traditional evaluation methods that cannot locate functions, and providing anatomical positioning basis for clinical diagnosis.

[0034] 2. Solved the evaluation difficulties in the critical care environment: The three-level anti-interference processing system was adopted to effectively overcome the complex interference problems in the critical care environment; at the same time, the non-invasive sound wave induction method was used to avoid the application limitations of traditional evaluation methods under special treatment conditions.

[0035] 3. An objective and quantitative evaluation system was established: quantitative evaluation indicators of brainstem function were established through multidimensional feature analysis and nonlinear feature extraction; a multi-parameter fusion risk assessment model was used to achieve standardization and objectivity of the evaluation results.

[0036] 4. Provides a reliable early warning mechanism: Based on a multi-level risk assessment model, combined with immediate warning and trend warning mechanisms, it can detect brainstem dysfunction early; through adaptive early warning optimization, the accuracy and reliability of the early warning are improved.

[0037] 5. Intelligent management of the system is realized: the hierarchical control architecture is adopted to realize efficient scheduling and management of the system; through standardized interface design, it has good scalability and interoperability, which is convenient for clinical application and system maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1A schematic diagram of the architecture of the system of the present invention is shown. DETAILED DESCRIPTION

[0039] Combination Figure 1 The present invention provides an intelligent neural monitoring system for real-time brain function assessment of critically ill patients, comprising:

[0040] The sound wave induction unit adopts a double-layer piezoelectric transducer array structure and realizes precise control of the sound field through the coordinated work of the main array and the auxiliary array. The unit selectively stimulates different functional areas of the brainstem through sound field modulation technology, can monitor the sound field in real time, and integrates an adaptive sound field control algorithm, which can dynamically adjust the sound wave parameters according to feedback data. In order to ensure the safety of stimulation, energy density limitation and frequency modulation limitation mechanisms are designed to realize full-process monitoring of sound wave output.

[0041] The oscillation detection unit adopts a differential acquisition structure and effectively solves the complex interference problems in critical care environments through a three-level processing method of hardware-level differential amplification, adaptive filtering, and interference identification. The unit has a signal enhancement processing function and extracts weak neural oscillation characteristics through time-frequency analysis and nonlinear enhancement algorithms. Data preprocessing uses signal standardization and segmentation processing to ensure data quality.

[0042] The functional evaluation unit realizes the quantitative evaluation of different functional areas of the brainstem; this unit includes time-frequency feature extraction and nonlinear feature extraction, and establishes specific evaluation indicators for the midbrain, pons and medulla oblongata functional areas respectively; a comprehensive evaluation of the functional status is achieved through a multidimensional feature fusion method, and a signal quality control and anomaly detection mechanism is provided.

[0043] The early warning diagnosis unit adopts a multi-level risk assessment model, realizes risk identification through two mechanisms: immediate warning and trend warning, and has a risk classification and response strategy matching mechanism. The unit has false alarm control and early warning optimization functions, and can adaptively adjust the early warning strategy according to the system operation status.

[0044] The control display unit receives and processes the system's control instructions, uniformly dispatches various modules, and realizes basic system status monitoring and instruction distribution functions; displays system status, important data information and alarm information in real time, and provides users with a concise visual interface; provides standardized external interfaces, and supports data interaction with other medical systems.

[0045] Furthermore, the acoustic wave induction unit is characterized by adopting a double-layer piezoelectric transducer array structure, including an 8×8 main array and a 4×4 auxiliary array; the main array is responsible for the output of the main acoustic wave energy, while the auxiliary array is used for fine adjustment of the sound field. The sound pressure distribution at the spatial point P (x, y, z) is described by the improved sound field superposition model p (x, y, z, t):

[0046] p(x, y, z, t) = ∑[A i ×D(θ i )×M(t)×sin(ωt-kr i +φ i )]×α(r i )×β(f)

[0047] in:

[0048] A i is the vibration amplitude of the i-th transducer element (Pa), ranging from 0 to 100 Pa;

[0049] ωt is the phase angle of the sound wave, ω is the angular frequency (rad / s), and t is the time (s);

[0050] k is the wave number, k = 2π / λ, where λ is the wavelength of the sound wave (m);

[0051] r i is the distance from the i-th transducer element to the target point (m);

[0052] φ i is the initial phase compensation value (rad);

[0053] α(r i ) is the spatial attenuation function, and its expression is Where η is the attenuation coefficient;

[0054] β(f) is the frequency response correction factor, the expression is β(f) = 1-((f-f0) / Δf) 2 , where f is the frequency, f0 is the center frequency, and Δf is the bandwidth;

[0055] M(t) is the modulation function, M(t)1+m×sin(ω m ×t+ψ), where m is the modulation depth, ranging from 0 to 0.5; ω m is the modulation angular frequency; ψ is the modulation phase;

[0056] D(θ i ) is a directional function, and an improved multi-order Bessel model is used to adaptively control the sound field to achieve precise stimulation of specific areas of the brainstem. It is defined as follows:

[0057] D(θ i )=J0(ka×sinθ i )×(1+γ×J1(ka×sinθ i ))

[0058] in:

[0059] J0 and J1 are zero-order and first-order Bessel functions;

[0060] ka is the product of wave number and array element spacing;

[0061] γ is the directional correction coefficient, ranging from 0.2 to 0.8;

[0062] θ i is the angle between the sound wave propagation direction of the i-th transducer element and the target direction (rad);

[0063] In order to achieve accurate focusing of the sound waves, the present invention also adopts a dual-loop feedback structure to automatically

[0064] Adaptive phase compensation, the phase compensation value of each transducer element φ i The basic compensation item and dynamic

[0065] The amendment consists of:

[0066] φ i = k(r i -r min )+φ s (t)+φ d (t)

[0067] in:

[0068] r min Minimum distance from all transducer elements to the target point (m);

[0069] μ adaptive step size factor, ranging from 0.01-0.1;

[0070] φ s (t) is the static compensation term, φ s (t) = φ a +φ b +φ c , acoustic medium compensation term φ a = k × ∫ (n (r) - 1) dr, k is the wave number, n (r) is the sound velocity distribution function, which represents the position-dependent sound velocity of the sound wave at each point in the path; the interface compensation term φ b =∑(θ i ×Δn i ), θ i is the incident angle of the i-th interface, indicating the incident angle of the sound wave on the interface, Δn i is the refractive index difference between the media on both sides of the interface; the structural compensation term φ c =2πd / λ, d is the optical path difference caused by the transducer structure, and λ is the wavelength of the sound wave.

[0071] φ d (t) is the dynamic compensation term, which follows the following differential equation:

[0072]

[0073] Where:

[0074] α is the proportionality coefficient, ranging from 0.1 to 1.0;

[0075] β is the differential coefficient, ranging from 0.01 to 0.1;

[0076] is the acoustic field energy focusing error gradient;

[0077] E(t) is the sound field energy focusing error, which represents the difference between the actual sound pressure at the target point and the expected sound pressure. E(t) = ∫∫∫|p(x, y, z, t)-p0(x, y, z)| 2 dV, where p0(x, y, z) is the target sound field distribution and dV is the spatial volume element.

[0078] To ensure the safety of stimulation, this unit is designed with multiple protection mechanisms:

[0079] A). Energy density limitation:

[0080]

[0081] in:

[0082] P max is the maximum allowable sound pressure, set to 100Pa;

[0083] r is the position of any point in the current sound wave propagation path;

[0084] r0 is the target point position;

[0085] λ is the attenuation characteristic length.

[0086] B). Frequency modulation limitations:

[0087] in

[0088] f0 is the center frequency;

[0089] κ is the rate-of-change limiting coefficient, with a typical value of 0.1.

[0090] Furthermore, the oscillation detection unit is characterized by adopting a differential acquisition structure, including a main acquisition channel and a reference channel, to effectively extract weak neural oscillation characteristics, and the original signal model x(t) is expressed as:

[0091] x(t)=s(t)+n1(t)+n2(t)+i(t)

[0092] in:

[0093] s(t) is the target neural oscillation signal;

[0094] n1(t) is the physiological background noise;

[0095] n2(t) is the interference of critical equipment;

[0096] i(t) is the external interference noise.

[0097] In view of the characteristics of critical care environment, a three-level anti-interference mechanism is designed, including:

[0098] A). Hardware suppression:

[0099] The differential signal processing formula is:

[0100] y1(t)=A×(x1(t)-x2(t))×H(t)

[0101] in:

[0102] A is the differential amplification factor;

[0103] H(t) is the hardware filter function, which adopts the cascade structure of high-pass + low-pass + notch, with 0.5Hz high-pass, 100Hz low-pass, and 50Hz notch;

[0104] x1(t) and x2(t) are differential input signals.

[0105] B). Adaptive filtering:

[0106] Using the improved LMS algorithm:

[0107] y2(t)=y1(t)-Σω(k,t)×r(tk)

[0108] Weight update equation:

[0109] ω(k,t+1)=ω(k,t)+μ(t)×e(t)×r(tk)

[0110] in:

[0111] ω(k, t) is the filter weight;

[0112] r(t) is the reference signal;

[0113] μ(t) is the adaptive step size;

[0114] e(t) is the error signal.

[0115] C). Pattern recognition:

[0116] Interference discrimination function based on pattern recognition:

[0117] D(t)=F(y2(t),θ(t))

[0118] in:

[0119] F is the discriminant function;

[0120] Θ(t) is the feature template set.

[0121] To enhance the signal, an improved wavelet transform is used for time-frequency analysis:

[0122] W(a,b)=∫y2(t)×ω(t)×ψ((tb) / a)dt

[0123] in:

[0124] The signal preprocessing step includes an improved wavelet threshold filtering algorithm to remove noise. First, the signal is decomposed by wavelet:

[0125] X(a,b)=∫x(t)×ψ((tb) / a)dt

[0126] in:

[0127] a is the scale parameter, corresponding to the frequency component;

[0128] b is the translation parameter, corresponding to the time position;

[0129] ψ(t) is the wavelet basis function, which serves as the basic waveform of the decomposed signal.

[0130] After wavelet decomposition, the adaptive threshold method is used for threshold denoising. The threshold formula is:

[0131]

[0132] in:

[0133] j is the decomposition scale, indicating the current number of decomposition levels;

[0134] σ j is the estimated value of the standard deviation of the noise in the jth layer, which is used to determine the size of the noise;

[0135] N is the signal length, which indicates the total number of sampling points of the signal;

[0136] α is the adjustment coefficient (ranging from 0.1 to 0.5), which is used to dynamically adjust the threshold.

[0137] Then the signal is enhanced, and the nonlinear enhancement function is:

[0138] S(t)=sign(W(t))×|W(t)| α ×e -β×|W(t)| )

[0139] in:

[0140] sign(W(t)) is the sign function, indicating the positive or negative nature of W(t);

[0141] W(t) is the characteristic value of the input signal;

[0142] α is the enhancement coefficient, which controls the strength of enhancement, and the typical range is 1.2 to 2.0;

[0143] β is the suppression coefficient, which controls the degree of signal suppression after enhancement, and its typical range is 0.1 to 0.5. Finally, the data is standardized and segmented, and the normalization uses amplitude normalization:

[0144] Z(t)=(S(t)-μ s ) / σ s

[0145] in:

[0146] z(t) is the normalized signal;

[0147] μ s is the local mean;

[0148] σ s is the local standard deviation;

[0149] S(t) is the enhanced signal value;

[0150] By normalizing the amplitude, the signal values ​​can be kept at a consistent scale under different measurement conditions, which is helpful for subsequent data segmentation and analysis.

[0151] In order to improve the accuracy of signal analysis and the ability to resist noise, the signal Z(t) is divided into multiple time segments by segmented processing. Each segment is analyzed independently to better capture the instantaneous changes of the signal. The length of each segment is 1024 data points, the overlap rate between segments is 50%, and the windowing function is a cosine window. After processing, the signal is used as the input data of the function evaluation unit for subsequent feature extraction and state evaluation.

[0152] Furthermore, the function evaluation unit is characterized by realizing accurate evaluation of different functional areas of the brainstem based on a multidimensional feature analysis method, specifically including:

[0153] A). Time-frequency feature extraction

[0154] The improved Hilbert-Huang transform (HHT) is used to capture the instantaneous characteristics of the signal. The formula is as follows:

[0155] H(t,ω)=∫Z(τ)×g(t-τ)×e (-jωτ) dτwhere:

[0156] H(t, ω) is the transformation result, which represents the amplitude and phase information of the signal Z(τ) at time t and frequency ω;

[0157] τ is the time-integrated variable;

[0158] Z(τ) is the signal data processed by the oscillation detection unit;

[0159] g(t) is an adaptive window function used to adjust the time resolution of signal analysis. The calculation method is: Where, T is the length of the window function, which controls the resolution of the signal at time t; σ is the shape parameter of the window function, which adjusts the width of the window function and affects the frequency resolution;

[0160] e (-jωτ) is a complex exponential term used to convert the signal into the frequency domain, j is an imaginary unit, representing the amplitude and phase information of the signal at the frequency ω;

[0161] The instantaneous frequency is calculated based on the transformation result, and the formula is:

[0162]

[0163] in:

[0164] arg(H(t, ω)) is the phase angle of the signal, and arg represents the phase angle of the complex number, that is, the angle part of the transformation result H(t, ω).

[0165] These time-frequency characteristics directly reflect the dynamic properties of neural oscillations and provide a basis for subsequent functional area analysis.

[0166] B). Nonlinear feature extraction

[0167] Calculate the complexity of a signal using approximate entropy:

[0168]

[0169] in:

[0170] m is the embedding dimension, with a value of [2,4], which controls the length of the signal segment;

[0171] r is the similarity threshold, which is 0.1 to 0.3 times the standard deviation and measures the similarity between signal segments;

[0172] N is the data length;

[0173] is the matching counting function, which is used to calculate the frequency of different modes in the signal and is defined as φ(m, r) = (∑ln(C i (m, r))) / N, where C i(m, r) = number of matches / (N-m+1), and the matching basis is d(X m (i), X m (j))≤r, d is the maximum distance function.

[0174] C) Functional area analysis

[0175] The oscillation synchronization index SI(t) was defined to evaluate midbrain function, and the formula is:

[0176]

[0177] in:

[0178] φ k (t) is the phase value of the kth channel;

[0179] N is the number of channels;

[0180] R(t) is the response intensity modulation factor, which is used to adjust according to the signal strength:

[0181] R(t)=1-e (-λ×|H(t，ω)| )

[0182] Where λ is the modulation coefficient, which regulates the effect of response intensity on the synchronization index.

[0183] The function of the pons region is evaluated by the time response characteristic vector TR(t), and the formula is:

[0184] TR(t)=[τd ( t), τ r (t), τ f (t)]

[0185] in:

[0186] τ d (t) is the response latency, the time from the onset of the stimulus to the time when the response exceeds 2 standard deviations of the baseline;

[0187] τ r (t) is the rise time, the time from the start of the response to the peak value;

[0188] τ f (t) is the recovery time, which is the time from the peak value to within 1 standard deviation of the baseline.

[0189] Then perform feature fusion processing:

[0190] F(t)=ω1×τ d (t)+ω2×τ r (t)+ω3×τ f (t)

[0191] in:

[0192] ω1, ω2, and ω3 are weight coefficients, and their sum is 1.

[0193] The activity index is defined for the assessment of bulbar function using the formula:

[0194]

[0195] in:

[0196] t and t0 are the current time point and the reference time point (the stimulus start time), respectively;

[0197] T is the characteristic time constant, with a value of [0.5,2];

[0198] Z(t) is the signal data processed by the oscillation detection unit.

[0199] D) Status Assessment

[0200] Construct the feature matrix X(t):

[0201] X(t)=[SI(t),F(t),AI(t)]

[0202] State mapping function S(t):

[0203] S(t)=ψ(X(t))×C(t)

[0204] in:

[0205] ψ is a nonlinear mapping operator, Ψ(x) = 1 / (1+e -kx ), where k is the mapping coefficient;

[0206] C(t) is the compensation matrix, C(t) = I + α × dX(t) / dt, which is used to adjust the dynamic change of state over time. In the formula, α is the dynamic compensation coefficient, ranging from [0.1, 0.5].

[0207] Define the scoring function Score(t) for real-time and cumulative state evaluation:

[0208] Score(t)=β×S(t)+(1-β)×∫S(τ)dτ

[0209] in:

[0210] β is the instantaneous weight coefficient;

[0211] The integral term represents the cumulative effect and reflects the overall change in state.

[0212] The signal quality index Q(t) is defined to measure the quality of the signal:

[0213] Q(t)=SNR(t)×COH(t)×STA(t)

[0214] in:

[0215] SNR(t) is the signal-to-noise ratio, which represents the ratio of signal energy to noise energy. It is used to measure the clarity and reliability of the signal. A high signal-to-noise ratio means that the signal contains more effective information components and less noise components, and the signal quality is better.

[0216] COH(t) is a coherence index that evaluates the consistency of the signal in the time domain and frequency domain. It is mainly used to measure the correlation of different frequency components at different time points. A high coherence indicates that the signal is consistent in frequency components and is suitable for the evaluation of stable signals.

[0217] STA(t) is a stability index that reflects the volatility and overall stability of the signal. Low volatility means that the signal is relatively stable and has higher reliability. It is obtained by calculating the mean square error of the signal and is used to determine whether the signal has drastic changes.

[0218] Thus, the comprehensive signal quality index Q(t) is obtained, which is used to judge whether the signal performance in all aspects meets expectations. If Q(t) is greater than the predetermined threshold, it means that the signal is good in clarity, consistency and stability; otherwise, it may be necessary to process the signal or remove the abnormal part. The following formula is used to identify abnormal signals:

[0219] E(t)=(||S(t)-S′(t)||) / ||S′(t)||

[0220] in:

[0221] S(t) is the current actual signal state value, reflecting the characteristic data or state information of the signal at a certain moment;

[0222] S′(t) is the predicted signal state value, obtained using Kalman filtering, which is used to estimate the normal state of the signal at the current moment;

[0223] ||·|| is the Euclidean norm, which indicates the length of the vector;

[0224] ‖S(t)-S′(t)‖ is the Euclidean distance between S(t) and S′(t). The larger the value, the greater the deviation between the actual state and the predicted state, indicating that an abnormal situation may have occurred. By standardizing this distance (i.e., dividing it by ‖S′(t)‖), a relative abnormality measure can be obtained to identify signal points that deviate significantly from the normal state.

[0225] Furthermore, the early warning diagnosis unit is characterized by firstly conducting risk assessment on each functional indicator, then considering the time accumulation effect, and finally realizing hierarchical early warning through multi-dimensional analysis, specifically including:

[0226] A) Risk Assessment

[0227] The immediate risk value R(t) of brainstem function is calculated by integrating multiple functional indicators, and corrections for signal quality and change trends are added to improve the accuracy of the assessment. The formula is:

[0228] R(t)=∑(ω i ×f i (t)×K(t)

[0229] in:

[0230] f i (t) are functional indicators, including SI(t) midbrain functional synchronization index, which measures the synchronization level of midbrain functional areas, F(t) pons functional characteristic value, which indicates the response characteristics of the pons area, and AI(t) medullary activity index, which measures the activity intensity of the medullary area;

[0231] K(t) is a correction factor used to make quality and trend corrections to the risk value, and is defined as:

[0232] K(t)=(1+α×|dScore(t) / dt|×e (-β×Q(t)) )

[0233] In the formula, α is the trend influence coefficient, ranging from 0.2 to 0.5, which indicates the sensitivity of the risk value to the rate of change of the functional score. A higher α value indicates that the trend change has a greater impact on the risk value. β is the quality influence coefficient, ranging from 0.1 to 0.3, which is used to reduce the impact of poor signal quality on the evaluation results. Q(t) is the signal quality indicator, which comes from the functional evaluation unit and indicates the reliability of the current signal. Low Q(t) will increase the attenuation of the risk value and reduce the impact of low-quality signals on the evaluation results.

[0234] B) Risk accumulation effect

[0235] The risk accumulation effect takes into account the progressiveness of brainstem functional impairment and generates the time cumulative risk RA(t) by introducing the cumulative term of historical risk values:

[0236] RA(t)=R(t)+γ×∫R(τ)×e -((t-τ) / T) dτ

[0237] in:

[0238] γ is the cumulative weight, ranging from 0.3 to 0.6, which controls the impact of historical risk on the current risk value. A larger γ value indicates a greater impact of historical risk;

[0239] T is the characteristic decay time, which is set based on clinical experience and indicates the degree to which the impact of historical risk decays over time. A larger T means that historical risk will affect the current risk value for a longer period of time.

[0240] τ is a historical time point, which is used to calculate the cumulative effect of past risk values ​​on the current time.

[0241] C). Early warning strategy

[0242] Whether to trigger an early warning is determined based on the immediate risk and cumulative risk trends. The immediate early warning judgment formula is:

[0243] W1(t)=H(RA(t)-θ1)×D(dRA(t) / dt)

[0244] in:

[0245] H(RA(t)-θ1) is the threshold judgment function, which is used to judge whether the current cumulative risk exceeds the warning threshold θ1. When it exceeds the warning threshold, the warning is triggered;

[0246] D(dRA(t) / dt) is the change rate judgment function, which is used to adjust the warning intensity according to the risk change trend. When dRA(t) / dt>0, the risk is on the rise, and the warning intensity is increased;

[0247] The trend warning analysis uses the following formula:

[0248] W2(t)=P(RA(t+Δt))-RA(t) where:

[0249] P is the prediction function, which predicts the future risk value RA(t+Δt) based on historical data;

[0250] Δt is the prediction time window, which is set to 5 to 10 minutes and represents the prediction range of future risks.

[0251] D) Risk grading

[0252] Risk grading L(t) determines the current risk level based on multi-dimensional analysis results:

[0253] L(t)=[RA(t), W1(t), W2(t)]

[0254] When RA(t)≥θ H Or W2(t)≥δ H , L(t) is level 1 (high risk);

[0255] When θ M ≤RA(t)<θ H And δ L ≤W2(t)<δ H , L(t) is level 2 (medium risk);

[0256] When RA(t)<θ M And W2(t)<δ L , L(t) is level 3 (low risk);

[0257] in:

[0258] θ H ,θ M are the high and medium thresholds of the immediate risk value;

[0259] δ H ,δ L These are the high and low thresholds for trend warning.

[0260] E). Early warning optimization mechanism

[0261] The early warning optimization mechanism ensures the accuracy of the early warning and further improves the early warning performance through credibility evaluation and adaptive adjustment; the credibility C(t) calculation formula is:

[0262]

[0263] in:

[0264] C(t) is the warning credibility, and the closer the value is to 1, the more credible the warning is;

[0265] λ is the sensitivity coefficient, which controls the impact of abnormal indicators on credibility;

[0266] E(t) is the abnormality index, which comes from the function evaluation unit and is used to determine the abnormality of the signal;

[0267] The warning threshold is dynamically updated according to the performance gradient:

[0268]

[0269] in:

[0270] θ(t) and θ(t+1) represent the warning thresholds at time t and time t+1 respectively;

[0271] μ is the learning rate, which controls the speed of threshold update;

[0272] It is the performance gradient, which indicates the change in the accuracy of the warning and is used to adjust the threshold to optimize the accuracy of the warning.

[0273] Finally, the warning information is transmitted through the control display unit and triggers the corresponding alarm mechanism:

[0274] Level 1 (high risk): triggers an audible and visual alarm, and immediate clinical intervention is recommended;

[0275] Level 2 (medium risk): Warning information, recommended to strengthen monitoring;

[0276] Level 3 (low risk): Keep records and maintain routine monitoring.

[0277] In summary, the present invention proposes an intelligent neural monitoring system for real-time brain function assessment of critically ill patients, which realizes efficient monitoring and intelligent early warning of brain function status through precise control of the sound field, differential acquisition anti-interference, multi-dimensional feature fusion analysis and hierarchical early warning optimization. The system adopts a double-layer piezoelectric transducer array structure, combined with adaptive sound field regulation and multi-level anti-interference strategy to effectively improve the signal acquisition quality; through multi-dimensional feature extraction and fusion analysis, quantitative evaluation indicators of brain functional areas are constructed, and based on the hierarchical risk assessment model, real-time risk warning and dynamic optimization mechanism are provided. The present invention can realize intelligent and reliable brain function monitoring in complex environments, and provide accurate and real-time decision support for the clinical management of critically ill patients.

[0278] In terms of the scope of rights protection, the present invention is not limited to the specific implementation methods described in the text. On the contrary, it should be clear that the scope of rights protection of the present invention covers all reasonable changes, modifications and equivalent implementation methods made based on the content of this document.

Claims

1. An intelligent neural monitoring system for real-time brain function assessment of critically ill patients, characterized in that: include: The sound wave induction unit adopts a double-layer piezoelectric transducer array structure, and realizes precise control of the sound field through the coordinated work of the main array and the auxiliary array; The oscillation detection unit adopts a differential acquisition structure to obtain neural oscillation signals; Functional assessment unit, which enables quantitative assessment of different functional areas of the brainstem; Early warning diagnosis unit, using a multi-level risk assessment model; The control and display unit is used for unified scheduling control, status monitoring and data information display of the system, and provides an external interface for data interaction with other medical systems.

2. The system according to claim 1, characterized in that: The sound wave inducing unit comprises: A main array and an auxiliary array, wherein the main array adopts an 8×8 layout to be responsible for the main output of sound wave energy, and the auxiliary array adopts a 4×4 layout to be used for fine adjustment of the sound field; and an improved sound field superposition model including a modulation function and a directivity function, the model is used in conjunction with an adaptive phase compensation mechanism of a dual-loop feedback structure, and is provided with multiple protection mechanisms including energy density limitation and frequency modulation limitation.

3. The system according to claim 1, characterized in that: The oscillation detection unit realizes signal acquisition through three-level anti-interference, which are hardware-level differential amplification, adaptive filtering and pattern recognition in sequence; adopts improved wavelet threshold filtering algorithm for noise reduction, extracts features through nonlinear signal enhancement processing, and ensures signal quality through standardized data preprocessing mechanism.

4. The system according to claim 1, characterized in that: The function evaluation unit adopts the improved Hilbert-Huang transform to extract time-frequency features, combines the nonlinear feature extraction method based on approximate entropy, establishes specific evaluation indicators for the midbrain, pons and medulla oblongata, and realizes a comprehensive evaluation of brainstem function through a state evaluation model with multi-dimensional feature fusion.

5. The system according to claim 1, characterized in that: The early warning diagnosis unit adopts a multi-parameter fusion risk assessment model, considers the time accumulation effect for risk analysis, realizes abnormal monitoring through the dual mechanisms of immediate warning and trend warning, and is provided with an early warning optimization mechanism based on credibility assessment to improve the accuracy of early warning.

Images