Non-invasive intracranial pressure continuous monitoring teaching device and fluctuation trend prediction method

Abstract

The invention relates to the technical field of medical teaching, and discloses a noninvasive intracranial pressure continuous monitoring teaching device and a fluctuation trend prediction method.The noninvasive intracranial pressure continuous monitoring teaching device comprises a protective shell and a closed-loop fluid system integrated in the protective shell, and the closed-loop fluid system is composed of a pump body, an intracranial simulation block, a first connecting pipe, a second connecting pipe, a third connecting pipe and a pressure regulating valve; the output end of the pump body is communicated with an input cavity of the intracranial simulation block through a first connecting pipe, the backflow end of the pump body forms a closed loop through a second connecting pipe, a pressure adjusting valve and a third connecting pipe, a detector used for collecting simulated intracranial pressure data in real time is arranged at the top end of the intracranial simulation block, and a display screen is arranged on the outer surface of the protective shell. A closed-loop fluid loop is constructed through a pump body, an intracranial simulation block and a pressure regulating valve regulated and controlled by PID, cerebrospinal fluid circulation is dynamically simulated, and the problem that traditional teaching lacks real pathological environment simulation and operation safety is effectively solved in combination with an embedded training lumbar puncture hole.

Description

Technical Field

[0001] The present invention relates to the field of medical teaching technology, and in particular to a non-invasive intracranial pressure continuous monitoring teaching device and a fluctuation trend prediction method. Background Art

[0002] In the field of medical education and technology, intracranial pressure monitoring and related clinical operation training are important aspects of neurosurgery education. With the development of non-invasive monitoring technology, how to simulate the dynamic changes of intracranial pressure and conduct safe operation training through simulation models has become a key direction for improving medical students' clinical skills.

[0003] After searching, the invention application with publication number CN117079514A discloses an intracranial blood bag drainage training simulation device and method. It provides an intracranial blood bag drainage training simulation device for conducting intracranial blood bag drainage simulation training, including: a simulated human brain containing a simulated blood bag; a liquid level detection sensor for detecting the liquid level in the simulated blood bag; a calculation unit for obtaining the changing trend of intracranial pressure; and a display unit for visually displaying the changing trend of intracranial pressure obtained by the calculation unit to the trainee. Another technical solution of the invention provides an intracranial blood bag drainage training simulation method. The invention facilitates medical staff to use a simulated drainage tube to insert into a simulated human brain and a simulated blood bag to simulate the blood bag drainage process, thereby facilitating medical staff to conduct intracranial blood bag drainage simulation training. Moreover, during the simulation training process, the liquid level of the intracranial blood bag can be quickly detected by the detection sensor, thereby facilitating medical staff to understand the current drainage situation during the drainage simulation training and improving the simulation training effect.

[0004] However, in the process of implementing the technical solution, the inventors of the present application discovered that the above-mentioned existing technology has the following technical problems: it is impossible to achieve accurate simulation of dynamic pathological intracranial pressure waveforms and safe closed-loop training of lumbar puncture operations in teaching scenarios. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a non-invasive intracranial pressure continuous monitoring teaching device and a fluctuation trend prediction method, which solves the problem that the existing technology cannot achieve accurate simulation of dynamic pathological intracranial pressure waveforms and safe closed-loop training of lumbar puncture operations in teaching scenarios.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a non-invasive intracranial pressure continuous monitoring teaching device, including a protective shell and a closed-loop fluid system integrated therein, the closed-loop fluid system consisting of a pump body, an intracranial simulation block, connecting pipe one, connecting pipe two, connecting pipe three and a pressure regulating valve, the output end of the pump body is connected to the input cavity of the intracranial simulation block through connecting pipe one, and its return end forms a closed loop through connecting pipe two, the pressure regulating valve and connecting pipe three, a detector is provided at the top of the intracranial simulation block for real-time collection of simulated intracranial pressure data, a display screen is provided on the outer surface of the protective shell, and an internal integrated control center processes data and drives the pressure regulating valve to dynamically adjust the pressure.

[0007] By adopting the above technical solution, a dynamic hydraulic cycle is formed using the pump body, intracranial simulation block and PID-controlled pressure regulating valve to accurately reproduce pathological waveforms such as C-wave and plateau wave. At the same time, the pressure time series data, optic nerve diameter change rate and cerebrospinal fluid release volume are integrated through the AI ​​risk assessment algorithm to establish a multi-parameter coupling model to predict the intracranial pressure range in the next 5 minutes. The training lumbar puncture hole embedded in the protective shell adopts an elastic silicone anti-reflux plate design. The channel is automatically closed within 0.3 seconds after the puncture needle is pulled out to prevent leakage. Combined with the flow sensor and the sound and light alarm mechanism, the display screen can visualize the intracranial pressure waveform, AI prediction curve and three-color risk level in real time, forming a teaching closed loop of pathological simulation-operation training-data monitoring-risk warning, which effectively solves the problem of traditional teaching lacking real pathological environment simulation and operational safety.

[0008] Preferably, a training lumbar puncture hole is fixedly embedded in the inner wall of the protective shell, and the puncture channel thereof passes through the protective shell and is connected with the fluid cavity of the intracranial simulation block, so as to simulate the clinical cerebrospinal fluid extraction operation.

[0009] Preferably, the training lumbar puncture hole includes a shell, the outside of the shell is fixedly connected to the inside of the detector through a protective shell, a flow sensor is provided inside the shell, a backflow prevention plate 1 is fixedly connected to the inside of one side of the shell, and a backflow prevention plate 2 is fixedly connected to the inside of the other side of the shell, the positions of the backflow prevention plate 1 and the backflow prevention plate 2 are adjacent, and the backflow prevention plate 1 and the backflow prevention plate 2 are both made of elastic silicone material, which is used to automatically close the channel after the puncture needle is pulled out to prevent leakage of simulated cerebrospinal fluid.

[0010] Preferably, the control center is integrated with a B-ultrasound image analysis module for real-time monitoring of the optic nerve diameter simulation unit associated with the intracranial simulation block, and inverting the intracranial pressure fluctuation trend through the diameter change.

[0011] Preferably, the control center is deployed with an AI risk assessment algorithm, and the input parameters of the AI ​​risk assessment algorithm include: The real-time cumulative amount of cerebrospinal fluid released is collected by training the flow sensor in the lumbar puncture hole; The simulated intracranial pressure data collected by the detector is calculated using the formula: ; The optic nerve diameter change rate output by the B-ultrasound image analysis module; The risk determination logic of the AI ​​risk assessment algorithm includes: when 、 or When the alarm is triggered, the sound and light alarm will be triggered; when and When the alarm is on, a secondary warning signal is output to the display screen.

[0012] Preferably, the AI ​​risk assessment algorithm fuses the pressure sensor time series data with the B-ultrasound optic nerve diameter data to establish a multi-parameter coupled intracranial pressure prediction model, and the multi-parameter coupled intracranial pressure prediction model satisfies:

[0013] in, Historical pressure time series data window; Predict the intracranial pressure value at time t+1; The instantaneous rate of change of optic nerve diameter; Adaptive weight coefficient, dynamically adjusted according to pathological type, , initial value 0; function : A three-layer fully connected neural network is used, and the hidden layer activation function is , the output layer is Sigmoid; Structure: Input layer: 128 neurons, hidden layer: double dense layer, output layer: linear activation function, output ; Weight adaptive mechanism: When When Dynamically increase the weight of operational factors.

[0014] Preferably, the pressure regulating valve is electrically connected to a PID controller in a control center, and is used to adjust the opening in real time according to the pressure value fed back by the detector, so as to simulate the dynamic fluctuation of intracranial pressure under pathological conditions.

[0015] Preferably, the display screen simultaneously and visually displays the real-time waveform of intracranial pressure, AI prediction trend curve and operation risk assessment level.

[0016] Preferably, a method for predicting fluctuation trends of a non-invasive intracranial pressure continuous monitoring teaching device comprises the following steps: S1: The PID controller in the control center drives the pressure regulating valve to adjust the pipeline resistance of the closed-loop fluid system in real time, generating simulated intracranial pressure waveforms with nine clinical pathological characteristics; S2. The detector collects real-time pressure time series data of the intracranial simulation block to form a 128-dimensional vector with a 128-second sliding window. The B-ultrasound image analysis module tracks the optic nerve diameter simulation unit, calculates the instantaneous rate of change of the diameter, and trains the flow sensor of the lumbar puncture hole to monitor the cumulative amount of cerebrospinal fluid released V in real time. S3 uses a three-layer fully connected neural network. The 128 neurons in the input layer receive pressure time series data and adaptive weight coefficient α. The hidden layer fuses pathological and operational features through a double dense layer. The weight α is dynamically initialized according to the pathological type. The output layer outputs the current intracranial pressure estimate. S4. Predict the trend of current intracranial pressure estimates based on real-time operational data and assess operational risks.

[0017] Preferably, the AI ​​risk assessment algorithm adopts an LSTM neural network, and the LSTM neural network implements fluctuation trend prediction according to the following configuration: Input sequence: dimensional matrix, the row vector is ; Network Architecture: LSTM Layer ;Dropout layer;Fully connected layer; The input sequence is first processed in sequence by two LSTM layers to capture temporal features. The two LSTM layers are connected in series and the hidden state of the previous layer is used as the input of the next layer. The processed features flow through the Dropout layer for random inactivation to suppress overfitting. The features are then integrated through the fully connected layer, and finally the future 5-minute forecast interval is generated by the dual-channel fully connected output layer, where nonlinear mapping is achieved between the fully connected layer and the output layer through a weight matrix.

[0018] Output layer: dual-channel fully connected layer, outputting the next 5-minute forecast interval: ,in, is the predicted mean, is the standard deviation (calculated from historical residuals), the upper limit of the interval When the patient is in the emergency room, the risk of decompensation of intracranial hypertension is determined and a teaching warning is triggered.

[0019] The present invention provides a non-invasive intracranial pressure continuous monitoring teaching device and a fluctuation trend prediction method. It has the following beneficial effects: 1. The present invention constructs a closed-loop fluid circuit through a pump body, an intracranial simulation block, and a PID-controlled pressure regulating valve to dynamically simulate cerebrospinal fluid circulation. The control center adjusts the pipeline resistance in real time. Combined with the embedded training lumbar puncture hole, zero leakage of simulated cerebrospinal fluid is achieved during the puncture operation, effectively solving the problem of traditional teaching lacking real pathological environment simulation and operational safety.

[0020] 2. The present invention uses a pressure sensor and a B-ultrasound imaging module to synchronously collect data on changes in intracranial pressure and optic nerve diameter. When the cerebrospinal fluid drainage volume exceeds 20 ml, or the pressure drops by more than 15 mmHg, or the optic nerve diameter changes by more than 8% per minute, an audible and visual alarm is immediately triggered. If the drainage volume exceeds 10 ml and the pressure drops by more than 10 mmHg, a secondary warning is displayed on the screen to help trainees identify operational risks in a timely manner.

[0021] 3. The display screen presents the pressure waveform, AI-predicted curve, and red, yellow, and green risk levels in real time: the left area marks the pathological waveform characteristics, the right area displays the predicted interval, and the bottom area dynamically indicates operational errors. The entire device forms a teaching cycle that combines pathological simulation, training data, monitoring, and risk warning. Students can practice repeatedly in a safe environment and make immediate corrections, improving their skill acquisition efficiency.

[0022] 4. The system of the present invention combines a three-layer neural network with an LSTM time series algorithm. When the drainage volume exceeds 15 ml, the model automatically increases the influence weight of the operational factors and ultimately outputs a pressure prediction range for the next 5 minutes. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a stereoscopic diagram of a non-invasive intracranial pressure continuous monitoring teaching device of the present invention; Figure 2 It is a schematic diagram of the local structure of the pump body of the present invention; Figure 3 It is a schematic diagram of the local structure of the detector of the present invention; Figure 4 It is a schematic cross-sectional view of the internal structure of the housing of the present invention; Figure 5 This is a schematic diagram of the operation of the B-ultrasound image analysis module of the present invention; Figure 6 This is the logic block diagram of the AI ​​risk assessment algorithm of the present invention.

[0024] Among them, 1. Protective shell; 2. Pump body; 3. Connecting pipe 1; 4. Intracranial simulation block; 5. Detector; 6. Connecting pipe 2; 7. Pressure regulating valve; 8. Connecting pipe 3; 9. Control center; 10. Display screen; 11. Training lumbar puncture hole; 111. Outer shell; 112. Anti-backflow plate 1; 113. Anti-backflow plate 2. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Please see the attached Figure 1 -Attached Figure 6 An embodiment of the present invention provides a non-invasive intracranial pressure continuous monitoring teaching device, including a protective shell 1 and a closed-loop fluid system integrated therein. The closed-loop fluid system is composed of a pump body 2, an intracranial simulation block 4, a connecting pipe 1 3, a connecting pipe 2 6, a connecting pipe 3 8 and a pressure regulating valve 7. The output end of the pump body 2 is connected to the input cavity of the intracranial simulation block 4 through the connecting pipe 1 3, and its return end forms a closed loop through the connecting pipe 2 6, the pressure regulating valve 7 and the connecting pipe 3 8. A detector 5 is provided at the top of the intracranial simulation block 4 for real-time collection of simulated intracranial pressure data. A display screen 10 is provided on the outer surface of the protective shell 1, and an internal integrated control center 9 processes data and drives the pressure regulating valve 7 to dynamically adjust the pressure.

[0027] Specifically, the simulated cerebrospinal fluid is driven by the pump body 2 to flow into the intracranial simulation block 4 through the connecting pipe 1 3, and then flows to the pressure regulating valve 7 through the connecting pipe 2 6, and finally flows back to the pump body 2 through the connecting pipe 3 8 in a closed loop, forming a dynamic hydraulic cycle. The PID controller of the control center 9 controls the pressure regulating valve 7, dynamically adjusts the pipeline resistance to simulate pathological intracranial pressure waveforms such as C-waves and plateau waves, and the detector 5 at the top of the intracranial simulation block 4 collects pressure data in real time. At the same time, the B-ultrasound image analysis module of the control center 9 monitors the changes in the diameter of the optic nerve. By correlating and inverting the pressure trend, the inner wall of the device is integrated with the training data. Lumbar puncture port 11, backflow plate 1 12, and backflow plate 2 113 automatically close the channel after the puncture needle is removed, enabling zero-leakage operation training. A flow sensor within the port simultaneously collects cerebrospinal fluid release. An AI risk assessment algorithm is deployed within the control center 9. This algorithm integrates three parameters: real-time pressure fluctuation (ΔP), cumulative cerebrospinal fluid release (V), and optic nerve diameter change rate (∂D / ∂t). It implements a graded early warning system to predict the intracranial pressure range (μ-1.96σ, μ+1.96σ) for the next five minutes. If the upper limit is >25 mmHg, the risk of intracranial hypertension decompensation is determined. Display screen 10 simultaneously visualizes the real-time intracranial pressure waveform, the AI ​​prediction curve, and the three-color risk level.

[0028] A training lumbar puncture hole 11 is fixedly embedded in the inner wall of the protective shell 1, and its puncture channel passes through the protective shell 1 and is connected to the fluid cavity of the intracranial simulation block 4, which is used to simulate the clinical cerebrospinal fluid extraction operation; the training lumbar puncture hole 11 includes an outer shell 111, the outside of the outer shell 111 is fixedly connected to the inside of the detector 5 through the protective shell 1, and a flow sensor is provided inside the outer shell 111. A backflow prevention plate 112 is fixedly connected to the inside of one side of the outer shell 111, and a backflow prevention plate 2 113 is fixedly connected to the inside of the other side of the outer shell 111. The positions of the backflow prevention plate 112 and the backflow prevention plate 2 113 are adjacent to each other. The backflow prevention plate 112 and the backflow prevention plate 2 113 are both made of elastic silicone material, which is used to automatically close the channel after the puncture needle is pulled out to prevent leakage of simulated cerebrospinal fluid.

[0029] Specifically, the puncture hole 11 is fixedly embedded in the inner wall of the protective shell 1, and its channel passes through the protective shell and is connected to the fluid cavity of the intracranial simulation block 4. A backflow prevention plate 1 112 and a backflow prevention plate 2 113 are arranged inside the shell 111. When the puncture needle is inserted, the two plates are simultaneously stretched open to form a fluid channel. After the needle is pulled out, the two plates automatically close within 0.3 seconds due to elastic recovery force, blocking the leakage of simulated cerebrospinal fluid. During the puncture process, the flow sensor in the hole collects the cumulative amount of cerebrospinal fluid released (V) in real time. When a dangerous operation is detected, the control center 9 immediately stops the pump body 2 and triggers an audible and visual alarm, and at the same time highlights the operation error point on the display screen 10.

[0030] The control center 9 integrates a B-ultrasound image analysis module for real-time monitoring of the optic nerve diameter simulation unit associated with the intracranial simulation block 4, and inverts the intracranial pressure fluctuation trend through diameter changes. The control center 9 deploys an AI risk assessment algorithm, and the input parameters of the AI ​​risk assessment algorithm include: The real-time cumulative amount of cerebrospinal fluid released is collected by training the flow sensor in the lumbar puncture hole 11; The simulated intracranial pressure data collected by detector 5 is calculated using the formula: ; The optic nerve diameter change rate output by the B-ultrasound image analysis module; The risk determination logic of the AI ​​risk assessment algorithm includes: when 、 or When the alarm is triggered, the sound and light alarm will be triggered; when and When the alarm is on, a secondary warning signal is output to the display screen 10.

[0031] Specifically, the B-ultrasound image analysis module integrated in the control center 9 monitors the optic nerve diameter simulation unit linked to the intracranial simulation block 4 in real time. Based on the inverse relationship between the optic nerve sheath diameter and the intracranial pressure, the intracranial pressure fluctuation trend is dynamically inverted through the instantaneous change rate of the optic nerve diameter. At the same time, the AI ​​risk assessment algorithm deployed by the system performs graded warnings based on multi-source real-time data: input parameters Cumulative volume of cerebrospinal fluid released (V): The total volume of simulated cerebrospinal fluid drainage during the puncture operation is collected by the flow sensor embedded in the training lumbar puncture hole 11; Intracranial pressure fluctuation difference (ΔP): the deviation between the current pressure and the baseline value is calculated in real time by the detector 5 (unit: mmHg); Optic nerve diameter change rate (∂D / ∂t): The diameter change rate output by the B-ultrasound module.

[0032] Risk determination and execution logic Level 1 sound and light alarm: When any of the following conditions is met: ΔP>20mmHg (sudden drop in intracranial pressure), V>15ml (excessive drainage), or ∂D / ∂t>0.5mm / min (rapid swelling of the optic nerve), the sound and light alarm will be triggered immediately and the operation of pump 2 will be stopped urgently. Level 2 screen warning: When ΔP>15mmHg and ∂D / ∂t>0.3mm / min (continuous increase in intracranial pressure accompanied by slow swelling of the optic nerve), a highlighted warning signal is output to the display screen 10 to clearly indicate the operation risk point.

[0033] The AI ​​risk assessment algorithm integrates pressure sensor time series data with B-ultrasound optic nerve diameter data to establish a multi-parameter coupled intracranial pressure prediction model. The multi-parameter coupled intracranial pressure prediction model meets the following requirements:

[0034] in, Historical pressure time series data window; Predict the intracranial pressure value at time t+1; The instantaneous rate of change of optic nerve diameter; Adaptive weight coefficient, dynamically adjusted according to pathological type, , initial value 0; function : A three-layer fully connected neural network is used, and the hidden layer activation function is , the output layer is Sigmoid; Structure: Input layer: 128 neurons, Hidden layer: double dense layer, Output layer: linear activation function, output output ; Weight adaptive mechanism: When When Dynamically increase the weight of operational factors.

[0035] Specifically, the multi-parameter coupling prediction model constructed by the AI ​​risk assessment algorithm is constructed through the function Dynamic intracranial pressure estimation is achieved, and its mathematical expression is: in: The historical pressure time series data window is formed by the detector (5) collecting 128 seconds of data at a sampling frequency of 1 time per second to form a 128-dimensional vector, covering the typical cycle of intracranial pressure fluctuations, and then input into the model after sliding average filtering and normalization processing; The instantaneous rate of change of the optic nerve sheath diameter (mm / min) is obtained by tracking the optic nerve diameter simulation unit of the intracranial simulation block 4 at a frame rate of 30 fps using the B-ultrasound image analysis module. The inter-frame difference algorithm combined with morphological filtering is used to eliminate tissue artifacts. The rate of change is nonlinearly related to the intracranial pressure, and this characteristic is encoded into the activation function of the hidden layer of the neural network. is the pathology-operation adaptive weight coefficient, with an initial value of 0 and a dynamic adjustment value: Pathology adaptation: preset weights based on simulated etiologies; Procedure intensification: The amount of cerebrospinal fluid released during lumbar puncture When Update the weights so that the weight of the operation factor in the feature fusion layer is increased to 1.8-2.5 times the base value. The neural network architecture uses a three-layer fully connected structure to achieve end-to-end prediction; input layer: 128 neurons directly receive Vector, synchronous access and α as independent feature channels; Hidden layer: The first dense layer (256 neurons, ReLU activation function) extracts the time domain features of the pressure sequence, and its negative value truncation mechanism is strengthened (∂D / ∂t)1.8 nonlinear effect; The second dense layer will α and optic nerve features are spliced ​​and fused with a weight ratio of 3:1. Pass when V>10ml Alpha increments enhance operational risk sensitivity; Output layer: A single neuron linear activation function directly outputs the predicted pressure value (unit: mmHg).

[0036] The pressure regulating valve 7 is electrically connected to the PID controller of the control center 9, and is used to adjust the opening in real time according to the pressure value feedback from the detector 5, simulating the dynamic fluctuation of intracranial pressure under pathological conditions; the display screen 10 synchronously and visually displays the real-time waveform of intracranial pressure, AI prediction trend curve and operation risk assessment level.

[0037] Specifically, the pressure regulating valve 7 dynamically simulates pathological intracranial pressure waveforms using a PID closed-loop control system. A detector 5 collects real-time pressure data from the intracranial simulation block 4 at a 10Hz frequency and feeds it back to the control center 9. The PID controller dynamically adjusts the pressure regulating valve 7 to alter the pipeline resistance, reproducing the characteristic clinical waveform. For example, when generating a C-type wave, the PID adjusts the valve opening based on the error integral term, causing the pressure to oscillate upward at a slope of 0.5 mmHg / s. When simulating a plateau wave, the proportional term dominates the valve opening, causing it to drop sharply to 15%, triggering a sudden increase in pressure. The display screen 10 simultaneously implements three-dimensional teaching visualization: the left panel dynamically plots the real-time pressure waveform (including the pathological characteristic markers generated by the PID), the right panel overlays the LSTM prediction curve, and a three-color operational risk indicator is embedded at the bottom. A red alarm is triggered when the cerebrospinal fluid release volume V>15ml, a yellow level 2 warning is displayed when ΔP>15mmHg and ∂D / ∂t>0.3mm / min, and a green warning is displayed when there is no risk.

[0038] A method for predicting fluctuation trends of a non-invasive intracranial pressure continuous monitoring teaching device comprises the following steps: S1, driving the pressure regulating valve 7 through the PID controller of the control center 9, adjusting the pipeline resistance of the closed-loop fluid system in real time, and generating simulated intracranial pressure waveforms of 9 clinical pathological characteristics; S2, the detector 5 collects the real-time pressure time series data of the intracranial simulation block 4, forms a 128-dimensional vector of a 128-second sliding window, the B-ultrasound image analysis module tracks the optic nerve diameter simulation unit, calculates the instantaneous rate of change of the diameter, and trains the flow sensor of the lumbar puncture hole 11 to monitor the cumulative amount of cerebrospinal fluid released V in real time; S3 uses a three-layer fully connected neural network. The 128 neurons in the input layer receive pressure time series data and adaptive weight coefficient α. The hidden layer fuses pathological and operational features through a double dense layer. The weight α is dynamically initialized according to the pathological type. The output layer outputs the current intracranial pressure estimate.

[0039] S4. Predict the trend of current intracranial pressure estimates based on real-time operational data and assess operational risks.

[0040] Specifically, the pressure regulating valve 7 generates a pathological characteristic waveform under the drive of the PID controller, and outputs 9 clinical waveforms with an accuracy of ±0.8 mmHg based on the feedback pressure of the detector 5; the detector 5 synchronously collects the real-time pressure sequence, and the B-ultrasound imaging module tracks the changes in the optic nerve diameter at a frame rate of 30 fps. The dual-channel data is pre-processed by the control center 9 and input into the AI ​​model. The real-time risk model adopts a fully connected neural network. When the lumbar puncture release volume V>10 ml, the α weight increment is triggered, and the pressure estimate for the next 5 seconds is output. The trend prediction model deploys the LSTM network, which is processed from the LSTM layer to the Dropout layer to the fully connected layer. The dual-channel output generates a prediction interval of [μ-1.96σ, μ+1.96σ] for the next 5 minutes. When the upper limit of the interval is >25 mmHg, the display screen 10 is automatically stopped and the intracranial hypertension decompensation is marked.

[0041] The AI ​​risk assessment algorithm uses an LSTM neural network. The LSTM neural network implements fluctuation trend prediction according to the following configuration: Input sequence: dimensional matrix, the row vector is ; in, : is the simulated intracranial pressure value at the moment, collected in real time by the detector (5); : is the simulated value of the optic nerve diameter at the moment, monitored by the B-ultrasound image analysis module; : is the cumulative amount of cerebrospinal fluid released at each moment, collected by the flow sensor at the training lumbar puncture hole (11); : is the instantaneous rate of change of the optic nerve diameter at time .

[0042] : is the time window range, representing the 59 time points before the current time t to the current time t.

[0043] Network Architecture: LSTM Layer ;Dropout layer;Fully connected layer; The input sequence is first processed in sequence by two LSTM layers to capture temporal features. The two LSTM layers are connected in series, and the hidden state of the previous layer is used as the input of the next layer. The processed features flow through the Dropout layer for random inactivation to suppress overfitting. The features are then integrated through the fully connected layer, and finally the dual-channel fully connected output layer generates the next 5-minute forecast interval, where nonlinear mapping is achieved between the fully connected layer and the output layer through the weight matrix.

[0044] Output layer: dual-channel fully connected layer, outputting the next 5-minute forecast interval: ,in, is the predicted mean, is the standard deviation, is the lower limit of the predicted intracranial pressure in the next 5 minutes. The upper limit of the predicted intracranial pressure in the next 5 minutes, the upper limit of the interval When the patient is in the emergency room, the risk of decompensation of intracranial hypertension is determined and a teaching warning is triggered.

[0045] Specifically, the AI ​​risk assessment algorithm uses an LSTM neural network to predict intracranial pressure fluctuation trends. The neural network's input sequence is a 128-dimensional matrix, where each row represents a feature vector at a time point, such as pressure data or other relevant physiological parameters from the past few seconds. This captures the temporal characteristics of intracranial pressure. The network architecture consists of an LSTM layer to model temporal dynamic dependencies, a dropout layer to prevent overfitting by randomly dropping neurons, and a fully connected layer for deep feature extraction and transformation. The output layer is a two-channel fully connected structure that directly generates a 5-minute intracranial pressure prediction interval. This interval is defined by the predicted mean μ and standard deviation σ, where μ represents the model's predicted mean intracranial pressure, and σ represents the standard deviation, derived from the calculation of historical prediction residuals and reflecting the model's prediction uncertainty range. The output is the upper and lower bounds of the prediction interval. When the upper bound exceeds 25 mmHg, the system determines the risk of intracranial hypertension decompensation and immediately triggers the teaching warning mechanism to alert the operator to intervene promptly.

[0046] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A non-invasive intracranial pressure continuous monitoring teaching device, characterized in that: The invention comprises a protective shell (1) and a closed-loop fluid system integrated therein, wherein the closed-loop fluid system is composed of a pump body (2), an intracranial simulation block (4), a connecting pipe 1 (3), a connecting pipe 2 (6), a connecting pipe 3 (8) and a pressure regulating valve (7). The output end of the pump body (2) is connected to the input cavity of the intracranial simulation block (4) via the connecting pipe 1 (3), and the return end thereof forms a closed loop via the connecting pipe 2 (6), the pressure regulating valve (7) and the connecting pipe 3 (8). A detector (5) is provided at the top of the intracranial simulation block (4) for collecting simulated intracranial pressure data in real time. A display screen (10) is provided on the outer surface of the protective shell (1), and an internal integrated control center (9) processes data and drives the pressure regulating valve (7) to dynamically adjust the pressure.

2. A non-invasive intracranial pressure continuous monitoring teaching device according to claim 1, characterized in that: The inner wall of the protective shell (1) is fixedly embedded with a training lumbar puncture hole (11), the puncture channel of which passes through the protective shell (1) and is connected to the fluid cavity of the intracranial simulation block (4), for simulating a clinical cerebrospinal fluid extraction operation.

3. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 2, characterized in that: The training lumbar puncture hole (11) includes a shell (111), the exterior of the shell (111) is fixedly connected to the interior of the detector (5) through a protective shell (1), a flow sensor is provided inside the shell (111), a backflow prevention plate 1 (112) is fixedly connected to the interior of one side of the shell (111), and a backflow prevention plate 2 (113) is fixedly connected to the interior of the other side of the shell (111), the backflow prevention plate 1 (112) and the backflow prevention plate 2 (113) are adjacent to each other, and the backflow prevention plate 1 (112) and the backflow prevention plate 2 (113) are both made of elastic silicone material, and are used to automatically close the channel after the puncture needle is pulled out to prevent leakage of simulated cerebrospinal fluid.

4. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 1, characterized in that: The control center (9) is integrated with a B-ultrasound image analysis module for real-time monitoring of the optic nerve diameter simulation unit associated with the intracranial simulation block (4), and inverting the intracranial pressure fluctuation trend through the diameter change.

5. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 4, characterized in that: The control center (9) is equipped with an AI risk assessment algorithm, and the input parameters of the AI ​​risk assessment algorithm include: The real-time cumulative amount of cerebrospinal fluid release was collected by a flow sensor trained on the lumbar puncture hole (11); The simulated intracranial pressure data collected by the detector (5) is calculated as follows: ; The optic nerve diameter change rate output by the B-ultrasound image analysis module; The risk determination logic of the AI ​​risk assessment algorithm includes: when 、 or When the alarm is triggered, the sound and light alarm will be triggered; when and When the alarm is triggered, a secondary warning signal is output to the display screen (10).

6. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 5, characterized in that: The AI ​​risk assessment algorithm fuses the pressure sensor time series data with the B-ultrasound optic nerve diameter data to establish a multi-parameter coupled intracranial pressure prediction model. The multi-parameter coupled intracranial pressure prediction model satisfies the following requirements: in, Historical pressure time series data window; Predict the intracranial pressure value at time t+1; The instantaneous rate of change of optic nerve diameter; Adaptive weight coefficient, dynamically adjusted according to pathological type, , initial value 0; function : A three-layer fully connected neural network is used, and the hidden layer activation function is , the output layer is Sigmoid; Structure: Input layer: 128 neurons, hidden layer: double dense layer, output layer: linear activation function, output ; Weight adaptive mechanism: When When Dynamically increase the weight of operational factors.

7. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 1, characterized in that: The pressure regulating valve (7) is electrically connected to a PID controller of a control center (9) and is used to adjust the opening in real time according to the pressure value fed back by the detector (5), simulating the dynamic fluctuation of intracranial pressure under pathological conditions.

8. The non-invasive intracranial pressure continuous monitoring teaching device according to claim 1, characterized in that: The display screen (10) synchronously and visually displays the real-time waveform of intracranial pressure, the AI ​​prediction trend curve, and the operation risk assessment level.

9. A method for predicting fluctuation trends of a non-invasive intracranial pressure continuous monitoring teaching device, characterized in that: A non-invasive continuous intracranial pressure monitoring teaching device according to any one of claims 1 to 8 comprises the following steps: S1, driving the pressure regulating valve (7) through the PID controller of the control center (9), adjusting the pipeline resistance of the closed-loop fluid system in real time, and generating simulated intracranial pressure waveforms of 9 clinical pathological characteristics; S2, the detector (5) collects the real-time pressure time series data of the intracranial simulation block (4), forms a 128-dimensional vector of a 128-second sliding window, the B-ultrasound image analysis module tracks the optic nerve diameter simulation unit, calculates the instantaneous rate of change of the diameter, and trains the flow sensor of the lumbar puncture hole (11) to monitor the cumulative amount of cerebrospinal fluid released V in real time; S3 uses a three-layer fully connected neural network. The 128 neurons in the input layer receive pressure time series data and adaptive weight coefficient α. The hidden layer fuses pathological and operational features through a double dense layer. The weight α is dynamically initialized according to the pathological type. The output layer outputs the current intracranial pressure estimate. S4. Predict the trend of current intracranial pressure estimates based on real-time operational data and assess operational risks.

10. The method for predicting fluctuation trends of a non-invasive intracranial pressure continuous monitoring teaching device according to claim 9, characterized in that: The control center (9) is deployed with an AI risk assessment algorithm, which uses an LSTM neural network. The LSTM neural network implements fluctuation trend prediction according to the following configuration: Input sequence: dimensional matrix, the row vector is ; in, : is the simulated intracranial pressure value at the moment, collected in real time by the detector (5); : is the simulated value of the optic nerve diameter at the moment, monitored by the B-ultrasound image analysis module; : is the cumulative amount of cerebrospinal fluid released at each moment, collected by the flow sensor at the training lumbar puncture hole (11); : is the instantaneous rate of change of the optic nerve diameter at the moment; : is the time window range, representing the 59 time points before the current time t to the current time t; Network Architecture: LSTM Layer ;Dropout layer;Fully connected layer; The input sequence is first processed sequentially by two LSTM layers to capture temporal features. The two LSTM layers are connected in series, with the hidden state of the previous layer serving as the input to the next layer. The processed features flow through a Dropout layer for random deactivation to prevent overfitting. The features are then integrated through a fully connected layer, and finally a dual-channel fully connected output layer generates a 5-minute forecast interval. The nonlinear mapping between the fully connected layer and the output layer is achieved through a weight matrix. Output layer: dual-channel fully connected layer, outputting the next 5-minute forecast interval: ,in, is the predicted mean, is the standard deviation, is the lower limit of the predicted intracranial pressure in the next 5 minutes, The upper limit of the predicted intracranial pressure in the next 5 minutes, the upper limit of the interval When the patient is in the emergency room, the risk of decompensation of intracranial hypertension is determined and a teaching warning is triggered.

Images