Cerebrospinal fluid shunt monitoring system

Abstract

The embodiment of the invention provides a cerebrospinal fluid shunt monitoring system. The cerebrospinal fluid shunt monitoring system comprises a data acquisition unit and a data processing unit, the data acquisition unit is used for acquiring temperatures of upstream and downstream areas of a cerebrospinal fluid shunt tube, and the data transmission component is used for transmitting the temperatures acquired by the temperature detection component to the data processing unit; and the data processing unit is used for processing the temperature acquired by the data acquisition unit so as to determine the flow of cerebrospinal fluid. According to the cerebrospinal fluid shunt monitoring system disclosed by the embodiment of the invention, the quantitative precision can be improved, and the monitoring lower limit reaches 0.01 ml / min; meanwhile, the depth adaptability is achieved, measurement of the subcutaneous tissue thickness of 1-6 mm is supported, and the false detection rate is reduced to tilt; 5%.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical devices, and particularly relates to a cerebrospinal fluid shunt monitoring system. BACKGROUND

[0002] Cerebrospinal fluid (CSF) is a colorless and transparent liquid that exists in the ventricular system, subarachnoid space and central canal of the spinal cord, surrounds the brain and spinal cord, and is mainly secreted by the choroid plexus (a special structure) in the ventricle. The composition is similar to plasma, but contains less protein, different ion concentrations (such as sodium, chlorine, potassium, magnesium), and some glucose and trace cells (mainly lymphocytes). When the production, circulation and absorption of cerebrospinal fluid are disturbed (such as hydrocephalus, some cystic cavities or tumors pressing and blocking the circulation path, absorption disorders after hemorrhage or infection, etc.), excessive accumulation of cerebrospinal fluid in the ventricular system causes increased intracranial pressure and threatens brain function. Therefore, cerebrospinal fluid shunt is needed. The cerebrospinal fluid shunt technique is a neurosurgical procedure aimed at draining excess cerebrospinal fluid from the ventricular system or subarachnoid space to other parts of the body where fluid can be absorbed, thereby reducing intracranial pressure and protecting brain tissue.

[0003] According to statistics, the average effectiveness of the cerebrospinal fluid shunt technique is about 50%, and the incidence of cerebrospinal fluid shunt tube obstruction, poor drainage, excessive shunt, shunt failure and poor shunt position is about 40%, which is the main reason for surgical failure.

[0004] At present, the widely used cerebrospinal fluid shunt technique is the ShuntCheck technology of NeuroDX laboratory. This technology uses a non-invasive monitoring device based on the principle of thermal dilution to evaluate the cerebrospinal fluid flow status of patients after ventriculoperitoneal (VP) shunt surgery, so as to quickly identify shunt obstruction problems. The principle of thermal dilution is that local cooling (such as an ice bag) is applied near the cerebrospinal fluid shunt tube on the skin surface to form a transient temperature change. If the cerebrospinal fluid shunt tube is unobstructed, the cooled cerebrospinal fluid will be detected by a thermal sensor when it flows downstream, resulting in a transient temperature drop. If there is no flow, there will be no temperature change signal. The advantages of this technology are non-invasive, fast (5-10 minutes), repeatable, suitable for children and the elderly. The disadvantages are that the signal may be weakened in obese or thick subcutaneous tissue; it needs to be operated by professional doctors or nursing staff, which is difficult to operate; it may miss detection in low-speed flow or partial obstruction, and needs to be combined with clinical symptoms, imaging (ultrasound / CT) or isotope shunt imaging for comprehensive judgment, and it cannot monitor the flow at present.

[0005] The non-invasive monitoring flow sensor based on heat conduction in the prior art can only simulate the detection of cerebrospinal fluid shunt conditions, but it cannot accurately detect individual differences, and the flow sensor system is complex, which is not convenient for patients to wear and use. SUMMARY

[0006] To solve the above technical problems, the present application provides a cerebrospinal fluid shunt monitoring system, comprising a data acquisition unit and a data processing unit, The data acquisition unit comprises a temperature detection component and a data transmission component, the temperature detection component comprises a temperature sensor array and a thermal actuator, the thermal actuator is arranged in the center of the plurality of temperature sensors arranged in a spiral as a whole; the temperature sensor array comprises a plurality of temperature sensors for collecting the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt, and the data transmission component transmits the temperature collected by the temperature detection component to the data processing unit; The data processing unit processes the temperature collected by the data acquisition unit to determine the cerebrospinal fluid flow.

[0007] In some embodiments, the plurality of temperature sensors are arranged in a spiral as a whole and in a straight line along the direction of cerebrospinal fluid flow detection.

[0008] In some embodiments, the plurality of temperature sensors are arranged in pairs, two temperature sensors in each pair are arranged on a straight line along the direction of cerebrospinal fluid flow detection, respectively arranged on both sides of the thermal actuator, and the distance of the two temperature sensors to the thermal actuator is equal, so that the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt can be obtained .

[0009] In some embodiments, the data processing unit processes the temperature collected by the data acquisition unit to determine the cerebrospinal fluid flow comprises: According to the arrangement of the plurality of temperature sensors of the temperature sensor array of the data acquisition unit and the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt collected by the data acquisition unit, a heat map is reconstructed; According to the subcutaneous tissue thickness of the measurement site, the heating time of the thermal actuator is determined and heated; Based on the relationship between the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt and the cerebrospinal fluid flow , a physical guide and a convolutional neural network feature fusion network structure are designed to predict the cerebrospinal fluid flow in the cerebrospinal fluid shunt .

[0010] In some embodiments, the heat map reconstruction comprises: According to the n×m pixel matrix A obtained from the arrangement of the plurality of temperature sensors of the temperature sensor array of the data acquisition unit, each pixel in the pixel matrix A corresponds to the collected temperature of a temperature sensor, the pixel matrix A is scaled by K times, K>1, and a double cubic interpolation algorithm is used to obtain a processed Kn×Km pixel matrix B.

[0011] In some embodiments, the conversion of the value B(X, Y) of the pixel point (X, Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A includes: The corresponding point P of the pixel point (X, Y) in the pixel matrix B is located at (x, y) = (X / K, Y / K), taking the corresponding point P as the reference point A(0, 0) in the pixel matrix A, selecting a plurality of pixel points closest to the reference point A(0, 0) to form a pixel array A(i, j), the decimal part of X / K is u, and the decimal part of Y / K is v; The BiCubic function is constructed as follows: , wherein a is -0.5, and d represents the difference between the row or column value of the selected pixel point and the row or column value of the reference point P; The weight value of each selected pixel point is calculated as follows: The weight values of all selected pixel points are summed to obtain the value B(X, Y) of the pixel point (X, Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A.

[0012] In some embodiments, the diffusion of thermal disturbance to the subcutaneous tissue thickness is determined by the following formula: The required heating time t: , wherein, is the thermal diffusivity of the skin.

[0013] In some embodiments, the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt under steady state and the cerebrospinal fluid flow is as follows: , wherein, , is the density of cerebrospinal fluid, is the specific heat capacity of cerebrospinal fluid, is the temperature difference between the liquid inlet and outlet, k is the thermal conductivity of the subcutaneous tissue, A is the effective area of the skin surface under the patch participating in heat conduction, and r is the distance from the temperature sensor corresponding to the temperature sampling point to the thermal actuator.

[0014] In some embodiments, the fusion network structure includes a physical guide channel, a fully connected layer, a heat map channel, a convolutional neural network, a feature splicing layer, and a regression output layer, the physical guide channel is used to receive the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt and output it to the fully connected layer; The full connection layer is connected with the physical guide channel, and is used for outputting a physical feature vector ; The heat map channel is used for inputting a steady-state heat map obtained based on temperature data obtained by the data acquisition unit into the convolutional neural network ; The multi-level convolutional neural network extracts spatial features in the steady-state heat map , and obtains a heat map feature vector ; The heat map feature vector is consistent in dimension with the physical feature vector The feature splicing layer adopts a gated weighting fusion mechanism to optimize and fuse the heat map feature vector and the physical feature vector , and obtains a weighted fusion feature vector ; The regression output layer outputs a cerebrospinal fluid flow value according to the weighted fusion feature vector .

[0015] In some embodiments, the temperature difference upstream and downstream of the cerebrospinal fluid shunt is expanded into a four-dimensional vector before being output to the full connection layer , , wherein is a temperature change value at a measurement position where the cerebrospinal fluid flow state changes significantly, The full connection layer includes n layers of full connection layers, and all layers before the n-1 layer adopt a ReLU activation function to introduce a nonlinear transformation to capture the physical mapping of flow and temperature difference; the n layer, that is, the last layer, adopts a linear activation.

[0016] Advantages of the present application: The cerebrospinal fluid shunt monitoring system of the embodiment of the present application can improve the quantitative accuracy, and the monitoring lower limit reaches 0.01 ml / min, which is higher than 0.05 ml / min of ShuntCheck, and the embodiment of the present application can distinguish between partial obstruction (0.027 ml / min) and complete obstruction (0.01 ml / min) of the cerebrospinal fluid shunt.

[0017] At the same time, the cerebrospinal fluid shunt monitoring system based on an artificial intelligence model of the embodiment of the present application has deep adaptability, supports measurement of subcutaneous tissue thickness of 1-6 mm, and the false detection rate is reduced to <5%, while the upper limit of the measurement of the subcutaneous tissue thickness in the prior art is 3 mm.

[0018] In addition, the cerebrospinal fluid shunt monitoring system of the embodiment of the present application is convenient to operate, and the single monitoring time is 10 minutes, which is less than 30 minutes of ShuntCheck, and supports real-time monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a structural block diagram of the cerebrospinal fluid shunt monitoring system of the embodiment of the present application; Figure 2a is a schematic diagram of a temperature sensor array of the cerebrospinal fluid shunt monitoring system of the embodiment of the present application; Figure 2b is a schematic diagram of a plurality of sensors in a serpentine arrangement of a temperature sensor array in the prior art; Figure 3 is a table of parameter comparison between the temperature sensor array in a spiral arrangement of the embodiment of the present application and the prior art; Figure 4 is a schematic diagram of a target pixel point and adjacent 16 pixel points in a bicubic interpolation of the embodiment of the present application; Figure 5a is an initial thermal map obtained by the temperature sensor of the embodiment of the present application; Figure 5b is a thermal map after fitting by a bicubic interpolation algorithm of the embodiment of the present application; Figure 6 is a physical guidance and convolutional neural network (CNN) feature fusion network structure of the embodiment of the present application; Figure 7 is a schematic diagram of extracting spatial features in the steady-state thermal map by a multi-level convolutional neural network (CNN) of the embodiment of the present application; Figure 8 is a graph of judging the working state of the cerebrospinal fluid shunt according to the cerebrospinal fluid flow value of the embodiment of the present application; Figure 9 is a flow block diagram of the cerebrospinal fluid shunt monitoring system of the embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below, combined with specific embodiments and with reference to the drawings. However, those skilled in the art know that the present application is not limited to the drawings and the following embodiments.

[0022] ​​As described herein, the term "comprising" and its various variants are to be understood to be open terms that mean "including but not limited to". The term "based on" and its conjugations are to be understood as "based at least in part on". The terms "first", "second", "third", etc. are used only to distinguish different features and do not have a substantive meaning. The terms "left", "right", "middle" and the like are used only to represent the positional relationship between the relative objects.

[0023] The present application embodiment proposes a cerebrospinal fluid (CSF) shunt monitoring system, which comprises a data acquisition unit and a data processing unit, as shown in Figure 1 The data acquisition unit transmits the collected data to the data processing unit.

[0024] The present application embodiment is based on the heat diffusion principle to monitor the flow of cerebrospinal fluid in the cerebrospinal fluid shunt, and can further calculate the flow of cerebrospinal fluid in the cerebrospinal fluid shunt. The heat diffusion principle refers to the process of heat transmission through a material, which is affected by factors such as thermal conductivity, thermal diffusion coefficient and material thickness. In biological tissues, heat diffusion is affected by factors such as blood flow and tissue hydration, especially in the skin and the tissues below it, where heat conduction often shows significant directional differences. When fluid flows through subcutaneous tissue (such as cerebrospinal fluid flowing through the cerebrospinal fluid shunt), due to the difference in thermal conductivity of the flowing liquid and the surrounding tissue, heat diffusion shows anisotropy. In the present application embodiment, this anisotropic heat conduction effect needs to be accurately detected by a high-density temperature sensor array. By analyzing the differences in the detected temperature data, a heat diffusion directionality map can be generated to determine whether cerebrospinal fluid is flowing.

[0025] In the present application embodiment, the data acquisition unit is used to acquire the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt. The data acquisition unit includes a temperature detection component, a data transmission component and a power supply component.

[0026] The temperature detection component includes a temperature sensor array and a flexible covering shell, and the flexible covering shell wraps the temperature sensor array. The flexible covering shell is a flexible structure composed of flexible materials, which is suitable for close contact with the monitoring site (human skin). The preferred flexible material is, for example, Ecoflex silicone, which has a thickness of 100 μm and a modulus of 70 kPa, which is more conducive to close and firm contact with the human skin and has no pressure marks after long-term wear.

[0027] In the present application embodiment, the temperature sensor array includes a plurality of temperature sensors for measuring the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt, and the plurality of temperature sensors are arranged in a spiral as a whole and arranged in a straight line along the direction of cerebrospinal fluid flow detection, as shown inFigure 2a As shown, Figure 2a The direction of cerebrospinal fluid flow detection is vertical. Currently known technologies employ multiple temperature sensors arranged in a serpentine pattern for temperature measurement, such as... Figure 2b As shown in the figure. Compared with the prior art, the temperature sensor array of the present invention has significant advantages in measurement accuracy, cerebrospinal fluid flow resolution, and anti-interference performance. The parameter comparison table is shown below. Figure 3 As shown.

[0028] In this embodiment of the invention, the arrangement of multiple temperature sensors (arranged in a spiral pattern and in a straight line along the direction of cerebrospinal fluid flow detection) improves measurement and calculation accuracy while reducing the difficulty of wearing them. When measuring cerebrospinal fluid, the temperature detection component needs to be placed above the cerebrospinal fluid shunt according to the marked direction (related to the sensors arranged in a straight line along the direction of cerebrospinal fluid flow detection, ensuring that the sensors arranged in a straight line along the direction of cerebrospinal fluid flow detection are parallel to the direction of the cerebrospinal fluid shunt), eliminating the need for professional personnel to wear it.

[0029] Non-invasive monitoring of cerebrospinal fluid is based on the principle of thermal diffusion, and the main factor affecting measurements using this principle is the thickness of the subcutaneous tissue. Therefore, in this embodiment of the invention, to improve the measurement accuracy of the cerebrospinal fluid shunt flow and provide higher spatial resolution, the temperature detection component further includes a thermal actuator T, which is positioned at the center of a plurality of temperature sensors arranged in a spiral pattern. Figure 3 As shown, the thermal actuator T provides a reference temperature during temperature detection by the temperature sensing array. This improves the signal-to-noise ratio of the temperature sensing array, facilitating the acquisition of temperatures upstream and downstream of the cerebrospinal fluid shunt and thus determining the flow of cerebrospinal fluid. Since the thermal anisotropy of human skin is related to factors such as race, age, sex, body type, and location, clinical medical monitoring typically uses methods like CT, ultrasound, or MRI to measure subcutaneous tissue thickness. Therefore, the activation time of the thermal actuator is adjusted based on the subcutaneous tissue thickness.

[0030] In this embodiment of the invention, a deep learning model is established based on race, age, gender, body type, location, etc. The deep learning model predicts the subcutaneous tissue thickness of the monitored patient, thereby enabling adaptive adjustment of the start-up time of the thermal actuator, which will be explained later.

[0031] The temperature sensor is preferably a resistance temperature detector (RTD), using 100 (10) resistors. 10) Assemble a temperature sensor array, such as Figure 2aThe RTD temperature sensor is usually made of high-purity platinum, nickel, copper or other metals, and the Cr / Au resistance temperature sensor is more preferred in the embodiment of the present application. The working principle is as follows: when the temperature rises, the lattice vibration of the metal intensifies, the free electron movement is blocked, and the resistance value increases; on the contrary, when the temperature decreases, the resistance decreases, satisfying the following formula: , Wherein: : temperature T , : reference temperature (typically 0℃) under the standard resistance (such as PT100 =100 ), : temperature coefficient (typical value of platinum is about 0.00685 / ℃).

[0032] The signal conversion process is as follows: the temperature sensor is powered by a constant current source, the voltage drop across the temperature sensor and the current value flowing through the resistance I are measured, the resistance value is calculated according to Ohm's law , and the temperature value is inversely calculated from the above formula T .

[0033] The measurement accuracy of the RTD temperature sensor is within ±0.1℃, and the long-term stability is good, but the response speed is slow, and an external excitation current is required. In the temperature sensor array equipped with a thermal actuator in the embodiment of the present application, the external excitation current can be synchronized and controlled according to the calculated thermal actuation response time according to the subcutaneous tissue thickness.

[0034] In the embodiment of the present application, the plurality of temperature sensors are a plurality of temperature sensors arranged in pairs, two temperature sensors in each pair of temperature sensors are arranged on a straight line along the cerebrospinal fluid flow detection direction, respectively arranged on both sides of the thermal actuator T, and the distances from the two temperature sensors to the thermal actuator T are equal, so that the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt can be obtained . Therefore, the temperature detection component only needs to measure the voltage drop of each pair of temperature sensors, and transmit it to the data processing unit through the transmission protocol. The data processing unit can obtain the temperature change of the upstream and downstream of the cerebrospinal fluid shunt through the above algorithm.

[0035] That is, as Figure 2aThe temperature sensor array is divided into upstream and downstream according to the direction of cerebrospinal fluid flow detection, and the temperature of the upstream and downstream sensors can detect the current temperature, and the temperature detection component sends the temperature data collected by the temperature sensor to the data processing unit in real time for processing.

[0036] In the embodiment of the present application, the data transmission component is preferably a device using a short-range transmission protocol, including Bluetooth, ZigBee, RFID, NFC (Near Field Communication), etc., and the temperature data collected by the temperature detection component is transmitted to the data processing unit.

[0037] The power supply component needs to meet the power supply requirements of the sensor, and can use a lithium battery module or a magnetic resonance wireless charging power supply form, and the present application does not have special requirements for it.

[0038] The data processing unit determines the flow of cerebrospinal fluid in the shunt according to the temperature data from the data acquisition unit. The working process of the data processing unit of the embodiment of the present application for processing the temperature data collected by the data acquisition unit to determine the cerebrospinal fluid flow is described in detail.

[0039] Those skilled in the art can understand that the data acquisition device used in the present application is not limited to the preferred data acquisition device described above, as long as it is a device that can collect temperature data of the upstream and downstream areas of the cerebrospinal fluid flow monitoring site.

[0040] The working process of the data processing unit for processing the temperature data collected by the data acquisition unit to determine the cerebrospinal fluid flow includes: S1: According to the arrangement mode of the plurality of temperature sensors of the temperature sensor array of the data acquisition unit and the temperature data of the upstream and downstream of the cerebrospinal fluid shunt collected by the data acquisition unit, heat map reconstruction is performed.

[0041] In the embodiment of the present application, the real-time temperature obtained by the temperature sensor array can be theoretically regarded as a plane image, and the points on the image are discontinuous and have numerical ambiguity, so it is necessary to convert the discrete temperature data into a continuous function representation, and the purpose is to simulate the flow of cerebrospinal fluid in the cerebrospinal fluid shunt to the greatest extent.

[0042] The thermal map reconstruction comprises: obtaining an n*m pixel matrix A (each pixel can be processed into a gray value of 0-255) according to a plurality of temperature sensor arrangements of a temperature sensor array of the data acquisition unit, each pixel in the pixel matrix A corresponds to an acquisition temperature of a temperature sensor, performing K times scaling on the pixel matrix A, K>1, obtaining a processed Kn*Km pixel matrix B by using a bicubic interpolation algorithm, so as to achieve the effect of improving the image resolution.

[0043] Specifically, the value A(x, y) of each pixel point of the original pixel matrix A is known, and the pixel point value B(X, Y) in the pixel matrix B needs to be found first, and the pixel point value A(x, y) corresponding to the pixel point value B(X, Y) in the pixel matrix A is selected according to the pixel point value A(x, y) of the pixel matrix A. The pixel point value B(X, Y) in the pixel matrix B is calculated according to the pixel point value A(x, y) of the pixel matrix A, and the weight of the plurality of pixel points is calculated by using the BiCubic base function, and then the value of B(X, Y) is equal to the weighted superposition of the pixel point values of the plurality of pixel points.

[0044] The conversion process of the pixel point value B(X, Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A is as follows: According to the proportional relationship, x / X = m / Km = 1 / K, first obtain the position of the corresponding point P of the pixel point (X, Y) in the pixel matrix B in the pixel matrix A as (x, y) = (X / K, Y / K), and take the corresponding point P as the reference point A(0, 0) in the pixel matrix A, and select a plurality of pixel points closest to the reference point to form a pixel array A(i, j), for example, as shown in Figure 4 Since X and Y are not necessarily divisible by K, there is a decimal in the P coordinate, and the decimal parts are u and v, that is, the decimal part of X / K is u, and the decimal part of Y / K is v.

[0045] At this time, the BiCubic function is constructed as follows: , Wherein, a is generally-0.5, and d represents the difference between the row or column value of the selected pixel point and the row or column value of the reference point P.

[0046] For example, the distance between A1(A1(-1, -1)) and P is (1+u, 1+v), so the horizontal coordinate weight of A1 is W(1+u), and the vertical coordinate weight of A1 is W(1+v), and the contribution value of A1 to B(X, Y) is A(-1, -1) W(1+u) W(1+v). Therefore, the calculation formula of B(X, Y) is as follows: .

[0047] That is, the weight value of each selected pixel point is calculated The weight values of all selected pixel points are summed to obtain the value B(X, Y) of the pixel point (X, Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A.

[0048] In the embodiment of the application, the above algorithm can reduce the blur and detail loss during amplification by interpolating the temperature sensor array temperature value through the subdivided grid.

[0049] Figure 5a The initial thermal map obtained by the temperature sensor is shown, and it cannot be clearly judged whether there is cerebrospinal fluid flow in the cerebrospinal fluid shunt. After fitting by the bicubic interpolation algorithm, a continuous image is obtained, as shown in Figure 5b It can be clearly seen whether the cerebrospinal fluid shunt occurs.

[0050] S2: According to the subcutaneous tissue thickness of the measurement site, the heating time of the thermal actuator is determined, and heating is performed.

[0051] In order to measure the flow of cerebrospinal fluid in the cerebrospinal fluid shunt, the test site of the measured person needs to be provided with an ambient temperature, that is, for example, the subcutaneous tissue on the skin surface of the test site is heated by the thermal actuator in the temperature sensor array of the embodiment of the application, so as to reduce the interference of the external environment on the temperature. Due to the differences in race and individual characteristics (age, gender, BMI, etc.), the tissue thickness and thermal characteristics (thermal conductivity k and thermal diffusivity a) above the cerebrospinal fluid shunt at the clavicle are indefinite, so the heating time of the thermal actuator needs to be adaptively adjusted.

[0052] In order to obtain the time required for the thermal disturbance to penetrate the skin thickness, a multi-task learning model (MTL-NN) is used, and a large amount of labeled data (race, age, gender, BMI, etc. Characteristics and subcutaneous tissue thickness , thermal conductivity k corresponding data set) is used as the training set to obtain the prediction model to provide the reference value of the subcutaneous tissue thickness , thermal conductivity k at the clavicle.

[0053] The heater time is the characteristic time scale solution of the semi-infinite body transient heat conduction model, and the heating time t of the thermal actuator can be solved by the following formula: , Where t is the time required for the thermal disturbance to diffuse to the subcutaneous tissue thickness , thermal diffusivity of the skin, which characterizes the rate of heat propagation , including thermal conductivity k, density and specific heat capacity ).

[0054] Assume the skin parameters are 0.003 m (3 mm), 1.5 x 10 -7 m 2 / s (typical biological tissue value), the calculated thermal actuator start-up time is 15 s. That is, it indicates that after the skin surface is heated, about 15 seconds are needed to make the thermal disturbance significantly affect the temperature at 3 mm below the skin, which is consistent with actual physiological observations.

[0055] It should be noted that the reference values of the skin parameters are given here. In real clinical environments, individual pathological skin state variations, physiological structural differences (such as obese patients), local interference factors, etc. may occur, and some calibration mechanisms need to be considered on the basis of the reference values.

[0056] S3: Based on the relationship between the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt under steady state and the cerebrospinal fluid flow , a physical guide and a convolutional neural network (CNN) feature fusion network structure are designed, and the input steady-state temperature difference and the stable heat map are trained to predict the cerebrospinal fluid flow in the cerebrospinal fluid shunt.

[0057] After the heating of the thermal actuator is completed, the power is maintained to provide a constant temperature environment for the measurement site when measuring the flow, that is, the cerebrospinal fluid shunt embedded in the skin is not disturbed by other heat sources, and is relatively in a steady-state environment. At this time, when the cerebrospinal fluid flows through the cerebrospinal fluid shunt, it will take away the heat flow . The heat flow taken away per unit time and the cerebrospinal fluid volume flow have the following mathematical relationship: (5) where is the density of cerebrospinal fluid, is the specific heat capacity of cerebrospinal fluid, is the temperature difference between the inlet and outlet of the liquid, and because the pipeline is small, it can be approximated as the average temperature difference between the cerebrospinal fluid and the skin surface below the patch.

[0058] When there is no cerebrospinal fluid flowing through the cerebrospinal fluid shunt, the skin surface temperature is distributed radially symmetrically (such as the temperature at a distance from the center ). When there is cerebrospinal fluid flowing through the cerebrospinal fluid shunt, the cerebrospinal fluid takes away the heat flow temperature at a certain point of flow direction lower than the temperature at the symmetric point (opposite direction) , forming the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt = As the cerebrospinal fluid flows through, the heat map received by the temperature sensor reaches a steady state , and the temperature difference also reaches a stable value. At this time, according to the Fourier law derivation combined with finite element analysis, under the steady state , and the heat flow have the following proportional relationship: (6) Where k is the thermal conductivity of subcutaneous tissue, A is the effective area of the skin surface under the patch participating in heat conduction, and r is the distance (geometric constant) from the temperature sensor corresponding to the temperature measurement point to the thermal actuator.

[0059] Solving (5) and (6) gives the following relationship between and the cerebrospinal fluid flow to be measured under steady state: (7) Where, (8) The value depends on: ① , : the liquid properties of cerebrospinal fluid itself, the liquid with high density and specific heat capacity can carry away more heat under the same flow, and is larger; ② k, A: the thermal conductivity of subcutaneous tissue and the heat conduction area of the skin surface under the patch obtained from S3, the better the heat conduction performance, the more sensitive to flow changes; ③ r: a geometric parameter, the farther the sensor is from the center, the gentler the temperature gradient, and the smaller

[0060] Based on the above analysis of the relationship between the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt and the cerebrospinal fluid flow , a physical guide and convolutional neural network (CNN) feature fusion network structure is designed to train input steady-state temperature difference and stable heat map to predict the cerebrospinal fluid flow in the cerebrospinal fluid shunt , as shown in Figure 6 .

[0061] The network structure includes a physical guide channel, a fully connected layer, a heat map channel, a convolutional neural network (CNN), a feature splicing layer, and a regression output layer.

[0062] Specifically, the physical guidance channel is used to receive the temperature difference upstream and downstream of the cerebrospinal fluid shunt , and output it to the fully connected layer. The physical guidance channel is a multi-stage feedforward network. In this embodiment, before the temperature difference upstream and downstream of the cerebrospinal fluid shunt is output to the fully connected layer, the temperature difference upstream and downstream of the cerebrospinal fluid shunt is expanded into a four-dimensional vector

[0063] As follows: , wherein, From top to bottom, in order, are a linear term (basic heat transfer effect) , a scaled term (physical constraint coupled with tissue thermal properties) , a squared term (capture non-linear heat dissipation caused by turbulence) , and a threshold term (identify flow state mutation) ; The temperature change value at the measurement position where the cerebrospinal fluid flow state changes significantly, where “significantly” refers to, for example, the cerebrospinal fluid changing from being blocked to flowing, or the flow rate increasing by a factor of two. For the case of the cerebrospinal fluid changing from being blocked to flowing, when the cerebrospinal fluid changes from a blocked steady state to a flowing steady state, the temperature changes by at least 0.1℃, = 0.1℃.

[0064] The fully connected layer is connected to the physical guidance channel, and includes multiple layers (assuming n layers) of fully connected layers. All layers up to the n-1 layer use a ReLU activation function to introduce a non-linear transformation to capture the physical mapping of flow rate and temperature difference. The n layer, i.e., the last layer, uses a linear activation to output a physical feature vector . The embodiment of the present application expands the features through a multi-dimensional vector, and forms a corresponding relationship between the input and the output through network learning.

[0065] The heat map channel is used to input a steady-state heat map obtained based on the temperature data obtained by the data acquisition unit to the multi-stage convolutional neural network.

[0066] The multi-stage convolutional neural network extracts spatial features in the steady-state heat map to obtain a heat map feature vector , and the heat map feature vector and the physical feature vector Dimension consistency. In this embodiment, 3x3 convolution kernel and ReLU activation function are used in each stage, edge expansion and pooling are optional operations, and the features of the last stage are flattened by the Softmax activation function to obtain the heat map feature vector , as shown in Figure 7 .

[0067] The feature concatenation layer optimally fuses the heat map feature vector and the physical feature vector by using a gated weighting fusion mechanism.

[0068] First, the feature gating value vector is calculated: , wherein is the feature gating value vector, representing the trustworthiness of the physical characteristics; , are the gating weight matrix and the bias, respectively; represents element-wise addition; represents the Sigmoid activation function.

[0069] According to the gating fusion formula, the weighted fused feature vector is obtained:

[0070] wherein is the identity matrix, is a diagonal matrix with all diagonal elements being alpha and others being zero.

[0071] The regression output layer outputs the cerebrospinal fluid flow value according to the weighted fused feature vector . The regression output layer is a multi-stage feedforward network, which is different from the feedforward network in the above process. It realizes dimension compression through layer-by-layer feature refinement, and the use of activation function is consistent with the physical guided channel (ReLU activation function is used in the first n-1 layers and all previous layers to introduce nonlinear transformation, and linear activation is used in the nth layer, i.e., the last layer). Finally, the cerebrospinal fluid flow value is output.

[0072] Figure 8 is a graph for judging the working state of the cerebrospinal fluid shunt according to the cerebrospinal fluid flow value .

[0073] For the cerebrospinal fluid flow value , the network parameters are updated by the following loss function through back propagation: in the above formula is an actual cerebrospinal fluid flow value, which is obtained through pre-training data collection; is an adjustable hyperparameter, has been given by equation (8).

[0074] Figure 9 A cerebrospinal shunt monitoring flowchart of an embodiment of the present application is shown as described above.

[0075] The cerebrospinal shunt monitoring system of the embodiment of the present application can improve the quantitative accuracy, and the monitoring lower limit reaches 0.01 ml / min, which is higher than 0.05 ml / min of ShuntCheck, and the embodiment of the present application can distinguish between partial obstruction (0.027 ml / min) and complete obstruction (0.01 ml / min) of the cerebrospinal shunt.

[0076] At the same time, the cerebrospinal shunt monitoring system of the embodiment of the present application has deep adaptability, supports measurement of subcutaneous tissue thickness of 1-6 mm, and the false detection rate is reduced to <5%, while the upper limit of the measurement of the subcutaneous tissue thickness of the prior art is 3 mm.

[0077] In addition, the cerebrospinal shunt monitoring system of the embodiment of the present application is convenient to operate, and the single monitoring time is 10 minutes, which is less than 30 minutes of ShuntCheck, and supports real-time monitoring.

[0078] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0079] The above describes the embodiments of the present application. However, the present application is not limited to the above-described embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A cerebrospinal fluid shunt monitoring system, comprising: The cerebrospinal fluid shunt monitoring system comprises a data acquisition unit and a data processing unit, The data acquisition unit comprises a temperature detection component and a data transmission component, the temperature detection component comprises a temperature sensor array and a thermal actuator, the thermal actuator is arranged in the center of the plurality of temperature sensors arranged in a spiral as a whole; the temperature sensor array comprises a plurality of temperature sensors for collecting the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt, and the data transmission component transmits the temperature collected by the temperature detection component to the data processing unit; The data processing unit processes the temperature collected by the data acquisition unit to determine the cerebrospinal fluid flow.

2. The cerebrospinal fluid shunt monitoring system of claim 1, wherein, The plurality of temperature sensors are arranged in a spiral as a whole and are arranged in a straight line along the cerebrospinal fluid flow detection direction.

3. The cerebrospinal fluid shunt monitoring system of claim 2, wherein The plurality of temperature sensors are arranged in pairs, and the two temperature sensors in each pair are arranged on a straight line along the cerebrospinal fluid flow detection direction, respectively arranged on the two sides of the thermal actuator, and the distances of the two temperature sensors to the thermal actuator are equal, so that the temperature difference δT of the upstream and downstream of the cerebrospinal fluid shunt can be obtained.

4. The cerebrospinal fluid shunt monitoring system of any of claims 1-3, wherein The data processing unit processes the temperature collected by the data acquisition unit to determine the cerebrospinal fluid flow, comprising: According to the arrangement mode of the plurality of temperature sensors of the temperature sensor array of the data acquisition unit and the temperature of the upstream and downstream regions of the cerebrospinal fluid shunt collected by the data acquisition unit, a thermal map is reconstructed; According to the subcutaneous tissue thickness of the measurement site, the heating time of the thermal actuator is determined, and heating is performed. Based on the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt under the steady state With the relationship between the cerebrospinal fluid flow , the physical guide and the convolutional neural network feature fusion network structure are designed to predict the cerebrospinal fluid flow in the cerebrospinal fluid shunt .

5. The cerebrospinal fluid shunt monitoring system of claim 4, wherein The thermal map reconstruction comprises: According to the n×m pixel matrix A obtained from the arrangement mode of the plurality of temperature sensors of the temperature sensor array of the data acquisition unit, each pixel in the pixel matrix A corresponds to the collected temperature of a temperature sensor, the pixel matrix A is scaled by K times, K>1, and a double cubic interpolation algorithm is used to obtain a processed Kn×Km pixel matrix B.

6. The cerebrospinal fluid shunt monitoring system of claim 5, wherein The conversion of the value B(X,Y) of the pixel point (X,Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A comprises: The corresponding point P of the pixel point (X,Y) in the pixel matrix B in the pixel matrix A is located at (x,y)=(X / K, Y / K), taking the corresponding point P as the reference point A(0,0) in the pixel matrix A, selecting a plurality of pixel points closest to the reference point A(0,0) to form a pixel array A(i,j), and the decimal part of X / K is u and the decimal part of Y / K is v; Constructing a BiCubic function: ， Wherein, a is-0.5, and d represents the difference between the row or column value of the selected pixel point and the row or column value of the reference point P. calculating the weight value of each selected pixel point summing the weight values of all selected pixel points to obtain the value B(X, Y) of the pixel point (X, Y) in the pixel matrix B corresponding to the pixel point (x, y) in the pixel matrix A.

7. The cerebrospinal fluid shunt monitoring system of claim 4, wherein , the heat disturbance diffusion to the subcutaneous tissue thickness is determined by the following equation the required heating time t: ， Wherein, α is the thermal diffusivity of the skin.

8. The cerebrospinal fluid shunt monitoring system of claim 4, wherein , the temperature difference between the upstream and downstream of the cerebrospinal fluid shunt in the steady state and the cerebrospinal fluid flow rate is: ， wherein, , is the density of cerebrospinal fluid, is the specific heat capacity of cerebrospinal fluid, is the temperature difference between the inlet and outlet of the liquid, k is the thermal conductivity of the subcutaneous tissue, A is the effective area of the skin surface under the patch involved in heat conduction, and r is the distance from the temperature sensor corresponding to the temperature measurement point to the thermal actuator.

9. The cerebrospinal fluid shunt monitoring system of claim 4, wherein The fusion network structure comprises a physical guidance channel, a fully connected layer, a thermal map channel, a convolutional neural network, a feature splicing layer, and a regression output layer, The physical guide channel is used to receive temperature difference upstream and downstream of a cerebrospinal fluid shunt and outputs it to the fully connected layer; The full connection layer is connected with the physical guide channel, and is used for outputting a physical feature vector ; the heat map channel is configured to input, to the convolutional neural network, a steady-state heat map derived based on temperature data obtained by the data acquisition unit ; The multi-level convolutional neural network extracts the steady-state heatmap. Spatial features are used to obtain heatmap feature vectors. The heatmap feature vector With the physical feature vector Consistent dimensions; The feature splicing layer employs a gated weighted fusion mechanism to process the heatmap feature vectors. and physical feature vectors Optimize and fuse to obtain a weighted fused feature vector. ; The regression output layer outputs a cerebrospinal fluid flow value , according to the weighted fused feature vector .

10. The cerebrospinal fluid shunt monitoring system of claim 9, wherein , the temperature difference upstream and downstream of the cerebrospinal fluid shunt the temperature difference upstream and downstream of the cerebrospinal fluid shunt extended to a four-dimensional vector , ， wherein To measure the temperature change value at which the flow state of the cerebrospinal fluid at the position is significantly changed, The full connection layer includes n layers of full connection layers, all layers before the n-1th layer adopt a ReLU activation function to introduce a nonlinear transformation to capture the physical mapping of the flow and the temperature difference; the nth layer, i.e., the last layer, adopts a linear activation.

Images