Operation method of an automatic needle insertion device combining a blood flow optimization model and a neural network

Abstract

To provide an automatic needle insertion method combining a blood flow optimization model and a neural network.SOLUTION: First, a position of a blood vessel and blood flow velocity distribution of a region waiting for needle insertion are calculated by blood flow information measured by automatic ultrasonic scanning using a Doppler ultrasonic principle. Then, a blood flow relative velocity optimization model and a neural network having a selection function are established, and optimum needle insertion point and needle insertion angle are determined on the basis of a result of physical measurement. Last, a needle insertion route is designed by simulating a needle insertion method of specialized staff using a neural network having a learning function. This method can be used for needle insertion for a subcutaneous blood vessel of a body part such as limbs, and can provide technical assistance for a self-help emergency situation in an environment where specialized staff is not enough while overcoming the difficulty of needle insertion by a human when blood vessel visibility by naked eyes is low.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to the fields of artificial intelligence and medicine, and more specifically to an automatic needle insertion method that combines a blood flow optimization model and a neural network. [Background technology]

[0002] Needle insertion into blood vessels is a common practice in modern medicine, from small ones such as venous blood sampling during medical examinations to large ones such as blood transfusions during first aid. However, it is often difficult for medical staff to accurately locate veins or arteries and determine their direction of travel. For example, some people have thin veins, tightly wrapped in a layer of muscle subcutaneous fat, or have thick hair on the surface of their bodies, making them less visible to the human eye. On the other hand, the veins of animals are wrapped in animal hair, making it difficult for the human eye to recognize the direction of the veins and arteries. Emergency care is often delayed due to the inability to accurately locate veins that can be used for needle insertion. Poor needle insertion methods can cause patients or animals to receive unnecessary injections multiple times, resulting in pain, bleeding, and even different degrees of infection, and can even cause medical accidents and serious conflicts between patients' families and doctors.

[0003] Ultrasound can be used to detect superficial and deep tissues that are invisible to the human eye, and is particularly suitable for accurately detecting and locating blood vessels in the human and animal bodies. In nature, vampire bats use their ultrasonic capabilities to accurately determine the distribution of blood vessels in animals and livestock that are invisible to the human eye, and then achieve blood-sucking. In artificially developed medical equipment, color Doppler ultrasound imaging devices can provide two-dimensional or three-dimensional image information such as the direction and velocity of blood flow in a given cross section.

[0004] However, currently there is still no imaging-free automated needle insertion method based on Doppler ultrasound principles and blood flow optimization models. Summary of the Invention

[0005] In view of the shortcomings and deficiencies of the prior art methods, the present invention proposes an automatic needle insertion method that combines a blood flow optimization model and a neural network. The present invention is specifically realized by the following technical aspects. An automatic needle insertion method that combines a blood flow optimization model and a neural network, comprising the following steps: Step 1: Using the principle of Doppler ultrasound, the spatial and angular relationship between the blood vessel position and blood flow velocity distribution in the needle insertion area is calculated according to the blood flow feedback signal measured by automatic ultrasound scanning, and ultrasonic blood flow information is obtained. Step 2: A blood flow relative velocity optimization model is constructed, and the instantaneous minimization model equation, which is a model of the instantaneous minimum value, is as follows: JPEG2024170321000002.jpg44170 Step 3: The above preferred needle insertion method is combined with a neural network with a selection function to determine the optimal needle insertion point and needle insertion angle, and an optimal needle insertion method is obtained. Step 4: Based on the optimal needle insertion method described above, a neural network with learning function is combined to learn and simulate the needle insertion technique of medical staff to obtain the needle insertion path.

[0006] Furthermore, the above step 1 includes automatic ultrasound scanning with multiple sets of frequencies and adjustable angles, and calculating the blood vessel position and blood flow velocity based on ultrasound Doppler.Usually, ultrasound examination in medical treatment is manual scanning, but in order to obtain the variable values ​​required for the blood flow relative velocity optimization model and achieve the purpose of automatic needle insertion, an ultrasound scan with automatically adjustable emission angle and frequency is performed on the part of the body waiting for needle insertion, and automatic measurement is performed using the Doppler principle.

[0007] Ultrasonic scanning can be mobile or stationary, mobile ultrasound scanning requires the ultrasound probe to automatically rotate at variable angles within a local range, and mobile translation scanning along a movable support, while stationary ultrasound scanning employs an electronically controlled phased array mode.

[0008] Based on the collected ultrasound information, irrelevant information is filtered out, and the necessary information, including the local spatial distribution and blood flow direction of the subcutaneous major blood vessels at the needle insertion site, is demodulated and calculated according to the Doppler ultrasound principle. The method of measuring and marking the blood flow spatial angle using the ultrasound Doppler principle is as follows:

[0009] The red blood cells move along the X-axis, i.e., the blood flow direction, at a speed v, and W is a reference axis, which may be perpendicular to the direction of gravity. The Z-axis, X-axis and W-axis are in the same plane, of which the Z-axis is perpendicular to the X-axis and the intersection angle between the Z-axis and W is φ. Ultrasound with a radiation frequency of f0 from the ultrasonic probe travels at a sound speed c in the internal medium. S The scattered signal propagates to the detection target at f and returns to the receiving probe. The received frequency is f r The ultrasonic incidence plane is the plane P1 where the incident sound beam axis and the Z axis are located, and the axis perpendicular to the Z axis in P1 is Y. The included angle between the ultrasonic incidence plane and the blood flow direction is α, that is, the included angle between the X axis and the Y axis is α, and the included angle between the incident sound beam axis and the XY plane is β, that is, the included angle between the incident sound beam axis and the Y axis is β. The scattering plane where the scattered sound beam axis of the red blood cell (group) along the receiver direction and the Z axis is located is P2, the axis perpendicular to the Z axis in P2 is Y', the included angle between the Y' axis and the blood flow direction is α', and the included angle between the scattered sound beam axis and the Y' axis is β'. The intersection angle between the incident sound beam axis and the W axis is γ, and the intersection angle between the scattered sound beam axis and the W axis is γ', and the values ​​of γ and γ' can be measured directly.

[0010] JPEG2024170321000003.jpg118170According to the effective ultrasonic feedback band extracted by collecting and filtering after the scanning angle is changed, a set of Doppler shifts is obtained by directional demodulation, and the blood vessel position and blood flow velocity parameters are calculated by using the frequency shift change equation set, the corresponding delay change and formula (3) to obtain the relative flow velocity and relative angle parameters (μ,α,β,φ), and further obtain the distribution and running direction of the local blood vessels in the required needle insertion area relative to the reference coordinate. When the ultrasonic transmitter and receiver are located in the same probe, i.e., at the same spatial position, the above frequency shift change equation set formula is as follows: In each equation (8a)-(9), the ± sign ensures that the calculation quantity on the right side is consistent with the positive and negative values ​​of cosβ or cosα on the left side. Under the condition that △α and △β can be accurately controlled, the relative angle parameters are obtained, and the local relative flow velocity is obtained from the frequency shift change equation. Furthermore, the included angle φ between the Z axis and the reference axis w is calculated and recorded according to equation (3). Also, even if the parameter α is held, the value of B can be obtained by calculating the delay due to the pitch change. JPEG2024170321000005.jpg28170

[0011] JPEG2024170321000006.jpg96170

[0012] JPEG2024170321000007.jpg81170JPEG2024170321000008.jpg68170

[0013] JPEG2024170321000009.jpg22170

[0014] JPEG2024170321000010.jpg81170

[0015] JPEG2024170321000011.jpg54170

[0016] Furthermore, in the above step 3, the preferred needle insertion plan and the actual clinical expression score obtained from the physical measurement and the blood flow relative velocity optimization model are used to train a neural network with a selection function, and integrated into a learning chip. The above neural network with a selection function includes multiple intermediate layers, and the parameters (μ1, α1, β1, φ1), ... (μ5, α5, β5, φ5) corresponding to the blood flow relative velocity and coordinates of the candidate injection points and the serial numbers of (1, 2, 3, 4, 5) are used as input data for the input layer, and evaluation feedback is performed on the needle insertion coordinates and needle insertion angle physically calculated by experienced medical personnel. Among them, μ is the ultrasonic sound speed c in the tissue medium of human or animal body. s p is the ratio of blood flow velocity to the direction of blood flow. α, β, φ determine the included angle between the incident sound beam and the blood flow direction. The evaluation feedback contents of the learning group include the preferred injection position and blood vessel running direction when the needle is actually inserted, which are pre-marked on the experimental sample by the medical staff using other medical equipment, and the serial number ranking given based on the priority. The preferred injection position and blood vessel running direction noted in the experiment are fed back to the neural network with the selection function in the form of data through scanning and calculation by the system. After multiple use evaluations, the learned neural network with the selection function can output the optimal needle insertion coordinate and needle insertion angle in the form of parameters (p, α0, β0, φ0, d) based on the input data.

[0017] JPEG2024170321000012.jpg23170

[0018] JPEG2024170321000013.jpg26170

[0019] The beneficial effects of the present invention are as follows: (1) By combining a blood flow optimization model with the learning and discrimination functions of a neural network, the present invention automatically selects the optimal needle insertion position on a blood vessel and designs a needle insertion path that imitates the needle insertion technique of professional staff, overcoming the difficulty of finding blood vessels with the naked eye under special circumstances and improving the survival rate of self-help emergency patients in environments where there is a shortage of professional staff. (2) The present invention is useful in avoiding or reducing the suffering caused to humans and animals by repeated failed needle insertion attempts. (3) The present invention is based on the ultrasonic Doppler principle, but only requires the calculation and optimization of spatial coordinates within a local range using relevant data of blood flow velocity, and does not require three-dimensional ultrasonic imaging. This eliminates the huge amount of calculation and development equipment required for three-dimensional ultrasonic imaging, and provides the basis for a novel technical method for realizing an automatic needle insertion device that is small in volume, light in weight and easy to carry. [Brief description of the drawings]

[0020] [Figure 1] FIG. 1 is a schematic diagram showing the overall flow of an automatic needle insertion method that combines a blood flow optimization model and a neural network. [Diagram 2] FIG. 2 is a schematic diagram showing the relationship between angular parameters in a calculation formula for Doppler shift. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] The present invention will now be described in more detail with reference to the accompanying drawings and examples, in which: The specific embodiments described herein are merely illustrative of the invention and are not intended to be limiting of the invention. As shown in Figure 1, the automatic needle insertion method combining the blood flow optimization model and neural network includes three main parts in the entire implementation process: obtaining ultrasonic blood flow information, constructing a blood flow relative velocity optimization model and determining the neural network, and automatic needle insertion. The specific steps are as follows:

[0022] Step 1: Acquiring ultrasonic blood flow information. Using the principle of Doppler ultrasound, the blood vessel position and blood flow velocity distribution in the injection area are calculated based on the blood flow feedback signal measured by automatic ultrasound scanning.

[0023] In order to obtain the variable values ​​required for the blood flow relative velocity optimization model and achieve the purpose of automatic needle insertion, ultrasonic scanning is performed for the area waiting for needle insertion, which can automatically adjust the emission angle and frequency, and further uses the Doppler principle to automatically measure. Ultrasonic scanning can be performed in a mobile or stationary manner. Taking the needle insertion of an arm vein as an example, the mobile manner means that the ultrasonic probe can automatically translate along the arm axis in the parallel cut surface of the skin within a certain distance around the injection site, and can also rotate axially within a certain angle range. Therefore, a holder of an appropriate length must be provided to accommodate the small range of activity of the probe. The stationary probe can achieve focus using phased array, electronically controlled phase difference, and is more applicable to areas on the body with limited moving space. In addition, the ultrasonic emitter needs to be automatically adjusted within a certain range (7.5MHz or more) and emit multiple sets of ultrasonic waves with different frequencies. Since the area to be needled is generally located shallowly under the skin, a relatively high ultrasonic frequency stage can be used, which improves the accuracy of sampling. Ultrasound scanning should be performed when the heart rate is stable, thereby ensuring that the cardiac cycle is relatively stable and avoiding or reducing turbulence in blood flow.

[0024] Considering the privacy protection of the human body that is the needle insertion target, it is necessary to complete the sampling within a limited space. The entire needle insertion device is fixed to the body part by a fixing device and keeps a relative stillness with respect to the area that needs needle insertion, so that the required accuracy of ultrasonic detection and needle insertion can be guaranteed. Selecting a good sampling volume is very important and challenging for obtaining accurate blood flow information, for example, by increasing the number of phased arrays, the density of the sampling grid is improved and the accuracy is improved. However, how to select the sampling volume accurately is not within the consideration scope of the above method of the present invention. In the following calculations, it is assumed that the sampling volume of each point can be well and automatically determined within an extremely short time and a sufficiently accurate measurement result can be obtained.

[0025] According to the installation position of the needle insertion device, a reference coordinate is selected, and the blood vessel position, blood flow velocity and other variables are automatically measured by the Doppler ultrasound principle. The following describes a method for measuring and marking the spatial angle of blood flow direction by using the ultrasonic Doppler principle.

[0026] As shown in Figure 2, the red blood cells move along the X-axis (blood flow direction) at a speed of v. W is the reference axis, and the vertical axis is the direction of gravity. The Z-axis, X-axis and W-axis are on the same plane, the Z-axis is perpendicular to the X-axis, and the intersection angle between the Z-axis and W is φ. The emission frequency of the ultrasonic probe is f0. Ultrasound travels at a sound speed of c in the medium inside the body. s The scattered signal propagates to the target at frequency f r The ultrasonic wave is received at the receiver. The ultrasonic incident plane is the plane P1 where the incident sound beam axis and the Z axis lie, and the axis perpendicular to the Z axis in P1 is Y. The included angle between the ultrasonic incident plane and the blood flow direction is α, that is, the included angle between the X axis and the Y axis is α. The included angle between the incident sound beam axis and the XY plane is β, that is, the included angle between the incident sound beam axis and the Y axis is β. The scattering plane where the scattered sound beam axis and the Z axis along the receiver direction of the red blood cell (group) are located is P2, the axis perpendicular to the Z axis in P2 is Y', the included angle between the Y' axis and the blood flow direction is α', and the included angle between the scattered sound beam axis and the Y' axis is β'. The intersection angle between the incident sound beam axis and the W axis is γ, and the intersection angle between the scattered sound beam axis and the W axis is γ', where the values ​​of γ and γ' can be measured directly. The angle between the incident sound beam axis and the blood flow direction should not be too large, that is, the initial values ​​of α, β, α' and β' are the quantities to be determined, but they should not exceed 60 degrees to avoid inaccurate measurement results. Ultrasonic reception frequency f r The relation between (t+Δt) and the incident frequency f0(t) is as follows: JPEG2024170321000014.jpg94170JPEG2024170321000015.jpg135170JPEG2024170321000016.jpg101170In equation (10), 1 is the distance between the probe and the target detection point, t is the time when the ultrasonic wave is transmitted, and t kindicates the signal reception time when β becomes β+k△β and α remains unchanged. If Δα and Δβ can be precisely controlled, the relative angle parameters can be obtained, and the local relative flow velocity can be obtained from the frequency shift change equation. j and △β k One alternative method for accurately controlling the value of is to transmit a set of ultrasonic pulses with fixed relative angles (which can be easily controlled accurately using a phased array) to each site where blood flow velocity is to be detected. In setting the frequency shift change equation set, the selected frequency shift intensity is higher than a certain threshold, that is, a large amount of low-intensity frequency shift information corresponds to non-blood flow related information, and can be preliminarily eliminated and not considered.

[0027] Step 2: Establish a blood flow relative velocity optimization model, and obtain a set of preferred needle insertion positions and needle insertion angles based on the physical measurements and the blood flow relative velocity optimization model.

[0028] The ideal needle insertion position is considered to be a place near the skin surface where the change in the blood flow direction is gradual (corresponding to a thick vein that does not branch out) and the blood flow rate is relatively large. The five hypotheses are denoted as p1, p2, ..., p5. Here, p1 is the optimal needle insertion point obtained by physical calculation, and p2 is the next best needle insertion point. Let us proceed in order.

[0029] JPEG2024170321000017.jpg75170JPEG2024170321000018.jpg86170JPEG2024170321000019.jpg61170

[0030] JPEG2024170321000020.jpg32170

[0031] JPEG2024170321000021.jpg70170JPEG2024170321000022.jpg64170JPEG2024170321000023.jpg34170

[0032] JPEG2024170321000024.jpg27170

[0033] JPEG2024170321000025.jpg57170

[0034] Step 3: The preferred needle insertion plan obtained in step 2 is combined with a neural network having a selection function to determine the optimal needle insertion point and needle insertion angle, and an optimal needle insertion plan is obtained.

[0035] In order to improve clinical efficacy, a neural network (network 1) with a selection function is trained using optimal values ​​obtained from physical measurements and mathematical optimization models and actual clinical expression scores, and integrated into a single learning chip. A simple example is to generate a convolutional neural network with multiple hidden layers, and use parameters (μ1,α1,β1,φ1)...(μ5,α5,β5,φ5) ​​corresponding to candidate injection points and serial numbers (1,2,3,4,5) as input data for the input layer, and have experienced specialist staff evaluate and feed back the physically calculated needle insertion coordinates and needle insertion angle. The evaluation feedback of the learning group can include the preferred needle insertion position and the blood vessel running direction at the time of actual needle insertion that the specialist staff has marked on the experimental sample in advance using other devices, as well as a ranking given according to priority, such as (2,1,3,5,4). The optimal needle insertion position and blood vessel running direction marked during the experiment are fed back to the neural network in the form of data through scanning and calculation by the system. After multiple rounds of evaluation, the neural network trained with the selection function can provide a better output scheme based on the input data and determine the optimal blood vessel to use for needle insertion, the needle insertion point p, and the exact direction of blood flow (α0, β0, φ0) at that point.

[0036] The above physical calculation, the calculation of the blood flow optimization model, and the learning of the neural network can be integrated into a small chip or a small computing device (rather than an external independent personal computer or a large computer), so that the entire needle insertion device can be easily carried. Therefore, the selection of the optimization model and the selection of the number of nodes and layers of the neural network should meet the practical computing capability.

[0037] It should be further noted that the device required by the present invention utilizes the Doppler ultrasound principle, but has the following essential differences compared with color Doppler blood flow imaging systems: The present invention uses the blood flow optimization model and Doppler principle to calculate the blood flow velocity distribution, calculates the exact position and angle of the sampling volume relative to the reference coordinate, prepares the needle, does not need imaging, has small volume, small weight, and is easy to carry. Color Doppler blood flow imaging is intuitive and qualitative, and the current color Doppler blood flow imaging system requires the cooperation of a display and a computer, and the collected ultrasound signal is processed through digital ultrasound DSC to generate a color image, which consumes a large amount of time and is related to processing a large amount of data required to generate a three-dimensional image.

[0038] Step 4: According to the above optimal needle insertion method, the movement of the needle head is controlled, and the needle head is accurately contacted with the optimal needle insertion point, and needle insertion is started. During the needle insertion process, a neural network with learning function is combined to learn and simulate the needle insertion techniques of professionals, and the needle insertion angle and speed are automatically adjusted to obtain the needle insertion path.

[0039] JPEG2024170321000026.jpg32170

[0040] JPEG2024170321000027.jpg26170

[0041] The present invention uses ultrasound to accurately measure local blood flow information, and combines a blood flow velocity optimization model and a neural network to automatically adjust the needle insertion position and needle insertion angle, and achieves automatic needle insertion by imitating the needle insertion technique of professional staff. The automatic needle insertion technology and method of the present invention provides a technical basis for the production of a portable automatic needle insertion device, and provides convenience for the subsequent use of portable automatic devices to achieve injection and blood collection in an environment with a shortage of medical personnel.

[0042] The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still understand that the technical solutions described in the above embodiments can be modified or some of the technical features can be replaced with equivalents. Any modifications, equivalent replacements, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

【Request 1】 【Request 2】 【Request 3】 【Request 4】 【Request 5】 [Request 6] 【Request 7】

Images