Vein puncture target recommendation method based on image processing and artificial intelligence algorithm

Abstract

The invention relates to a venipuncture target recommendation method based on image processing and an artificial intelligence algorithm. The venipuncture target recommendation method comprises the following steps: S1, performing image enhancement processing on an input infrared image; s2, executing a hand image blood vessel segmentation task; s3, judging whether a blood vessel region exists in the hand graph or not; s4, segmenting the center line of the blood vessel, and dividing the blood vessel into a plurality of independent sections according to the bending characteristics of the blood vessel; and S5, summarizing a target spot rule suitable for puncture by utilizing the database model, and determining key characteristics of the puncture target spot according to the target spot rule. According to a blood vessel puncture rule, a linear regression model is established, the weight of each feature of a puncture target is calculated, the score of each potential puncture target is obtained in a weighted summation mode, and the puncture target with the highest score is selected for recommendation. The target spot and the puncture direction which accord with the puncture habit of a nurse can be more accurately selected. And the accuracy of puncture operation is improved, so that the success rate and the safety of puncture are remarkably improved in practical application.

Description

Technical Field

[0001] The present invention belongs to the technical field of venipuncture, and in particular relates to a venipuncture target recommendation method based on image processing and artificial intelligence algorithms. Background Art

[0002] Venipuncture is a common medical procedure that involves inserting a puncture needle through the skin into a vein to draw blood samples for various laboratory tests, such as blood routine tests and biochemical tests. In addition, venipuncture can also be used to establish a venous channel to facilitate the infusion of liquids such as drugs, nutrient solutions, blood products, or intravenous infusion therapy. In current medical practice, venipuncture is mainly performed manually by nurses. When performing venipuncture, medical staff usually choose suitable vein sites, such as the back of the hand, forearm, elbow, etc., where the veins are more obvious and thick and straight, which is convenient for operation. Afterwards, medical staff will disinfect the selected vein site and perform operations such as needle insertion to minimize patient discomfort and reduce the probability of complications.

[0003] At present, some technical solutions have begun to use neural networks and image processing technology to assist in the identification of intravenous puncture targets. The specific steps are as follows: Step 1: Use a binocular camera to obtain images of the left and right arms respectively, and save these images as left camera images and right camera images respectively. Step 2: Perform support vector machine detection, convolutional neural network feature matching and grayscale histogram analysis on the left camera images and the right camera images in turn. Through these steps, the rectangular envelopes, matching feature points and grayscale information of the left and right arms can be obtained. Step 3: Based on the rectangular envelopes, matching feature points and grayscale information of the left and right arms, the optimal matching feature points for intravenous puncture are selected. Then, the three-dimensional coordinates of the optimal puncture target are calculated using the calibration parameters of the binocular camera, thereby realizing the identification of the puncture target.

[0004] However, the current technical solutions have some obvious shortcomings in practical applications, which are specifically manifested in the following aspects:

[0005] 1. First, the existing methods have obvious deficiencies in the target calculation speed. In the process of selecting targets, the accuracy of the recommended targets is not ideal, which may lead to a decrease in the efficiency of the puncture operation and increase the discomfort of the patient. This inefficient calculation process not only wastes precious time, but may also have a negative impact on the patient's treatment effect.

[0006] 2. Secondly, the current technical solutions mainly rely on a single visual algorithm, which may limit the accuracy of recognition. In a complex clinical environment, a single algorithm may not be able to cope with a variety of different situations and challenges, thus affecting the success rate and safety of puncture. For example, in the case of insufficient light or changes in patient position, the recognition ability of a single algorithm may be greatly reduced, resulting in the inability to ensure the accuracy and safety of the puncture operation.

[0007] 3. Finally, the target points obtained by the existing technology are not very consistent with the target points and puncture directions that are in line with the nurses' puncture habits. This may cause the nurses to feel uncomfortable in the actual operation, thus affecting the accuracy and efficiency of the puncture. If the puncture target point and direction cannot match the nurses' operating habits, it may increase the nurses' workload and may even lead to the failure of the puncture operation.

[0008] In summary, the current technical solutions have defects in target calculation speed, algorithm dependence, and target selection that conforms to nurses' operating habits. These problems urgently require further improvements by technical personnel in this field to improve the efficiency, accuracy and safety of puncture operations. Summary of the invention

[0009] In view of the shortcomings of the current related existing technologies, the present invention provides a method for recommending intravenous puncture targets based on image processing and artificial intelligence algorithms. The present invention uses imaging technology and algorithms to perform image processing and calculation on images of the patient's hand area. The present invention can obtain puncture targets that meet the puncture rules of medical staff, thereby improving the puncture success rate, alleviating patient pain, shortening operation time, reducing the risk of complications, and improving medical quality.

[0010] To achieve the above object, the technical solution adopted by the present invention is: a method for recommending venous puncture targets based on image processing and artificial intelligence algorithm, which at least includes the following steps:

[0011] Step S1, performing image enhancement processing on the input infrared image; using a limited contrast adaptive histogram equalization method to enhance the image, thereby improving the contrast of the blood vessel area in the image, thereby optimizing the accuracy of blood vessel segmentation;

[0012] Step S2, performing the hand image blood vessel segmentation task;

[0013] Step S3, determining whether there is a blood vessel region in the hand image; if there is a blood vessel region, further performing a blood vessel centerline extraction operation; in this process, the blood vessels in the connected area are refined to a single pixel width to facilitate feature extraction and clear expression of the target topological structure;

[0014] Step S4, segment the blood vessel centerline and divide it into multiple independent segments according to the curvature of the blood vessel. This process uses a sliding window method to perform statistical analysis on the data in the window and determine the location of the segmentation point by judging the number of non-zero data in the window;

[0015] Step S5: using the database model, summarize the rules of target points suitable for puncture, and clarify the key characteristics of the puncture target points. These characteristics are ranked in order of importance: vascular filling, vascular straightness, vascular diameter and vascular puncture length;

[0016] According to the vascular puncture rules, a linear regression model was established to calculate the weights of each feature of the puncture target. The score of each potential puncture target was obtained by weighted summation, and the puncture target with the highest score was selected for recommendation.

[0017] Furthermore, in step S1, the image is enhanced by using a limited contrast adaptive histogram equalization method to improve the contrast of the blood vessel area in the image, so as to better perform blood vessel segmentation. The method comprises at least the following steps:

[0018] S11, dividing the entire image into multiple small blocks, usually using 8x8 pixel blocks; this step is to process the local area of ​​the image independently to better adapt to the local characteristics of the image;

[0019] S12, for each small block, calculating its histogram, that is, counting the number of pixels at each gray level;

[0020] S13. When calculating the cumulative distribution, a contrast limiting strategy is implemented to prevent the histogram from being particularly steep. The specific operation is: the values ​​in the histogram that are higher than a certain threshold a form a set B, and then each element b in B is changed to a, the parts higher than a are summed to obtain c, c is divided by 256 to obtain d, and d is added to each gray level;

[0021] S14, performing equalization processing on the pruned histogram, that is, calculating the equalized grayscale value according to the cumulative distribution probability of each grayscale level;

[0022] S15, traversing the image blocks and performing bilinear interpolation between blocks to smoothly transition the contrast changes between different blocks;

[0023] S16, mixing the processed image with the original image to obtain the final enhanced image; this step is to retain some details and features of the original image while increasing the contrast and clarity of the image; through the above steps, the limited contrast adaptive histogram equalization method can suppress noise while enhancing the image contrast.

[0024] Furthermore, in step S2, the hand image blood vessel segmentation algorithm is based on the Hessian matrix and implements Frangi filtering to achieve the purpose of blood vessel segmentation; the implementation process of the Frangi filter is as follows:

[0025] S21, image preprocessing stage: denoising the original image and converting it into a grayscale image for subsequent processing;

[0026] S22, Hessian matrix construction: Use the second-order derivative calculation of the image to construct the Hessian matrix to capture the local structural information of the image;

[0027] S23, eigenvalue and eigenvector analysis: Calculate the eigenvalue and eigenvector of the Hessian matrix to further analyze the structural characteristics of the image;

[0028] S24, constructing a vascular response function: constructing a vascular response function based on the analysis results of the eigenvalues ​​and eigenvectors to highlight the vascular structure;

[0029] S25, parameter optimization selection: select appropriate parameters according to image characteristics and requirements to optimize the blood vessel segmentation effect;

[0030] S26, filtering operation and result generation: applying the constructed vascular response function to the image, and finally obtaining an enhanced vascular image through filtering operation.

[0031] Furthermore, in step S3, the blood vessel centerline extraction operation is performed as follows: the blood vessel centerline extraction adopts a skeleton extraction algorithm, which can refine a connected area into a pixel width for feature extraction and target topology representation.

[0032] The algorithm steps are:

[0033] S31, obtain the first address of the original image and the width and height of the image, and set the loop flag 1;

[0034] S32, corroding the original image using the structural element, and saving the corrosion result;

[0035] S33, setting the loop flag to 0, if there is a point in the erosion result that is 255, that is, the original image has not been completely eroded into an empty set, then setting the loop flag to 1;

[0036] S34, using the structure element to perform an opening operation on the eroded image to eliminate the small white area, and calculating the difference between the erosion operation and the opening operation to obtain the eliminated white area;

[0037] S35, performing a union operation on the result obtained in step S34 and the previously obtained skeleton to obtain the skeleton obtained in this cycle;

[0038] S36, assigning the corrosion result saved in this cycle to the original image;

[0039] S37, repeating steps S32 to S36 until the original image is corroded into an empty set;

[0040] This process ensures accurate extraction of the vessel centerline; the resulting skeleton is the result.

[0041] Furthermore, in step S4, the blood vessel centerline is segmented as follows: the blood vessel centerline segmentation mainly decomposes the blood vessel into multiple segments according to its curvature. The segmentation algorithm used is to perform statistical analysis on the data in the window by means of a sliding window, and determine whether a point is a segmentation point according to the number of non-zero data in the window.

[0042] The algorithm steps are:

[0043] S41, setting the window size, sliding step, and non-zero point number threshold;

[0044] S42, calculating the number of non-zero points in the window and determining and recording the segmentation point;

[0045] S43, connecting the segmentation points and arranging them into blood vessel segments.

[0046] Furthermore, in step S5, the implementation method of selecting the puncture target with the highest score for recommendation is as follows:

[0047] S51, collecting relevant data information of the puncture target; the data information includes the fullness, straightness, diameter and puncturable length of the blood vessel where the target is located, and scoring each target;

[0048] S52, establishing a regression model, wherein the model input parameters include the fullness, straightness, diameter and punctureable length of the target blood vessel; the model output is the score of the target; through this model, the collected data can be converted into a specific score value;

[0049] S53. In order to make the regression model more accurate, the collected target data is used to train the regression model, and the weight values ​​of the fullness, straightness, diameter, and punctureable length of the blood vessel where the target is located are obtained through training.

[0050] S54, performing regression analysis on multiple target points calculated in one image according to the weights calculated in step S53, and calculating the score of each target point, thereby providing a basis for the next step of selection;

[0051] S55. After obtaining the scores of all targets, sort the scores of the targets; find the target with the highest score through sorting, so as to determine the target with the most possibility of successful puncture.

[0052] S56. The target with the highest score and its puncture direction and length are recommended as the most reasonable puncture target.

[0053] The present invention adopts the above technical solution, which has at least the following beneficial effects:

[0054] By adopting the above-mentioned technical scheme, the present invention can achieve the following significant beneficial effects: First, compared with the existing technology, the present invention shows higher performance in terms of target calculation speed and target recommendation accuracy. Specifically, the present invention comprehensively uses three different methods, namely image processing technology, lightweight neural network and database model, to recommend targets, thereby significantly improving the processing speed. After careful processing and accurate recognition of the image, the present invention further uses the database model to strictly screen the identified puncture targets. Compared with the traditional single visual algorithm, the present invention can more accurately select targets and puncture directions that meet the nurses' puncture habits. This optimization not only improves the accuracy of the puncture operation, but also can better adapt to the nurses' operating habits, thereby significantly improving the success rate and safety of the puncture in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of the method for recommending target points for venous puncture of the present invention. DETAILED DESCRIPTION

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0058] See also Figure 1 This embodiment provides a method for recommending a target for venous puncture based on image processing and artificial intelligence algorithm, which at least includes the following steps:

[0059] Step S1, first perform image enhancement processing on the input infrared image; in this process, the image is enhanced by using the limited contrast adaptive histogram equalization method to improve the contrast of the blood vessel area in the image. This operation helps to optimize the accuracy of blood vessel segmentation, thereby providing clearer image data for subsequent steps.

[0060] Step S2, performing the hand image blood vessel segmentation task; using image processing technology, the blood vessel area in the image is separated from the background for further analysis and processing.

[0061] Step S3: After completing the blood vessel segmentation, it is necessary to determine whether there is a blood vessel area in the hand figure; if there is a blood vessel area, the blood vessel centerline extraction operation is further performed; in this process, the blood vessels in the connected area are refined to a single pixel width to facilitate feature extraction and clear expression of the target topological structure; this refinement operation helps to more accurately identify the centerline of the blood vessel and provide a basis for subsequent blood vessel analysis.

[0062] Step S4: segment the blood vessel centerline and divide it into multiple independent segments according to the curvature of the blood vessel. This process uses a sliding window method to perform statistical analysis on the data in the window and determine the location of the segmentation point by judging the number of non-zero data in the window. In this way, the complex blood vessel network can be decomposed into smaller segments that are easier to handle, thereby improving the accuracy of puncture target recommendation.

[0063] Step S5, using the database model (neural network), summarize the rules of target points suitable for puncture, and clarify the key characteristics of the puncture target points accordingly. A series of rules can be extracted, which are helpful to clarify the key characteristics of the puncture target points. These characteristics are ranked in order of importance: vascular filling, vascular straightness, vascular diameter and vascular puncture length; in order to more accurately determine the best puncture target point, these characteristics are comprehensively evaluated and ranked according to their importance. First, vascular filling is crucial because it is directly related to the success rate and safety of puncture. Secondly, vascular straightness is also very important because straight blood vessels are easier to puncture, reducing the complexity and risk of operation. Then, the diameter of the blood vessel is also a key factor. The appropriate diameter can ensure that the puncture needle can enter the blood vessel smoothly, without being unable to puncture due to being too thin, or causing unnecessary damage due to being too thick. Finally, the puncture length of the blood vessel is also an important characteristic that needs to be considered. A longer puncture length can provide more operating space, thereby improving the success rate of puncture. Through the comprehensive evaluation of these characteristics, the optimal puncture target can be determined more accurately, thereby improving the efficiency and safety of the entire puncture process.

[0064] According to the vascular puncture rules, a linear regression model is established to calculate the weights of each feature of the puncture target. Through weighted summation, the score of each potential puncture target is obtained, and the puncture target with the highest score is selected for recommendation. This scoring mechanism ensures that the recommended target has the highest puncture success rate and safety.

[0065] The auxiliary puncture device in this embodiment can help medical staff better find the blood vessel area and recommend puncture targets and puncture directions suitable for puncture to medical staff. Medical staff can reduce the possibility of secondary puncture by puncturing according to the targets given by the equipment, thereby improving the puncture success rate and the work efficiency of medical staff. In addition, this method also helps to reduce the possibility of local infection, systemic infection, hematoma, blood vessel rupture, nerve damage, etc. in patients due to puncture, thereby improving overall medical safety and patient satisfaction.

[0066] As an implementation mode, in step S1, a limited contrast adaptive histogram equalization method is used to enhance the image to improve the contrast of the blood vessel area in the image, so as to better perform blood vessel segmentation. The method includes at least the following steps:

[0067] S11. In image processing, in order to better adapt to the local characteristics of the image, the entire image is divided into multiple small blocks, usually 8x8 pixel blocks; this step is to process the local area of ​​the image independently in order to better adapt to the local characteristics of the image;

[0068] S12, for each small block, calculating its histogram, that is, counting the number of pixels at each gray level;

[0069] S13. In order to prevent the histogram from being particularly steep, a contrast limitation strategy is implemented when calculating the cumulative distribution. The specific operation is: the values ​​in the histogram that are higher than a certain threshold a form a set B, and then each element b in B is changed to a, and the parts higher than a are summed to obtain c, c is divided by 256 to obtain d, and d is added to each gray level;

[0070] S14, performing equalization processing on the pruned histogram, that is, calculating the equalized grayscale value according to the cumulative distribution probability of each grayscale level;

[0071] S15, traversing the image blocks and performing bilinear interpolation between blocks to smoothly transition the contrast changes between different blocks;

[0072] S16, mixing the processed image with the original image to obtain the final enhanced image; this step is to retain some details and features of the original image while increasing the contrast and clarity of the image; through the above steps, the limited contrast adaptive histogram equalization method can suppress noise while enhancing the image contrast.

[0073] As an implementation method, in step S2 of this embodiment, the hand image blood vessel segmentation algorithm is based on the Hessian matrix and implements Frangi filtering to achieve the purpose of blood vessel segmentation; the implementation process of the Frangi filter is as follows:

[0074] S21, image preprocessing stage: denoising the original image and converting it into a grayscale image for subsequent processing;

[0075] S22, Hessian matrix construction: Use the second-order derivative calculation of the image to construct the Hessian matrix to capture the local structural information of the image;

[0076] S23, eigenvalue and eigenvector analysis: Calculate the eigenvalue and eigenvector of the Hessian matrix to further analyze the structural characteristics of the image;

[0077] S24, constructing a vascular response function: constructing a vascular response function based on the analysis results of the eigenvalues ​​and eigenvectors to highlight the vascular structure;

[0078] S25, parameter optimization selection: select appropriate parameters according to image characteristics and requirements to optimize the blood vessel segmentation effect;

[0079] S26, filtering operation and result generation: applying the constructed vascular response function to the image, and finally obtaining an enhanced vascular image through filtering operation.

[0080] As an implementation mode, in step S3 of this embodiment, the blood vessel centerline extraction operation is performed in the following manner: the blood vessel centerline extraction adopts a skeleton extraction algorithm, which can refine a connected area into a pixel width for feature extraction and target topology representation;

[0081] The algorithm steps are:

[0082] S31, obtain the first address of the original image and the width and height of the image, and set the loop flag 1;

[0083] S32, corroding the original image using the structural element, and saving the corrosion result;

[0084] S33, setting the loop flag to 0, if there is a point in the erosion result that is 255, that is, the original image has not been completely eroded into an empty set, then setting the loop flag to 1;

[0085] S34, using the structure element to perform an opening operation on the eroded image to eliminate the small white area, and calculating the difference between the erosion operation and the opening operation to obtain the eliminated white area;

[0086] S35, performing a union operation on the result obtained in step S34 and the previously obtained skeleton to obtain the skeleton obtained in this cycle;

[0087] S36, assigning the corrosion result saved in this cycle to the original image;

[0088] S37, repeating steps S32 to S36 until the original image is corroded into an empty set;

[0089] This process ensures accurate extraction of the vessel centerline; the resulting skeleton is the result.

[0090] As an implementation mode, in step S4 of this embodiment, the specific implementation mode of segmenting the blood vessel centerline is as follows: the blood vessel centerline segmentation mainly decomposes the blood vessel into multiple segments according to its curvature, and the segmentation algorithm used is to perform statistical analysis on the data in the window by means of a sliding window, and judge whether a point is a segmentation point according to the number of non-zero data in the window;

[0091] The algorithm steps are:

[0092] S41, setting the window size, sliding step, and non-zero point number threshold;

[0093] S42, calculating the number of non-zero points in the window and determining and recording the segmentation point;

[0094] S43, connecting the segmentation points and arranging them into blood vessel segments.

[0095] As an implementation mode, in this embodiment, the method of selecting the puncture target with the highest score for recommendation in step S5 is as follows:

[0096] S51, collecting relevant data information of the puncture target; the data information includes the fullness, straightness, diameter and puncturable length of the blood vessel where the target is located, and scoring each target;

[0097] S52, establishing a regression model, wherein the model input parameters include the fullness, straightness, diameter and punctureable length of the target blood vessel; the model output is the score of the target; through this model, the collected data can be converted into a specific score value;

[0098] S53. In order to make the regression model more accurate, the collected target data is used to train the regression model, and the weight values ​​of the fullness, straightness, diameter, and punctureable length of the blood vessel where the target is located are obtained through training.

[0099] S54, performing regression analysis on multiple target points calculated in one image according to the weights calculated in step S53, and calculating the score of each target point, thereby providing a basis for the next step of selection;

[0100] S55. After obtaining the scores of all targets, sort the scores of the targets; find the target with the highest score through sorting, so as to determine the target with the most possibility of successful puncture.

[0101] S56. The target with the highest score and its puncture direction and length are recommended as the most reasonable puncture target.

[0102] Compared with the existing technology, the present invention shows higher performance in terms of target calculation speed and target recommendation accuracy. Specifically, the present invention uses a combination of three different methods, namely image processing technology, neural network and database model, to recommend targets, thereby significantly improving the processing speed. After careful processing and accurate identification of the image, the present invention further uses the database model to strictly screen the identified puncture targets. Compared with the traditional single visual algorithm, the present invention can more accurately select targets and puncture directions that meet the nurses' puncture habits. This optimization not only improves the accuracy of the puncture operation, but also can better adapt to the nurses' operating habits, thereby significantly improving the success rate and safety of the puncture in practical applications.

[0103] Although the embodiments of the present invention have been shown and described above, it will be appreciated that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, substitute and vary the above embodiments within the scope of the present invention.

Claims

1. A method for recommending venipuncture target points based on image processing and artificial intelligence algorithms, characterized in that: At least the following steps are included: Step S1, performing image enhancement processing on the input infrared image; using a limited contrast adaptive histogram equalization method to enhance the image, thereby improving the contrast of the blood vessel area in the image, thereby optimizing the accuracy of blood vessel segmentation; Step S2, performing the hand image blood vessel segmentation task; Step S3, determining whether there is a blood vessel region in the hand image; if there is a blood vessel region, further performing a blood vessel centerline extraction operation; in this process, the blood vessels in the connected area are refined to a single pixel width to facilitate feature extraction and clear expression of the target topological structure; Step S4, segment the blood vessel centerline and divide it into multiple independent segments according to the curvature of the blood vessel. This process uses a sliding window method to perform statistical analysis on the data in the window and determine the location of the segmentation point by judging the number of non-zero data in the window; Step S5: using the database model, summarize the rules of target points suitable for puncture, and clarify the key characteristics of the puncture target points. These characteristics are ranked in order of importance: vascular filling, vascular straightness, vascular diameter and vascular puncture length; According to the vascular puncture rules, a linear regression model was established to calculate the weights of each feature of the puncture target. The score of each potential puncture target was obtained by weighted summation, and the puncture target with the highest score was selected for recommendation.

2. The method according to claim 1, characterized in that: In step S1, the image is enhanced by using a limited contrast adaptive histogram equalization method to improve the contrast of the blood vessel area in the image, so as to better perform blood vessel segmentation. The method includes at least the following steps: S11, dividing the entire image into multiple small blocks, using 8x8 pixel blocks; this step is to process the local area of ​​the image independently, so as to better adapt to the local characteristics of the image; S12, for each small block, calculating its histogram, that is, counting the number of pixels at each gray level; S13. When calculating the cumulative distribution, a contrast limiting strategy is implemented to prevent the histogram from being particularly steep. The specific operation is: the values ​​in the histogram that are higher than a certain threshold a form a set B, and then each element b in B is changed to a, the parts higher than a are summed to obtain c, c is divided by 256 to obtain d, and d is added to each gray level; S14, performing equalization processing on the pruned histogram, that is, calculating the equalized grayscale value according to the cumulative distribution probability of each grayscale level; S15, traversing the image blocks and performing bilinear interpolation between blocks to smoothly transition the contrast changes between different blocks; S16, mixing the processed image with the original image to obtain the final enhanced image; this step is to retain some details and features of the original image while increasing the contrast and clarity of the image; through the above steps, the limited contrast adaptive histogram equalization method can suppress noise while enhancing the image contrast.

3. The method according to claim 2, characterized in that: In step S2, the hand image blood vessel segmentation algorithm is based on the Hessian matrix and implements Frangi filtering to achieve the purpose of blood vessel segmentation; the implementation process of the Frangi filter is as follows: S21, image preprocessing stage: denoising the original image and converting it into a grayscale image for subsequent processing; S22, Hessian matrix construction: Use the second-order derivative calculation of the image to construct the Hessian matrix to capture the local structural information of the image; S23, eigenvalue and eigenvector analysis: Calculate the eigenvalue and eigenvector of the Hessian matrix to further analyze the structural characteristics of the image; S24, constructing a vascular response function: constructing a vascular response function based on the analysis results of the eigenvalues ​​and eigenvectors to highlight the vascular structure; S25, parameter optimization selection: select appropriate parameters according to image characteristics and requirements to optimize the blood vessel segmentation effect; S26, filtering operation and result generation: applying the constructed vascular response function to the image, and finally obtaining an enhanced vascular image through filtering operation.

4. The method according to claim 3, characterized in that: In step S3, the blood vessel centerline extraction operation is performed as follows: the blood vessel centerline extraction adopts a skeleton extraction algorithm, which can refine a connected area into a pixel width for feature extraction and target topology representation. The algorithm steps are: S31, obtain the first address of the original image and the width and height of the image, and set the loop flag 1; S32, corroding the original image using the structural element, and saving the corrosion result; S33, setting the loop flag to 0, if there is a point in the erosion result that is 255, that is, the original image has not been completely eroded into an empty set, then setting the loop flag to 1; S34, using the structure element to perform an opening operation on the eroded image to eliminate the small white area, and calculating the difference between the erosion operation and the opening operation to obtain the eliminated white area; S35, performing a union operation on the result obtained in step S34 and the previously obtained skeleton to obtain the skeleton obtained in this cycle; S36, assigning the corrosion result saved in this cycle to the original image; S37, repeating steps S32 to S36 until the original image is corroded into an empty set; This process ensures accurate extraction of the vessel centerline; the resulting skeleton is the result.

5. The method according to claim 4, characterized in that: In step S4, the blood vessel centerline is segmented as follows: the blood vessel centerline segmentation mainly decomposes the blood vessel into multiple segments according to its curvature. The segmentation algorithm used is to perform statistical analysis on the data in the window by means of a sliding window, and determine whether a point is a segmentation point according to the number of non-zero data in the window. The algorithm steps are: S41, setting the window size, sliding step, and non-zero point number threshold; S42, calculating the number of non-zero points in the window and determining and recording the segmentation point; S43, connecting the segmentation points and arranging them into blood vessel segments.

6. The method according to claim 5, characterized in that: The implementation method of selecting the puncture target with the highest score for recommendation in step S5 is as follows: S51, collecting relevant data information of the puncture target; the data information includes the fullness, straightness, diameter and puncturable length of the blood vessel where the target is located, and scoring each target; S52, establishing a regression model, wherein the model input parameters include the fullness, straightness, diameter and punctureable length of the target blood vessel; the model output is the score of the target; through this model, the collected data can be converted into a specific score value; S53. In order to make the regression model more accurate, the collected target data is used to train the regression model, and the weight values ​​of the fullness, straightness, diameter, and punctureable length of the blood vessel where the target is located are obtained through training. S54, performing regression analysis on multiple target points calculated in one image according to the weights calculated in step S53, and calculating the score of each target point, thereby providing a basis for the next step of selection; S55. After obtaining the scores of all targets, sort the scores of the targets; find the target with the highest score through sorting, so as to determine the target with the most possibility of successful puncture. S56. The target with the highest score and its puncture direction and length are recommended as the most reasonable puncture target.

Images