CN112950578B - Vascular identification and positioning method and device based on two-dimensional image enhancement - Google Patents

Abstract

The present invention provides a blood vessel recognition and positioning method and device based on two-dimensional image enhancement, wherein the method includes projecting the original blood vessel image to the blood sampling target area to form a first-stage blood vessel image; obtaining the first-stage blood vessel image and training the projection area part thereof to obtain the projection area ROI semantic segmentation model; inputting the original blood vessel image into the projection area ROI semantic segmentation model to extract the ROI area, and forming a second-stage blood vessel image based on the extracted ROI area; performing enhancement processing based on the second-stage blood vessel image to obtain an optimized blood vessel image, wherein the optimized blood vessel image contains the needle insertion target blood vessel; obtaining the preferred needle insertion point of the needle insertion target blood vessel and the two-dimensional position parameters of the preferred needle insertion point in the first-stage blood vessel image. The present invention can intelligently identify blood vessels and accurately locate the needle insertion point of the blood sampling operation, so that the blood sampling operation can be completed more effectively and accurately, reduce the psychological burden of patients, alleviate the pain of the entire process, and improve the blood sampling experience of patients.

Description

Blood vessel recognition and positioning method and device based on two-dimensional image enhancement

Technical Field

The present invention belongs to the technical field of blood vessel identification and positioning, and in particular relates to a blood vessel identification and positioning method and device based on two-dimensional image enhancement.

Background Art

The current clinical blood collection situation basically relies on the experience of medical staff themselves, who identify blood vessels and determine the needle entry point by observing the person having blood drawn.

However, medical staff cannot guarantee that blood collection is accurate and effective every time. This may be due to the lack of experience of the medical staff themselves, or the thick subcutaneous fat of the person whose blood is being drawn, or the older age of the person whose blood is being drawn or the fragile blood vessels. The success rate of manual blood collection and patient experience have been repeatedly criticized.

Summary of the invention

Therefore, the technical problem to be solved by the present invention is to provide a blood vessel identification and positioning method and device based on two-dimensional image enhancement, which can intelligently identify blood vessels and accurately locate the needle insertion point of the blood collection operation, so that the blood collection operation can be completed more effectively and accurately, and can also reduce the psychological burden of patients and alleviate the pain of the entire process, thereby improving the patient's blood collection experience.

In order to solve the above problems, the present invention provides a blood vessel recognition and positioning method based on two-dimensional image enhancement, comprising:

Acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image onto the blood sampling target area to form a first-stage blood vessel image;

Acquire the blood vessel image of the first stage and obtain a projection region ROI semantic segmentation model based on the projection region part training;

Inputting the original blood vessel image into the projection area ROI semantic segmentation model to extract the ROI area, and forming a second-stage blood vessel image based on the extracted ROI area;

Performing enhancement processing based on the second-stage vascular image to obtain an optimized vascular image, wherein the optimized vascular image contains a target vascular for needle insertion;

The preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage are obtained.

Preferably, the first-stage blood vessel image is acquired and the projection region ROI semantic segmentation model is obtained based on the projection region part training in the image by the following method:

The projection area in the first-stage vascular image is marked as a target area ROI and prepared as a training set. The deeplabV3+ semantic segmentation model of MobilenetV2 is used for model training to obtain the projection area ROI semantic segmentation model.

Preferably, performing enhancement processing based on the second-stage vascular image to obtain an optimized vascular image specifically includes:

The optimized blood vessel image is obtained by sequentially performing grayscale conversion, preprocessing operation, and binarization operation on the second-stage blood vessel image.

Preferably,

The preprocessing operation includes a histogram equalization process with limited contrast and a median filter performed successively; the binarization operation includes an edge detection operation, corrosion expansion, a secondary median filter, and a minimum circumscribed ellipse fitting operation performed successively.

Preferably,

Acquiring the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage specifically includes:

Perform a refinement operation on the target needle insertion vessel, extract the skeleton portion of the target needle insertion vessel, obtain the single-pixel contour of the target needle insertion vessel, and then obtain the center line of the single-pixel contour to determine the direction of the target needle insertion vessel; perform straight line detection on the target needle insertion vessel to obtain the two endpoints of the fitting straight line of the target needle insertion vessel, and use the first endpoint of the two endpoints of the fitting straight line that is in the insertion movement direction of the blood collection needle as the preferred needle insertion point.

The present invention also provides a blood vessel identification and positioning device based on two-dimensional image enhancement, comprising:

An acquisition unit, used for acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image to the blood sampling target area to form a first-stage blood vessel image;

A model training unit, used for acquiring the first-stage blood vessel image and obtaining a projection region ROI semantic segmentation model based on the projection region part of the image;

A ROI region acquisition unit, configured to input the original blood vessel image into the projection region ROI semantic segmentation model to extract a ROI region, and form a second-stage blood vessel image based on the extracted ROI region;

A positioning unit, configured to perform enhancement processing based on the second-stage blood vessel image to obtain an optimized blood vessel image, wherein the optimized blood vessel image contains a target blood vessel for needle insertion;

The preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage are obtained.

Preferably, the first-stage blood vessel image is acquired and the projection region ROI semantic segmentation model is obtained based on the projection region part training in the image by the following method:

The projection area in the first-stage vascular image is marked as a target area ROI and prepared as a training set. The deeplabV3+ semantic segmentation model of MobilenetV2 is used for model training to obtain the projection area ROI semantic segmentation model.

Preferably, performing enhancement processing based on the second-stage vascular image to obtain an optimized vascular image specifically includes:

The optimized blood vessel image is obtained by sequentially performing grayscale conversion, preprocessing operation, and binarization operation on the second-stage blood vessel image.

Preferably,

The preprocessing operation includes a histogram equalization process with limited contrast and a median filter performed successively; the binarization operation includes an edge detection operation, corrosion expansion, a secondary median filter, and a minimum circumscribed ellipse fitting operation performed successively.

Preferably,

Acquiring the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage specifically includes:

Perform a refinement operation on the target needle insertion vessel, extract the skeleton portion of the target needle insertion vessel, obtain the single-pixel contour of the target needle insertion vessel, and then obtain the center line of the single-pixel contour to determine the direction of the target needle insertion vessel; perform straight line detection on the target needle insertion vessel to obtain the two endpoints of the fitting straight line of the target needle insertion vessel, and use the first endpoint of the two endpoints of the fitting straight line that is in the insertion movement direction of the blood collection needle as the preferred needle insertion point.

The present invention provides a blood vessel identification and positioning method and device based on two-dimensional image enhancement, which obtains a first-stage blood vessel image formed by projection, uses a trained projection area ROI semantic segmentation model to extract the ROI area to obtain a second-stage blood vessel image, and enhances the second-stage blood vessel image to obtain an optimized blood vessel image, and finally obtains the two-dimensional position parameters of the target blood vessel for needle insertion and the corresponding preferred needle insertion point, thereby realizing accurate identification and positioning of the target blood vessel for needle insertion and the preferred needle insertion point. This changes the working mode of the prior art that originally relied entirely on the personal experience of medical staff to identify blood vessels and find needle insertion points. The technical solution of the present invention can facilitate the realization of intelligent and automated blood collection, and can bring beneficial effects in improving the success rate of blood collection operations, alleviating the pain of patients, and improving the blood collection experience.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the steps of a method for blood vessel recognition and positioning based on two-dimensional image enhancement according to an embodiment of the present invention;

FIG2 is a schematic diagram of the steps of an image-based blood vessel recognition and positioning method according to another embodiment of the present invention;

FIG. 3 is a schematic diagram of the structure of a blood vessel recognition and positioning device based on two-dimensional image enhancement according to an embodiment of the present invention.

DETAILED DESCRIPTION

With reference to FIGS. 1 to 3 , according to an embodiment of the present invention, a method for identifying and locating blood vessels based on two-dimensional image enhancement is provided, comprising:

Acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image onto the blood sampling target area to form a first-stage blood vessel image;

Acquire the first-stage blood vessel image and obtain a projection region ROI (region of interest) semantic segmentation model based on the projection region part training;

Inputting the original blood vessel image into the projection area ROI semantic segmentation model to extract the ROI area, and forming a second-stage blood vessel image based on the extracted ROI area;

Performing enhancement processing based on the second-stage vascular image to obtain an optimized vascular image, wherein the optimized vascular image contains a target vascular for needle insertion;

The preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage are obtained.

In this technical solution, a first-stage vascular image formed by projection is obtained, and the ROI area is extracted using the trained projection area ROI semantic segmentation model to obtain a second-stage vascular image, and the second-stage vascular image is enhanced to obtain an optimized vascular image, and finally the two-dimensional position parameters of the target blood vessel for needle insertion and the corresponding preferred needle insertion point are obtained, thereby achieving accurate identification and positioning of the target blood vessel for needle insertion and the preferred needle insertion point. This changes the working mode of the prior art that originally relied entirely on the personal experience of medical staff to identify blood vessels and find needle insertion points. The use of the technical solution of the present invention can facilitate the realization of intelligent and automated blood collection, and can bring beneficial effects in improving the success rate of blood collection operations, alleviating the pain of patients, and improving the blood collection experience.

In some embodiments, the first-stage vascular image is obtained and the projection area ROI semantic segmentation model is obtained based on the projection area part of the image. The projection area part of the first-stage vascular image is marked as the target area ROI and prepared as a training set, and the deeplabV3+ semantic segmentation model of MobilenetV2 is used to perform model training to obtain the projection area ROI semantic segmentation model. The projection area ROI semantic segmentation model obtained by training can further improve the recognition accuracy of blood vessels in the image.

In some embodiments, performing enhancement processing based on the second-stage vascular image to obtain the optimized vascular image specifically includes: performing grayscale conversion, preprocessing operations, and binarization operations on the second-stage vascular image in sequence to obtain the optimized vascular image.

Specifically, the grayscale conversion is to grayscale the second-stage blood vessel image, and convert the color three-channel image into a grayscale single-channel image, which can speed up the processing of image data.

The preprocessing operation includes a histogram equalization process with limited contrast and a median filter. This step mainly preprocesses the grayscale image after the grayscale conversion, eliminates the noise caused by the randomness of the photon flux, and denoises the image.

The binarization operation includes edge detection, corrosion expansion, secondary median filtering, and minimum circumscribed ellipse fitting operations. The edge detection can identify the blood vessel part in the projection area and extract it; the corrosion expansion and secondary median filtering can eliminate the noise points in the blood vessel contour image in the extracted blood vessel part; the minimum circumscribed ellipse fitting operation is the aforementioned blood vessel contour. Through this operation, the average length, average width, area, relative position, and factors of the blood vessel contour can be quantitatively analyzed according to the fitting results, and the data of the above four dimensions can be normalized. The preferred index is calculated according to the weight of each factor, and the most suitable blood vessel for needle insertion is found from the blood vessel contour according to the preferred index as the needle insertion target blood vessel.

In some embodiments, obtaining the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameter of the preferred needle entry point in the blood vessel image at the first stage specifically includes:

A thinning operation is performed on the target needle insertion vessel to extract the skeleton portion of the target needle insertion vessel, obtain the single-pixel contour of the target needle insertion vessel, and then obtain the center line of the single-pixel contour to determine the direction of the target needle insertion vessel, which can ensure that the preferred needle insertion point and the needle insertion trajectory are always inside the blood vessel area; a straight line detection is performed on the target needle insertion vessel to obtain the two endpoints of the fitting straight line of the target needle insertion vessel, and the first endpoint of the two endpoints of the fitting straight line that is in the insertion movement direction of the blood collection needle is used as the preferred needle insertion point.

According to an embodiment of the present invention, there is also provided a blood vessel recognition and positioning device based on two-dimensional image enhancement, comprising:

An acquisition unit, used for acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image to the blood sampling target area to form a first-stage blood vessel image;

A model training unit, used for acquiring the first-stage blood vessel image and obtaining a projection region ROI (region of interest) semantic segmentation model based on the projection region part of the image;

A ROI region acquisition unit, configured to input the original blood vessel image into the projection region ROI semantic segmentation model to extract a ROI region, and form a second-stage blood vessel image based on the extracted ROI region;

A positioning unit, configured to perform enhancement processing based on the second-stage blood vessel image to obtain an optimized blood vessel image, wherein the optimized blood vessel image contains a target blood vessel for needle insertion;

The preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage are obtained.

In this technical solution, a first-stage vascular image formed by projection is obtained, and the ROI area is extracted using the trained projection area ROI semantic segmentation model to obtain a second-stage vascular image, and the second-stage vascular image is enhanced to obtain an optimized vascular image, and finally the two-dimensional position parameters of the target blood vessel for needle insertion and the corresponding preferred needle insertion point are obtained, thereby achieving accurate identification and positioning of the target blood vessel for needle insertion and the preferred needle insertion point. This changes the working mode of the prior art that originally relied entirely on the personal experience of medical staff to identify blood vessels and find needle insertion points. The use of the technical solution of the present invention can facilitate the realization of intelligent and automated blood collection, and can bring beneficial effects in improving the success rate of blood collection operations, alleviating the pain of patients, and improving the blood collection experience.

In some embodiments, the first-stage vascular image is obtained and the projection area ROI semantic segmentation model is obtained based on the projection area part of the image. The projection area part of the first-stage vascular image is marked as the target area ROI and prepared as a training set, and the deeplabV3+ semantic segmentation model of MobilenetV2 is used to perform model training to obtain the projection area ROI semantic segmentation model. The projection area ROI semantic segmentation model obtained by training can further improve the recognition accuracy of blood vessels in the image.

In some embodiments, performing enhancement processing based on the second-stage vascular image to obtain the optimized vascular image specifically includes: performing grayscale conversion, preprocessing operations, and binarization operations on the second-stage vascular image in sequence to obtain the optimized vascular image.

Specifically, the grayscale conversion is to grayscale the second-stage blood vessel image, and convert the color three-channel image into a grayscale single-channel image, which can speed up the processing of image data.

The preprocessing operation includes a histogram equalization process with limited contrast and a median filter. This step mainly preprocesses the grayscale image after the grayscale conversion, eliminates the noise caused by the randomness of the photon flux, and denoises the image.

The binarization operation includes edge detection, corrosion expansion, secondary median filtering, and minimum circumscribed ellipse fitting operations. The edge detection can identify the blood vessel part in the projection area and extract it; the corrosion expansion and secondary median filtering can eliminate the noise points in the blood vessel contour image in the extracted blood vessel part; the minimum circumscribed ellipse fitting operation is the aforementioned blood vessel contour. Through this operation, the average length, average width, area, relative position, and factors of the blood vessel contour can be quantitatively analyzed according to the fitting results, and the data of the above four dimensions can be normalized. The preferred index is calculated according to the weight of each factor, and the most suitable blood vessel for needle insertion is found from the blood vessel contour according to the preferred index as the needle insertion target blood vessel.

In some embodiments, obtaining the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameter of the preferred needle entry point in the blood vessel image at the first stage specifically includes:

A thinning operation is performed on the target needle insertion vessel to extract the skeleton portion of the target needle insertion vessel, obtain the single-pixel contour of the target needle insertion vessel, and then obtain the center line of the single-pixel contour to determine the direction of the target needle insertion vessel, which can ensure that the preferred needle insertion point and the needle insertion trajectory are always inside the blood vessel area; a straight line detection is performed on the target needle insertion vessel to obtain the two endpoints of the fitting straight line of the target needle insertion vessel, and the first endpoint of the two endpoints of the fitting straight line that is in the insertion movement direction of the blood collection needle is used as the preferred needle insertion point.

It is easy for those skilled in the art to understand that, under the premise of no conflict, the above-mentioned advantageous methods can be freely combined and superimposed.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the present invention. The above are only preferred embodiments of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and variations can be made without departing from the technical principles of the present invention, and these improvements and variations should also be regarded as the protection scope of the present invention.

Claims (2)

1. A blood vessel recognition and positioning method based on two-dimensional image enhancement, characterized by comprising: Acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image onto the blood sampling target area to form a first-stage blood vessel image; Acquire the blood vessel image of the first stage and obtain a projection region ROI semantic segmentation model based on the projection region part training; Inputting the original blood vessel image into the projection area ROI semantic segmentation model to extract the ROI area, and forming a second-stage blood vessel image based on the extracted ROI area; An optimized vascular image is obtained by performing enhancement processing based on the second-stage vascular image, wherein the optimized vascular image contains a target blood vessel for needle insertion, and the second-stage vascular image is sequentially subjected to grayscale conversion, preprocessing operation, and binarization operation to obtain the optimized vascular image, wherein the binarization operation includes edge detection operation, corrosion expansion, secondary median filtering, and minimum circumscribed ellipse fitting operation performed successively; the edge detection can identify the vascular part in the projection area and extract it; the corrosion expansion and secondary median filtering can eliminate the noise points in the vascular contour image in the extracted vascular part; the operation object of the minimum circumscribed ellipse fitting operation is the aforementioned vascular contour, and through this operation, the average length, average width, area, and relative position factors of the vascular contour can be quantitatively analyzed according to the fitting result, the data of the above four dimensions are normalized, and the preferred index is calculated according to the weight of each factor, and the most suitable blood vessel for needle insertion is found from the vascular contour according to the preferred index as the target blood vessel for needle insertion; Acquiring the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage specifically includes: Performing a thinning operation on the target blood vessel, extracting the skeleton of the target blood vessel, obtaining a single-pixel outline of the target blood vessel, and then obtaining a center line of the single-pixel outline to determine the direction of the target blood vessel; performing straight line detection on the target blood vessel to obtain two endpoints of a fitting straight line of the target blood vessel, and taking the first endpoint of the two endpoints of the fitting straight line that is in the insertion direction of the blood collection needle as the preferred needle insertion point; The first-stage vascular image is obtained and the projection region ROI semantic segmentation model is obtained based on the projection region part training in the projection region image. The following method is used: The projection area in the first-stage blood vessel image is marked as a target area ROI and prepared as a training set, and the deeplabV3+ semantic segmentation model of MobilenetV2 is used for model training to obtain the projection area ROI semantic segmentation model; The preprocessing operation includes a histogram equalization process for limiting contrast and a median filter performed successively. 2. A blood vessel identification and positioning device based on two-dimensional image enhancement, characterized in that it comprises: An acquisition unit, used for acquiring an image of a blood sampling target area to form an original blood vessel image, and projecting the original blood vessel image to the blood sampling target area to form a first-stage blood vessel image; A model training unit, used for acquiring the first-stage blood vessel image and obtaining a projection region ROI semantic segmentation model based on the projection region part of the image; A ROI region acquisition unit, configured to input the original blood vessel image into the projection region ROI semantic segmentation model to extract a ROI region, and form a second-stage blood vessel image based on the extracted ROI region; A positioning unit is used to perform enhancement processing based on the two-stage vascular image to obtain an optimized vascular image, wherein the optimized vascular image contains a target blood vessel for needle insertion, and the two-stage vascular image is sequentially subjected to grayscale conversion, preprocessing operation, and binarization operation to obtain the optimized vascular image, wherein the binarization operation includes an edge detection operation, corrosion expansion, secondary median filtering, and minimum circumscribed ellipse fitting operation performed successively; the edge detection can identify the vascular part in the projection area and extract it; the corrosion expansion and secondary median filtering can eliminate noise points in the vascular contour image in the extracted vascular part; the operation object of the minimum circumscribed ellipse fitting operation is the aforementioned vascular contour, and through this operation, the average length, average width, area, and relative position factors of the vascular contour can be quantitatively analyzed according to the fitting result, the data of the above four dimensions are normalized, and the preferred index is calculated according to the weight of each factor, and the most suitable blood vessel for needle insertion is found from the vascular contour according to the preferred index as the target blood vessel for needle insertion; Acquiring the preferred needle entry point of the needle entry target blood vessel and the two-dimensional position parameters of the preferred needle entry point in the blood vessel image of the first stage specifically includes: Performing a thinning operation on the target blood vessel, extracting the skeleton of the target blood vessel, obtaining a single-pixel outline of the target blood vessel, and then obtaining a center line of the single-pixel outline to determine the direction of the target blood vessel; performing straight line detection on the target blood vessel to obtain two endpoints of a fitting straight line of the target blood vessel, and taking the first endpoint of the two endpoints of the fitting straight line that is in the insertion direction of the blood collection needle as the preferred needle insertion point; The first-stage vascular image is obtained and the projection region ROI semantic segmentation model is obtained based on the projection region part training in the projection region image. The following method is used: The projection area in the first-stage blood vessel image is marked as a target area ROI and prepared as a training set, and the deeplabV3+ semantic segmentation model of MobilenetV2 is used for model training to obtain the projection area ROI semantic segmentation model; The preprocessing operation includes a histogram equalization process for limiting contrast and a median filter performed successively.