Blood vessel detection device and blood vessel detection method based on image

Through the multi-stage machine learning detection model combined with time series information, the problem of subjective deviation and insufficient accuracy in existing vascular detection technologies is solved, and more efficient vascular detection and diagnostic results are achieved.

CN114596247BActive Publication Date: 2024-11-26ACER INC +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110074836.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-11-30
Filing Date
2021-01-20
Publication Date
2024-11-26
Estimated Expiration
2041-01-20

AI Technical Summary

Technical Problem

The existing vascular detection technology has subjective factors that lead to deviations in diagnostic results, which consumes time and money, and cannot effectively process large-scale information in vascular images, resulting in insufficient recognition accuracy.

Method used

Using a multi-stage vascular detection method based on machine learning, medical images are detected through the first detection model and the second detection model, and the characteristics of pixels are analyzed using machine learning algorithms, combined with time series information, to improve recognition accuracy and reduce false positives. .

Benefits of technology

It improves the accuracy and recognition efficiency of vascular detection, reduces subjective deviations during the diagnosis process, enhances the continuity and accuracy of vascular contours, and reduces the false positive rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114596247B_ABST
    Figure CN114596247B_ABST
Patent Text Reader

Abstract

The present invention provides a blood vessel detection device and an image-based blood vessel detection method. In the method, the first data to be evaluated is detected by a first detection model to obtain a first detection result; the second data to be evaluated is detected by a second detection model to obtain a second detection result. The first data to be evaluated includes one or more medical images obtained by photographing the blood vessel, and the first detection result output by the first detection model includes one or more pixels in the medical image belonging to the blood vessel. The first and second detection models are constructed based on a machine learning algorithm, the second data to be evaluated includes the first detection result, and the second detection result output by the second detection model includes one or more pixels in the medical image belonging to the blood vessel. Thereby, the accuracy of blood vessel recognition can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an image recognition technology, and in particular to a blood vessel detection device and an image-based blood vessel detection method. Background Art

[0002] Vascular images are an important basis for determining the lesions of many diseases. When doctors make diagnoses based on vascular images, they often include subjective factors, which leads to deviations in the diagnosis results. In addition, manual interpretation is also a time-consuming and costly process. From the patient's perspective, the entire diagnostic process is like a black box. Patients cannot understand how the diagnosis results are obtained, which leads to disputes due to poor communication.

[0003] Existing blood vessel segmentation technologies mainly include calculation methods. The calculation method analyzes the intensity of each pixel on the image. For example, the second-order partial differential of the pixel with respect to the two dimensions of the image is calculated, and then the eigenvector and eigenvalue of the obtained second-order partial differential matrix are used to determine whether the pixel belongs to a blood vessel. However, due to the large number of blood vessel styles, and the fact that the blood vessels in vascular medical images are likely to be pathological, the degree of blood vessel difference is increased, making it impossible to cover various blood vessel types through the determination of eigenvectors and eigenvalues. In addition, this method is limited to small-scale determination, and does not consider large-scale information. Summary of the invention

[0004] The present invention is directed to a blood vessel detection device and a blood vessel detection method based on an image, which can accurately detect blood vessels in a blood vessel image.

[0005] According to an embodiment of the present invention, an image-based blood vessel detection method includes (but is not limited to) the following steps: detecting first data to be evaluated by a first detection model to obtain a first detection result; detecting second data to be evaluated by a second detection model to obtain a second detection result. The first detection model is constructed based on a machine learning algorithm, the first data to be evaluated includes one or more medical images taken of blood vessels, and the first detection result output by the first detection model includes one or more pixels in the medical image belonging to the blood vessel. The second detection model is constructed based on a machine learning algorithm, the second data to be evaluated includes the first detection result, and the second detection result output by the second detection model includes one or more pixels in the medical image belonging to the blood vessel.

[0006] According to an embodiment of the present invention, a blood vessel detection device includes (but is not limited to) a memory and a processor. The memory stores program codes. The processor is coupled to the memory, and is configured to load and execute those program codes to detect the first data to be evaluated through a first detection model to obtain a first detection result, and detect the second data to be evaluated through a second detection model to obtain a second detection result. The first detection model is constructed based on a machine learning algorithm, the first data to be evaluated includes one or more medical images obtained by taking a blood vessel, and the first detection result output by the first detection model includes one or more pixels in the medical image belonging to the blood vessel. The second detection model is constructed based on a machine learning algorithm, the second data to be evaluated includes the first detection result, and the second detection result output by the second detection model includes one or more pixels in the medical image belonging to the blood vessel.

[0007] Based on the above, the blood vessel detection device and the image-based blood vessel detection method according to the embodiments of the present invention can provide a multi-stage blood vessel detection mechanism and allow previous detection results to be used as input data for subsequent detection models to improve recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings are included to provide a further understanding of the present invention and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the present invention and together with the description serve to explain the principles of the present invention.

[0009] Figure 1 is a component block diagram of a blood vessel detection device according to an embodiment of the present invention;

[0010] Figure 2 is a flow chart of an image-based blood vessel detection method according to an embodiment of the present invention;

[0011] Figure 3 is a schematic diagram of a flow chart of a blood vessel detection method according to an embodiment of the present invention;

[0012] Figure 4A and Figure 4B An example of comparing the test results is shown;

[0013] Figure 5 is a schematic diagram of target tracking based on time series according to an embodiment of the present invention;

[0014] Figure 6A to Figure 6C Shows a comparison example of medical images and test results;

[0015] Figure 7 is a schematic diagram of a flow chart of a main blood vessel detection method according to an embodiment of the present invention;

[0016] Figure 8 An example of major blood vessel detection is shown;

[0017] Fig. 9 is a schematic diagram of a flow chart of a target blood vessel detection method according to an embodiment of the present invention;

[0018] Fig.10 Examples showing different projection positions;

[0019] Fig.11 An example of blood vessel tip detection is shown.

[0020] Description of Figure Numbers

[0021] 100: blood vessel detection device;

[0022] 110: memory;

[0023] 130: processor;

[0024] 150: Display;

[0025] S210-S230, S301, S710, S901-S905: steps;

[0026] D1~D3: data to be evaluated;

[0027] M1~M3: detection model;

[0028] R1~R7, R31~R33: test results;

[0029] SD1–SD5: Supplementary data;

[0030] R11, R12, R21, R22: segmented image;

[0031] Sk: backbone;

[0032] O11~O13, MI: medical images;

[0033] Vh, Lh1, Lh2: vascular head;

[0034] P: position;

[0035] LCX: left circumflex artery;

[0036] LAD-proximal: left anterior descending artery-proximal;

[0037] Diagonal: diagonal branch;

[0038] OM: obtuse marginal ramus;

[0039] RCA: right coronary artery;

[0040] AcM: acute marginal branch;

[0041] Box: Tracking area;

[0042] FP: False Positive. DETAILED DESCRIPTION

[0043] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0044] Figure 1 FIG. 1 is a block diagram of components of a blood vessel detection device 100 according to an embodiment of the present invention. Figure 1 The blood vessel detection device 100 includes (but is not limited to) a memory 110 and a processor 130. The blood vessel detection device 100 may be a desktop computer, a laptop computer, a smart phone, a tablet, a server, a medical detection instrument, or other computing devices.

[0045] The memory 110 may be any type of fixed or removable random access memory (RAM), read only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or similar components. In one embodiment, the memory 110 is used to record program codes, software modules, configurations, data, or files.

[0046] The processor 130 is coupled to the memory 110, and the processor 130 may be a central processing unit (CPU), a graphics processing unit (GPU), or other programmable general-purpose or special-purpose microprocessor, digital signal processor (DSP), programmable controller, field programmable gate array (FPGA), application-specific integrated circuit (ASIC), neural network accelerator or other similar components or a combination of the above components. In one embodiment, the processor 130 is used to execute all or part of the operations of the blood vessel detection device 100, and can load and execute various program codes, software modules, files and data recorded in the memory 110.

[0047] In some embodiments, the blood vessel detection device 100 further includes a display 150. The display 150 may be a liquid crystal display (LCD), a light-emitting diode (LED), an organic light-emitting diode (OLED), a quantum dot display, or other types of displays. In one embodiment, the display 150 is used to display medical images or processed medical images.

[0048] Hereinafter, the method described in the embodiment of the present invention will be described in conjunction with various devices, components and modules in the blood vessel detection device 100. Each process of the method can be adjusted according to the implementation situation, and is not limited thereto.

[0049] Figure 2 FIG. 1 is a flow chart of an image-based blood vessel detection method according to an embodiment of the present invention. Figure 2 , the processor 130 is configured to detect the first data to be evaluated through the first detection model to obtain a first detection result (step S210). Specifically, the first detection model is constructed based on a machine learning algorithm. The machine learning algorithm can be a convolutional neural network (CNN), a recurrent neural network (RNN), a multi-layer perceptron (MLP), a support vector machine (SVM) or other algorithms. The machine learning algorithm can analyze the training samples to obtain rules from them, so as to predict unknown data through the rules. The detection model is the machine learning model constructed after learning, and is used to infer the data to be evaluated. In one embodiment, the first detection model uses the medical images (or vascular images) taken of the blood vessels and the results marked as belonging to the blood vessels as training samples.

[0050] The first data to be evaluated includes one or more medical images obtained by photographing one or more blood vessels. It should be noted that the embodiment of the present invention does not limit the body part where the blood vessels are located. In addition, the first detection model is configured to analyze the first data to be evaluated and output a first detection result based on the analysis. The first detection result can be that one or more pixels in the medical image belong to a blood vessel or do not belong to a blood vessel.

[0051] In another embodiment, the first data to be evaluated may also include other features of the blood vessel, such as contour, size, or edge, etc. In some embodiments, the first data to be evaluated may also include features of medical images taken of the same part at different time points.

[0052] The processor 130 is configured to detect the second data to be evaluated through the second detection model to obtain a second detection result (step S230). Specifically, the second detection model is also constructed based on the machine learning algorithm. The first and second detection models may use the same or different machine learning algorithms, and the present invention is not limited. The difference from the first data to be evaluated is that the second data to be evaluated used as input data for the second detection model includes the first detection result output by the first detection model. In addition, the second detection result output by the second detection model may be that one or more pixels in the medical image belong to a blood vessel or do not belong to a blood vessel.

[0053] In one embodiment, the processor 130 may perform image processing on the medical image, and the image processing is related to enhancing the contour of the blood vessel. For example, edge enhancement processing, contrast adjustment or other processing to improve contour recognition. The second data to be evaluated may also include medical images that have been image processed. Thereby, combining the features of the calculation method (for example, improving contour recognition) and the learning method (for example, machine learning) will improve the accuracy of the recognition result with less data. In addition, enhancing the contour of the blood vessel will help reduce the situation where the blood vessel detection area is discontinuous.

[0054] For example, Figure 3 FIG. 1 is a flow chart of a blood vessel detection method according to an embodiment of the present invention. Figure 3 , a medical image is used as the data to be evaluated D1 and input into the detection model M1 to obtain the detection result R1. Then, the detection result R1 and two supplementary data SD1, SD2 (for example, the original medical image and the edge-enhanced image) are used as the data to be evaluated D2 and input into the detection model M2 to obtain the detection result R2.

[0055] In another embodiment, the second data to be evaluated may also include other features of the blood vessel, such as contour, size, thickness, etc. In some embodiments, the second data to be evaluated may also include features of medical images taken of the same part at different time points.

[0056] In some embodiments, the processor 130 may segment one or more pixels belonging to blood vessels in each medical image according to the first or second detection result of the medical image to form a segmented image. For example, the grayscale of the pixels belonging to the blood vessels is minimized by binarization, and the grayscale of the pixels not belonging to the blood vessels is maximized.

[0057] Figure 4A and Figure 4BAn example of comparing test results is shown. Figure 4A and Figure 4B , the left figure shows segmented images R21 and R22 formed according to the detection result R2, and the right figure shows segmented images R11 and R12 formed according to the detection result R1. As can be seen from the box in the figure, the segmented images R11 and R12 have discontinuous blood vessels, but the blood vessel contours in the segmented images R21 and R22 are continuous.

[0058] Please refer to Figure 3 , the processor 130 may perform time series-based background removal on the detection result R2 (step S301) to obtain a detection result R3. In one embodiment, those medical images are sorted according to a time series. That is, medical images taken at the same projection position and / or angle but at different time points (or medical images of different frames). The processor 130 may segment one or more vascular regions belonging to the blood vessels in each medical image based on the second detection results of those medical images to obtain multiple segmented images of those medical images corresponding to different frames.

[0059] It is worth noting that when interpreting medical images, doctors often refer to the changes in frames in different time series to improve the accuracy of the interpretation results. The processor 130 can determine the displacement of each pixel of each blood vessel region in the segmented image according to the time series. For example, the processor 130 can use optical flow or other target tracking technology to analyze the displacement (i.e., dynamic change) of each pixel between consecutive frames.

[0060] The processor 130 can classify those vascular regions according to the displacement corresponding to each vascular region and the comparison result of the vascular region in the segmented image of the previous frame. The comparison result is related to the overlap size of the displacement and those vascular regions. For example, the processor 130 can use the obtained displacement to compare with the vascular region in the segmented image of the previous or previous frame. The comparison result is that the two vascular regions with the largest overlap in the position of the displacement (or trajectory) will be classified into the same category (for example, combined into one region), and then the change in the size of each vascular region in the time series is obtained. In this way, the background noise in the segmented image can be removed (for example, regions of other categories are regarded as noise). That is, the segmented image of the detection result R3 has less background noise than the segmented image of the detection result R2.

[0061] For example, Figure 5 FIG. 1 is a schematic diagram of target tracking based on time series according to an embodiment of the present invention. Figure 5, using optical flow background removal, the shape of the blood vessel area changes in different frames. In this way, the complete blood vessel contour can be gradually segmented according to the time series, and the noise outside the blood vessel contour can be removed. However, if optical flow background removal is not used, there are still some spots outside the blood vessel contour.

[0062] Figure 6A to Figure 6C An example of comparing medical images O11 to O13 and detection results R31 to R33 is shown. FIG. 6A to FIG. 6C Taking the blood vessel segmentation in cardiovascular photography as an example, the blood vessel contours of the segmented images obtained by the detection results R31~R33 are quite close to the medical images O11~O13, and the blood vessels are continuous and uninterrupted.

[0063] It is worth noting that there may be many vascular branches in the segmented image, and some of the branches are not helpful for diagnosis. For the main blood vessels, in one embodiment, the processor 130 can detect the third data to be evaluated through the third detection model to obtain a third detection result. Specifically, the third detection model is also constructed based on one of those machine learning algorithms. The difference from the first and second data to be evaluated is that the third data to be evaluated includes the second detection result of the medical image of the current frame and the second detection result of the medical image of the previous frame (i.e., the output of the previous order detection model). Similarly, based on the correlation of blood vessels in the time series, the medical images of the previous and next frames will help blood vessel detection. The third detection result output by the third detection model includes the main blood vessels in the medical image. In the learning stage, the third detection model is trained based on a variety of training samples marked as specific main blood vessels (for example, the left circumflex branch of the coronary artery, the left anterior descending branch, etc.).

[0064] For example, Figure 7 FIG. 1 is a flow chart of a method for detecting main blood vessels according to an embodiment of the present invention. Figure 7 ,exist Figure 3 After the detection result R3 is obtained, the detection result R3 and three supplementary data SD3-SD5 (for example, the detection result R3 of the previous frame, the previous and next frames after optical flow extrapolation and blurring plus the boundary of the current frame, and the original medical image of the current frame) are used as the data to be evaluated D3 and input into the detection model M3 to obtain the detection result R4. At this time, the segmented image corresponding to the detection result R4 only includes the main blood vessels, and other branches will be regarded as background and removed.

[0065] It should be noted that the main blood vessel can be determined based on the weight of evaluating the complexity of the lesions of multiple vascular segments. Taking the syntax score as an example, the right coronary artery can be divided into the following vascular segments: the proximal right coronary artery, the middle right coronary artery, the distal right coronary artery, the right coronary-posterior descending branch, the right coronary-posterior lateral branch, etc. The complexity of the lesion is quantitatively evaluated according to the anatomical characteristics such as the location, severity, bifurcation, and calcification of the lesion, and each vascular segment is assigned a corresponding weight. The size of these weights is related to the identification of the main blood vessels. For example, those with a weight greater than a specific threshold value (for example, 0.8, 1, or 1.2, etc.) are the main blood vessels. Those with a weight less than a specific threshold value are not the main blood vessels. However, the main blood vessels are not limited to the coronary arteries.

[0066] In some embodiments, the processor 130 may further perform time series-based background removal (step S710 ) on the detection result R4 as described in step S301 to remove additional noise.

[0067] In one embodiment, the processor 130 may determine the backbone of the main blood vessel according to the detection results output by the second detection model and the detection results output by the third detection model. Specifically, the detection results of the second detection model may be used as a global feature reference to further modify the blood vessel contour of the detection results output by the third detection model.

[0068] For example, Figure 8 An example of major blood vessel detection is shown. Figure 8 For example, the processor 130 may derive other branches from the detection result R2 and remove these branches from the corresponding positions of the detection result R4 output by the detection model M3 to form the detection result R5. Compared with the detection result R4, the contour of the main blood vessel of the detection result R5 may be continuous and have no side branches.

[0069] Next, the processor 130 may search for the shortest path from the head end of the blood vessel to the tail end of the blood vessel and use the shortest path as the backbone. Fig. 9 FIG. 1 is a flow chart of a target blood vessel detection method according to an embodiment of the present invention. Fig. 9 In one embodiment, the processor 130 may determine the target vessel according to the projection position mapped by the medical image to perform scene / view classification (step S901) for identifying the vessel head. Specifically, according to the relevant documents of medical image diagnostics (Radiology), different projection positions and / or angles correspond to different target vessels. The target vessel may be a specific vessel that a doctor or other observer wants to view in the medical image, or may be the aforementioned main vessel.

[0070] Taking the coronary arteries as an example, Fig.10 Examples showing different projection positions. Fig.10, the left image is a subcostal view with a right anterior oblique (RAO) of 10 degrees and a caudate position of 30 degrees, and the left circumflex artery LCX, the left anterior descending artery-proximal LAD, the diagonal branch, and the obtuse marginal branch OM can be observed. However, the left circumflex artery LCX and the left anterior descending artery-proximal LAD are more distinguishable, but part of the left circumflex artery LCX and the obtuse marginal branch OM are not easy to distinguish. The right image is a view with a left anterior oblique (LAO) of 10 degrees and a cephalic position of 30 degrees, and the right coronary artery RCA and the acute marginal branch AcM can be observed. Although the large C-shape formed by the RCA can be clearly presented, its distal end may overlap with other branches and be difficult to distinguish. It can be seen that under a specific view, some blood vessels can be clearly distinguished (these blood vessels can be used as the target vessels of this view), but other parts may overlap with others or have unclear outlines.

[0071] It should be noted that there are many views for tubular arteries or other parts, and the embodiment of the present invention does not limit the type of views.

[0072] In one embodiment, the processor 130 may classify the scene / view to which the medical image belongs by a classifier associated with a machine learning algorithm, and infer the target blood vessel corresponding to the scene / view. In another embodiment, the processor 130 may use other image recognition techniques (e.g., Scale-Invariant Feature Transform (SIFT) or edge detection) to infer the scene / view to which the medical image belongs.

[0073] Next, the processor 130 may determine the vascular head of the target blood vessel in the medical image (step S903). Specifically, the processor 130 may identify the category of the object in one or more candidate regions in the medical image MI based on object detection technology (e.g., Region Convolutional Neural Network (R-CNN), fast R-CNN, or YOLO (You only look once)), and determine the candidate region whose category is the vascular head Vh. For example, Fig.11 An example of blood vessel tip detection is shown. Fig.11 In the medical image of the Tth frame, the blood vessel head Lh1 can be detected.

[0074] It should be noted that the program code can load an application programming interface (API) provided by TensorFlow, PyTorch, Keras or other libraries to implement object detection.

[0075] In one embodiment, the processor 130 may utilize target tracking technology (eg, optical flow method) to track the region of the blood vessel head, and accordingly reduce false positive and false negative detection results. Fig.11 For example, in the medical image of the T+1th frame, two blood vessel heads Lh1 and Lh2 can be detected. Based on the optical flow method, the medical images of the Tth and T+1th frames can be analyzed to form a tracking area Box (for example, by comparing the pixel displacement and the overlap size of the area), and the blood vessel head Lh2 can be regarded as a false positive FP. Finally, in the medical image of the T+2th frame, only the tracking area Box (corresponding to the blood vessel head Lh1) can be retained, and the blood vessel head Lh2 of the false positive FP can be removed.

[0076] The processor 130 may map the target blood vessel to a segmented image corresponding to the detection result R2 according to the blood vessel tip (step S905) (i.e., image post-processing). Specifically, the segmented image R2 corresponding to the detection result includes pixels belonging to the blood vessel. The processor 130 may perform a more detailed detection on the detection result of the blood vessel tip and the second detection result (e.g., the detection result R2) using an object detection technology (e.g., R-CNN, fast R-CNN, or YOLO), thereby determining the blood vessel tip, and mapping it to a corresponding position P on the segmented image and generating a detection result R7 (i.e., the blood vessel tip is marked on the segmented image of the blood vessel).

[0077] After the blood vessel head is determined, in the detection result R7, the processor 130 can search for all possible paths that may be the backbone among the multiple paths from the tail end of the blood vessel to the head end of the blood vessel. Then, the processor 130 can determine the backbone of the main blood vessel based on the width and length of the main blood vessel. For example, the ratio of the square of the edge length of any point in the path to the square of the total width is related to the path decision. Please refer to Figure 8 Finally, the processor 130 generates a detection result R6. The detection result R6 includes the backbone Sk of the main blood vessel. The backbone Sk may be the shortest path from the tail end of the blood vessel to the head end of the blood vessel.

[0078] In some embodiments, the processor 130 may further present the aforementioned detection results R1-R7 on the display 150 for doctors or other viewers to view. In addition, considering the changes of blood vessels in time series, the display 150 may display the dynamic changes of blood vessels in real time or on the fly.

[0079] In summary, in the blood vessel detection device and the image-based blood vessel detection method of the embodiments of the present invention, a multi-stage machine learning-based detection model is provided, and the main blood vessels can be further identified. In addition, the embodiments of the present invention also refer to the factors of time change. Thereby, the accuracy of blood vessel detection can be improved.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An image-based blood vessel detection method, characterized in that: include: The first data to be evaluated is detected by a first detection model to obtain a first detection result, wherein the first detection model is constructed based on one of a plurality of machine learning algorithms, the first data to be evaluated includes one of a plurality of medical images obtained by taking a blood vessel and sorted according to a time series as a medical image of a current frame, and the first detection result output by the first detection model includes at least one pixel in the medical image of the current frame belonging to the blood vessel; detecting second data to be evaluated by a second detection model to obtain a second detection result, wherein the second detection model is constructed based on one of the machine learning algorithms, the second data to be evaluated includes the first detection result, and the second detection result output by the second detection model includes that the at least one pixel in the medical image of the current frame belongs to the blood vessel; Performing time series-based background removal on the second detection result to obtain an intermediate detection result; as well as The third data to be evaluated is detected by a third detection model to obtain a third detection result, wherein the third detection model is constructed based on one of the machine learning algorithms, and the third data to be evaluated includes the intermediate detection result of the medical image of the current frame, the intermediate detection result of the medical image of the previous frame, the previous and next frames after optical flow extrapolation and blurring plus the boundary of the current frame, and the original medical image of the current frame, and the third detection result output by the third detection model includes the main blood vessels in the medical image of the current frame, and the main blood vessels are determined based on the weights of evaluating the lesion complexity of multiple blood vessel segments.

2. The image-based blood vessel detection method according to claim 1, characterized in that: Also includes: Image processing is performed on the medical image of the current frame, wherein the image processing is related to enhancing the contour of the blood vessel, and the second data to be evaluated also includes an image obtained after the image processing.

3. The image-based blood vessel detection method according to claim 1, characterized in that: After the step of obtaining the second detection result, the method further includes: Segmenting a plurality of blood vessel regions belonging to the blood vessel in the medical image of the current frame according to the second detection result of the medical image of the current frame, so as to obtain a plurality of segmented images; determining a displacement of each pixel of the blood vessel region in the segmented image according to the time series; and The blood vessel regions are classified according to a comparison result of the displacement corresponding to each blood vessel region with the blood vessel region in the previous segmented image, wherein the comparison result is related to the overlap size between the displacement and the blood vessel region.

4. The image-based blood vessel detection method according to claim 1, characterized in that: After the step of obtaining the third detection result, the method further includes: The backbone of the main blood vessel is determined according to the second detection result and the third detection result, wherein the backbone is determined according to its width and length.

5. The image-based blood vessel detection method according to claim 1, characterized in that: After the step of obtaining the second detection result, the method further includes: Determine the target blood vessel according to the projection position mapped by the medical image of the current image frame; and Determine the blood vessel head of the target blood vessel in the medical image of the current frame.

6. The image-based blood vessel detection method according to claim 5, characterized in that: After the step of determining the blood vessel head of the target blood vessel in the medical image of the current frame, the method further includes: The target blood vessel is mapped to a segmented image corresponding to the second detection result according to the blood vessel head, wherein the segmented image includes the at least one pixel belonging to the blood vessel.

7. A blood vessel detection device, comprising: A memory storing a plurality of program codes; as well as A processor is coupled to the memory and configured to load and execute the program code to: The first data to be evaluated is detected by a first detection model to obtain a first detection result, wherein the first detection model is constructed based on one of a plurality of machine learning algorithms, the first data to be evaluated includes one of a plurality of medical images obtained by taking a blood vessel and sorted according to a time series as a medical image of a current frame, and the first detection result output by the first detection model includes at least one pixel in the medical image of the current frame belonging to the blood vessel; detecting second data to be evaluated by a second detection model to obtain a second detection result, wherein the second detection model is constructed based on one of the machine learning algorithms, the second data to be evaluated includes the first detection result, and the second detection result output by the second detection model includes that the at least one pixel in the medical image of the current frame belongs to the blood vessel; Performing time series-based background removal on the second detection result to obtain an intermediate detection result; as well as The third data to be evaluated is detected by a third detection model to obtain a third detection result, wherein the third detection model is constructed based on one of the machine learning algorithms, and the third data to be evaluated includes the intermediate detection result of the medical image of the current frame, the intermediate detection result of the medical image of the previous frame, the previous and next frames after optical flow extrapolation and blurring plus the boundary of the current frame, and the original medical image of the current frame, and the third detection result output by the third detection model includes the main blood vessels in the medical image of the current frame, and the main blood vessels are determined based on the weights of evaluating the lesion complexity of multiple blood vessel segments.

8. The blood vessel detection device according to claim 7, characterized in that: The processor is also configured to: Image processing is performed on the medical image of the current frame, wherein the image processing is related to enhancing the contour of the blood vessel, and the second data to be evaluated also includes an image obtained after the image processing.

9. The blood vessel detection device according to claim 7, characterized in that: The processor is also configured to: Segmenting a plurality of blood vessel regions belonging to the blood vessel in the medical image of the current frame according to the second detection result of the medical image of the current frame, so as to obtain a plurality of segmented images; determining a displacement of each pixel of the blood vessel region in the segmented image according to the time series; as well as The blood vessel regions are classified according to a comparison result of the displacement corresponding to each blood vessel region with the blood vessel region in the previous segmented image, wherein the comparison result is related to the overlap size between the displacement and the blood vessel region.

10. The blood vessel detection device according to claim 7, characterized in that: The processor is also configured to: The backbone of the main blood vessel is determined according to the second detection result and the third detection result, wherein the backbone is determined according to its width and length.

11. The blood vessel detection device according to claim 7, characterized in that: The processor is also configured to: Determine the target blood vessel according to the projection position mapped by the medical image of the current image frame; and Determine the blood vessel head of the target blood vessel in the medical image of the current frame.

12. The blood vessel detection device according to claim 11, characterized in that: The processor is also configured to: The target blood vessel is mapped to a segmented image corresponding to the second detection result according to the blood vessel head, wherein the segmented image includes the at least one pixel belonging to the blood vessel.