DSA image video sequence automatic segmentation method, system and device and medium

Abstract

The invention discloses a DSA image video sequence automatic segmentation method, system and device and a medium. The method comprises the steps of obtaining a DSA image video sequence sample data set and performing preprocessing; key frames are extracted from the continuous DSA image video sequence and marked, and a key structure is identified and marked; inputting the labeled sample data set into an SAM2 model; and carrying out key feature extraction on the SAM2 model, extracting geometric features of the coronary artery and relative positions of the coronary artery and other anatomical structures, identifying boundaries of the coronary artery based on the geometric features and the relative positions of the coronary artery and the other anatomical structures, carrying out segmentation display on the coronary artery and the other anatomical structures, and outputting a segmentation result map. According to the method, efficient and accurate automatic segmentation of the DSA image video sequence is realized, dynamic changes of coronary arteries, view angle transformation and complex scenes in an operation process can be effectively dealt with, the consistency and robustness of segmentation results are kept, and the method has very high processing precision and efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method, system, device and computer-readable storage medium for automatic segmentation of digital subtraction angiography (DSA) image video sequences based on a SAM2 model. Background Art

[0002] Percutaneous coronary intervention (PCI) is an important and commonly used surgical procedure for treating patients with coronary heart disease. Its success or failure depends on accurate visualization of the coronary artery anatomy. Coronary heart disease is a heart disease caused by narrowing or blockage of the coronary arteries. Through PCI surgery, doctors can deliver stents or drug-coated balloons into diseased coronary arteries through catheters to restore blood flow. Digital subtraction angiography (DSA) is a key imaging technology used to display coronary arteries during PCI surgery. DSA image video sequence segmentation is to highlight the arterial structure by removing background interference.

[0003] Traditional DSA image video sequence segmentation usually relies on manual operations, which is time-consuming and easily affected by subjective factors, resulting in inconsistent results, which may affect the accuracy of surgery and patient prognosis.

[0004] With the rapid development of deep learning and computer vision, researchers have begun to study the use of artificial intelligence (AI) technology to automate the processing of medical imaging video sequences. Automated image segmentation can not only greatly reduce the time of manual operations, but also improve the consistency and accuracy of segmentation results. However, medical images, especially DSA images, are characterized by high complexity and high noise, which makes artificial intelligence technology face many challenges in processing these images. For example, the morphology of coronary arteries is complex and varied, making it difficult to accurately identify and segment these structures. In addition, in specific robot-assisted surgery scenarios, the dynamic changes of arteries during surgery also increase the difficulty of the segmentation task.

[0005] Therefore, how to accurately perform automatic segmentation of DSA image video sequences has always been a difficult problem that those skilled in the art have been committed to solving. Summary of the invention

[0006] The embodiment of the present application solves the problem of inaccurate DSA image video sequence segmentation results in the existing method by providing a DSA image video sequence automatic segmentation method.

[0007] In order to solve the above problems, in a first aspect, an embodiment of the present application provides a method for automatic segmentation of a DSA image video sequence, the method comprising the following steps:

[0008] Acquire a sample data set of DSA image video sequences during PCI surgery, and preprocess the sample data set;

[0009] Extract key frames from a continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, and catheters;

[0010] The labeled sample data set is input into the SAM 2 model; the SAM 2 model extracts key features, extracts the geometric features of the coronary arteries, and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments the coronary arteries and other anatomical structures, and outputs the segmentation result map.

[0011] Preferably, the sample data set is diverse, including different patients, different PCI surgery stages and different pathological conditions.

[0012] Preferably, preprocessing the sample data set includes:

[0013] Frame alignment of DSA image video sequences to correct motion blur caused by heart beats or patient movement;

[0014] De-noising the DSA video sequence to improve image quality;

[0015] The DSA video sequence is processed with enhanced contrast to make the coronary artery vascular structure more clearly visible.

[0016] Preferably, when labeling the key frames, special attention is paid to the branching structure of the blood vessels, the relative positions of the trunk and the branches, and the relationship between the coronary arteries and other structures of the heart.

[0017] Preferably, the SAM 2 model processing process specifically includes the following steps:

[0018] Use image and video encoders to extract key features from DSA image video sequences, understand the visual content at each time step, and extract the geometric features of the coronary arteries and their relative positions to other anatomical structures;

[0019] The hint encoder is used to process the annotation content on the key frames in the DSA image video sequence to guide the segmentation task and lock the coronary arteries in the DSA image video sequence;

[0020] The coronary arteries are tracked using a memory mechanism; when the coronary arteries move, their characteristics are recorded in a memory bank; when the coronary arteries are occluded, the memory bank is relied upon to predict the position and appearance of the coronary arteries when they reappear;

[0021] Using the mask decoder, based on the encoded image features of the image and video encoders, the hints of the hint encoder, and the tracking results of the memory mechanism, the final segmentation mask is generated to segment the coronary arteries and other anatomical structures and output the segmentation result map.

[0022] Preferably, the method further comprises a fine-tuning step for specific needs:

[0023] On the basis of the SAM 2 model in step S3, a low-rank adaptation layer is added, and the low-rank adaptation layer is trained on a specific DSA image video sequence sample data set that needs to be matched, so as to fine-tune the SAM 2 model.

[0024] In a second aspect, an embodiment of the present application provides a DSA image video sequence automatic segmentation device, comprising:

[0025] A data acquisition and preprocessing module is used to acquire a sample data set of DSA image video sequences during PCI surgery and preprocess the sample data set;

[0026] A key frame extraction and annotation module is used to extract key frames from a continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, and catheters;

[0027] The coronary artery segmentation module is used to input the labeled sample data set into the SAM 2 model; the SAM 2 model extracts key features, extracts the geometric features of the coronary arteries, and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments and displays the coronary arteries and other anatomical structures, and outputs the segmentation result map.

[0028] Preferably, the device further comprises:

[0029] The fine-tuning module is used to add a low-rank adaptation layer on the basis of the SAM 2 model, and train the low-rank adaptation layer on a specific DSA image video sequence sample data set that needs to be matched, so as to fine-tune the SAM 2 model.

[0030] In a third aspect, an embodiment of the present application provides a DSA image video sequence automatic segmentation module, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned DSA image video sequence automatic segmentation method when executing the program.

[0031] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned DSA image video sequence automatic segmentation method.

[0032] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0033] This embodiment provides a method for automatic segmentation of DSA image video sequences. By extracting key frames from a diverse sample data set and appropriately annotating them, based on the memory mechanism and mask decoder of the SAM 2 model, the method realizes efficient and accurate automatic segmentation of DSA image video sequences in PCI surgery, and can effectively cope with the dynamic changes of coronary arteries, perspective changes, and complex scenes during surgery, and maintain the consistency and robustness of the segmentation results. At the same time, it has high processing accuracy and efficiency, can provide effective auxiliary decision support for doctors, and is expected to be widely used in future robot-assisted surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0035] Figure 1 This is a flow chart of the DSA image video sequence automatic segmentation method provided in Example 1 of the present application;

[0036] Figure 2 The original image of the DSA image video sequence provided in Example 1 of the present application;

[0037] Figure 3 This is the automatic segmentation effect diagram provided in Example 1 of the present application;

[0038] Figure 4 This is a structural block diagram of the DSA image video sequence automatic segmentation device provided in Example 2 of the present application;

[0039] Figure 5 This is a structural block diagram of the DSA image video sequence automatic segmentation module provided in Example 3 of the present application;

[0040] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0041] The embodiment of the present application solves the problem of inaccurate DSA image video sequence segmentation results in the existing method by providing a DSA image video sequence automatic segmentation method.

[0042] In order to better understand the above technical solution, exemplary embodiments will be described in detail here, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of devices consistent with some aspects of the present application as detailed in the attached claims.

[0043] Embodiment 1

[0044] Figure 1 This is a flowchart of the DSA image video sequence automatic segmentation method provided in Example 1 of the present application. The DSA image video sequence automatic segmentation method comprises the following steps:

[0045] Step S1: Obtain a sample data set of DSA image video sequences during PCI surgery, and preprocess the sample data set.

[0046] When collecting sample data sets, it is necessary to ensure that the sample data sets are diverse, including different patients, different PCI surgery stages, and different pathological conditions.

[0047] Preprocessing the sample data set includes:

[0048] Frame alignment of DSA image video sequences to correct motion blur caused by heart beats or patient movement;

[0049] De-noising the DSA image video sequence to improve the image quality; the denoising process can apply a denoising algorithm, such as non-local mean denoising or 3D denoising;

[0050] The DSA video sequence is processed with enhanced contrast to make the coronary artery vascular structure more clearly visible.

[0051] Step S2: extract key frames from the continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, catheters, etc.

[0052] Among them, key frames should clearly show important moments during the procedure, such as stent deployment, catheter insertion, etc.

[0053] To annotate the keyframes, you can use existing annotation tools, such as ITK-SNAP or 3D Slicer, to assist in this process. When annotating the keyframes, make sure that the annotation operation is consistent under different viewing angles and lighting conditions.

[0054] During the annotation process, special attention should be paid to the branching structure of blood vessels (such as the branches of the left and right coronary arteries), the relative positions of the trunk and branches, and the relationship between the coronary arteries and other cardiac structures (such as the left atrium, right atrium, and aorta).

[0055] Of course, it is also possible to use annotated datasets with known anatomical features to ensure that the model understands key characteristics of the coronary arteries (such as bifurcations and stenoses).

[0056] Step S3: Input the labeled sample data set into the SAM 2 model; the SAM 2 model performs key feature extraction, extracts the geometric features of the coronary arteries and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments the coronary arteries and other anatomical structures, and outputs the segmentation result map.

[0057] The SAM 2 model (Segment Anything Model 2) is a general segmentation model launched in July 2024. It has attracted much attention for its outstanding performance in image and video segmentation tasks. The SAM 2 model not only performs well in processing static images, but also effectively solves the problem of target occlusion and tracking in video sequences through its built-in memory mechanism and mask decoder, especially achieving excellent results in interactive segmentation tasks. These features of the SAM 2 model give it a significant advantage in image segmentation tasks in dynamic scenes.

[0058] In this embodiment, the SAM 2 model processing process specifically includes the following steps:

[0059] Image and video encoders are used to extract key features from DSA image video sequences, understand the visual content at each time step, and extract the geometric features of the coronary arteries and the relative positions of the coronary arteries and other anatomical structures; the geometric features of the coronary arteries include the curvature of the coronary arteries, diameter changes, etc.

[0060] The hint encoder is used to process the annotation content on the key frames in the DSA image video sequence to guide the segmentation task and lock the coronary arteries in the DSA image video sequence;

[0061] The coronary arteries are tracked using a memory mechanism; when the coronary arteries move, their characteristics are recorded in a memory bank; when the coronary arteries are occluded, the memory bank is relied upon to predict the position and appearance of the coronary arteries when they reappear;

[0062] Using the mask decoder, based on the encoded image features of the image and video encoders, the hints of the hint encoder, and the tracking results of the memory mechanism, the final segmentation mask is generated to segment the coronary arteries and other anatomical structures and output the segmentation result map.

[0063] In a preferred embodiment, in the process of extracting key features, information such as the thickness of the vessel wall and the properties of the plaque (such as calcification and fibrosis) are combined to improve the model's ability to identify diseased vessels. At the same time, the model is enhanced with multi-viewing views and depth information to identify complex vascular morphology, and the boundary features of the coronary artery are extracted using image gradient and edge detection technology.

[0064] Furthermore, the memory mechanism includes a memory encoder, a memory bank, and a memory attention module. These components together store and utilize information from past frames, enabling the model to maintain consistent target tracking over time. The memory mechanism allows the SAM 2 model to handle temporal dependencies and occlusions in video data, and as the coronary arteries move and interact, the SAM 2 model records their features in the memory bank. When a coronary artery is occluded, the model can rely on this memory bank to predict its location and appearance when it appears again. In this way, even if the coronary artery is temporarily occluded or exits and re-enters the scene, the continuity of model recognition can be ensured.

[0065] Furthermore, the SAM 2 model also uses memory context to ensure accurate tracking across frames. Special attention is paid to the model learning the shape and size changes of the coronary arteries and their relationship with the surrounding tissues. In this process, prior knowledge such as anatomical features (such as the anatomical structure of the heart, the branching of the aorta, the blood flow path, etc.) and key parameters of the surgical scene are used to enhance the accuracy of tracking.

[0066] Furthermore, the accuracy of pattern recognition is improved through first-frame hints. On the first frame of the video sequence, 1-point hints or bounding box hints are applied to clarify the key feature locations of the coronary arteries (such as bifurcation points, stenosis areas, etc.). To improve the accuracy of hints, consider combining shape priors and contextual information (such as different stages of the cardiac cycle) to guide the model's focus area.

[0067] At the same time, using the memory mechanism of SAM 2, the segmentation results of the first frame are used as a reference to dynamically track the coronary arteries in subsequent frames. In combination with prior knowledge, such as the movement of the coronary arteries (such as the contraction and relaxation of the heart) and the deformation patterns commonly seen during surgery, the model is ensured to maintain high accuracy in dynamic changes. Special attention is paid to the twisting and bending of the blood vessels, which are common during surgery.

[0068] Furthermore, dynamic adjustments and optimizations are performed during the model learning process. When processing each frame of the video sequence, the SAM 2 model dynamically adjusts the mask and feature extraction process based on the segmentation results of the current frame and the previous frame, paying special attention to the morphological changes, perspective changes, and possible occlusions of the coronary arteries. In addition, the model combines real-time data monitoring information during surgery (such as hemodynamic parameters, electrocardiogram signals, etc.) to further optimize the segmentation results. Through the real-time feedback mechanism, the model can adaptively adjust the feature extraction and segmentation strategies to ensure the best segmentation effect at different surgical stages. At the same time, the adversarial training strategy in deep learning is used to improve the robustness of the model to interference and noise, thereby improving the stability and reliability of segmentation.

[0069] In the actual use process, according to specific needs, such as the needs of a specified pathological feature, the trained SAM 2 model can also be fine-tuned. The specific method is as follows:

[0070] Based on the SAM 2 model in step S3, a low-rank adaptation layer (LoRA) is added, and the low-rank adaptation layer is trained on a specific DSA image video sequence sample dataset (such as a DSA image video sequence sample dataset of a specified pathological feature) to fine-tune the SAM 2 model.

[0071] Through fine-tuning, the model can strengthen the learning of changes in coronary artery features in DSA images of a specified pathological feature without the need to retrain the entire model. This can reduce the computing resources required for training and improve the model's adaptability to DSA image video sequences.

[0072] Application Examples

[0073] In a complex PCI surgery case, the patient's coronary artery showed multiple lesions, and the operation caused frequent changes in perspective due to instrument operation. The method provided in this embodiment successfully tracked and segmented the coronary artery structure in the entire video sequence. Even in the case of drastic changes in perspective or overlapping blood vessels, the model can still accurately maintain the continuity and stability of the segmentation.

[0074] Figure 2 The original DSA image of a key frame is shown. Figure 3 The figure shows the segmentation result of the method of this embodiment. By comparing and analyzing the original DSA image and the segmentation result, it can be seen that the segmentation result clearly shows the contour of the blood vessels and accurately divides the boundary between the coronary artery and the surrounding tissue. Specifically, the segmentation result not only retains the subtle features of the blood vessels, such as branching, bending, and stenosis, but also effectively removes background noise. This effect significantly improves the readability of the image, allowing doctors to more clearly identify the lesion area during diagnosis and intervention.

[0075] Further analysis shows that the segmentation results of the method in this embodiment are highly consistent with the manual annotations of the doctor on the original image, and the Dice similarity coefficient reaches 0.92. This high-precision segmentation provides doctors with an accurate reference and helps to formulate more effective interventional treatment strategies.

[0076] In the comparative experiment, the SAM 2 model obtained by the method of the present invention was compared with the traditional UNet model in the above surgical case. The following are the comparison results of several key performance indicators:

[0077] Performance Indicators SAM 2 Model of the Present Invention Traditional UNet model Dice similarity coefficient 0.92 0.85 Intersection over Union (IoU) 0.85 0.75 Processing time 0.5 sec / frame 1.5 seconds / frame Robustness (under high noise) high Low

[0078] As can be seen from the above table, the segmentation performance of the SAM 2 model obtained by the method of the present invention is significantly better than that of the traditional UNet model. It has higher accuracy and speed, and is also more robust.

[0079] In summary, this embodiment proposes an automatic segmentation method for DSA image video sequences based on the SAM 2 model, and uses the memory mechanism and mask decoder of the SAM 2 model to achieve efficient and accurate segmentation of DSA image video sequences in PCI surgery. The method of this embodiment can effectively cope with the dynamic changes of coronary arteries, perspective changes, and complex scenes during surgery, and maintain the consistency and robustness of the segmentation results.

[0080] Clinical tests have shown that the method of this embodiment not only improves the accuracy and efficiency of DSA image segmentation, but also significantly reduces human errors during the operation. It has significant advantages in practical applications, can provide doctors with effective auxiliary decision support, and is expected to be widely used in future robot-assisted surgery.

[0081] Embodiment 2

[0082] Based on the same inventive concept as the DSA image video sequence automatic segmentation method in the aforementioned embodiment 1, this embodiment also provides a DSA image video sequence automatic segmentation device, such as Figure 4 As shown, the device comprises:

[0083] The data acquisition and preprocessing module 10 is used to acquire a sample data set of a DSA image video sequence during PCI surgery and preprocess the sample data set;

[0084] A key frame extraction and annotation module 20 is used to extract key frames from a continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, catheters, etc.;

[0085] The coronary artery segmentation module 30 is used to input the labeled sample data set into the SAM 2 model; the SAM 2 model extracts key features, extracts the geometric features of the coronary arteries, and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments and displays the coronary arteries and other anatomical structures, and outputs a segmentation result map.

[0086] Furthermore, the DSA image video sequence automatic segmentation device also includes:

[0087] A fine-tuning module is used to add a low-rank adaptation layer (LoRA) based on the SAM 2 model, and train the low-rank adaptation layer on a specific DSA image video sequence sample dataset (such as a DSA image video sequence sample dataset of a specified pathological feature) to fine-tune the SAM 2 model.

[0088] The fine-tuned device is particularly suitable for the automatic segmentation of DSA image video sequences in PCI surgery with a specified pathological feature.

[0089] The various specific processes and specific examples of the DSA image video sequence automatic segmentation method in the aforementioned embodiment 1 are also applicable to the DSA image video sequence automatic segmentation system in this embodiment. Through the detailed description of embodiment 1, those skilled in the art can clearly know the implementation method of the DSA image video sequence automatic segmentation system in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0090] The device provided in this embodiment can be integrated into an existing surgical assistance system. The device can be seamlessly connected with a surgical robot, image-guided equipment, etc., so that doctors can obtain high-precision segmentation results in real time when performing PCI surgery, providing effective auxiliary decision support for doctors.

[0091] Embodiment 3

[0092] Based on the same inventive concept as the DSA image video sequence automatic segmentation method in the aforementioned embodiment 1, this embodiment further provides a DSA image video sequence automatic segmentation module, such as Figure 5 As shown, a computer program is stored thereon, and when the program is executed by a processor, the steps of the automatic segmentation method of DSA image video sequence described in Example 1 are implemented.

[0093] Among them, Figure 5In the embodiment of the present invention, a bus architecture (represented by bus 500) is shown, which may include any number of interconnected buses and bridges, and bus 500 links various circuits including one or more processors represented by processor 502 and memory represented by memory 504. Bus 500 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are not further described herein. Bus interface 506 provides an interface between bus 500 and receiver 501 and transmitter 503. Receiver 501 and transmitter 503 may be the same element, namely a transceiver, which provides a unit for communicating with various other devices over a transmission medium.

[0094] The processor 502 is responsible for managing the bus 500 and general processing, while the memory 504 may be used to store data used by the processor 502 when performing operations.

[0095] Embodiment 4

[0096] Based on the same inventive concept as the DSA image video sequence automatic segmentation method in the aforementioned embodiment 1, this embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the DSA image video sequence automatic segmentation method described in embodiment 1 are implemented.

[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0101] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0102] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for automatic segmentation of a DSA image video sequence, the method comprising the following steps: Acquire a sample data set of DSA image video sequences during PCI surgery, and preprocess the sample data set; Extract key frames from a continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, and catheters; The labeled sample data set is input into the SAM 2 model; the SAM 2 model extracts key features, extracts the geometric features of the coronary arteries, and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments the coronary arteries and other anatomical structures, and outputs the segmentation result map.

2. The DSA image video sequence automatic segmentation method according to claim 1, characterized in that: The sample dataset is diverse, including different patients, different PCI surgery stages, and different pathological conditions.

3. The DSA image video sequence automatic segmentation method according to claim 1, characterized in that: Preprocessing the sample data set includes: Frame alignment of DSA image video sequences to correct motion blur caused by heart beats or patient movement; De-noising the DSA video sequence to improve image quality; The DSA video sequence is processed with enhanced contrast to make the coronary artery vascular structure more clearly visible.

4. The DSA image video sequence automatic segmentation method according to claim 1, characterized in that: When labeling the key frames, special attention is paid to the branching structure of the blood vessels, the relative positions of the trunk and the branches, and the relationship between the coronary arteries and other cardiac structures.

5. The DSA image video sequence automatic segmentation method according to claim 1, characterized in that: The SAM 2 model processing process specifically includes the following steps: Use image and video encoders to extract key features from DSA image video sequences, understand the visual content at each time step, and extract the geometric features of the coronary arteries and their relative positions to other anatomical structures; The hint encoder is used to process the annotation content on the key frames in the DSA image video sequence to guide the segmentation task and lock the coronary arteries in the DSA image video sequence; The coronary arteries are tracked using a memory mechanism; when the coronary arteries move, their characteristics are recorded in a memory bank; when the coronary arteries are occluded, the memory bank is relied upon to predict the position and appearance of the coronary arteries when they reappear; Using the mask decoder, based on the encoded image features of the image and video encoders, the hints of the hint encoder, and the tracking results of the memory mechanism, the final segmentation mask is generated to segment the coronary arteries and other anatomical structures and output the segmentation result map.

6. The DSA image video sequence automatic segmentation method according to claim 1, characterized in that: The method also includes fine-tuning steps for specific needs: On the basis of the SAM 2 model in step S3, a low-rank adaptation layer is added, and the low-rank adaptation layer is trained on a specific DSA image video sequence sample data set that needs to be matched, so as to fine-tune the SAM 2 model.

7. A DSA image video sequence automatic segmentation device, characterized in that: include: A data acquisition and preprocessing module is used to acquire a sample data set of DSA image video sequences during PCI surgery and preprocess the sample data set; A key frame extraction and annotation module is used to extract key frames from a continuous DSA image video sequence, annotate the key frames, and identify and mark key structures, wherein the key structures include coronary arteries, stents, and catheters; The coronary artery segmentation module is used to input the labeled sample data set into the SAM 2 model; the SAM 2 model extracts key features, extracts the geometric features of the coronary arteries, and the relative positions of the coronary arteries and other anatomical structures, and based on this, identifies the boundaries of the coronary arteries, segments and displays the coronary arteries and other anatomical structures, and outputs the segmentation result map.

8. The DSA image video sequence automatic segmentation device according to claim 7, characterized in that: The device also includes: The fine-tuning module is used to add a low-rank adaptation layer on the basis of the SAM 2 model, and train the low-rank adaptation layer on a specific DSA image video sequence sample data set that needs to be matched, so as to fine-tune the SAM 2 model.

9. A DSA image video sequence automatic segmentation module, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the DSA image video sequence automatic segmentation method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the DSA image video sequence automatic segmentation method as claimed in any one of claims 1 to 6 are implemented.

Images