A photoacoustic blood vessel image segmentation method, device, equipment and storage medium

Abstract

The application discloses a photoacoustic blood vessel image segmentation method and device, equipment and storage medium, relates to photoacoustic image processing technical field, and the method comprises the steps that a photoacoustic blood vessel image is acquired;The blood vessel topology graph composed of nodes and blood vessel segments is determined according to the photoacoustic blood vessel image;According to the length of each blood vessel segment, the width of each blood vessel segment, the included angle between the main direction of each blood vessel segment and the reference direction, the connection relationship of each blood vessel segment in the blood vessel topology graph and the local small blood vessel density around the blood vessel segment, an initial classification result is obtained;Wherein, the initial classification result includes the first type of blood vessel extending along the reference direction, the second type of blood vessel extending outward from the first type of blood vessel, and the third type of blood vessel in the form of point or cluster;The initial classification result is corrected based on the blood vessel topology graph to obtain a target segmentation result.The method improves the segmentation accuracy of blood vessels in photoacoustic images.

Description

Technical Field

[0001] This application relates to the field of photoacoustic image processing technology, and in particular to a photoacoustic blood vessel image segmentation method, apparatus, device and storage medium. Background Technology

[0002] Photoacoustic imaging utilizes the absorption of pulsed light by hemoglobin to generate ultrasound signals, which can non-invasively display the distribution, morphology and microcirculation structure of blood vessels in superficial tissues such as toes and fingers, and has important application value in blood flow function assessment.

[0003] Traditional methods for processing blood vessel images include thresholding, region growing, edge detection, morphological processing, blood vessel enhancement filtering, and deep learning-based semantic segmentation techniques. These methods can extract candidate regions of blood vessels under ideal conditions, but they usually only output binary results of blood vessels / background.

[0004] Existing methods cannot perform fine-grained segmentation of blood vessels when there is noise, breaks, artifacts, and uneven intensity in photoacoustic images, resulting in poor segmentation accuracy of blood vessels in photoacoustic images. Summary of the Invention

[0005] This application provides a photoacoustic blood vessel image segmentation method, apparatus, device, and storage medium, which can improve the accuracy of blood vessel segmentation in photoacoustic images.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a photoacoustic blood vessel image segmentation method, the method comprising: Acquire photoacoustic vascular images; determine a vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular images; obtain an initial classification result based on the length of each vascular segment, the width of each vascular segment, the angle between the main direction and the reference direction of each vascular segment, the connection relationship of each vascular segment in the vascular topology map, and the local small vessel density around the vascular segment; wherein, the initial classification result includes a first type of vessel extending along the reference direction, a second type of vessel extending outward from the first type of vessel, and a third type of vessel in the form of dots or clusters; correct the initial classification result based on the vascular topology map to obtain the target segmentation result.

[0007] In some possible implementations, the initial classification result is obtained based on the length of each vessel segment, the width of each vessel segment, the angle between the principal direction and the reference direction of each vessel segment, the connectivity of each vessel segment in the vascular topology map, and the local small vessel density around the vessel segment, including: When the length of the vessel segment is greater than a preset first length threshold, the width of the vessel segment is greater than a preset first width threshold, and the angle between the main direction and the reference direction of the vessel segment is less than an angle threshold, the vessel segment is initially classified as a first type of vessel extending along the reference direction; when the vessel segment is connected to a previously classified first type of vessel and the angle between the main direction and the reference direction of the vessel segment is greater than a branch angle threshold, the vessel segment is initially classified as a second type of vessel extending outward from the first type of vessel; when the length of the vessel segment is less than a preset second length threshold, the width of the vessel segment is less than a preset second width threshold, or the density of local small vessels around the vessel segment is greater than a density threshold, the vessel segment is initially classified as a third type of vessel, either dot-like or cluster-like.

[0008] In some possible implementations, the step of correcting the initial classification result based on the vascular topology map to obtain the target segmentation result includes: When multiple blood vessel segments are arranged end-to-end along the reference direction in the blood vessel topology map, the main direction of each blood vessel segment is consistent with the reference direction, the angle between the main direction of each blood vessel segment and the reference direction is less than the angle threshold, and the gap between adjacent blood vessel segments is less than the gap threshold, the multiple blood vessel segments are merged into a blood vessel segment of the first type of blood vessel to obtain the target segmentation result.

[0009] In some possible implementations, the method further includes: When a blood vessel segment is initially classified as a Class I or Class II blood vessel but is not connected to other blood vessel segments in the blood vessel topology map, the blood vessel segment is corrected to background. When a blood vessel segment is connected to a Class I blood vessel and the angle between the main direction and the reference direction of the blood vessel segment is greater than the branch angle threshold, the blood vessel segment is corrected to Class II blood vessel. When the distance between two endpoints is less than the distance threshold, the angle between the local directions of the two endpoints is less than the angle threshold, and the average value of the blood vessel enhancement response on the connection path between the two endpoints is greater than the response threshold, a completion path is generated between the two endpoints and the region corresponding to the completion path is incorporated into the blood vessel candidate region to obtain the target segmentation result.

[0010] In some possible implementations, determining the vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular image includes: The target tissue region is determined based on the photoacoustic vascular image; multi-scale vascular enhancement processing is performed on the photoacoustic vascular image to obtain a vascular enhancement map; vascular candidate regions are extracted within the target tissue region based on the vascular enhancement map; a skeleton is extracted from the vascular candidate regions, and a vascular topology map composed of nodes and vascular segments is constructed based on the skeleton.

[0011] In some possible implementations, the multi-scale vascular enhancement processing of the photoacoustic vascular image to obtain a vascular enhancement map includes: Multiple scale parameters and multiple orientation parameters are preset; based on the photoacoustic vascular image, the vascular enhancement response value is calculated under each scale parameter and each orientation parameter; the vascular enhancement response values ​​under the same scale parameter in multiple directions are fused to obtain the scale response map of that scale parameter; the scale response maps of all scale parameters are fused to generate a vascular enhancement map.

[0012] In some possible implementations, acquiring photoacoustic vascular images includes: Acquire an initial photoacoustic vascular image; preprocess the initial photoacoustic vascular image to obtain a photoacoustic vascular image.

[0013] Secondly, this application provides a photoacoustic blood vessel image segmentation device, the device comprising: The acquisition module is used to acquire photoacoustic vascular images and determine a vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular images. The classification module is used to obtain an initial classification result based on the length of each blood vessel segment, the width of each blood vessel segment, the angle between the main direction and the reference direction of each blood vessel segment, the connection relationship of each blood vessel segment in the blood vessel topology map, and the local small blood vessel density around the blood vessel segment; wherein, the initial classification result includes a first type of blood vessel extending along the reference direction, a second type of blood vessel extending outward from the first type of blood vessel, and a third type of blood vessel in the form of dots or clusters. The correction module is used to correct the initial classification result based on the blood vessel topology map to obtain the target segmentation result.

[0014] Thirdly, this application provides a computing device, including a memory and a processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of the first aspects.

[0015] Fourthly, this application provides a computer-readable storage medium for storing a computer program for performing the method as described in any one of the first aspects.

[0016] Fifthly, this application provides a computer program product comprising one or more computer instructions, wherein when the computer instructions are executed by a computer, the computer performs the method as described in any one of the first aspects.

[0017] As can be seen from the above technical solution, this application has at least the following beneficial effects: In this application, photoacoustic vascular images are acquired; a vascular topology map composed of nodes and vascular segments is determined based on the photoacoustic vascular images; an initial classification result is obtained based on the length of each vascular segment, the width of each vascular segment, the angle between the principal direction and the reference direction of each vascular segment, the connectivity of each vascular segment in the vascular topology map, and the density of local small vessels around the vascular segment; wherein, the initial classification result includes a first type of vessel extending along the reference direction, a second type of vessel extending outward from the first type of vessel, and a third type of vessel in the form of dots or clusters; the initial classification result is corrected based on the vascular topology map to obtain the target segmentation result. In existing technical solutions, binary segmentation of vessels and non-vessels is often used as the target, which makes it difficult to distinguish vessels of different semantic types and is sensitive to noise and artifacts, easily producing breaks and false branches. Therefore, this application can use vascular topology relationships to distinguish vessels into three semantic categories, and reduce breaks and misclassifications through topology correction, thereby improving the accuracy and stability of segmentation.

[0018] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0019] Figure 1 A schematic diagram illustrating an application scenario provided in an embodiment of this application; Figure 2 A flowchart illustrating a photoacoustic blood vessel image segmentation method provided in this application embodiment; Figure 3 A schematic diagram of a photoacoustic blood vessel image segmentation device provided in an embodiment of this application; Figure 4 This is a schematic diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," and "third," etc., used in this application specification and accompanying drawings are used to distinguish different objects, not to limit a specific order.

[0021] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] Currently, processing photoacoustic vascular images presents several challenges. These challenges primarily stem from the inherent characteristics of photoacoustic imaging technology and the limitations of existing image processing methods.

[0023] First, most methods only output binary results of blood vessels and background, failing to distinguish between main blood vessels extending along the long axis of the tissue, outward-extending branch vessels, and punctate or clustered small blood vessels. This single output cannot meet the semantic requirements of medical visualization and subsequent quantitative analysis. Furthermore, traditional thresholding, region growing, and edge detection methods are highly sensitive to speckle noise, reconstruction artifacts, and intensity inhomogeneity in photoacoustic images, easily leading to missed, over-segmented, or artifact-induced missegmentation of blood vessels. Conventional pixel-level segmentation mainly relies on local grayscale or texture features, lacking utilization of topological information such as the vascular skeleton, endpoints, bifurcation points, and connectivity relationships, resulting in segmentation results often containing broken vessels, false branches, and isolated noise. Distinguishing between main vessels, branches, and small blood vessel clusters currently relies heavily on manual interpretation, which is inefficient and lacks automated processes that can be performed in batches and are reproducible. Deep learning methods require extensive pixel-level annotation and are easily affected by cross-device differences, case morphology differences, and pathological heterogeneity in small sample photoacoustic images, resulting in insufficient segmentation stability and interpretability.

[0024] In view of this, embodiments of this application provide a photoacoustic blood vessel image segmentation method, which aims to solve one or more of the above-mentioned problems and improve the accuracy and stability of segmenting and classifying blood vessels in photoacoustic images.

[0025] To make the technical solution of this application clearer and easier to understand, the application scenarios of the technical solution of this application are described below with reference to the accompanying drawings. Figure 1 As shown in the figure, this figure is a schematic diagram of an application scenario provided by an embodiment of this application.

[0026] In this application scenario, the terminal can be a photoacoustic imaging device, a smartphone, a tablet, or a medical imaging workstation, used to acquire photoacoustic vascular images. The communication network can be a wireless network or a wired network, used to transmit data between the terminal and the server. The server can be a local server or a cloud server, used to perform photoacoustic vascular image segmentation tasks. In one example, after the terminal acquires a photoacoustic cross-sectional image of a toe or finger, it sends the image to the server via the communication network. After receiving the photoacoustic vascular image, the server executes the photoacoustic vascular image segmentation method provided in this application embodiment to obtain target segmentation results containing a first type of blood vessel, a second type of blood vessel, and a third type of blood vessel, and returns the results to the terminal for display or storage.

[0027] To make the technical solution of this application clearer and easier to understand, the following describes a photoacoustic blood vessel image segmentation method provided by an embodiment of this application, in conjunction with the above application scenarios. For example... Figure 2 As shown, this figure is a flowchart of a photoacoustic blood vessel image segmentation method provided in an embodiment of this application. The method includes: S201, The server acquires photoacoustic images of blood vessels.

[0028] Photoacoustic vascular images are images obtained using photoacoustic imaging technology that reflect the distribution of hemoglobin and the morphology of blood vessels in tissues. Examples include photoacoustic vascular images of a cross-section of the toe, photoacoustic vascular images of the fingertip, and photoacoustic vascular images of the skin of the palm.

[0029] The server acquires an initial photoacoustic vascular image; the initial photoacoustic vascular image is preprocessed to obtain a photoacoustic vascular image. Preprocessing may include one or more of the following operations: background subtraction, i.e., subtracting the dark field signal or average background value of the tissueless region; bandpass filtering, used to preserve the frequency components corresponding to the blood vessels and suppress high-frequency noise and low-frequency drift; median filtering, used to eliminate isolated noise points and preserve blood vessel edges; quantile normalization, i.e., linearly mapping the intensity range from the p1th percentile to the p2th percentile in the image to a uniform range, where p1 and p2 can be 1 and 99 respectively, or can be adaptively determined according to image quality. After preprocessing, the initial photoacoustic vascular image yields a photoacoustic vascular image in which the contrast between the tissue region and the background region is enhanced, and speckle noise and intensity unevenness are suppressed.

[0030] Preprocessing photoacoustic vascular images can reduce the impact of speckle noise, intensity inhomogeneity, and reconstruction artifacts on subsequent processing, thereby improving the accuracy of vascular segmentation.

[0031] S202. The server determines the vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular image.

[0032] A vascular topology graph is a graph structure consisting of nodes and vascular segments connecting the nodes, used to describe the connectivity and morphological characteristics of vascular networks.

[0033] The server determines the target tissue region based on the photoacoustic vascular image. The target tissue region refers to the set of pixels or voxels in the photoacoustic vascular image that belong to the tissue to be analyzed (e.g., a toe or finger), and can be determined based on the tissue contour, imaging field of view, or preset acquisition coordinates. In some embodiments, the server performs edge detection or thresholding on the photoacoustic vascular image to obtain the boundary between the tissue and the background, and uses the connected region inside the boundary as the target tissue region. In addition, it can directly crop the region within the preset acquisition coordinates of the imaging device, or receive the region range manually drawn by the user. After determining the target tissue region, the boundary of the target tissue region is expanded or contracted by 5 to 20 pixels to remove edge artifacts and fully include the internal structure of the tissue.

[0034] The server performs multi-scale vessel enhancement processing on the photoacoustic vascular image to obtain a vessel enhancement map. Let the preprocessed photoacoustic vascular image be I(x), where x represents the pixel position in the image. The specific steps include: presetting multiple scale parameters and multiple orientation parameters; calculating the vessel enhancement response value under each scale parameter and each orientation parameter based on the photoacoustic vascular image; fusing the vessel enhancement response values ​​of multiple directions under the same scale parameter to obtain the scale response map of that scale parameter; and fusing the scale response maps of all scale parameters to generate the vessel enhancement map.

[0035] In some embodiments, multiple scale parameters and multiple orientation parameters are preset. The scale parameters are denoted as... The unit is mm or pixel, used to control the size of the filter kernel. Let the scale set be... Where K is the number of scales. The first scale, The second scale, For the Kth scale, the value ranges from 0.05 mm to 1.50 mm. The direction parameter is denoted as... This is used to control the orientation of the directional filter; let the set of directions be... Where M is the number of directions. Indicates the first direction. Indicates the second direction. This represents the Mth direction, with a value ranging from 0° to 180°.

[0036] Based on photoacoustic vascular images, vascular enhancement response values ​​are calculated for each scale parameter and each orientation parameter. For the k-th scale and m-th orientation, this application employs the Frangi Vesselness Filter (multi-scale vascular enhancement filtering) method to calculate the vascular enhancement response value, denoted as . The vascular enhancement response value represents the value at location x and scale x. ,direction The image below shows the intensity of tubular structures; the higher the value, the more likely it is to be a blood vessel.

[0037] Vascular enhancement response values ​​from multiple directions under the same scale parameter are fused to obtain a scale response map for that scale parameter. The fusion method is to select the response of the largest direction; that is, for each location x, the largest response value among M directions is selected as the response at that scale. The calculation formula is:

[0038] in, For scale The scale response value at position x below. For scale Down, direction The scale response value at position x below.

[0039] The scale response maps at each scale are normalized. Let the target tissue region be... For the scale response map at the k-th scale Calculate located at Low quantiles within and high quantile For example, the 1st and 99th percentiles, or the 2nd and 98th percentiles, normalized scale response plots. Calculate using the following formula:

[0040] in, For the normalized first Scale response plots at various scales; To prevent small constants with a denominator of zero; This means limiting the value to the range of 0 to 1; It is the lower quantile. For high quantiles, pixels or voxels outside the target tissue region can be set to... This is to prevent artifacts from outside the organization from entering subsequent fusion.

[0041] All normalized scale response maps are fused to generate a vascular enhancement map. This embodiment uses a maximum response fusion method, where at each location x, the maximum value of the response across all scales is taken as the comprehensive vascular enhancement response at that location, expressed as:

[0042] It can also record the corresponding scale:

[0043] The result after fusion This is recorded as a vascular enhancement image.

[0044] The embodiments of this application can also employ other fusion methods to implement the above fusion steps, such as weighted fusion, which involves summing the response maps at each scale according to preset weights; or adaptive fusion based on local signal-to-noise ratio, which involves calculating the signal-to-noise ratio at each scale within the local neighborhood of each pixel, generating adaptive weights accordingly, and then summing them using weighted methods. These alternative methods will not be elaborated upon further in this application.

[0045] The server extracts candidate vascular regions within the target tissue area based on the enhanced vascular map; it then extracts a skeleton from the candidate vascular regions and constructs a vascular topology map composed of nodes and vascular segments based on the skeleton.

[0046] In some embodiments, this embodiment employs a dual-threshold hysteresis connection method to extract candidate blood vessel regions. The server sets a high threshold. and low threshold ,satisfy .

[0047] The server will display the enhanced blood vessel image. Median value higher than The region is selected as a candidate region for strong blood vessels, and the value is located in and Regions that are connected to strong vessel candidate regions and are between them are retained as weak vessel candidate regions, while regions with values ​​lower than [a certain value] are [retained]. The region is used as background and removed. The high and low thresholds can be adaptively determined based on the grayscale distribution of the angiography within the target tissue region, for example, using the 80th percentile as the threshold. Take the 50th percentile as After passing through a double-threshold hysteresis connection, a binary mask of the blood vessel candidate region is obtained. ,in This indicates that the location belongs to the candidate region for blood vessels. Indicates background.

[0048] Binary mask for blood vessel candidate regions Skeleton extraction is performed. The skeleton refers to the set of centerlines of the candidate blood vessel region, which can preserve the topological structure and connectivity of the original region. This embodiment uses a morphological thinning algorithm, which iteratively deletes boundary pixels or voxels that satisfy the topological invariance condition until a centerline with a single pixel width is obtained. The extracted skeleton is denoted as S, which is a set of discrete points.

[0049] Constructing a vascular topology map based on the skeleton Let V represent the set of nodes and E represent the set of vessel segments. Nodes include endpoints and bifurcation points. An endpoint is a point in the skeleton with one adjacent point. A bifurcation point is a point in the skeleton with three or more adjacent points. A vessel segment is a continuous skeleton line segment between two nodes, that is, a sequence of intermediate points traversed when walking along the skeleton from one node to another, which does not contain other nodes. For each vessel segment e, record its skeleton point sequence, the positions of its two endpoints, and the corresponding vessel candidate region.

[0050] This application employs multi-scale vascular enhancement processing, which can simultaneously enhance vascular structures of varying thicknesses. Large-scale responses enhance main vessels, while small-scale responses enhance smaller branches. Dual-threshold hysteresis connections are used to extract candidate vascular regions, which can suppress isolated noise while preserving weak-signal vessels. Skeleton extraction and topology map construction explicitly express the connectivity relationships of the vascular network, helping to reduce errors caused by vascular rupture and spurious branches.

[0051] S203. The server obtains the initial classification result based on the length of each blood vessel segment, the width of each blood vessel segment, the angle between the main direction and the reference direction of each blood vessel segment, the connection relationship of each blood vessel segment in the blood vessel topology map, and the local small blood vessel density around the blood vessel segment.

[0052] The reference direction refers to the long axis direction of the tissue being analyzed. For photoacoustic images of toes or fingers, the reference direction can be obtained directly from the acquisition coordinate system of the imaging device, or it can be determined through the principal axis direction of the tissue contour or the results of principal component analysis.

[0053] In some embodiments, the server is based on a vascular topology map. Each segment of blood vessel Calculate the length of vessel segment e, the width of vessel segment e, the angle between the principal direction and the reference direction of vessel segment e, and the local microvascular density around vessel segment e. Let the centerline point sequence of vessel segment e be... .

[0054] The formula for calculating the length of vascular segment e is:

[0055] The formula for calculating the width of vessel segment e is:

[0056] The formula for calculating the angle between the principal direction and the reference direction of vessel segment e is:

[0057] The formula for calculating the local small vessel density around vessel segment e is:

[0058] in, Let e ​​be the length of the blood vessel segment. The number of centerline points contained in vascular segment e. Let i be the coordinates of the (i+1)th centerline point. Let be the coordinates of the i-th centerline point. To indicate and The Euclidean distance between them Let e ​​be the width of the blood vessel segment. Let x be the distance from position x to the nearest background boundary. Let be the angle between the principal direction and the reference direction of vessel segment e. Let be the principal direction angle of vessel segment e. For reference direction, This refers to the local small vessel density surrounding vessel segment e. Representing the neighborhood The number of inner pixels, This represents the vascular enhancement response value at a small scale.

[0059] The connectivity of a vessel segment in a vascular topology map refers to whether vessel segment e is directly connected to or adjacent to a vessel segment already marked as a type I vessel. In the vascular topology map... In this system, each vessel segment is connected to a node (endpoint or bifurcation point) at both ends. If two different vessel segments share the same node, then the two vessel segments are directly connected.

[0060] Let the set of the first class of blood vessel segments be . For the vessel segment e to be classified, examine its two endpoint nodes. If either endpoint node contains another vessel segment... If so, then vascular segment e is considered to be connected to a type I vessel. This connection is denoted as... The value can be either true or false.

[0061] After completing the above calculations, the server performs an initial classification of each vessel segment e according to the following method, obtaining the initial classification result. The specific classification method is as follows: When the length of a blood vessel segment exceeds a preset first length threshold, the width of the blood vessel segment exceeds a preset first width threshold, and the angle between the main direction of the blood vessel segment and the reference direction is less than an angle threshold, the blood vessel segment is initially classified as a first type of blood vessel extending along the reference direction. The preset first length threshold can be between 0.5 mm and 2 mm; the preset first width threshold can be between 0.1 mm and 0.3 mm; and the angle threshold can be between 15 degrees and 45 degrees. These thresholds can be adaptively adjusted based on image resolution, tissue size, and blood vessel distribution.

[0062] When a vascular segment is connected to a previously classified first-category vessel, i.e. If the condition is true and the angle between the main direction and the reference direction of the vessel segment is greater than the branch angle threshold, the vessel segment is initially classified as a second type of vessel extending outward from the first type of vessel; where the branch angle threshold can be taken as 30 degrees to 60 degrees.

[0063] When the length of a vessel segment is less than a preset second length threshold, the width of a vessel segment is less than a preset second width threshold, or the density of local small vessels around the vessel segment is greater than a density threshold, the vessel segment is initially classified as a third type of vessel, either punctate or clustered. The preset second length threshold can be between 0.1 mm and 0.5 mm; the preset second width threshold can be between 0.03 mm and 0.1 mm; and the density threshold can be between 0.3 and 0.7. If a blood vessel segment meets the criteria for multiple categories, it can be determined according to priority. For example, the first category can be determined first, then the second, and finally the third. Alternatively, the category that best matches can be selected based on the principle of maximum confidence.

[0064] This application introduces multiple attributes into vascular segments, enabling them to be classified into three categories with different semantic meanings. The first category corresponds to main vessels, the second category to branch vessels, and the third category to punctate or clustered small vessels. This classification method does not rely on a large amount of labeled data, has clear rules, and is easy to implement and interpret.

[0065] S204. The server corrects the initial classification result based on the vascular topology map to obtain the target segmentation result.

[0066] In S203, the server has performed an initial classification of each vessel segment based on its length, width, directional angle, connectivity, and local small vessel density. However, due to potential vessel breaks, weak signals, artifacts, or noise in the photoacoustic image, the initial classification results may contain inconsistencies. For example, a complete main vessel may be segmented into multiple short segments due to local signal attenuation, leading to their incorrect classification as category II or III; or, an isolated artifact region may be misclassified as category I or II. This application utilizes information such as connectivity, spatial arrangement, and directional consistency in the vessel topology map to correct the initial classification results, thereby obtaining more accurate and coherent target segmentation results.

[0067] The correction process includes several sub-operations, which can be executed in any order or iteratively.

[0068] In some embodiments, when multiple vessel segments are arranged end-to-end along a reference direction in a vessel topology map, the principal direction of each vessel segment is consistent with the reference direction, the angle between the principal direction of each vessel segment and the reference direction is less than an angle threshold, and the gap between adjacent vessel segments is less than a gap threshold, the multiple vessel segments are merged into a single vessel segment of the first type of vessel, thereby obtaining the target segmentation result. The gap threshold can be between 0.2 mm and 0.5 mm.

[0069] In some embodiments, when a vessel segment is initially classified as a Class I or Class II vessel but is not connected to other vessel segments in the vascular topology map, the vessel segment is corrected to background; because real vascular networks are usually connected, isolated long vessels or branching vessels are likely misclassified due to artifacts or noise. This operation helps to eliminate isolated areas caused by extra-tissue bright spots, edge highlighting artifacts, or random noise.

[0070] In some embodiments, when a vessel segment is not initially classified as a Class II vessel, but the segment is connected to a Class I vessel (sharing the same node), and the angle between the segment's principal direction and the reference direction is greater than a branch angle threshold (e.g., 45 degrees), it is corrected to a Class II vessel. This operation can correct vessel segments that were not initially classified as Class II due to their length or width being slightly below the threshold, but should actually belong to branch vessels.

[0071] In some embodiments, when the distance between two endpoints is less than a distance threshold (e.g., 0.1 mm to 0.5 mm), the local orientation angle between the two endpoints is less than an angle threshold, and the average value of the vascular enhancement response on the connection path between the two endpoints is greater than a response threshold (e.g., 0.3 to 0.5), a completion path is generated between the two endpoints, and the region corresponding to the completion path is incorporated into the vascular candidate region to obtain the target segmentation result. The connection path can be a straight line, a spline curve, or a minimum cost path based on the vascular enhancement map. After completion, the pixels on the connection path are added to the vascular candidate region, and the skeleton is re-extracted and the topology map is updated. The completion operation is mainly used to repair minor breaks caused by weak signals or local noise, thereby improving the integrity of the vascular network.

[0072] In addition to the operations described above, the server can also make other corrections based on topology consistency. For example, if a vessel segment is classified as a type II vessel, but its connected parent vessel is classified as type III instead of type I, it can be corrected based on the category confidence of adjacent slices. Furthermore, in areas with high local small vessel density, if a vessel segment is classified as either type I or type II while its surrounding segments are all type III, it can be adjusted to type III to maintain local category consistency.

[0073] After the above corrections, the server obtains the final target segmentation result. The target segmentation result includes segmentation images for three types of blood vessels: the first type, the second type, and the third type. Each type of blood vessel segmentation image can be a binary mask, where a pixel value of 1 indicates that the location belongs to that type of blood vessel, and a pixel value of 0 indicates that it does not belong. Alternatively, the three results can be merged into a single color-coded segmentation image, with different colors representing different categories.

[0074] In addition, the server can output optional results, such as category confidence plots, uncertain region marker plots, skeleton plots, or topology plots, to assist manual review or subsequent analysis.

[0075] This application utilizes the connectivity and spatial arrangement information in vascular topology maps to merge broken main vessels, eliminate isolated misclassified artifacts, supplement branch vessel categories, and repair minor breaks, thereby significantly improving the continuity of vessel segmentation and the accuracy of classification. The corrected target segmentation results have clear semantic categories, facilitating medical visualization, quantitative analysis, and manual review.

[0076] The above text combined Figures 1 to 2 The photoacoustic blood vessel image segmentation method provided in the embodiments of this application has been described in detail. The apparatus and equipment provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0077] This application also provides a photoacoustic blood vessel image segmentation device, such as... Figure 3 As shown in the figure, this is a schematic diagram of a photoacoustic blood vessel image segmentation device provided in an embodiment of this application. The device includes: The acquisition module 301 is used to acquire photoacoustic vascular images and determine a vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular images. The classification module 302 is used to obtain an initial classification result based on the length of each blood vessel segment, the width of each blood vessel segment, the angle between the main direction and the reference direction of each blood vessel segment, the connection relationship of each blood vessel segment in the blood vessel topology map, and the local small blood vessel density around the blood vessel segment; wherein, the initial classification result includes a first type of blood vessel extending along the reference direction, a second type of blood vessel extending outward from the first type of blood vessel, and a third type of blood vessel in the form of dots or clusters; The correction module 303 is used to correct the initial classification result based on the blood vessel topology map to obtain the target segmentation result.

[0078] In some possible implementations, the classification module 302 is specifically used to initially classify the blood vessel segment as a first type of blood vessel extending along the reference direction when the length of the blood vessel segment is greater than a preset first length threshold, the width of the blood vessel segment is greater than a preset first width threshold, and the angle between the main direction and the reference direction of the blood vessel segment is less than an angle threshold; when the blood vessel segment is connected to the classified first type of blood vessel and the angle between the main direction and the reference direction of the blood vessel segment is greater than a branch angle threshold, the blood vessel segment is initially classified as a second type of blood vessel extending outward from the first type of blood vessel; when the length of the blood vessel segment is less than a preset second length threshold, the width of the blood vessel segment is less than a preset second width threshold, or the density of local small blood vessels around the blood vessel segment is greater than a density threshold, the blood vessel segment is initially classified as a third type of blood vessel, which is dot-like or cluster-like.

[0079] In some possible implementations, the correction module 303 is specifically used to merge the multiple blood vessel segments into a blood vessel segment of a first type of blood vessel when multiple blood vessel segments are arranged end to end along the reference direction in the blood vessel topology map, the main direction of each blood vessel segment is consistent with the reference direction, the angle between the main direction of each blood vessel segment and the reference direction is less than an angle threshold, and the gap between adjacent blood vessel segments is less than a gap threshold, thereby obtaining the target segmentation result.

[0080] In some possible implementations, the correction module 303 is further configured to: correct a blood vessel segment to background when it is initially classified as a first-class or second-class blood vessel but is not connected to other blood vessel segments in the blood vessel topology map; correct a blood vessel segment to second-class blood vessel when it is connected to a first-class blood vessel and the angle between the main direction and the reference direction of the blood vessel segment is greater than the branch angle threshold; and generate a completion path between the two endpoints and incorporate the region corresponding to the completion path into the blood vessel candidate region to obtain the target segmentation result when the distance between the two endpoints is less than the distance threshold, the angle between the local directions of the two endpoints is less than the angle threshold, and the average value of the blood vessel enhancement response on the connection path between the two endpoints is greater than the response threshold.

[0081] In some possible implementations, the acquisition module 301 is specifically used to determine the target tissue region based on the photoacoustic vascular image; perform multi-scale vascular enhancement processing on the photoacoustic vascular image to obtain a vascular enhancement map; extract vascular candidate regions within the target tissue region based on the vascular enhancement map; extract a skeleton from the vascular candidate regions; and construct a vascular topology map composed of nodes and vascular segments based on the skeleton.

[0082] In some possible implementations, the acquisition module 301 is specifically used to preset multiple scale parameters and multiple direction parameters; calculate the vascular enhancement response value under each scale parameter and each direction parameter based on the photoacoustic vascular image; fuse the vascular enhancement response values ​​of multiple directions under the same scale parameter to obtain the scale response map of that scale parameter; and fuse the scale response maps of all scale parameters to generate a vascular enhancement map.

[0083] In some possible implementations, the acquisition module 301 is specifically used to acquire an initial photoacoustic vascular image; and to preprocess the initial photoacoustic vascular image to obtain a photoacoustic vascular image.

[0084] The photoacoustic blood vessel image segmentation apparatus according to the embodiments of this application can correspondingly execute the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the photoacoustic blood vessel image segmentation apparatus are respectively for implementing Figure 2 For the sake of brevity, the corresponding processes of each method in the illustrated embodiments will not be described in detail here.

[0085] This application also provides a computing device. For example... Figure 4 As shown in the figure, this is a schematic diagram of a computing device provided in an embodiment of this application. The computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0086] Bus 401 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0087] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0088] The communication interface 403 is used for communication with external devices. For example, if the computing device is a first switch, the communication interface 403 can be used for communication between the first switch and a first user terminal, or for communication between the first switch and a second switch.

[0089] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0090] The memory 404 stores executable code, and the processor 402 executes the executable code to perform the aforementioned photoacoustic blood vessel image segmentation method.

[0091] Specifically, in achieving Figure 3 In the case of the illustrated embodiment, and Figure 3 When the modules or units of the photoacoustic blood vessel image segmentation device described in the embodiments are implemented by software, the following steps are performed: Figure 3 The software or program code required for the functions of each module / unit can be partially or entirely stored in memory 404. Processor 402 executes the program code corresponding to each unit stored in memory 404 to perform the aforementioned photoacoustic blood vessel image segmentation method.

[0092] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium capable of being stored by a computing device, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct a computing device to perform the aforementioned photoacoustic vessel image segmentation method.

[0093] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0094] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0095] When the computer program product is executed by a computer, the computer performs any of the aforementioned photoacoustic blood vessel image segmentation methods. The computer program product can be a software installation package; when any of the aforementioned photoacoustic blood vessel image segmentation methods needs to be used, the computer program product can be downloaded and executed on the computer.

[0096] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A photoacoustic blood vessel image segmentation method, characterized in that, The method includes: Acquire photoacoustic images of blood vessels; Based on the photoacoustic vascular image, a vascular topology consisting of nodes and vascular segments is determined; An initial classification result is obtained based on the length of each vessel segment, the width of each vessel segment, the angle between the main direction and the reference direction of each vessel segment, the connection relationship of each vessel segment in the vessel topology map, and the local small vessel density around the vessel segment; wherein, the initial classification result includes a first type of vessel extending along the reference direction, a second type of vessel extending outward from the first type of vessel, and a third type of vessel in the form of dots or clusters. The initial classification result is corrected based on the aforementioned vascular topology map to obtain the target segmentation result.

2. The method according to claim 1, characterized in that, The initial classification result is obtained based on the length of each vessel segment, the width of each vessel segment, the angle between the principal direction and the reference direction of each vessel segment, the connectivity of each vessel segment in the vascular topology map, and the density of local small vessels around the vessel segment. This includes: When the length of the blood vessel segment is greater than a preset first length threshold, the width of the blood vessel segment is greater than a preset first width threshold, and the angle between the main direction of the blood vessel segment and the reference direction is less than an angle threshold, the blood vessel segment is initially classified as the first type of blood vessel extending along the reference direction. When the vessel segment is connected to the first type of vessel that has been classified and the angle between the main direction and the reference direction of the vessel segment is greater than the branch angle threshold, the vessel segment is initially classified as the second type of vessel extending outward from the first type of vessel. When the length of the blood vessel segment is less than a preset second length threshold, the width of the blood vessel segment is less than a preset second blood vessel width threshold, or the density of local small blood vessels around the blood vessel segment is greater than a density threshold, the blood vessel segment is initially classified as the third type of blood vessel, which is either dot-like or cluster-like.

3. The method according to claim 1, characterized in that, The step of correcting the initial classification result based on the vascular topology map to obtain the target segmentation result includes: When multiple blood vessel segments are arranged end-to-end along the reference direction in the blood vessel topology map, the main direction of each blood vessel segment is consistent with the reference direction, the angle between the main direction of each blood vessel segment and the reference direction is less than the angle threshold, and the gap between adjacent blood vessel segments is less than the gap threshold, the multiple blood vessel segments are merged into a blood vessel segment of the first type of blood vessel to obtain the target segmentation result.

4. The method according to claim 3, characterized in that, The method further includes: When a vessel segment is initially classified as a Class I or Class II vessel but is not connected to other vessel segments in the vessel topology map, the vessel segment is corrected to background. When a vessel segment is connected to a classified first-class vessel and the angle between the main direction and the reference direction of the vessel segment is greater than the branch angle threshold, the vessel segment is corrected to a second-class vessel. When the distance between two endpoints is less than the distance threshold, the local directional angle between the two endpoints is less than the angle threshold, and the average value of the enhanced blood vessel response on the connection path between the two endpoints is greater than the response threshold, a completion path is generated between the two endpoints and the region corresponding to the completion path is incorporated into the blood vessel candidate region to obtain the target segmentation result.

5. The method according to claim 1, characterized in that, The step of determining the vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular image includes: The target tissue region is determined based on the photoacoustic vascular image; The photoacoustic vascular image is subjected to multi-scale vascular enhancement processing to obtain a vascular enhancement map; Based on the enhanced vascular map, candidate vascular regions are extracted within the target tissue region; A skeleton is extracted from the candidate blood vessel region, and a blood vessel topology map consisting of nodes and blood vessel segments is constructed based on the skeleton.

6. The method according to claim 5, characterized in that, The process of performing multi-scale vascular enhancement processing on the photoacoustic vascular image to obtain a vascular enhancement map includes: Multiple scale parameters and multiple orientation parameters are preset; Based on the photoacoustic vascular image, the vascular enhancement response value is calculated for each scale parameter and each orientation parameter; By fusing the vascular enhancement response values ​​in multiple directions under the same scale parameter, a scale response map for that scale parameter is obtained. The scale response maps of all scale parameters are fused to generate a vascular enhancement map.

7. The method according to claim 1, characterized in that, The acquisition of photoacoustic vascular images includes: Acquire initial photoacoustic vascular images; The initial photoacoustic vascular image is preprocessed to obtain a photoacoustic vascular image.

8. A photoacoustic blood vessel image segmentation device, characterized in that, The device includes: The acquisition module is used to acquire photoacoustic vascular images and determine a vascular topology map composed of nodes and vascular segments based on the photoacoustic vascular images. The classification module is used to obtain an initial classification result based on the length of each blood vessel segment, the width of each blood vessel segment, the angle between the main direction and the reference direction of each blood vessel segment, the connection relationship of each blood vessel segment in the blood vessel topology map, and the local small blood vessel density around the blood vessel segment; wherein, the initial classification result includes a first type of blood vessel extending along the reference direction, a second type of blood vessel extending outward from the first type of blood vessel, and a third type of blood vessel in the form of dots or clusters. The correction module is used to correct the initial classification result based on the blood vessel topology map to obtain the target segmentation result.

9. A computing device, characterized in that, Including memory and processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method as described in any one of claims 1 to 7.

Images