A three-dimensional photoacoustic representative cross-section determination method, device, equipment and medium

Abstract

The application discloses a three-dimensional photoacoustic representative cross section determination method, device, equipment and medium, and relates to the technical field of medical imaging. The method comprises the following steps: acquiring a candidate cross section sequence of a tissue to be analyzed; performing similarity calculation, blood vessel structure feature calculation, anatomical position assessment, adjacent slice consistency assessment and artifact detection on each candidate cross section in the candidate cross section sequence to obtain principal component similarity, a blood vessel structure score, a position score, a consistency score and an artifact penalty value of each candidate cross section; determining a comprehensive score of each candidate cross section based on the principal component similarity, the blood vessel structure score, the position score, the consistency score and the artifact penalty value; and screening a representative cross section based on the comprehensive score of each candidate cross section. The method can select a representative cross section.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and in particular to a method, apparatus, device and medium for determining a three-dimensional photoacoustic representative section. Background Technology

[0002] Photoacoustic imaging can acquire the three-dimensional vascular structure of peripheral tissues. For tissues with a distinct long axis, such as fingers and toes, it is often necessary to determine a representative cross-section from the three-dimensional data for quantitative analysis and case comparison.

[0003] The existing methods for determining representative sections mainly include manual selection based on experience, selection based on fixed slice number or geometric center, selection based on maximum intensity or maximum vascular area, or selection after quality scoring of a single slice.

[0004] However, the above method relies on a single indicator for judgment, which cannot guarantee the consistency of the anatomical location of the selected section among different samples, and is prone to misjudging abnormal structures such as boundary artifacts, ring artifacts, and local saturation as representative sections, ultimately resulting in the selected representative section not being representative. Summary of the Invention

[0005] This application provides a method, apparatus, device, and medium for determining a three-dimensional photoacoustic representative cross section, which can select a representative cross section.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a method for determining a three-dimensional photoacoustic representative section, including: Obtain candidate cross-sectional sequences of the tissue to be analyzed; Similarity is calculated for each candidate section in the candidate section sequence to obtain the principal component similarity; vascular structure features are calculated for each candidate section in the candidate section sequence to obtain the vascular structure score; anatomical location is evaluated for each candidate section in the candidate section sequence to obtain the location score; consistency between adjacent slices is evaluated for each candidate section in the candidate section sequence to obtain the consistency score; artifact detection is performed on each candidate section in the candidate section sequence to obtain the artifact penalty value. The comprehensive score for each candidate section is determined based on principal component similarity, vascular structure score, location score, consistency score, and artifact penalty value. Based on the comprehensive scores of each candidate section, a representative section is selected.

[0007] In some possible implementations, a sequence of candidate cross-sections of the tissue to be analyzed is obtained, including: Acquire three-dimensional photoacoustic data of the tissue to be analyzed; Threshold segmentation is performed on the three-dimensional photoacoustic volume data to obtain an initial binary mask; Morphological closing operations, hole filling, and maximum connected component filtering are performed on the initial binary mask to obtain a three-dimensional tissue mask. Based on the three-dimensional mask of the tissue, the candidate cross-section sequence is determined.

[0008] In some possible implementations, candidate section sequences are determined based on a 3D tissue mask, including: Based on the tissue 3D mask, the target photoacoustic body data is determined; Based on the target photoacoustic body data, the long axis and spatial range of the long axis of the tissue are determined; Within the long axis space, multiple sections orthogonal to the long axis of the tissue are identified as candidate section sequences.

[0009] In some possible implementations, similarity is calculated for each candidate cross-section in the candidate cross-section sequence to obtain the principal component similarity of each candidate cross-section, including: Each candidate section in the candidate section sequence is converted into a one-dimensional vector, and a slice matrix is ​​constructed. The slice matrix is ​​centered to obtain a centered matrix; Principal component analysis is performed on the centered matrix to obtain the principal component vectors and their eigenvalues. The weights of each principal component vector are calculated based on the eigenvalues ​​corresponding to each principal component vector. For each candidate cross section, the similarity between the candidate cross section and each principal component vector is calculated. Then, each similarity is multiplied by the weight of the corresponding principal component vector and summed to obtain the principal component similarity of the candidate cross section.

[0010] In some possible implementations, a comprehensive score for each candidate cross-section is determined based on principal component similarity, vascular structure score, location score, consistency score, and artifact penalty value, including: For each candidate cross section, the principal component similarity, vascular structure score, location score, and consistency score of the candidate cross section are weighted and summed to obtain the weighted total score of the candidate cross section. The overall score of a candidate section is obtained by subtracting its artifact penalty value from its weighted total score.

[0011] In some possible implementations, representative sections are selected based on the comprehensive scores of each candidate section, including: The candidate cross section with the highest comprehensive score is selected as the initial cross section from all candidate cross sections; Determine whether the overall score of the initially selected section is not lower than the preset threshold; If yes, the initially selected section will be determined as the representative section; otherwise, an insufficient image quality message will be displayed.

[0012] In some possible implementations, artifact detection is performed on each candidate cross-section in the candidate cross-section sequence to obtain an artifact penalty value for each candidate cross-section, including: For each candidate section, calculate the edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-extra-tissue highlight index, and reconstruction missing region index for that candidate section. The edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-external highlight index, and reconstruction missing region index are weighted and fused to obtain the artifact penalty value for the corresponding candidate section.

[0013] Secondly, this application provides a three-dimensional photoacoustic representative section determination device, comprising: The acquisition module is used to acquire candidate cross-sectional sequences of the tissue to be analyzed. The calculation module is used to calculate the similarity of each candidate section in the candidate section sequence to obtain the principal component similarity of each candidate section; to calculate the vascular structure features of each candidate section in the candidate section sequence to obtain the vascular structure score of each candidate section; to evaluate the anatomical location of each candidate section in the candidate section sequence to obtain the location score of each candidate section; to evaluate the consistency between adjacent slices of each candidate section in the candidate section sequence to obtain the consistency score of each candidate section; and to detect artifacts of each candidate section in the candidate section sequence to obtain the artifact penalty value of each candidate section. The section determination module is used to determine the comprehensive score of each candidate section based on principal component similarity, vascular structure score, location score, consistency score and artifact penalty value; and to select representative sections based on the comprehensive score of each candidate section.

[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, candidate cross-sectional sequences of the tissue to be analyzed are obtained, and the principal component similarity, vascular structure score, anatomical location score, adjacent slice consistency score, and artifact penalty value are calculated for each candidate cross-section. These results are then fused to obtain a comprehensive score, and representative cross-sections are selected based on their scores. This method no longer relies on manual judgment, fixed slices, or a single indicator to determine the cross-section, but evaluates it from multiple perspectives, including overall structural changes, vascular visualization, anatomical location rationality, adjacent slice continuity, and artifact effects. This approach reduces the differences in results caused by human selection, resulting in more consistent results from different users and batches of data; it adapts to differences in tissue length, curvature, and placement angles, ensuring consistent anatomical location in the selected cross-sections; it avoids incorrectly selecting representative cross-sections due to artifacts, localized high-brightness noise, or abnormally isolated structures; it fully utilizes the continuity between adjacent slices, reducing the impact of single-layer noise, localized vascular rupture, or motion artifacts; and it considers more factors affecting representativeness, making the evaluation process more comprehensive and ultimately enabling the selection of representative cross-sections from three-dimensional photoacoustic volume data.

[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 An application environment diagram for a method for determining a three-dimensional photoacoustic representative section provided in this application embodiment; Figure 2 A flowchart illustrating a method for determining a three-dimensional photoacoustic representative section provided in an embodiment of this application; Figure 3 A structural diagram of a three-dimensional photoacoustic representative section determination 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] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the related technologies is given first: Photoacoustic imaging is a hybrid imaging modality that combines the high contrast of optics with the high penetration depth of ultrasound. It uses short-pulse lasers to irradiate tissue, causing light absorbers (such as hemoglobin) within the tissue to absorb light energy and undergo thermoelastic expansion, generating ultrasonic signals. These signals are then received by an ultrasound probe and reconstructed to create an image of the light absorption distribution within the tissue. This method can clearly display the morphology and distribution of blood vessels without the use of external contrast agents and is suitable for the observation and quantitative analysis of blood vessels in peripheral tissues such as toes, fingers, auricles, and skin flaps.

[0023] In clinical and research analysis, it is usually necessary to select a representative cross section (i.e., a representative cross section) from the three-dimensional photoacoustic volume data, or to obtain a representative thin layer by fusing adjacent multilayers, for vascular segmentation, morphological quantification, case control, and equipment demonstration.

[0024] Current methods for selecting representative sections mainly include: manual selection based on experience, selection based on fixed layer number / geometric center, selection based on maximum photoacoustic intensity or maximum vessel area, and selection based solely on the quality score of a single layer image. While these methods are simple to implement, they cannot guarantee the representativeness of the selection results when there is noise, artifacts, local vascular rupture, or differences in tissue posture and anatomical length in the three-dimensional photoacoustic data.

[0025] In view of this, embodiments of this application provide a method for determining a three-dimensional photoacoustic representative section. 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 an application environment diagram provided by an embodiment of this application.

[0026] In this application environment, the server acts as the execution entity, communicating with the photoacoustic imaging device and the terminal device. The photoacoustic imaging device is used to acquire three-dimensional photoacoustic volume data of the tissue to be analyzed and upload the three-dimensional photoacoustic volume data to the server. The terminal device is used to submit a representative section acquisition request to the server and display the finally determined representative section and related scoring results. The server is used to receive the three-dimensional photoacoustic volume data and the corresponding request, perform preprocessing, tissue mask extraction, tissue long axis determination, candidate section generation, multi-dimensional scoring calculation and comprehensive screening, output standardized representative sections, representative thin layers and section scoring tables, and return the results to the terminal device.

[0027] To make the technical solution of this application clearer and easier to understand, the following describes a method for determining a three-dimensional photoacoustic representative section provided by an embodiment of this application, with the server as the execution entity, in conjunction with the above application scenario. Figure 2 As shown in the figure, this is a flowchart illustrating a method for determining a three-dimensional photoacoustic representative section according to an embodiment of this application. The method for determining a three-dimensional photoacoustic representative section includes: S201. The server obtains the candidate cross-sectional sequence of the tissue to be analyzed.

[0028] The tissue to be analyzed is a biological soft tissue with a local long-axis morphology, such as toes, fingers, auricles, skin flaps, or superficial vascular tissue. The candidate cross-sectional sequence is a set of orthogonal cross-sectional images extracted along the long axis of the tissue within a stable anatomical region.

[0029] One possible approach is for the server to acquire three-dimensional photoacoustic volume data of the tissue to be analyzed, perform threshold segmentation on the three-dimensional photoacoustic volume data to obtain an initial binary mask, perform morphological closing operation, hole filling and maximum connected component filtering on the initial binary mask to obtain a three-dimensional tissue mask, and determine a candidate cross-section sequence based on the three-dimensional tissue mask.

[0030] The 3D photoacoustic volume data is data D2 acquired by a photoacoustic imaging device. It is formed by stacking multiple layers of 2D photoacoustic tomographic images along the spatial direction and contains information on tissue light absorption distribution and vascular structure. The initial binary mask is preliminary binary data generated after thresholding the 3D photoacoustic volume data, used to roughly distinguish tissue regions from background regions. Morphological closing operation is an image processing operation performed on the initial binary mask, involving dilation followed by erosion, used to eliminate small gaps in the mask, smooth tissue contours, and connect minor fracture areas. Hole filling is an image processing operation that fills the closed cavities inside the mask with effective tissue regions. Maximum connected component filtering is a filtering operation that retains the largest connected regions in 3D space and removes isolated noise points. The 3D tissue mask is regularized 3D binary data representing the location and extent of the effective tissue region.

[0031] The process of determining the candidate section sequence based on the tissue 3D mask can be as follows: Based on the tissue 3D mask, determine the target photoacoustic volume data; based on the target photoacoustic volume data, determine the tissue long axis and the long axis spatial range; within the long axis spatial range, determine multiple sections orthogonal to the tissue long axis as the candidate section sequence.

[0032] Target photoacoustic volume data is obtained by constraining and filtering the original 3D photoacoustic volume data, retaining only the pure 3D photoacoustic data of the effective tissue region. For example, background and noise that do not belong to the tissue to be analyzed are removed from the original volume data, leaving only the 3D data of the main finger region. The tissue long axis is the direction along which the tissue extends the longest in 3D space, calculated based on the 3D coordinate distribution of the tissue 3D mask. The long axis spatial range is the standardized, stable anatomical region on the tissue long axis that is retained after removing unstable regions such as tip and proximal artifacts, and can be used for image selection.

[0033] For example, the server controls an electric translation stage to move the laser excitation module and the ultrasound probe layer by layer along the long axis of the tissue to be analyzed or a preset scanning direction. At each scanning position zc, the server controls the acquisition of one or more frames of two-dimensional photoacoustic tomography images Rc(a,b), where a and b are the pixel orientation indices within the two-dimensional image, and zc = z0 + c × Δz, where Δz is the translation stage step distance or the interval between adjacent slices recorded by the system. The server stacks the continuously acquired two-dimensional tomographic images according to their corresponding spatial positions to form initial three-dimensional photoacoustic volume data D0(a,b,c), with the following correspondence: D0(a,b,c)=Rc(a,b) xa = x0 + a × Δx yb=y0+b×Δy zc = z0 + c × Δz Δx and Δy are obtained from the 2D reconstructed pixel dimensions or system calibration file, while Δz is obtained from the translation stage step value, mechanical encoder reading, or reconstructed voxel spacing. If the 3D photoacoustic volume data is obtained directly from array acquisition and 3D reconstruction, then D0(a,b,c) is the reconstructed 3D voxel intensity, where a is the horizontal pixel index within the 2D tomographic image; b is the vertical pixel index within the 2D tomographic image; and c is the slice index in the scanning direction (depth direction), corresponding to the c-th layer of the 2D tomographic image. Δx, Δy, and Δz are given by the reconstruction parameters. If it is multi-wavelength data, the server can execute this method separately for each wavelength, or select a single-wavelength channel with strong hemoglobin absorption, or obtain a structural channel for slice selection by weighted averaging of multiple wavelengths.

[0034] The server can sequentially perform correction, normalization, and denoising on the initial 3D photoacoustic volume data D0(a,b,c) to obtain the preprocessed 3D photoacoustic volume data D2. For example, the server can first perform laser energy and system gain correction, assuming... This represents the laser energy value recorded during the acquisition of layer c. The receive gain or reconstruction gain value used for layer c. The reference value for laser energy correction is the median of the laser energy across all acquisition layers, or a user-preset fixed reference energy value. The reference value for system gain correction is the median of the receiving gain or reconstruction gain of all acquisition layers, or a user-preset fixed reference gain value; the server corrects the intensity of each layer to a uniform energy and uniform gain scale, using the following correction formula:

[0035] in, This represents the 3D photoacoustic volume data after laser energy and system gain correction; ε is a very small positive number to prevent the denominator from being zero. If there is no layer-by-layer energy or gain recording, then... and One can be chosen.

[0036] The server can also perform background subtraction. The background can be estimated from laser-free frames, extratissue regions, low quantile regions of volume data, or user-specified background regions. The background subtraction formula is:

[0037] in, The background signal intensity estimate can be obtained from a laser-free frame, an area outside the tissue, a low quantile region of volume data, or a user-specified background region. This is the 3D photoacoustic volume data after background subtraction processing.

[0038] To reduce intensity scale differences caused by different subjects, different scanning batches, and different coupling conditions, the server... Perform quantile normalization. Let... for The plow quantile within the valid volume data region; for Within the effective volume data region, the phigh quantile is set to 1% to 5% as the lower limit of normalization, and phigh to 95% to 99.9% as the upper limit of normalization. The normalization formula is:

[0039] in, This is the 3D photoacoustic volume data after quantile normalization; clip() is the truncation function used to restrict the calculation results to the specified interval [0,1].

[0040] Then, the server performs 3D denoising on Dn(a,b,c). Non-local mean filtering can be used. Let v be the voxel index to be processed, v′ be the neighboring voxel index within the search window Ω(v), P(v) be the 3D image patch centered at v, and H be the filtering intensity parameter of non-local mean filtering. Then, the formula for calculating the voxel value after denoising is:

[0041] in, This represents the voxel value corresponding to voxel v after 3D denoising. voxel v and neighboring voxels The weighting coefficients between them are used for nonlocal mean filtering calculation.

[0042] The weight calculation formula is:

[0043] Where exp() is an exponential function with the natural constant e as its base; denoted as L2 norm (Euclidean distance) between 3D image patches P(v) and P(v′), used to measure the similarity between the two patches. The search window size, image patch size, and H can be preset by the system or automatically determined based on the standard deviation of background noise. If computational resources are limited, the server can also use 3D median filtering, anisotropic diffusion filtering, or deblurring based on the system point spread function as alternatives; the output after these alternatives is still denoted as D2.

[0044] The server can also perform 3D thresholding on the preprocessed 3D photoacoustic volume data D2, or perform 2D thresholding on D2 after projecting the maximum intensity in one direction, and then map the 2D result back to 3D. The threshold Tmask can be automatically obtained by the server using the Otsu method, the background mean plus κ times the background standard deviation, or a fixed quantile of D2. The server obtains an initial binary mask M0(a,b,c) based on the thresholding. The segmentation rule is as follows: for each voxel of the preprocessed 3D photoacoustic volume data D2, a threshold judgment is performed. When the voxel intensity value is greater than or equal to the preset threshold Tmask, the voxel is marked as a tissue region (assigned a value of 1) in the initial binary mask M0; when the voxel intensity value is less than the preset threshold Tmask, the voxel is marked as a background region (assigned a value of 0) in the initial binary mask M0, thus generating the initial binary mask M0(a,b,c).

[0045] The server performs morphological closing operations, hole filling, and maximum connected component filtering on the initial binary mask M0(a,b,c) to remove small-area floating noise, resulting in the tissue 3D mask M(a,b,c). Based on the tissue 3D mask M(a,b,c), the set of voxel coordinates where M(a,b,c)=1 is selected, which is the target photoacoustic volume data R. Let v be the three-dimensional physical coordinates corresponding to voxel v, and μ be the centroid of the tissue voxel coordinates, i.e., the origin of the calculation of the tissue's major axis. Then the formula for calculating the centroid is:

[0046] Where |R| is the total number of voxels, i.e., the number of elements in R; Σ represents the summation of the coordinates of all voxels. In other words, the server calculates the average of the three-dimensional physical coordinates of all voxels in the tissue region to obtain the tissue centroid μ.

[0047] The server determines the major axis of the tissue based on the 3D coordinate covariance matrix of the tissue 3D mask. The calculation formula is:

[0048] That is, the server calculates the covariance matrix of the three-dimensional coordinates of tissue voxels. This matrix reflects the shape distribution and principal orientation of the organization in three-dimensional space; the server's covariance matrix Perform eigenvalue decomposition and take the eigenvector corresponding to the largest eigenvalue as the unit vector eL of the tissue's long axis. This vector represents the direction of the tissue's long axis. That is, the tissue's long axis = a straight line (or the centerline of curved tissue) with μ as the origin along the direction eL.

[0049] The distal and proximal orientations are determined by the server based on the mechanical scanning direction, manually selected distal and proximal markers, the position of the image's outer contour tip, or a preset orientation. The server projects the tissue voxels onto the long axis to obtain the long axis coordinates and normalized positions. The corresponding calculation formula is as follows:

[0050]

[0051]

[0052]

[0053] in, Let v be the projected coordinates (scalar) of voxel v on the long axis of the tissue, representing the position of voxel along the long axis of the tissue; For the three-dimensional physical coordinates (spatial coordinates) corresponding to voxel v; The first percentile of the projected coordinates of the tissue voxel on the long axis is used as a reference value for the lower limit of normalization. () is the percentile function, used to calculate the first percentile of data in a set; The 99th percentile of the projected coordinates of tissue voxels on the long axis is used as the upper limit reference value for normalization. ( ) is the percentile function, used to calculate the 99th percentile of data in a set; ηv is the normalized position of voxel v on the tissue long axis, with a value range of [0,1], representing the relative position of the voxel on the tissue long axis. That is, the long axis spatial range = along the eL direction, from arrive A one-dimensional interval.

[0054] The server operates within the preset stable range of the normalized major axis coordinate η. , Candidate cross sections are generated within the [section]. ηmax is used to exclude artifact regions at the distal tip and proximal end. and It can be preset by experience, such as the toes being preferred. =0.10、 =0.85. Let the total number of candidate sections be N, and the interval between candidate sections be Δη. Then the normalized position of the q-th candidate section is calculated by the server:

[0055]

[0056] in, Let q be the normalized major axis position of the q-th candidate section; This represents the normalized positional interval between adjacent candidate sections.

[0057] For an approximately linear tissue, the server uses the tissue centroid μ, the tissue major axis unit vector eL, and the major axis spatial range [ , Based on this, calculate the center point of the q-th candidate section. , can be represented as:

[0058] The server takes the normal vector of the q-th candidate section as the unit vector eL of the tissue major axis, ensuring that the candidate section is orthogonal to the tissue major axis; the server constructs two unit basis vectors orthogonal to eL. and The candidate cross-sectional plane Πq is determined as follows: Πq={ +α× +β× |α∈[-W / 2,W / 2],β∈[-H / 2,H / 2]} Where W is the physical width of the candidate cross-section; H is the physical height of the candidate cross-section. W and H can be obtained by adding a safety margin to the outer frame size of the tissue 3D mask on the local cross-section, or they can be preset by the system. α and β are the local coordinates in the candidate cross-section plane, with α varying along the eq1 direction and β varying along the eq2 direction. For curved tissue, the server takes the center point of the q-th candidate cross-section as a sampling point on the centerline L(τq), and the candidate cross-section normal vector is taken as the local tangent vector tq=dL(τq) / dτ, and a local orthogonal plane is constructed at this point. τq is the arc length normalization parameter of the curved tissue centerline, corresponding to the sampling position of the q-th candidate cross-section; The server performs trilinear interpolation on the preprocessed 3D photoacoustic volume data D2 to extract candidate cross-sectional images. The interpolation formula is:

[0059] in, is the two-dimensional image of the q-th candidate cross-section, corresponding to the voxel intensity value at the local coordinates (α,β) on the plane Πq; TriInterp is a trilinear interpolation function used to extract image values ​​on a specified plane from three-dimensional volume data.

[0060] The server obtains the candidate cross-section sequence through the above processing. , ,..., }

[0061] S202. Calculate the similarity of each candidate cross section in the candidate cross section sequence to obtain the principal component similarity of each candidate cross section.

[0062] Principal component similarity is calculated by comparing each candidate cross-section with the principal component vectors, then weighting and summing the results according to their eigenvalues. This yields the degree of consistency between the cross-section and the typical features of the overall data. For example, calculating the similarity between a cross-section and the first few principal components and weighting the results, a higher score indicates that the cross-section is more representative of the overall data.

[0063] One possible approach is to convert each candidate cross-section in the candidate cross-section sequence into a one-dimensional vector to construct a slice matrix; to center the slice matrix to obtain a centered matrix; to perform principal component analysis on the centered matrix to obtain each principal component vector and its eigenvalue; to calculate the weight of each principal component vector based on its eigenvalue; and to calculate the similarity between each candidate cross-section and each principal component vector, and then multiply each similarity by its corresponding weight and sum them to obtain the principal component similarity of the candidate cross-section.

[0064] The slice matrix is ​​a two-dimensional feature matrix formed by flattening each two-dimensional cross-sectional image in the candidate cross-sectional sequence into a one-dimensional vector and then concatenating them sequentially by row / column. The centering matrix is ​​obtained by removing the mean from the slice matrix; its purpose is to eliminate overall bias and make principal component analysis more accurate. Each principal component vector is a set of orthogonal basis vectors extracted after principal component analysis of the centering matrix, representing the most prevalent morphological changes in the candidate cross-sectional data. The eigenvalues ​​of each principal component vector are their variance contributions, indicating how much data variation this principal component can explain; larger eigenvalues ​​indicate greater importance.

[0065] For example, the server standardizes the candidate cross-section sequence and constructs a slice matrix. The server resamples each candidate cross-section to a uniform pixel size H0×W0. H0 and W0 are determined based on the workstation display resolution, the input size of the segmentation algorithm, or a fixed physical sampling interval. The server rotates the candidate cross-sections according to the two-dimensional principal axis or eq1 / eq2 direction of the tissue mask within the cross-section, so that the short and long axes of the tissue in different samples are consistent.

[0066] The server performs local intensity normalization on each candidate cross-section. Let Iqr(s) be the intensity of the q-th cross-section at pixel s after resampling and orientation normalization, Qq_low be the low quantile within the tissue mask of this cross-section, Qq_high be the high quantile within the tissue mask of this cross-section, and clip be the cutoff function. Then, the q-th normalized cross-section image J... q The formula for calculating (s) is: J q (s)=clip((Iqr(s)-Qq_low) / (Qq_high-Qq_low+ε),0,1) The server will J q (s) is expanded into a vector according to a fixed pixel order. The vector length is P = H0 × W0. The server will arrange all candidate section vectors by row to construct the slice matrix X, calculated using the following formula:

[0067]

[0068] Where vec() is the vectorization operator. For vectors The transpose of , where N is the total number of candidate sections.

[0069] The server centers the slice matrix to obtain a centered matrix. The server then calculates the mean vector of all candidate slice vectors. The calculation formula is:

[0070] The server will generate each candidate cross-section vector Subtract the mean vector , thus obtaining the centered vector The calculation formula is:

[0071] The matrix consisting of all the centered vectors is the centered matrix Z.

[0072] The server performs principal component analysis on the centralized matrix Z to obtain the principal component vectors. and the eigenvalues ​​corresponding to each principal component vector The server calculates the weights of each principal component vector based on its corresponding eigenvalues. The calculation formula is:

[0073] Where K is the number of principal components selected. Let be the weight of the t-th principal component vector. Let be the eigenvalue corresponding to the t-th principal component vector.

[0074] For each candidate cross-section, the server calculates the similarity between that cross-section and each principal component vector. Each similarity is then multiplied by its corresponding principal component vector weight and summed to obtain the principal component similarity of that candidate cross-section. The calculation formula is:

[0075]

[0076] Among them, cos( , ) is the centered vector With principal component vectors cosine similarity, × For vector dot product, For vectors L2 norm, Principal Component Vector The L2 norm.

[0077] S203. Calculate the vascular structure features of each candidate section in the candidate section sequence to obtain the vascular structure score of each candidate section.

[0078] Vascular structure feature calculation is the process of extracting and quantifying features such as morphology, distribution, and integrity of blood vessels within a candidate cross-section. For example, on a toe cross-section image, the vascular signal is first enhanced, then the blood vessels are extracted, and their area, length, and intensity are calculated. The vascular structure score is a quantitative indicator used to measure the richness and structural integrity of blood vessels in a candidate cross-section. For example, the area, length, and energy of blood vessels in a cross-section are fused according to weights to obtain a score between 0 and 1. The higher the score, the better the vascular display of this cross-section and the more suitable it is as a representative cross-section.

[0079] For example, the server performs multi-scale vessel enhancement processing on the standardized cross-section, calculates the second derivative matrix of the cross-section and its eigenvalues ​​at multiple scales σ∈Σσ, and obtains the vessel enhancement value of each pixel based on the linear structure response. The calculation formula is:

[0080] Server-side vascular enhancement value Adaptive threshold segmentation is performed to obtain the initial blood vessel candidate region. The segmentation rules are as follows: If (s)=1, then ≥ And s∈ ;otherwise (s)=0 in, The threshold for blood vessel segmentation. Let q be the two-dimensional tissue mask corresponding to the q-th candidate cross-section.

[0081] The server initializes the blood vessel candidate region. Perform small connected component removal and skeletonization, where the area to be removed is smaller than the minimum area A. min Or the skeleton length is less than the minimum length L min The noisy connected domain is used to obtain the final vascular region. (s).

[0082] The server calculates the vascular response energy. Blood vessel area ratio R qarea Length of the vascular skeleton L qskel and the length L of the main connected component qmain The calculation formulas are as follows:

[0083]

[0084] L qskel =pixel_size×length[skeleton(Bq)]

[0085] The server uses a robust normalization function to normalize each feature. The normalization function is as follows:

[0086] in, It is a placeholder variable, which could be the vascular response energy. The feature is any one of the following four features: blood vessel area ratio Rqarea, blood vessel skeleton length Lqskel, and main connected region length Lqmain. Q5 is the 5th quantile of the feature in all candidate cross sections, and Q95 is the 95th quantile of the feature in all candidate cross sections.

[0087] The server performs a weighted summation of the normalized features to obtain the vascular structure score for the q-th candidate cross-section. The calculation formula is:

[0088] in, , , , The weights are non-negative and satisfy the following conditions: + + + =1.

[0089] S204. Perform anatomical position evaluation on each candidate section in the candidate section sequence to obtain the position score of each candidate section.

[0090] Anatomical location assessment is a process of evaluating the rationality and standardization of the location of a candidate section based on the anatomically stable region along the tissue's long axis. For example, it involves first calculating the normalized position of the candidate section along the tissue's long axis, and then determining whether it falls within a stable region excluding tip and proximal artifacts; the closer to the center of the region, the higher the score. The location score is a quantitative indicator characterizing the degree to which a candidate section is in a standard anatomical position and suitable as a representative section. For example, using a Gaussian weighted function, sections located at the very center of the stable region have a score close to 1; sections near tip or proximal artifact areas have a score close to 0.

[0091] For example, the server determines the candidate section based on its normalized position along the long axis of the tissue. Calculate position score A central weighted function is used to evaluate the cross-sectional location, giving higher scores to candidate cross-sections located in the middle of the anatomically stable region, and lower scores to candidate cross-sections near the distal tip, proximal joint, or areas with incomplete tissue masking. The calculation formula is as follows:

[0092] in, The position score for the q-th candidate section. Let q be the normalized position of the q-th candidate section on the long axis of the tissue. This is the center of the anatomical stable zone. , This represents the lower limit of the anatomical stability interval. This represents the upper limit of the stable anatomical region. This is the weighted width coefficient.

[0093] Piecewise functions can also be used for calculation: the candidate section is located at [ , The score is higher when the interval is within the stable interval, decreases linearly when the interval is at the edge of the stable interval, and is 0 when the interval is outside the stable interval.

[0094] S205. Perform adjacent slice consistency evaluation on each candidate section in the candidate section sequence to obtain the consistency score of each candidate section.

[0095] Adjacent slice consistency assessment compares the current candidate slice with several adjacent slices to determine whether the structure is continuous and the morphological transition is smooth. This process can be used to eliminate abnormal slices such as single-layer noise, vascular rupture, and local artifacts. For example, the 8th slice is compared with the 6th, 7th, 9th, and 10th slices to check the continuity of blood vessel position, contour, and signal. The consistency score is a quantitative indicator representing the degree of morphological matching and structural transition smoothness of a single candidate slice with its surrounding adjacent slices. For example, a slice that is highly consistent with its neighboring slices will have a score close to 1; a slice that is suddenly broken, deformed, or noisy will have a very low score.

[0096] For example, consistency score This is used to determine whether the q-th candidate section belongs to a stable tissue structure within its neighborhood, rather than being random noise generated by a single slice. The neighborhood set of the q-th candidate section is defined. The expression is:

[0097] in, is the index of the neighborhood section; N is the total number of candidate sections; m is the size of the neighborhood window, where m ranges from 1 to 5, or can be determined by calculating the preset physical thickness and the spacing between candidate sections.

[0098] The server calculates the correlation coefficient of cross-sectional intensity, the correlation coefficient of vascular enhancement image, and the Descein coefficient of vascular mask, respectively, using the following formulas:

[0099] in, The cross-sectional strength correlation coefficient measures the original strength similarity between the q-th and h-th cross sections. The correlation coefficient of the vascular enhancement map measures the similarity of the vascular enhancement results; is the Dice coefficient of the blood vessel mask, which measures the overlap of blood vessel regions; wec() is a vectorization operation that converts a two-dimensional image into a one-dimensional vector.

[0100] By combining the coefficients and their corresponding weights, the consistency score of the q-th candidate cross-section is calculated:

[0101] in, γI represents the consistency score of the q-th candidate section; γI, γV, and γB are weighting coefficients, all of which are non-negative and their sum is 1. If a candidate section has no corresponding neighboring slices, the mean is calculated based solely on the existing neighboring slices.

[0102] S206. Perform artifact detection on each candidate section in the candidate section sequence to obtain the artifact penalty value for each candidate section.

[0103] Artifact detection is the process of automatically identifying imaging defects in each candidate cross-section. It specifically detects various abnormal regions and distortion signals that interfere with image selection, preventing problematic cross-sections from being selected as representative cross-sections. The artifact penalty value is a quantitative indicator used to quantify the degree to which a cross-section is affected by artifacts and to deduct points from the score of cross-sections containing artifacts.

[0104] One possible approach is to calculate, for each candidate cross-section, the edge highlight ratio, annular artifact index, stripe artifact index, saturated pixel ratio, out-of-tissue highlight index, and reconstruction missing region index for that candidate cross-section; and to perform weighted fusion of the edge highlight ratio, annular artifact index, stripe artifact index, saturated pixel ratio, out-of-tissue highlight index, and reconstruction missing region index to obtain the artifact penalty value for the corresponding candidate cross-section.

[0105] Edge highlight ratio describes the proportion of abnormally bright pixels at the edges of a candidate cross-section, used to measure whether the cross-section has problems such as excessively bright edges and contour distortion. For example, by dividing the tissue edge into a boundary zone and counting the proportion of excessively bright pixels, a higher proportion indicates more severe edge artifacts. The ring artifact index is a quantized value calculated through polar coordinate transformation and circumferential intensity analysis, used to detect the presence of concentric circular distortion artifacts in the cross-section. For example, if bright rings appear outward from the center of the image, the intensity and coverage of the rings are calculated to obtain the ring artifact index. The stripe artifact index is a quantized index calculated through frequency domain analysis or Radon transform, used to detect the presence of linear, radial, or periodic stripe interference artifacts in the candidate cross-section. For example, if an image shows vertical / diagonal bright and dark stripes, frequency domain features are extracted and quantized; a higher value indicates more severe stripe artifacts. The saturation pixel ratio describes the proportion of pixels with extreme intensity within the candidate cross-section. For example, if a pixel value reaches its maximum range and no longer changes, it is considered saturated; the proportion of such pixels in the tissue is counted. The high-brightness index outside the tissue is a comprehensive indicator of the number, density, and number of connected components of abnormal bright spots detected outside the tissue area. It is used to eliminate background noise interference, such as isolated bright spots or strong light noise next to the tissue, and to calculate their degree of influence. The reconstruction missing area index is the proportion of the area of ​​the identified defective areas with no signal, low signal, or failed reconstruction within the tissue. It is used to determine whether the cross-section is complete and usable. For example, if there are black holes, faults, or unreconstructed areas inside the tissue, I calculate their size proportions to obtain the reconstruction missing area index.

[0106] For example, the artifact penalty value calculated by the server To avoid misselection of candidate sections caused by various imaging defects, the calculation process incorporates standardized sections. Two-dimensional tissue mask Enhanced vascular image and the original unnormalized cross section The extraction of various artifact metrics was completed. The server relies on a two-dimensional tissue mask. Divide the organization into boundary zones and a central region, using radius... Perform an erosion operation on the mask to obtain Boundary zone Obtained by subtracting the etched mask from the original mask, the central region This refers to the mask after etching, and selection... High quantiles within the tissue mask are used as the highlight threshold. Calculate the density of bright pixels in the boundary zone and the central region separately, and then calculate the edge highlight ratio using the following formula. :

[0107]

[0108]

[0109] If the boundary band area is too small, the server expands pixels bidirectionally along the outer contour of the mask to redefine the boundary band. Here, Jq(s) is the q-th normalized cross-sectional image; High-brightness pixel density within the boundary band; High-brightness pixel density in the central area.

[0110] server With the two-dimensional centroid as the origin of the polar coordinate system, convert the image to polar coordinate form. , Represents the polar radius. Representing the polar angle, the circumferential average strength is calculated per polar radius. Standardization intensity With high brightness corner coverage Combined with normalized threshold The ring artifact index is calculated using the following formula. :

[0111]

[0112]

[0113]

[0114] In the formula, To determine the median absolute deviation, the server can also use the circular Hough transform to calculate the arc response intensity and coverage as a basis for equivalence determination.

[0115] The server determines stripe interference based on image frequency domain features and then processes the image after removing the mean. Solving the two-dimensional Fourier amplitude spectrum Delineate high-frequency regions Narrow sector by direction Statistical energy proportion, combined with concentration threshold The stripe artifact index is calculated using the following formula. :

[0116]

[0117]

[0118] This indicator can also be calculated using Radon transform, based on the peak percentage of the projection angle. This represents the percentage of energy within a sector in a certain direction. This represents the total energy in the high-frequency region.

[0119] The server sets a saturation threshold based on imaging system parameters. Combined with the maximum amplitude of the original image To determine saturated pixels, use the formula... The percentage of saturated pixels within the organization mask is used to obtain the saturated pixel ratio. When using only the normalized cross-section, the server can set a fixed threshold to approximate the pixel saturation state. Among these, This is the original unnormalized cross-sectional image.

[0120] The server performs dilation calculations on a two-dimensional tissue mask. The area within the imaging field of view excluding the dilated region is selected as the extra-tissue evaluation area. Set the highlight threshold outside the organization. Filter out abnormally bright pixels within the selected area, and combine with the formula , , The external high-brightness index is calculated by combining the number of high-brightness pixels, the number of connected components, and the normalization parameters. .in, A set of abnormally bright pixels outside the organization; This is a weighting coefficient for the number of pixels; The number of connected components; For reference, the number of connected components.

[0121] The server relies on effectiveness masks and pixel intensity thresholds. neighborhood variance threshold Identify and reconstruct the set of missing pixels within the tissue region. Only areas with a minimum threshold are counted. The connected regions are used to calculate the missing region index using the following formula. :

[0122]

[0123] The server can further improve the accuracy of defect detection by combining the content of adjacent slices. Among them, For effectiveness masking; This represents local variance; AreaLargeComp() only counts areas greater than a certain size. The total area of ​​the connected domains.

[0124] The server maps the above six artifact metrics to the range of 0 to 1, and then performs weighted fusion according to the following formula to obtain the artifact penalty value:

[0125] in, All are weight coefficients corresponding to the artifact index, are non-negative, and satisfy the following conditions: The server can also add threshold triggering rules. When any artifact metric exceeds the corresponding limit, the upper limit of the artifact penalty value will be directly increased, or the current candidate section will be marked as an invalid section.

[0126] S207. Based on principal component similarity, vascular structure score, location score, consistency score, and artifact penalty value, determine the comprehensive score of each candidate section.

[0127] The comprehensive score is a result that combines multiple evaluation dimensions and the influence of artifacts to determine the overall quality of the candidate cross-section.

[0128] One possible approach is to perform a weighted summation of the principal component similarity, vascular structure score, location score, and consistency score of each candidate cross section to obtain a weighted total score for that candidate cross section; and then subtract the artifact penalty value of the candidate cross section from the weighted total score to obtain a comprehensive score for that candidate cross section.

[0129] The weighted total score is obtained by summing the scores after assigning weights to the principal component similarity, vascular structure score, location score, and consistency score.

[0130] For example, the server uniformly calculates the principal component similarity. Vascular structure score Position score Consistency score and artifact penalty value Normalized to the 0-1 range to ensure uniformity of dimensions and calculation standards for all evaluation indicators, and to achieve a comprehensive score. The specific calculation formula is as follows:

[0131] in, , , , , These are the weighting coefficients corresponding to the scores, all of which are greater than or equal to 0. For example, =0.2、 =0.3、 =0.2、 =0.2、 =0.5.

[0132] After calculating the comprehensive score for all candidate sections, the server first performs an initial screening based on preset quality thresholds to remove invalid sections that do not meet the imaging quality and structural requirements. Removal criteria include artifact penalty values ​​exceeding the preset artifact threshold, vascular structure scores below the vascular visualization threshold, tissue mask areas below the integrity threshold, and insufficient effective reconstruction area. From the selected valid candidate sections, the server selects the candidate section with the highest comprehensive score as the initial representative section. If the neighboring candidate sections of this optimal representative section simultaneously meet both the structural similarity threshold and the artifact quality threshold, the server can fuse the optimal section and its neighboring layers to generate the corresponding representative thin layer.

[0133] The final server output includes a representative cross-sectional image, a representative thin-layer image, the corresponding candidate cross-section number, volumetric spatial coordinates, normal vector, normalized position of the major axis, individual scores for each item, and a complete candidate cross-section scoring table.

[0134] S208. Based on the comprehensive score of each candidate section, select the representative section.

[0135] The representative cross section is selected from the candidate cross section sequence and is the one with the best overall imaging quality and vascular structure performance.

[0136] One possible approach is to select the candidate section with the highest comprehensive score from all candidate sections as the initial selection section; determine whether the comprehensive score of the initial selection section is not lower than a preset threshold; if so, determine the initial selection section as the representative section; if not, output an image quality insufficiency prompt.

[0137] The initial cross-section is the candidate cross-section with the highest overall score among all candidate cross-sections. The preset threshold is the minimum overall score standard used to determine whether a candidate cross-section meets the effective imaging quality requirements.

[0138] For example, a preset threshold for the comprehensive score can be set to 0.6. In this case, 20 candidate cross-sections are selected. The server calculates the comprehensive score for each candidate cross-section, with a score range of 0.25 to 0.82. The candidate cross-section with the highest comprehensive score of 0.82 is the initially selected cross-section. Since this initial cross-section score of 0.82 is greater than the preset threshold of 0.6, it meets the quality requirements, and the server determines it as the representative cross-section. In another example scenario, if the highest comprehensive score of all candidate cross-sections is 0.51, which is lower than the preset threshold of 0.6, it indicates that all current candidate cross-sections have quality problems such as severe artifacts, missing vascular structures, poor cross-sectional consistency, or positional misalignment. No valid cross-section meets the requirements, and the server immediately outputs an image quality insufficient warning, indicating that a valid representative cross-section cannot be selected.

[0139] Based on the above, the three-dimensional photoacoustic representative section determination method acquires a sequence of candidate sections of the tissue to be analyzed, calculates the principal component similarity, vascular structure score, anatomical position score, adjacent slice consistency score, and artifact penalty value for each candidate section, and then integrates the above results to obtain a comprehensive score. Representative sections are selected according to the score. This method no longer relies on manual judgment, fixed slices, or a single indicator to determine the section, but evaluates from multiple perspectives, including overall structural changes, vascular visualization, anatomical position rationality, adjacent slice continuity, and artifact effects. This approach reduces the results variability caused by human selection, resulting in more consistent results from different users and batches of data; it adapts to differences in tissue length, curvature, and placement angle, ensuring consistency in the selected sections' anatomical position; it avoids incorrectly selecting representative sections from artifacts, localized high-brightness noise, or abnormally isolated structures; it fully utilizes the continuity between adjacent slices, reducing the impact of single-layer noise, local vascular rupture, or motion artifacts; and it considers more factors affecting representativeness, making the evaluation process more comprehensive and ultimately selecting representative sections from the three-dimensional photoacoustic volume data.

[0140] The above text combined Figures 1 to 2 The method for determining the three-dimensional photoacoustic representative section 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.

[0141] This application also provides a three-dimensional photoacoustic representative section determination device, such as... Figure 3 As shown, this figure is a structural diagram of a three-dimensional photoacoustic representative section determination device provided in an embodiment of this application. The device includes: The acquisition module 301 is used to acquire the candidate cross-sectional sequence of the tissue to be analyzed.

[0142] The calculation module 302 is used to calculate the similarity of each candidate section in the candidate section sequence to obtain the principal component similarity of each candidate section; to calculate the vascular structure features of each candidate section in the candidate section sequence to obtain the vascular structure score of each candidate section; to evaluate the anatomical position of each candidate section in the candidate section sequence to obtain the position score of each candidate section; to evaluate the consistency between adjacent slices of each candidate section in the candidate section sequence to obtain the consistency score of each candidate section; and to detect artifacts of each candidate section in the candidate section sequence to obtain the artifact penalty value of each candidate section.

[0143] The section determination module 303 is used to determine the comprehensive score of each candidate section based on principal component similarity, vascular structure score, location score, consistency score and artifact penalty value; and to select representative sections based on the comprehensive score of each candidate section.

[0144] In some possible implementations, module 301 is specifically used for: Acquire three-dimensional photoacoustic volume data of the tissue to be analyzed; perform threshold segmentation on the three-dimensional photoacoustic volume data to obtain an initial binary mask; perform morphological closing operation, hole filling and maximum connected component screening on the initial binary mask to obtain a three-dimensional tissue mask; determine the candidate section sequence based on the three-dimensional tissue mask.

[0145] In some possible implementations, module 301 is specifically used for: Based on the tissue 3D mask, the target photoacoustic volume data is determined; based on the target photoacoustic volume data, the tissue long axis and its spatial range are determined; within the long axis spatial range, multiple cross sections orthogonal to the tissue long axis are determined as candidate cross section sequences.

[0146] In some possible implementations, the computing module 302 is specifically used for: Each candidate cross section in the candidate cross section sequence is converted into a one-dimensional vector to construct a slice matrix. The slice matrix is ​​then centered to obtain a centered matrix. Principal component analysis is performed on the centered matrix to obtain each principal component vector and its eigenvalue. Based on the eigenvalue of each principal component vector, the weight of each principal component vector is calculated. For each candidate cross section, the similarity between the candidate cross section and each principal component vector is calculated, and each similarity is multiplied by the weight of the corresponding principal component vector and then summed to obtain the principal component similarity of the candidate cross section.

[0147] In some possible implementations, the section determination module 303 is specifically used for: For each candidate section, the principal component similarity, vascular structure score, location score, and consistency score of the candidate section are weighted and summed to obtain the weighted total score of the candidate section; the artifact penalty value of the candidate section is subtracted from the weighted total score of the candidate section to obtain the comprehensive score of the candidate section.

[0148] In some possible implementations, the section determination module 303 is specifically used for: Select the candidate section with the highest comprehensive score from all candidate sections as the initial selection section; determine whether the comprehensive score of the initial selection section is not lower than the preset threshold; if so, determine the initial selection section as the representative section; if not, output an image quality insufficient prompt.

[0149] In some possible implementations, the computing module 302 is specifically used for: For each candidate section, calculate the edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-external highlight index, and reconstruction missing region index for that candidate section; then weight and fuse the edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-external highlight index, and reconstruction missing region index to obtain the artifact penalty value for the corresponding candidate section.

[0150] The three-dimensional photoacoustic representative section determination device according to the embodiments of this application can correspond to the execution of the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the three-dimensional photoacoustic representative section determination device 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.

[0151] 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.

[0152] 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.

[0153] 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).

[0154] Communication interface 403 is used for communication with external devices.

[0155] 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).

[0156] The memory 404 stores executable code, and the processor 402 executes the executable code to perform the aforementioned three-dimensional photoacoustic representative section determination method.

[0157] Specifically, in achieving Figure 3 In the case of the illustrated embodiment, and Figure 3 When the modules or units of the three-dimensional photoacoustic representative section determination device described in the embodiments are implemented by software, the execution... 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, and executes the aforementioned three-dimensional photoacoustic representative section determination method.

[0158] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, 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 execute the above-described three-dimensional photoacoustic representative section determination method.

[0159] 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.

[0160] 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.

[0161] When the computer program product is executed by a computer, the computer executes any of the aforementioned methods for determining the three-dimensional photoacoustic representative cross-section. The computer program product can be a software installation package; when any of the aforementioned methods for determining the three-dimensional photoacoustic representative cross-section is required, the computer program product can be downloaded and executed on the computer.

[0162] 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.

[0163] 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 method for determining a three-dimensional photoacoustic representative section, characterized in that, The method includes: Obtain candidate cross-sectional sequences of the tissue to be analyzed; Similarity is calculated for each candidate section in the candidate section sequence to obtain the principal component similarity of each candidate section; vascular structure features are calculated for each candidate section in the candidate section sequence to obtain the vascular structure score of each candidate section; anatomical location is evaluated for each candidate section in the candidate section sequence to obtain the location score of each candidate section; adjacent slice consistency is evaluated for each candidate section in the candidate section sequence to obtain the consistency score of each candidate section; artifact detection is performed on each candidate section in the candidate section sequence to obtain the artifact penalty value of each candidate section. Based on the principal component similarity, vascular structure score, location score, consistency score, and artifact penalty value, a comprehensive score for each candidate section is determined. Based on the comprehensive scores of each candidate section, a representative section is selected.

2. The method according to claim 1, characterized in that, The process of obtaining candidate cross-sectional sequences of the tissue to be analyzed includes: Acquire three-dimensional photoacoustic data of the tissue to be analyzed; Threshold segmentation is performed on the three-dimensional photoacoustic volume data to obtain an initial binary mask; Morphological closing operations, hole filling, and maximum connected component filtering are performed on the initial binary mask to obtain a three-dimensional tissue mask; Based on the three-dimensional mask of the tissue, a sequence of candidate cross sections is determined.

3. The method according to claim 2, characterized in that, The determination of the candidate cross-sectional sequence based on the tissue three-dimensional mask includes: Based on the three-dimensional mask of the tissue, the target photoacoustic body data is determined; Based on the target photoacoustic body data, the long axis and spatial range of the long axis of the tissue are determined. Within the long axis space, multiple sections orthogonal to the long axis of the tissue are identified as a candidate section sequence.

4. The method according to claim 1, characterized in that, The step of calculating the similarity of each candidate cross-section in the candidate cross-section sequence to obtain the principal component similarity of each candidate cross-section includes: Each candidate section in the candidate section sequence is converted into a one-dimensional vector to construct a slice matrix; The slice matrix is ​​centered to obtain a centered matrix; Principal component analysis is performed on the centered matrix to obtain each principal component vector and the eigenvalues ​​of each principal component vector; The weights of each principal component vector are calculated based on the eigenvalues ​​corresponding to each principal component vector. For each candidate cross section, the similarity between the candidate cross section and each principal component vector is calculated. Then, each similarity is multiplied by the weight of the corresponding principal component vector and summed to obtain the principal component similarity of the candidate cross section.

5. The method according to claim 1, characterized in that, The comprehensive score for each candidate section is determined based on the principal component similarity, vascular structure score, location score, consistency score, and artifact penalty value, including: For each candidate cross section, the principal component similarity, vascular structure score, location score, and consistency score of the candidate cross section are weighted and summed to obtain the weighted total score of the candidate cross section. The overall score of a candidate section is obtained by subtracting its artifact penalty value from its weighted total score.

6. The method according to claim 1, characterized in that, The representative cross-sections are selected based on the comprehensive scores of each candidate cross-section, including: The candidate cross section with the highest comprehensive score is selected as the initial cross section from all candidate cross sections; Determine whether the overall score of the initially selected section is not lower than a preset threshold; If yes, the initially selected section is determined as the representative section; otherwise, an insufficient image quality warning is output.

7. The method according to claim 1, characterized in that, The step of performing artifact detection on each candidate cross-section in the candidate cross-section sequence to obtain the artifact penalty value for each candidate cross-section includes: For each candidate section, calculate the edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-extra-tissue highlight index, and reconstruction missing region index for that candidate section. The edge highlight ratio, ring artifact index, stripe artifact index, saturated pixel ratio, tissue-external highlight index, and reconstruction missing region index are weighted and fused to obtain the artifact penalty value for the corresponding candidate section.

8. A three-dimensional photoacoustic representative section determination device, characterized in that, The device includes: The acquisition module is used to acquire candidate cross-sectional sequences of the tissue to be analyzed. The calculation module is used to calculate the similarity of each candidate section in the candidate section sequence to obtain the principal component similarity of each candidate section; to calculate the vascular structure features of each candidate section in the candidate section sequence to obtain the vascular structure score of each candidate section; to evaluate the anatomical location of each candidate section in the candidate section sequence to obtain the location score of each candidate section; to evaluate the consistency between adjacent slices of each candidate section in the candidate section sequence to obtain the consistency score of each candidate section; and to detect artifacts in each candidate section in the candidate section sequence to obtain the artifact penalty value of each candidate section. The section determination module is used to determine the comprehensive score of each candidate section based on the principal component similarity, vascular structure score, location score, consistency score and artifact penalty value; and to select representative sections based on the comprehensive score of each candidate section.

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