Acupuncture positioning method based on image processing

Abstract

The invention discloses an acupuncture positioning method based on image processing. The method comprises the following steps: S1, generating preprocessed medical human body image data; s2, generating a semantic segmentation image, and calibrating each anatomical region and a specific acupuncture point region of a human body from the semantic segmentation image; s3, generating a candidate acupoint region heat map based on the semantic segmentation image and the multi-scale standardized acupoint region heat map template; s4, performing alignment matching on the candidate acupoint region heat map and a standardized acupoint region heat map database by utilizing structure-guided heat map matching to obtain a matched acupoint result, wherein the standardized acupoint region heat map database comprises standard acupoint distribution information; and S5, integrating the matched acupuncture point result with clinical data of the patient to form a personalized acupuncture point candidate set, selecting a final acupuncture target meeting a heat map responsivity threshold and an individual adaptation condition, and outputting an acupuncture positioning result according to the final acupuncture target. According to the invention, the treatment relevance and clinical practicability of target recommendation are improved, and the operability and safety of intelligent acupuncture treatment are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of acupuncture, and in particular to an acupuncture positioning method based on image processing. Background Art

[0002] With the rapid development of computer vision and artificial intelligence technologies, medical image processing has shown broad application prospects in auxiliary diagnosis and treatment, intelligent identification and personalized medicine. Acupuncture, as an important means of clinical treatment in Traditional Chinese Medicine, has been widely verified to be effective in various chronic diseases, pain management and functional regulation. However, the selection of acupoints during acupuncture treatment is still highly dependent on the experience and judgment of traditional Chinese medicine practitioners. The subjectivity and variability of target selection when faced with different individual signs, complex diseases or the interaction of multiple treatment responses seriously affect the stability of efficacy and treatment standardization.

[0003] Currently, the common acupuncture point positioning methods used in clinical practice mainly include traditional anatomical calibration methods, expert experience methods, and auxiliary methods that partially rely on surface positioning instruments. Traditional methods generally have the following defects: First, the manual operation process requires reliance on rich knowledge background and operation experience of traditional Chinese medicine, and the accuracy is poor when facing changes in the anatomical structure of different patients; second, there is a lack of unified standardized positioning methods, and there are large differences in treatment plans between hospitals and doctors; third, existing auxiliary systems mostly focus on single-point recognition or rough recommendations based on static maps, and it is difficult to combine patients' real-time medical images with specific symptoms to make dynamic and personalized target recommendations.

[0004] Some studies have attempted to combine image recognition with traditional Chinese medicine theory to achieve automatic extraction and identification of acupoint areas, but most of these methods remain at the shallow feature matching or template overlap stage and cannot achieve precise adaptation to the patient's individual structure; at the same time, existing methods rarely consider the topological constraints of meridian structures and the dynamic feedback of clinical response data, resulting in low credibility of automatically recommended targets.

[0005] Therefore, there is an urgent need for an innovative method that combines image semantic segmentation and heat map matching mechanism to systematically solve the problems of automated target positioning, personalized treatment pathways, and intelligent models. Summary of the Invention

[0006] One purpose of the present invention is to propose an acupuncture positioning method based on image processing, which improves the treatment relevance and clinical practicality of target recommendation and enhances the operability and safety of intelligent acupuncture treatment.

[0007] An acupuncture positioning method based on image processing according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect unified medical human image data of the patient, perform image preprocessing on the unified medical human image data, and generate preprocessed medical human image data;

[0009] S2. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model processes the pre-processed medical human image data to generate semantic segmentation images. The semantic segmentation images are used to demarcate various anatomical regions of the human body and specific acupuncture points.

[0010] S3. Generate candidate acupoint heat maps based on the semantically segmented image and a multi-scale standardized acupoint heat map template. The multi-scale standardized acupoint heat map template is constructed based on Traditional Chinese Medicine meridian theory and clinical statistical data.

[0011] S4. Using structure-guided heatmap matching, the candidate acupoint area heatmap is aligned and matched with the standardized acupoint area heatmap database to obtain matched acupoint results, where the standardized acupoint area heatmap database contains standard acupoint distribution information;

[0012] S5. Integrate the matched acupoint results with the patient's clinical data, including disease classification, physiological parameters, and previous treatment responses, to form a personalized acupoint candidate set, and select the final acupuncture target that meets the heat map response threshold and individual adaptation conditions, and output the acupuncture positioning result based on the square root of the final acupuncture target.

[0013] Optionally, the S1 includes the following steps:

[0014] S11. Acquire unified medical human image data of the patient through an imaging device, the unified medical human image data including medical human images acquired by a magnetic resonance imaging device, a computed tomography device, an ultrasound imaging device, or an infrared thermal imaging device, and construct a unified medical human image dataset:

[0015]

[0016] Among them, I raw represents the unified medical human image dataset, I i represents the i-th medical human image, f(x,y,c i ) indicates that the horizontal coordinate is x, the vertical coordinate is y, and the channel is c in the medical human image i The pixel value of the medical human image, is the set of medical human image channels, and N is the total number of unified medical human image data;

[0017] S12. Using a two-dimensional Gaussian smoothing operator to smooth and suppress local random noise in the medical human image in the unified medical human image data, thereby generating noise-removed medical human image data. The noise-removed medical human image data maintains structural information of the original medical human image while reducing artifact interference from background and edges of the medical human image.

[0018] S13. Using the Laplace transform method to extract tissue structure boundary information from the medical human image, the edge of the medical human image is enhanced to obtain edge-enhanced medical human image data, which is used to highlight the contours of the human body structure and the edge features of the acupuncture points;

[0019] S14. Performing contrast adjustment on the edge-enhanced medical human image data, stretching the grayscale distribution of the medical human image using a histogram equalization method to uniformly distribute the brightness range of the medical human image, thereby generating contrast-adjusted medical human image data. The contrast-adjusted medical human image data improves the recognizability of dark and bright areas in the medical human image.

[0020] S15. Perform geometric alignment processing on the contrast-adjusted medical human image data. Based on the standard human anatomy template medical human image, each medical human image is geometrically transformed in spatial scale by constructing an affine transformation matrix. The affine transformation matrix contains rotation, scaling, and translation parameters to obtain a standardized pre-processed medical human image dataset I. pre .

[0021] Optionally, S2 includes the following steps:

[0022] S21. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model includes a feature extraction module, a multi-scale meridian guidance feature fusion module, an acupoint response enhancement module, and a semantic prediction module. The cross-modal multi-branch deep learning semantic segmentation model uses a standardized pre-processed medical human image dataset as input. Each standardized pre-processed medical human image The corresponding modal branches are input through the feature extraction module, including magnetic resonance modal branch, computed tomography modal branch, ultrasound modal branch and infrared modal branch, and the magnetic resonance features are output respectively. Computed tomography characteristics Ultrasound characteristics and infrared characteristics The four types of features together constitute the cross-modal original feature set

[0023] S22. Using multi-scale meridian-guided feature fusion module to fusion cross-modal original feature sets Perform fusion processing, and introduce the meridian anatomical structure constraint map G during the fusion process meridian =(Vacu ,E acu ), in the meridian anatomical structure constraint map, the acupoint node set V acu Represents the three-dimensional coordinate positions and meridian affiliation labels of all standardized human acupuncture points, and the edge set E acu Representing the structural connection relationship between acupoint nodes, under the guidance of the graph structure of the meridian anatomical structure constraint map, the multi-scale meridian guided feature fusion module performs weighted processing on each cross-modal original feature based on the graph attention mechanism to generate the fused acupuncture target feature representation The fusion weight of each cross-modal original feature is recorded as β k , used to measure the importance of each modality in the acupuncture target recognition task;

[0024] S23. Represent the fused acupuncture target features Input the acupoint response enhancement module for feature enhancement processing. The acupoint response enhancement module has a built-in acupoint response enhancement operator Φ acu The acupoint response enhancement operator is to perform weighted enhancement on the response value of the area close to the standard acupoint in the fusion feature representation after acupoint response enhancement based on the three-dimensional position distribution and anatomical adjacency relationship of each acupoint in the meridian anatomical structure constraint map in the spatial dimension, set the response control parameter γ to adjust the enhancement amplitude of the acupoint area, and set the spatial response scale parameter σ to limit the influence range of acupoint enhancement, and output the enhanced acupoint response fusion feature representation.

[0025] S24. Representing the fusion features after acupoint response enhancement The input is sent to the semantic prediction module, which consists of a multi-layer acupoint feature-guided deconvolution structure and outputs the acupuncture point semantic probability image S i (x,y,l):

[0026]

[0027] Among them, S i (x, y, l) represents the probability value of the spatial coordinate (x, y) of the i-th medical human image belonging to the semantic category l, Semantic category set Contains acupuncture point area, muscle area, bone area and skin area, Deconv h (·) represents the h-th layer feature-guided deconvolution operation, w h is the convolution layer weight obtained by training optimization, H is the number of deconvolution layers, and Softmax is the Softmax function;

[0028] S25. Based on the semantic probability image of acupuncture points S i(x, y, l) Assign the maximum probability of the semantic category to each pixel position and determine the semantic category label l x,y , obtain the acupuncture point semantic segmentation image set S seg :

[0029]

[0030] in, Represents the semantic category label determined at the coordinate (x, y) of the i-th medical human image.

[0031] Optionally, S3 includes the following steps:

[0032] S31. Image set S based on semantic segmentation of acupuncture points seg , extract the pixel set marked as acupuncture point area in each medical human image

[0033]

[0034] in, Indicates the acupuncture point label corresponding to the acupuncture point area in the semantic category label set, represents the set of pixel coordinates of all acupoint areas determined in the i-th medical human image;

[0035] S32. Constructing a multi-scale standardized acupuncture point heat map template set The scale set S contains the spatial scale levels of multiple human body regions, and the standardized acupoint heat map template T s It represents the standard acupoint thermal distribution map constructed at scale level s, which is constructed based on the statistical distribution of a large clinical sample:

[0036]

[0037] Among them, M s represents the number of standardized acupoints defined under scale level s, is the standard position coordinate of the jth standard acupuncture point at scale level s, is the activation intensity weight of the jth standard acupoint at this scale, σ s is the spatial diffusion coefficient at scale level s, and the standardized acupoint area heat map template T s (x,y) represents the probability density response value of the location as the activated target point;

[0038] S33. Extracting pixel sets of acupuncture point areas A collection of standardized acupuncture point heat map templates Perform regional projection and response fitting to construct a candidate acupoint area heat map

[0039]

[0040] Among them, ω s is the fusion weight of each scale level s in the candidate heat map, is a local weighted kernel function defined within a reference radius r, which is used to perform neighborhood diffusion in space based on the acupoint area response and improve the response coherence of the candidate area;

[0041] S34. Map the candidate acupoint area heat map back to the original image size and normalize it to obtain a normalized candidate acupoint area heat map

[0042] Optionally, the S4 includes the following steps:

[0043] S41. Construction of a standardized acupoint heat map database in, represents the kth standard acupoint area heat map, represents the mask area of ​​the standard acupoint area in the standard image space, The structural guidance map corresponding to the standard acupoint area includes the boundaries and topological relationships of the main anatomical structures in the standard acupoint area;

[0044] S42. Heat map of candidate acupoint area Each standard acupoint heat map in the standardized acupoint heat map database Perform structure-guided heatmap matching, perform structure-aware similarity calculation on each pair of standardized acupoint area heatmaps, and obtain the structure-guided matching similarity value Sim i,k ;

[0045] S43. Recording the structure-guided matching similarity values ​​of all standardized acupoint area heat maps and candidate acupoint area heat maps in sequence according to the standardized acupoint area heat map number k to form a matching result set, which is a combination set including the standardized acupoint area number and its corresponding structure-guided matching similarity value;

[0046] S44. The matching result set is traversed and sorted according to the structure-guided matching similarity value. A matching strength threshold δ is set. Standardized acupoint heatmap numbers having structure-guided matching similarity values ​​greater than or equal to the matching strength threshold δ are selected from the matching result set to obtain a set of matching standard acupoint numbers. The set of matching standard acupoint numbers represents the numbers of standard acupoint templates having a high degree of structure-guided matching similarity with the current candidate acupoint heatmap.

[0047] S45. Extract the spatial coordinate positions and meridian affiliation labels of all standardized acupoints in the standard acupoint area number set corresponding to the standard acupoint area heat map database as the current candidate acupoint area heat map. The set of successfully matched standard acupuncture points is used to obtain the matched acupuncture point result set The matching acupoint result set contains the two-dimensional coordinate position of each acupoint and its meridian information, x k represents the horizontal spatial coordinate of the kth candidate acupuncture point in the medical human image, y k represents the longitudinal spatial coordinate of the kth candidate acupuncture point in the medical human image, l k Indicates the standardized acupoint label to which the k-th candidate acupuncture point belongs.

[0048] Optionally, the structure guides the matching similarity value Sim i,k for:

[0049]

[0050] Among them, Sim i,k represents the similarity value of the structure-guided matching between the i-th candidate acupoint area heat map and the k-th standardized heat map, It represents the structural consistency score between the candidate acupoint area heat map and the standardized acupoint area heat map on the structural topology map, which is obtained by inverse mapping of the structural map matching distance.

[0051] Optionally, the S5 includes the following steps:

[0052] S51. Obtain patient-specific clinical data, including disease classification information C diag , physiological parameter set Θ phys and previous treatment response index R hist ;

[0053] S52. Build a personalized acupoint candidate set based on the matched acupoint result set and patient-specific clinical data, and combine it with the current candidate acupoint area heat map The patient's individual physiological adaptability, disease adaptability and previous treatment response indicators are used to set the following screening rules to construct a personalized acupoint candidate set

[0054] If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.7, it will be recorded as a first-level candidate acupoint;

[0055] If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.6, it will be recorded as a secondary candidate acupoint;

[0056] If the current candidate acupoint area heat map or standardized acupoint labels Or if the physiological parameter matching degree is <0.6, it will not be included in the candidate set;

[0057] in, Indicates the current condition C diag The matching acupoint label set, the physiological parameter matching degree is based on the patient's physiological parameter set Θ phys Calculation of adaptability to the anatomical region where the acupoint is located;

[0058] S53. Prioritize the first-level candidate acupoints in the personalized acupoint candidate set, and combine the previous treatment response index R hist Sort by acupuncture points with high efficacy scores and balanced spatial distribution as the final acupuncture target set

[0059] S54. Output the final acupuncture target set The final acupuncture target point set is mapped to the original medical human image space and superimposed to generate a visual acupuncture positioning result image.

[0060] Optionally, the disease classification information C diag Represents the standardized code of the current condition, the physiological parameter set Θ phys Including basic indicators such as body temperature, blood pressure, and physical fitness score, the previous treatment response index R hist Indicates the records of therapeutic effects on different acupoints.

[0061] Optionally, according to the final acupuncture target set Define acupuncture target selection strategy:

[0062] If the number of first-level candidate acupoints is ≥5, the top 5 acupoints with the highest efficacy scores will be selected first;

[0063] If the number of first-level candidate acupoints is less than 5, the second-level candidate acupoints with the highest efficacy scores will be added to make it 5;

[0064] If the efficacy scores of all candidate acupoints are lower than the preset threshold, only one first-level candidate acupoint with the highest responsiveness will be retained.

[0065] The beneficial effects of the present invention are:

[0066] (1) The semantic segmentation model constructed by the present invention adopts a cross-modal multi-branch structure, establishes independent branch networks for different types of medical images, and introduces the Chinese medicine meridian structure map as topological guidance information in the fusion stage. The feature weighting of the acupoint area is performed through the graph attention mechanism, which can accurately identify the multi-scale regional structure related to the acupuncture target in a complex anatomical background, significantly improves the accuracy and robustness of acupoint segmentation, and solves the problem of identifying the inconsistent spatial distribution of surface acupoints among different individuals.

[0067] (2) In the process of candidate target generation, the present invention constructs a multi-scale standardized acupoint area heat map template, introduces an anatomical structure map for guidance, and proposes a structure-guided similarity matching method. In the heat map matching, the heat map response intensity and structural topological consistency of the candidate area are comprehensively considered, avoiding the error accumulation problem of traditional methods that only ignore local anatomical features based on spatial intensity matching. The structural similarity is regulated by the matching weight function, so that the acupuncture point recommendation results achieve high consistency in both heat map response and anatomical structure, effectively improving the credibility and interpretability of target alignment.

[0068] (3) The present invention proposes a multi-dimensional fusion recommendation mechanism that combines disease classification, physiological parameters and past efficacy feedback. In the candidate acupoint construction stage, by setting the heat map response threshold, adaptability matching degree and efficacy ranking, a hierarchical screening standard for the first-level and second-level candidate acupoints is formed, and the final target selection path is further optimized in combination with historical response data. It can realize dynamic adjustment of the treatment area and personalized output of the target, improve the therapeutic relevance and clinical practicality of the target recommendation, and enhance the operability and safety of intelligent acupuncture treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0070] Figure 1 This is a flow chart of an acupuncture positioning method based on image processing proposed by the present invention. DETAILED DESCRIPTION

[0071] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0072] refer to Figure 1 , an acupuncture positioning method based on image processing, comprising the following steps:

[0073] S1. Collect unified medical human image data of the patient, perform image preprocessing on the unified medical human image data, and generate preprocessed medical human image data;

[0074] S2. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model processes the pre-processed medical human image data to generate semantic segmentation images. The semantic segmentation images are used to demarcate various anatomical regions of the human body and specific acupuncture points.

[0075] S3. Generate candidate acupoint heat maps based on the semantically segmented image and a multi-scale standardized acupoint heat map template. The multi-scale standardized acupoint heat map template is constructed based on Traditional Chinese Medicine meridian theory and clinical statistical data.

[0076] S4. Using structure-guided heatmap matching, the candidate acupoint area heatmap is aligned and matched with the standardized acupoint area heatmap database to obtain matched acupoint results, where the standardized acupoint area heatmap database contains standard acupoint distribution information;

[0077] S5. Integrate the matched acupoint results with the patient's clinical data, including disease classification, physiological parameters, and previous treatment responses, to form a personalized acupoint candidate set, and select the final acupuncture target that meets the heat map response threshold and individual adaptation conditions, and output the acupuncture positioning result based on the square root of the final acupuncture target.

[0078] In this embodiment, S1 includes the following steps:

[0079] S11. Acquire unified medical human image data of the patient through an imaging device, the unified medical human image data including medical human images acquired by a magnetic resonance imaging device, a computed tomography device, an ultrasound imaging device, or an infrared thermal imaging device, and construct a unified medical human image dataset:

[0080]

[0081] Among them, I raw represents the unified medical human image dataset, I i represents the i-th medical human image, f(x,y,c i ) indicates that the horizontal coordinate is x, the vertical coordinate is y, and the channel is c in the medical human image i The pixel value of the medical human image, is the set of medical human image channels, and N is the total number of unified medical human image data;

[0082] S12. Using a two-dimensional Gaussian smoothing operator to smooth and suppress local random noise in the medical human image in the unified medical human image data, thereby generating noise-removed medical human image data. The noise-removed medical human image data maintains structural information of the original medical human image while reducing artifact interference from background and edges of the medical human image.

[0083] S13. Using the Laplace transform method to extract tissue structure boundary information from the medical human image, the edge of the medical human image is enhanced to obtain edge-enhanced medical human image data, which is used to highlight the contours of the human body structure and the edge features of the acupuncture points;

[0084] S14. Performing contrast adjustment on the edge-enhanced medical human image data, stretching the grayscale distribution of the medical human image using a histogram equalization method to uniformly distribute the brightness range of the medical human image, thereby generating contrast-adjusted medical human image data. The contrast-adjusted medical human image data improves the recognizability of dark and bright areas in the medical human image.

[0085] S15. Perform geometric alignment processing on the contrast-adjusted medical human image data. Based on the standard human anatomy template medical human image, each medical human image is geometrically transformed in spatial scale by constructing an affine transformation matrix. The affine transformation matrix contains rotation, scaling, and translation parameters to obtain a standardized pre-processed medical human image dataset I. pre .

[0086] In this embodiment, S2 includes the following steps:

[0087] S21. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model includes a feature extraction module, a multi-scale meridian guidance feature fusion module, an acupoint response enhancement module, and a semantic prediction module. The cross-modal multi-branch deep learning semantic segmentation model uses a standardized pre-processed medical human image dataset as input. Each standardized pre-processed medical human image The corresponding modal branches are input through the feature extraction module, including magnetic resonance modal branch, computed tomography modal branch, ultrasound modal branch and infrared modal branch, and the magnetic resonance features are output respectively. Computed tomography characteristics Ultrasound characteristics and infrared characteristics The four types of features together constitute the cross-modal original feature set

[0088] S22. Using multi-scale meridian-guided feature fusion module to fusion cross-modal original feature sets Perform fusion processing, and introduce the meridian anatomical structure constraint map G during the fusion process meridian =(V acu ,E acu ), in the meridian anatomical structure constraint map, the acupoint node set V acu Represents the three-dimensional coordinate positions and meridian affiliation labels of all standardized human acupuncture points, and the edge set E acu Representing the structural connection relationship between acupoint nodes, under the guidance of the graph structure of the meridian anatomical structure constraint map, the multi-scale meridian guided feature fusion module performs weighted processing on each cross-modal original feature based on the graph attention mechanism to generate the fused acupuncture target feature representation The fusion weight of each cross-modal original feature is recorded as β k, used to measure the importance of each modality in the acupuncture target recognition task;

[0089] S23. Represent the fused acupuncture target features Input the acupoint response enhancement module for feature enhancement processing. The acupoint response enhancement module has a built-in acupoint response enhancement operator Φ acu The acupoint response enhancement operator is to perform weighted enhancement on the response value of the area close to the standard acupoint in the fusion feature representation after acupoint response enhancement based on the three-dimensional position distribution and anatomical adjacency relationship of each acupoint in the meridian anatomical structure constraint map in the spatial dimension, set the response control parameter γ to adjust the enhancement amplitude of the acupoint area, and set the spatial response scale parameter σ to limit the influence range of acupoint enhancement, and output the enhanced acupoint response fusion feature representation.

[0090] S24. Representing the fusion features after acupoint response enhancement The input is sent to the semantic prediction module, which consists of a multi-layer acupoint feature-guided deconvolution structure and outputs the acupuncture point semantic probability image S i (x,y,l):

[0091]

[0092] Among them, S i (x, y, l) represents the probability value of the spatial coordinate (x, y) of the i-th medical human image belonging to the semantic category l, Semantic category set Contains acupuncture point area, muscle area, bone area and skin area, Deconv h (·) represents the h-th layer feature-guided deconvolution operation, w h is the convolution layer weight obtained by training optimization, H is the number of deconvolution layers, and Softmax is the Softmax function;

[0093] S25. Based on the semantic probability image of acupuncture points S i (x, y, l) Assign the maximum probability of the semantic category to each pixel position and determine the semantic category label l x,y , obtain the acupuncture point semantic segmentation image set S seg :

[0094]

[0095] in, Represents the semantic category label determined at the coordinate (x, y) of the i-th medical human image.

[0096] In this embodiment, S3 includes the following steps:

[0097] S31. Image set S based on semantic segmentation of acupuncture points seg , extract the pixel set marked as acupuncture point area in each medical human image

[0098]

[0099] in, Indicates the acupuncture point label corresponding to the acupuncture point area in the semantic category label set, represents the set of pixel coordinates of all acupoint areas determined in the i-th medical human image;

[0100] S32. Constructing a multi-scale standardized acupuncture point heat map template set The scale set S contains the spatial scale levels of multiple human body regions, and the standardized acupoint heat map template T s It represents the standard acupoint thermal distribution map constructed at scale level s, which is constructed based on the statistical distribution of a large clinical sample:

[0101]

[0102] Among them, M s represents the number of standardized acupoints defined under scale level s, is the standard position coordinate of the jth standard acupuncture point at scale level s, is the activation intensity weight of the jth standard acupoint at this scale, σ s is the spatial diffusion coefficient at scale level s, and the standardized acupoint area heat map template T s (x,y) represents the probability density response value of the location as the activated target point;

[0103] S33. Extracting pixel sets of acupuncture point areas A collection of standardized acupuncture point heat map templates Perform regional projection and response fitting to construct a candidate acupoint area heat map

[0104]

[0105] Among them, ω s is the fusion weight of each scale level s in the candidate heat map, is a local weighted kernel function defined within a reference radius r, which is used to perform neighborhood diffusion in space based on the acupoint area response and improve the response coherence of the candidate area;

[0106] S34. Map the candidate acupoint area heat map back to the original image size and normalize it to obtain a normalized candidate acupoint area heat map

[0107] In this embodiment, S4 includes the following steps:

[0108] S41. Construction of a standardized acupoint heat map database in, represents the kth standard acupoint area heat map, represents the mask area of ​​the standard acupoint area in the standard image space, The structural guidance map corresponding to the standard acupoint area includes the boundaries and topological relationships of the main anatomical structures in the standard acupoint area;

[0109] S42. Heat map of candidate acupoint area Each standard acupoint heat map in the standardized acupoint heat map database Perform structure-guided heatmap matching, perform structure-aware similarity calculation on each pair of standardized acupoint area heatmaps, and obtain the structure-guided matching similarity value Sim i,k ;

[0110] S43. Recording the structure-guided matching similarity values ​​of all standardized acupoint area heat maps and candidate acupoint area heat maps in sequence according to the standardized acupoint area heat map number k to form a matching result set, which is a combination set including the standardized acupoint area number and its corresponding structure-guided matching similarity value;

[0111] S44. The matching result set is traversed and sorted according to the structure-guided matching similarity value. A matching strength threshold δ is set. Standardized acupoint heatmap numbers having structure-guided matching similarity values ​​greater than or equal to the matching strength threshold δ are selected from the matching result set to obtain a set of matching standard acupoint numbers. The set of matching standard acupoint numbers represents the numbers of standard acupoint templates having a high degree of structure-guided matching similarity with the current candidate acupoint heatmap.

[0112] S45. Extract the spatial coordinate positions and meridian affiliation labels of all standardized acupoints in the standard acupoint area number set corresponding to the standard acupoint area heat map database as the current candidate acupoint area heat map. The set of successfully matched standard acupuncture points is used to obtain the matched acupuncture point result set The matching acupoint result set contains the two-dimensional coordinate position of each acupoint and its meridian information, x k represents the horizontal spatial coordinate of the kth candidate acupuncture point in the medical human image, y k represents the longitudinal spatial coordinate of the kth candidate acupuncture point in the medical human image, l k Indicates the standardized acupoint label to which the k-th candidate acupuncture point belongs.

[0113] In this embodiment, the structure-guided matching similarity value Sim i,k for:

[0114]

[0115] Among them, Sim i,k represents the similarity value of the structure-guided matching between the i-th candidate acupoint area heat map and the k-th standardized heat map, It represents the structural consistency score between the candidate acupoint area heat map and the standardized acupoint area heat map on the structural topology map, which is obtained by inverse mapping of the structural map matching distance.

[0116] In this embodiment, S5 includes the following steps:

[0117] S51. Obtain patient-specific clinical data, including disease classification information C diag , physiological parameter set Θ phys and previous treatment response index R hist ;

[0118] S52. Build a personalized acupoint candidate set based on the matched acupoint result set and patient-specific clinical data, and combine it with the current candidate acupoint area heat map The patient's individual physiological adaptability, disease adaptability and previous treatment response indicators are used to set the following screening rules to construct a personalized acupoint candidate set

[0119] If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.7, it will be recorded as a first-level candidate acupoint;

[0120] If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.6, it will be recorded as a secondary candidate acupoint;

[0121] If the current candidate acupoint area heat map or standardized acupoint labels Or if the physiological parameter matching degree is <0.6, it will not be included in the candidate set;

[0122] in, Indicates the current condition C diag The matching acupoint label set, the physiological parameter matching degree is based on the patient's physiological parameter set Θ phys Calculation of adaptability to the anatomical region where the acupoint is located;

[0123] S53. Prioritize the first-level candidate acupoints in the personalized acupoint candidate set, and combine the previous treatment response index R hist Sort by acupuncture points with high efficacy scores and balanced spatial distribution as the final acupuncture target set

[0124] S54. Output the final acupuncture target set The final acupuncture target point set is mapped to the original medical human image space and superimposed to generate a visual acupuncture positioning result image.

[0125] In this embodiment, the disease classification information C diag Represents the standardized code of the current disease, the physiological parameter set Θ phys Including basic indicators such as body temperature, blood pressure, physical fitness score, previous treatment response index R hist Indicates the records of therapeutic effects on different acupoints.

[0126] In this embodiment, according to the final acupuncture target point set Define acupuncture target selection strategy:

[0127] If the number of first-level candidate acupoints is ≥5, the top 5 acupoints with the highest efficacy scores will be selected first;

[0128] If the number of first-level candidate acupoints is less than 5, the second-level candidate acupoints with the highest efficacy scores will be added to make it 5;

[0129] If the efficacy scores of all candidate acupoints are lower than the preset threshold, only one first-level candidate acupoint with the highest responsiveness will be retained.

[0130] Example 1:

[0131] The Rehabilitation Department of A City Traditional Chinese Medicine Hospital admitted a 52-year-old female patient named Li, whose chief complaint was "stiffness in the shoulder and back with limited movement for three months after surgery." She was diagnosed with right shoulder periarthritis with muscle adhesions. She had previously received conservative treatment with massage and electrotherapy without significant results. The patient hoped to try acupuncture treatment and hoped to develop a more precise acupoint plan to improve the efficacy.

[0132] In response to the hospital's pilot demand for "intelligent individualized acupuncture treatment", the Rehabilitation Department and the Intelligent Medicine Research Team deployed the system of the present invention and applied it to actual outpatient scenarios for the first time.

[0133] After the patient was admitted to the hospital, the nurse used the equipped infrared thermal imager, high-frequency ultrasound and low-dose CT equipment to collect local data of the right shoulder. All images were converted into a standard format through the hospital imaging workstation and uploaded to the system of the present invention.

[0134] After the unified medical human image dataset is constructed, the system completes image preprocessing, including Gaussian smoothing noise reduction, edge enhancement, contrast equalization and geometric registration with the standard shoulder template, and outputs the preprocessed image set I pre , the size is 512×512, with a total of 3 groups of modality images.

[0135] The system automatically calls the cross-modal multi-branch semantic segmentation model for processing at 10:33:

[0136] The CT channel extracts the clear boundary between the scapular spine and the scapular spine, which are marked as bony anatomical features;

[0137] Ultrasound images automatically identified the boundary between the supraspinatus and infraspinatus muscles, with complete muscle contours and an accuracy rate of 94.1%;

[0138] Infrared images revealed a low-temperature plaque with an average temperature of 2.6°C in the acromion region, which was automatically identified as an area of ​​insufficient blood perfusion.

[0139] The system generates semantic segmentation images of acupuncture points Among them, the "Jianjing", "Tianzong" and "Jianyu" acupoints are fully labeled, and the accuracy is higher than the expert manual annotation reference (IoU=0.91).

[0140] Then, the system automatically enters the S3 process, and the three-level scale template T is constructed based on the statistical data of Chinese medicine meridians. s (x,y) was called to perform candidate heat map fitting, and the candidate heat map was generated. The response value in the Jianjing acupoint area reached 0.87, and the system evaluated it as a first-level activation area.

[0141] Then it enters the S4 structure-guided thermal map matching stage. The system compares the candidate thermal map with the standard thermal map database. In the match numbered ACU-SD-042 (right shoulder standard template), the structural consistency score is 0.93. The system confirms that the position deviation between the current hot zone and the standard "shoulder well" is 3.4mm, automatically generates coordinate correction parameters and completes the target update.

[0142] The system calls historical patient data and integrates it with the current patient's electronic medical record, extracting Mr. Li's disease code as ICD-M79.10, corresponding to the recommended acupoint set

[0143] Combined with Li's physiological parameters (weight 65kg, BMI 26.8, normal blood pressure), the response record is empty, and the system uses the initialization matching logic to perform a first-level screening:

[0144] Jianjing: heat map response 0.87, adaptability score 0.91; Tianzong: heat map response 0.62, adaptability score 0.85; Jianzhongyu: heat map response 0.51, adaptability score 0.79; Quyuan: heat map response 0.36, adaptability score 0.68;

[0145] The final output first-level candidate acupoints are "Jianjing" and "Tianzong", and the number of recommended targets is set to 2. Output files and image files are generated for doctors to review and on-site calibration.

[0146] Treatment Record: Acupuncturist Zhang initiated treatment based on system recommendations. A 0.25mm x 40mm filiform needle was inserted at the "Jianjing" point, 3.4mm to the right, supplemented with the Tianzong Pingbu Pingxie method. Needles were retained for 30 minutes each time, for a total of five treatments. After the third treatment, the patient reported that "arm lifting became smoother," and after the fifth treatment, grip strength was measured and subjectively rated.

[0147] Table 1 Data before and after treatment

[0148]

[0149]

[0150] Table 2 compares the experimental group (experience group at the same time) data:

[0151]

[0152] The present invention can be efficiently integrated into real clinical processes, helping doctors to achieve accurate target identification, personalized recommendations and visual output of results. It not only improves treatment efficiency, but also enhances patients' treatment experience and response effects, fully verifying the feasibility and effectiveness of the present invention in actual application scenarios.

[0153] The semantic segmentation model constructed by the present invention adopts a cross-modal multi-branch structure, establishes independent branch networks for different types of medical images, and introduces the Chinese medicine meridian structure map as topological guidance information in the fusion stage. The feature weighting of the acupoint area is performed through the graph attention mechanism, which can accurately identify the multi-scale regional structure related to acupuncture targets in complex anatomical backgrounds, significantly improves the accuracy and robustness of acupoint segmentation, and solves the problem of identifying the inconsistent spatial distribution of surface acupoints among different individuals.

[0154] In the process of candidate target generation, the present invention constructs a multi-scale standardized acupoint area heat map template, introduces an anatomical structure map for guidance, and proposes a structure-guided similarity matching method. In the heat map matching, the heat map response intensity and structural topological consistency of the candidate area are comprehensively considered, avoiding the error accumulation problem of traditional methods that only ignore local anatomical features based on spatial intensity matching. The structural similarity is regulated by the matching weight function, so that the acupuncture point recommendation results achieve high consistency in both heat map response and anatomical structure, effectively improving the credibility and interpretability of target alignment.

[0155] The present invention proposes a multi-dimensional fusion recommendation mechanism that combines disease classification, physiological parameters and past efficacy feedback. In the candidate acupoint construction stage, by setting the heat map response threshold, adaptability matching degree and efficacy ranking, a hierarchical screening standard for first- and second-level candidate acupoints is formed, and the final target selection path is further optimized in combination with historical response data. It can dynamically adjust the treatment area and personalize the output of targets, improve the therapeutic relevance and clinical practicality of target recommendations, and enhance the operability and safety of intelligent acupuncture treatment.

[0156] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An acupuncture positioning method based on image processing, characterized in that: The steps include: S1. Collect unified medical human image data of the patient, perform image preprocessing on the unified medical human image data, and generate preprocessed medical human image data; S2. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model processes the pre-processed medical human image data to generate semantic segmentation images. The semantic segmentation images are used to demarcate various anatomical regions of the human body and specific acupuncture points. S3. Generate candidate acupoint heat maps based on the semantically segmented image and a multi-scale standardized acupoint heat map template. The multi-scale standardized acupoint heat map template is constructed based on Traditional Chinese Medicine meridian theory and clinical statistical data. S4. Using structure-guided heatmap matching, the candidate acupoint area heatmap is aligned and matched with the standardized acupoint area heatmap database to obtain matched acupoint results, where the standardized acupoint area heatmap database contains standard acupoint distribution information; S5. Integrate the matched acupoint results with the patient's clinical data to form a personalized acupoint candidate set, select the final acupuncture target that meets the heat map responsiveness threshold and individual adaptation conditions, and output the acupuncture positioning result by square root of the final acupuncture target.

2. The acupuncture positioning method based on image processing according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Acquire unified medical human image data of the patient through an imaging device, the unified medical human image data including medical human images acquired by a magnetic resonance imaging device, a computed tomography device, an ultrasound imaging device, or an infrared thermal imaging device, and construct a unified medical human image dataset: Among them, I raw represents the unified medical human image dataset, I i represents the i-th medical human image, f(x,y,c i ) indicates that the horizontal coordinate is x, the vertical coordinate is y, and the channel is c in the medical human image i The pixel value of the medical human image, is the set of medical human image channels, and N is the total number of unified medical human image data; S12. Using a two-dimensional Gaussian smoothing operator to smooth and suppress local random noise of the medical human image in the unified medical human image data to generate noise-removed medical human image data; S13. Using the Laplace transform method to extract the boundary information of the tissue structure in the medical human image, the edge of the medical human image is enhanced to obtain edge-enhanced medical human image data; S14. Perform contrast adjustment on edge-enhanced medical human image data, and use histogram equalization to stretch the grayscale distribution of the medical human image so that the brightness range of the medical human image is evenly distributed, generating contrast-adjusted medical human image data; S15. Perform geometric alignment processing on the contrast-adjusted medical human image data. Based on the standard human anatomy template medical human image, each medical human image is geometrically transformed in spatial scale by constructing an affine transformation matrix. The affine transformation matrix contains rotation, scaling, and translation parameters to obtain a standardized pre-processed medical human image dataset I. pre .

3. The acupuncture positioning method based on image processing according to claim 2, characterized in that: The S2 comprises the following steps: S21. Construct a cross-modal multi-branch deep learning semantic segmentation model. The cross-modal multi-branch deep learning semantic segmentation model includes a feature extraction module, a multi-scale meridian guidance feature fusion module, an acupoint response enhancement module, and a semantic prediction module. The cross-modal multi-branch deep learning semantic segmentation model uses a standardized pre-processed medical human image dataset as input. Each standardized pre-processed medical human image The corresponding modal branches are input through the feature extraction module, including magnetic resonance modal branch, computed tomography modal branch, ultrasound modal branch and infrared modal branch, and the magnetic resonance features are output respectively. Computed tomography characteristics Ultrasound characteristics and infrared characteristics The four types of features together constitute the cross-modal original feature set S22. Using multi-scale meridian-guided feature fusion module to fusion cross-modal original feature sets Perform fusion processing, and introduce the meridian anatomical structure constraint map G during the fusion process meridian =(V acu ,E acu ), in the meridian anatomical structure constraint map, the acupoint node set V acu Represents the three-dimensional coordinate positions and meridian affiliation labels of all standardized human acupuncture points, and the edge set E acu Representing the structural connection relationship between acupoint nodes, under the guidance of the graph structure of the meridian anatomical structure constraint map, the multi-scale meridian guided feature fusion module performs weighted processing on each cross-modal original feature based on the graph attention mechanism to generate the fused acupuncture target feature representation The fusion weight of each cross-modal original feature is recorded as β k , used to measure the importance of each modality in the acupuncture target recognition task; S23. Represent the fused acupuncture target features Input the acupoint response enhancement module for feature enhancement processing. The acupoint response enhancement module has a built-in acupoint response enhancement operator Φ acu The acupoint response enhancement operator is to perform weighted enhancement on the response value of the area close to the standard acupoint in the fusion feature representation after acupoint response enhancement based on the three-dimensional position distribution and anatomical adjacency relationship of each acupoint in the meridian anatomical structure constraint map in the spatial dimension, set the response control parameter γ to adjust the enhancement amplitude of the acupoint area, and set the spatial response scale parameter σ to limit the influence range of acupoint enhancement, and output the enhanced acupoint response fusion feature representation. S24. Representing the fusion features after acupoint response enhancement The input is sent to the semantic prediction module, which consists of a multi-layer acupoint feature-guided deconvolution structure and outputs the acupuncture point semantic probability image S i (x,y,l): Among them, S i (x, y, l) represents the probability value of the spatial coordinate (x, y) of the i-th medical human image belonging to the semantic category l, Semantic category set Contains acupuncture point area, muscle area, bone area and skin area, Deconv h (·) represents the h-th layer feature-guided deconvolution operation, w h is the convolution layer weight obtained by training optimization, H is the number of deconvolution layers, and Softmax is the Softmax function; S25. Based on the semantic probability image of acupuncture points S i (x, y, l) Assign the maximum probability of the semantic category to each pixel position and determine the semantic category label l x,y , obtain the acupuncture point semantic segmentation image set S seg : in, Represents the semantic category label determined at the coordinate (x, y) of the i-th medical human image.

4. The acupuncture positioning method based on image processing according to claim 3, characterized in that: The S3 includes the following steps: S31. Image set S based on semantic segmentation of acupuncture points seg , extract the pixel set marked as acupuncture point area in each medical human image in, Indicates the acupuncture point label corresponding to the acupuncture point area in the semantic category label set, represents the set of pixel coordinates of all acupoint areas determined in the i-th medical human image; S32. Constructing a multi-scale standardized acupuncture point heat map template set The scale set S contains the spatial scale levels of multiple human body regions, and the standardized acupoint heat map template T s It represents the standard acupoint thermal distribution map constructed at scale level s, which is constructed based on the statistical distribution of a large clinical sample: Among them, M s represents the number of standardized acupoints defined under scale level s, is the standard position coordinate of the jth standard acupuncture point at scale level s, is the activation intensity weight of the jth standard acupoint at this scale, σ s is the spatial diffusion coefficient at scale level s, and the standardized acupoint area heat map template T s (x,y) represents the probability density response value of the location as the activated target point; S33. Extracting pixel sets of acupuncture point areas A collection of standardized acupuncture point heat map templates Perform regional projection and response fitting to construct a candidate acupoint area heat map Among them, ω s is the fusion weight of each scale level s in the candidate heat map, is the local weighted kernel function defined within the reference radius r; S34. Map the candidate acupoint area heat map back to the original image size and normalize it to obtain a normalized candidate acupoint area heat map 5. The acupuncture positioning method based on image processing according to claim 4, characterized in that: The S4 comprises the following steps: S41. Construction of a standardized acupoint heat map database in, represents the kth standard acupoint area heat map, represents the mask area of ​​the standard acupoint area in the standard image space, The structural guidance map corresponding to the standard acupoint area includes the boundaries and topological relationships of the main anatomical structures in the standard acupoint area; S42. Heat map of candidate acupoint area Each standard acupoint heat map in the standardized acupoint heat map database Perform structure-guided heatmap matching, perform structure-aware similarity calculation on each pair of standardized acupoint area heatmaps, and obtain the structure-guided matching similarity value Sim i,k ; S43. Recording the structure-guided matching similarity values ​​of all standardized acupoint area heat maps and candidate acupoint area heat maps in sequence according to the standardized acupoint area heat map number k to form a matching result set, which is a combination set including the standardized acupoint area number and its corresponding structure-guided matching similarity value; S44. The matching result set is traversed, sorted according to the structure-guided matching similarity value, and a matching strength threshold δ is set. The standardized acupoint heat map numbers whose structure-guided matching similarity value is greater than or equal to the matching strength threshold δ are selected from the matching result set to obtain a set of matching standard acupoint numbers; S45. Extract the spatial coordinate positions and meridian affiliation labels of all standardized acupoints in the standard acupoint area number set corresponding to the standard acupoint area heat map database as the current candidate acupoint area heat map. The set of successfully matched standard acupuncture points is used to obtain the matched acupuncture point result set The matching acupoint result set contains the two-dimensional coordinate position of each acupoint and its meridian information, x k represents the horizontal spatial coordinate of the kth candidate acupuncture point in the medical human image, y k represents the longitudinal spatial coordinate of the kth candidate acupuncture point in the medical human image, l k Indicates the standardized acupoint label to which the k-th candidate acupuncture point belongs.

6. The acupuncture positioning method based on image processing according to claim 5, characterized in that: The structure guide matching similarity value Sim i,k for: Among them, Sim i,k represents the similarity value of the structure-guided matching between the i-th candidate acupoint area heat map and the k-th standardized heat map, It represents the structural consistency score between the candidate acupoint area heat map and the standardized acupoint area heat map on the structural topology map, which is obtained by inverse mapping of the structural map matching distance.

7. The acupuncture positioning method based on image processing according to claim 1, characterized in that: The S5 comprises the following steps: S51. Obtain patient-specific clinical data, including disease classification information C diag , physiological parameter set Θ phys and previous treatment response index R hist ; S52. Build a personalized acupoint candidate set based on the matched acupoint result set and patient-specific clinical data, and combine it with the current candidate acupoint area heat map The patient's individual physiological adaptability, disease adaptability and previous treatment response indicators are used to set the following screening rules to construct a personalized acupoint candidate set If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.7, it will be recorded as a first-level candidate acupoint; If the current candidate acupoint area heat map And standardized acupoint labels If the physiological parameter matching degree is ≥0.6, it will be recorded as a secondary candidate acupoint; If the current candidate acupoint area heat map or standardized acupoint labels Or if the physiological parameter matching degree is <0.6, it will not be included in the candidate set; in, Indicates the current condition C diag The matching acupoint label set, the physiological parameter matching degree is based on the patient's physiological parameter set Θ phys Calculation of adaptability to the anatomical region where the acupoint is located; S53. Prioritize the first-level candidate acupoints in the personalized acupoint candidate set, and combine the previous treatment response index R hist Sort by acupuncture points with high efficacy scores and balanced spatial distribution as the final acupuncture target set S54. Output the final acupuncture target set The final acupuncture target point set is mapped to the original medical human image space and superimposed to generate a visual acupuncture positioning result image.

8. The acupuncture positioning method based on image processing according to claim 7, characterized in that: The disease classification information C diag Represents the standardized code of the current condition, the physiological parameter set Θ phys Including basic indicators such as body temperature, blood pressure, and physical fitness score, the previous treatment response index R hist Indicates the records of therapeutic effects on different acupoints.

9. The acupuncture positioning method based on image processing according to claim 7, characterized in that: According to the final acupuncture target set Define acupuncture target selection strategy: If the number of first-level candidate acupoints is ≥5, the top 5 acupoints with the highest efficacy scores will be selected first; If the number of first-level candidate acupoints is less than 5, the second-level candidate acupoints with the highest efficacy scores will be added to make it 5; If the efficacy scores of all candidate acupoints are lower than the preset threshold, only one first-level candidate acupoint with the highest responsiveness will be retained.

Images