Human body acupoint automatic identification and projection method, system and device and medium

Abstract

The invention relates to a human body acupoint automatic identification and projection method, system and device and a medium, and belongs to the technical field of medical image processing, and the method comprises the steps: obtaining a real-time patient image, and carrying out the preprocessing; performing patient in-place detection on the preprocessed real-time patient image based on a pre-trained target detection model, and outputting an in-place posture detection result of the patient; when the in-position posture detection result conforms to the preset diagnosis and treatment posture, an acupuncture point coordinate template corresponding to the in-position posture detection result is loaded; based on the acupuncture point coordinate template and a pre-trained key point detection model, performing acupuncture point identification on the preprocessed real-time patient image, and outputting current coordinates of the acupuncture points of the patient in an image coordinate system; converting the current coordinate into a real position coordinate in a real coordinate system according to a preset projection correction value; and controlling the laser galvanometer to project light spots to the body of the patient based on the real position coordinates so as to mark the acupuncture point positions. According to the invention, the accuracy of acupoint positioning can be improved.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method, system, device and medium for automatic identification and projection of human acupoints. Background Technology

[0002] Since the beginning of the 21st century, my country's population has been aging rapidly, and the aging situation is becoming increasingly serious. The health care of this large elderly population has become a significant social issue, impacting my country's economic competitiveness and social security capabilities. The elderly population commonly suffers from chronic diseases such as pain, hypertension, diabetes, and urinary tract diseases (e.g., frequent urination, urgency, incomplete urination), requiring regular or long-term treatment, health maintenance, and rehabilitation. This consumes substantial medical resources, placing a significant burden on individuals, families, and society. Acupuncture, as a traditional Chinese medicine practice, has long been proven to help relieve pain and stress (especially for elderly people with chronic diseases). Therefore, acupuncture is an important and effective alternative medical therapy for elderly people with illnesses, disabilities, or impairments living in areas with low healthcare coverage.

[0003] With the deepening application of artificial intelligence and automation technologies in the medical field, automated auxiliary systems for traditional Chinese medicine acupuncture treatment have become a research hotspot, especially in acupoint identification and annotation. In existing technologies, the automation process of acupuncture treatment faces significant challenges, particularly in acupoint identification. Due to the dense distribution of acupoints on the human body, significant individual differences, and the fact that acupoint targets are typically small in images with indistinct surface features, traditional automated identification methods based on manually designed operators suffer from low robustness and insufficient accuracy. For example, these methods are prone to failure when dealing with different patient body shapes or changes in lighting, leading to significant acupoint location deviations and failing to meet the millimeter-level accuracy requirements of clinical practice. Furthermore, the diverse patient positioning postures (such as supine, lateral, or prone) during acupuncture further reduce the accuracy of acupoint location. Summary of the Invention

[0004] To improve the accuracy of acupoint location, this application provides a method, system, device, and medium for automatic identification and projection of human acupoints.

[0005] In a first aspect, this application provides a method for automatic identification and projection of human acupoints, employing the following technical solution: A method for automatic identification and projection of human acupoints, the method comprising: Acquire real-time patient images and perform preprocessing; Based on a pre-trained target detection model, the pre-processed real-time patient image is used to detect patient positioning and output the patient's positioning posture detection result. When the positioning posture detection result matches the preset treatment posture, the acupoint coordinate template corresponding to the positioning posture detection result is loaded. Based on the acupoint coordinate template and the pre-trained keypoint detection model, acupoint recognition is performed on the pre-processed real-time patient image, and the current coordinates of the patient's acupoints in the image coordinate system are output. The current coordinates are converted into real position coordinates in the real coordinate system according to the preset projection correction amount; The laser galvanometer is controlled to project light spots onto the patient's body based on the actual position coordinates to mark the acupoint locations.

[0006] By adopting the above technical solution and embodiments, image processing, target detection, key point recognition, coordinate correction, and laser projection technology are organically combined to provide an automated and high-precision method for locating acupoints on the human body. This application can reduce manual intervention in clinical diagnosis and treatment, improve treatment efficiency, and ensure the accuracy of treatment. The system automatically loads appropriate acupoint coordinate templates and performs real-time recognition and precise projection by detecting the patient's posture, avoiding the errors of traditional manual annotation. It also has strong operability and real-time performance, making it particularly suitable for modern medical equipment.

[0007] Optionally, the target detection model includes a CSPDarknet backbone network, a feature pyramid network, and a decoupled detection head: The steps for detecting patient placement in the preprocessed real-time patient image based on a pre-trained object detection model include: Multi-scale features of the real-time patient images are extracted using the CSPDarknet backbone network. The multi-scale features are fused using the feature pyramid network. The decoupled detection head outputs the pose category probability and predicted bounding box coordinates; wherein, the decoupled detection head uses anchor-free bounding box technology to compress the number of predicted bounding boxes to a single group; High-confidence prediction boxes are selected based on a global optimal matching strategy, and the in-place posture detection results are output.

[0008] Optionally, the step of performing acupoint recognition on the preprocessed real-time patient image based on the acupoint coordinate template and the pre-trained keypoint detection model, and outputting the current coordinates of the patient's acupoints in the image coordinate system, includes: The real-time patient image is input into the ResNet50 backbone network of the key point detection model to extract image features and output a feature map. The target acupoint region is determined based on the acupoint coordinate template. On the feature map with the resolution reduced to a preset ratio, the initial confidence score and initial coordinate offset of each acupoint within the target acupoint area are generated simultaneously. For each acupoint, identify at least one adjacent acupoint and calculate the coordinate offset of each adjacent acupoint. According to the preset weighting rules, fuse the initial coordinate offset of the current acupoint with the coordinate offsets of all adjacent acupoints to obtain the weighted fusion result. The coordinate offset of the current acupoint is updated based on the weighted fusion result; The coordinates of each acupoint in the image coordinate system are calculated based on the updated coordinate offset, and the coordinate set containing the coordinates of all acupoints is output as the current coordinates.

[0009] Optionally, the projection correction is a transformation matrix between the image coordinate system and the real coordinate system, and the transformation matrix is ​​defined by translation parameters (Δx, Δy) and rotation parameters θ. The projection correction amount satisfies the following formula: in, These are the correction values ​​for the x, y coordinates and angle θ in the projection correction, respectively. Δx, Δy, and Δθ are the x, y coordinate values ​​and angle values ​​in the translation and rotation parameters, respectively. i y i θ i The x, y coordinates and angle of the real-world coordinate point in the real-world coordinate system. d y d θ d These are the coordinates and angles of the real-world coordinates in the image coordinate system.

[0010] Optionally, the method further includes: Acquire the next frame of real-time patient image; Based on the target detection model, the patient's position is detected in the next frame of the real-time patient image, and it is determined whether the position posture detection result has changed. If so, terminate the current acupoint recognition process, update the acupoint coordinate template according to the changed positioning posture detection results, and continue to execute the subsequent acupoint recognition steps; If not, then based on the original acupoint coordinate template and the key point detection model, acupoint recognition is performed on the next frame of real-time patient image, and the coordinates of the patient's acupoints in the image coordinate system are updated to obtain the updated coordinates; Calculate the absolute value of the difference between the updated coordinates and the current coordinates; If the absolute value of the difference is greater than the preset error threshold, the updated coordinates are converted into new real position coordinates according to the projection correction amount, and the laser galvanometer is controlled to reproject the light spot.

[0011] Optionally, the method further includes: When the positioning posture detection result is no patient or incorrect treatment posture, a posture error prompt instruction is generated, and the acupoint recognition and laser projection process is paused; Waiting to acquire new real-time patient images and re-performing the in-place detection step.

[0012] Secondly, this application provides an automatic human acupoint recognition and projection system, which adopts the following technical solution: An automatic acupoint recognition and projection system, the system comprising: The image acquisition module is used to acquire real-time patient images and perform preprocessing. The positioning posture detection module is used to perform patient positioning detection on the pre-processed real-time patient image based on a pre-trained target detection model and output the patient positioning posture detection result. The acupoint template loading module is used to load the acupoint coordinate template corresponding to the positioning posture detection result when the positioning posture detection result matches the preset treatment posture. The acupoint coordinate recognition module is used to perform acupoint recognition on the preprocessed real-time patient image based on the acupoint coordinate template and the pre-trained key point detection model, and output the current coordinates of the patient's acupoints in the image coordinate system. The real coordinate transformation module is used to convert the current coordinates into real position coordinates in the real coordinate system according to a preset projection correction amount; The laser projection module is used to control the laser galvanometer to project light spots onto the patient's body based on the real position coordinates to mark the acupoint locations.

[0013] Thirdly, this application provides a computer device, which adopts the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0014] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the first process of an automatic human acupoint recognition and projection method according to one embodiment of this application.

[0016] Figure 2This is a second flowchart illustrating a method for automatic identification and projection of human acupoints according to one embodiment of this application.

[0017] Figure 3 This is a schematic diagram of the third process of an automatic human acupoint recognition and projection method according to one embodiment of this application.

[0018] Figure 4 This is a schematic diagram of the fourth process of an automatic human acupoint recognition and projection method according to one embodiment of this application.

[0019] Figure 5 This is a schematic diagram of the structure of a single-stage target detection network according to one embodiment of this application.

[0020] Figure 6 This is a schematic diagram of a scenario for automatic recognition and projection of human acupoints according to one embodiment of this application.

[0021] Figure 7 This is a schematic diagram of the structure of an optical scanning head according to one embodiment of this application. Detailed Implementation

[0022] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-7 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0023] This application discloses an automatic identification and projection method for human acupoints.

[0024] Reference Figure 1 A method for automatic identification and projection of human acupoints, the method comprising: Step S101: Acquire real-time patient images and perform preprocessing; This involves continuously acquiring images of the patient's body surface using a high-resolution industrial camera (typically configured with 2K or higher resolution). The camera must be fixed directly above the bed, maintaining a perpendicular projection relationship with the patient's body surface, ensuring that the image coordinate system is mappable to the real-world coordinate system. The acquisition process must meet real-time requirements (frame rate ≥ 30fps) to capture the patient's subtle movements.

[0025] Furthermore, the original image needs to be preprocessed before being transmitted to the processor, including pixel value normalization (to eliminate lighting differences), histogram equalization (to enhance contrast), and grayscale conversion (to reduce computational complexity). For example, the RGB image is converted into single-channel data using the weighting formula Gray = 0.299R + 0.587G + 0.114B, providing standardized input for subsequent model processing.

[0026] For example, the normalization algorithm can be expressed as: Among them, I max and I min The grayscale images I are respectively gray The maximum and minimum gray values ​​in I n This is for normalized output.

[0027] For example, in the gray-level histogram equalization algorithm, P(j) represents the frequency of a pixel with gray-level value j appearing in the image, and CDF is the cumulative distribution function. H(i) represents the pixel value after histogram equalization, M and N represent the image width and height, respectively, and L represents the range of pixel values. min represents the minimum pixel value of the image before histogram equalization.

[0028] Step S102: Based on the pre-trained target detection model, perform patient positioning detection on the pre-processed real-time patient image and output the patient's positioning posture detection result. The positioning posture test results include no patient, incorrect treatment posture, supine, lateral, or prone position; In one possible implementation, the object detection model can adopt the core design of the YOLOX (You Only Look OnceX) architecture, including the CSPDarknet backbone network, feature pyramid network, and decoupled detection head. By adopting the idea of ​​decoupling classification and regression tasks, combined with decoupled head technology and anchor-free bounding box technology, and with the SimOTA advanced label assignment strategy, data augmentation techniques such as Mosaic and MixUp, the output includes a posture category (no patient / incorrect posture / supine / lateral / prone) containing a probability distribution.

[0029] Step S103: When the positioning posture detection result matches the preset treatment posture, load the acupoint coordinate template corresponding to the positioning posture detection result; Different body positions lead to significant differences in the spatial distribution of acupoints on the body surface. For example, in the supine position, acupoints on the chest and abdomen such as Tanzhong and Zhongwan are exposed, while in the prone position, acupoints on the lower back such as Dazhui and Mingmen need to be located. The system pre-stores standardized digital templates, each template containing: a standard coordinate set of acupoints (normalized coordinates on a standard human body model) and spatial constraint relationships (topological connection diagram between acupoints).

[0030] Furthermore, the preset treatment posture can be set to supine or prone. For example, when the posture detection result is "supine", the chest and abdomen template (containing coordinates of 120 acupoints) is invoked; when "prone" is detected, the waist and back template (containing 80 acupoints) is loaded. The templates serve as prior knowledge to guide the key point detection model to focus on the effective area, avoiding invalid detection of non-exposed acupoints (such as acupoints on the back when supine).

[0031] In addition, when the positioning posture detection result is no patient or incorrect treatment posture, a posture error prompt instruction is generated, and the acupoint recognition and laser projection process is paused, waiting for new real-time patient images to be acquired and the above positioning detection steps to be re-executed.

[0032] Step S104: Based on the acupoint coordinate template and the pre-trained key point detection model, perform acupoint recognition on the preprocessed real-time patient image and output the current coordinates of the patient's acupoints in the image coordinate system. In one possible implementation, the keypoint detection model can be a PIPNet (Pixel-in-Pixel Net) network. Specifically, ResNet50 can be selected as the network backbone to extract features of the patient's acupoints. Unlike traditional heatmap regression, the PIPNet network eliminates the upsampling process and directly predicts scores and offsets on the low-resolution feature map. Furthermore, the keypoint detection model can use a nearest neighbor regression module (NRM) to further improve the model's detection performance. The NRM module can not only add its own offset to each keypoint but also predict the offsets of several of its nearest neighbors simultaneously for each keypoint.

[0033] Step S105: Convert the current coordinates into real position coordinates in the real coordinate system according to a preset projection correction amount. Specifically, the real position information of the patient's acupoints can be determined based on the projection correction amount and the current coordinates of the acupoints. Then, a laser galvanometer is used to project the acupoints onto their real positions using this real position information, thus marking the acupoints on the patient's body. The projection correction amount can be determined based on pre-aligned known image coordinates and real coordinates.

[0034] In this embodiment, the projection correction is the transformation matrix between the image coordinate system and the real coordinate system. The transformation matrix is ​​defined by the translation parameter (Δx, Δy) and the rotation parameter θ. Specifically, the projection correction amount satisfies the following formula: in, These are the correction values ​​for the x, y coordinates and angle θ in the projection correction, respectively. Δx, Δy, and Δθ are the x, y coordinate values ​​and angle values ​​in the translation and rotation parameters, respectively. i y i θ i The x, y coordinates and angle of the real-world coordinate point in the real-world coordinate system. d y d θ d These are the coordinates and angles of the points in the image coordinate system, corresponding to the real-world coordinates.

[0035] Step S106: Control the laser galvanometer to project light spots onto the patient's body based on the real position coordinates to mark the acupoint locations.

[0036] Specifically, laser projection technology relies on the precise control of a laser galvanometer to mark the patient's body surface. By inputting the converted real coordinates into the laser galvanometer control system, the system can precisely adjust the angle and position of the laser, ensuring that the light spot is accurately projected onto the patient's specific acupoints. This process can be adjusted in real time through a computer control system, ensuring high precision and real-time performance of the projection process.

[0037] The laser galvanometer system consists of X / Y axis galvanometers, whose deflection angles θx and θy satisfy the following conditions with respect to the true coordinates: θ x =k x ·x real +b x θ y =k y ·y real +b y ; Where k and b are pre-calibrated linear parameters, x real and y real These are the actual coordinates.

[0038] The above embodiments organically combine image processing, target detection, key point recognition, coordinate correction, and laser projection technology to provide an automated and high-precision method for locating acupoints on the human body. This application can reduce manual intervention in clinical diagnosis and treatment, improve treatment efficiency, and ensure the accuracy of treatment. The system automatically loads appropriate acupoint coordinate templates and performs real-time recognition and precise projection by detecting the patient's posture, avoiding the errors of traditional manual annotation. It also has strong operability and real-time performance, making it particularly suitable for modern medical equipment.

[0039] Reference Figure 2 As one implementation of step S102, the step of performing patient placement detection on the preprocessed real-time patient image based on a pre-trained target detection model includes: Step S201: Extract multi-scale features of real-time patient images through the CSPDarknet backbone network; The CSPDarknet backbone network optimizes gradient flow through Cross Stage Partial Connections, extracting multi-scale features while reducing computational cost.

[0040] Specifically, the real-time patient image first passes through a convolutional layer, transforming the number of input channels (e.g., 3) into the number of output channels in the network's first output (e.g., 32). Then, the output tensor of the convolutional layer is fed into the first residual block of CSPDarknet. In this residual block, the input tensor is fed into a series of convolutional and batch normalization layers, followed by a non-linear activation function (e.g., Leaky ReLU), and then added to the input tensor via a skip connection. Finally, the output tensor is fed to the next layer. This process continues, with the entire structure of CSPDarknet consisting of multiple residual blocks. The output tensor of each residual block has the same size as its input tensor. Finally, through the stacking of multiple residual blocks, the number of channels in the output tensor increases to 1024, and the size becomes 13x13, resulting in a 13×13×1024 high-dimensional feature map.

[0041] Step S202: Use the feature pyramid network to fuse multi-scale features; Among them, the Feature Pyramid Network (PAFPN) integrates deep semantic features and shallow location features through bidirectional aggregation paths from top to bottom and bottom to top. For example, it concatenates features from the Conv5 layer (large receptive field) and the Conv3 layer (high resolution) to improve the detection capability of small targets (such as the edges of a patient's limbs).

[0042] Step S203: The decoupled detection head outputs the pose category probability and the coordinates of the predicted bounding box; wherein, the decoupled detection head uses anchor-free bounding box technology to compress the number of predicted bounding boxes to a single group; Specifically, the decoupled detection head separates the classification task (pose category) from the regression task (box coordinates), avoiding feature interference between the two tasks. Anchor-free technology eliminates preset anchor boxes and directly predicts the target center point offset, compressing the number of predicted boxes by more than 80%.

[0043] Step S204: Based on the global optimal matching strategy, select high-confidence prediction boxes and output the in-place posture detection results.

[0044] Among them, the SimOTA label allocation strategy is used to dynamically match the predicted bounding boxes and the ground truth bounding boxes, and the cost function Cost = λ is used. cls ·L cls +λ reg ·L reg Filter high-confidence detection results (λ is the balance coefficient) and output the posture category (no patient / incorrect posture / supine / lateral / prone) containing the probability distribution.

[0045] In the above implementation, a highly efficient and reliable patient posture detection system was constructed based on a four-level progressive architecture of cross-stage feature optimization extraction, bidirectional multi-scale fusion, task decoupling and anchor box elimination, and global dynamic matching (SimOTA). In complex scenarios (such as partial patient occlusion or multi-angle lateral lying), the detection accuracy and efficiency are greatly improved, laying a precise posture benchmark for the spatial positioning of subsequent acupoint recognition and avoiding acupoint mapping deviations caused by incorrect basic posture.

[0046] Reference Figure 3 As one implementation of step S104, the step of performing acupoint recognition on the preprocessed real-time patient image based on the acupoint coordinate template and the pre-trained keypoint detection model, and outputting the current coordinates of the patient's acupoints in the image coordinate system includes: Step S301: Input the real-time patient image into the ResNet50 backbone network of the key point detection model to extract image features and output the feature map; Among them, ResNet50, a classic deep residual network with 50 layers, is capable of effectively extracting high-level features from images. This network addresses the vanishing gradient and information loss problems that occur during deep network training by introducing residual connections, enabling the model to learn and extract features from images at a deeper level. For the task of human acupoint recognition, ResNet50, through its powerful feature learning capabilities, can extract rich local and global features from images, such as the human body's contours and joint positions. These features are crucial for subsequent acupoint detection.

[0047] Step S302: Determine the target acupoint area based on the acupoint coordinate template; The system uses predefined acupoint coordinate templates to locate areas in an image that may contain acupoints. These templates are typically generated from expert experience or previously labeled data, and contain acupoint coordinate information for different body positions. By matching these templates with the input patient image, the system can identify the target area, i.e., the region where potential acupoints are located. This process needs to take into account factors such as the patient's different body positions, postures, and angles; therefore, the template matching and region identification must be sufficiently accurate to ensure the accuracy of acupoint identification in subsequent steps.

[0048] Step S303: On the feature map with the resolution reduced to a preset ratio, the initial confidence score and initial coordinate offset of each acupoint within the target acupoint area are generated synchronously. The reduced resolution aims to decrease computational burden, improve processing speed, and allow the model to capture more global features without losing crucial information in details. For example, the preset ratio can be set to 1 / 4 of the original image. The initial confidence score represents the model's confidence in the existence of a specific acupoint, typically expressed as the probability output by the neural network. The initial coordinate offset refers to the offset of the acupoint relative to its expected position in the target region. This offset is calculated based on the matching between the features extracted by the network from the image and the preset template.

[0049] Step S304: Identify at least one adjacent acupoint for each acupoint and calculate the coordinate offset corresponding to each adjacent acupoint. Specifically, in practical applications, acupoints on the human body follow certain patterns, and their spatial positions are relatively fixed. Therefore, when a system detects an acupoint, it needs to simultaneously identify its adjacent acupoints. These adjacent points can be defined by anatomical features of the human body, such as the relative positions of the shoulder and elbow, or the relative positions of the navel and spine. Based on the relationships between adjacent acupoints, the system can calculate the coordinate offset of each adjacent acupoint. These offsets provide crucial information for subsequent coordinate fusion.

[0050] Step S305: According to the preset weighting rules, the initial coordinate offset of the current acupoint is fused with the coordinate offsets of all adjacent acupoints to obtain the weighted fusion result. The purpose of weighted fusion is to improve the accuracy of coordinate offset by comprehensively considering the spatial relationships between multiple acupoints and the coordination between adjacent points. Weighting rules are typically determined based on factors such as the confidence score of each acupoint, its geometric relationship with adjacent acupoints, and historical data. Through reasonable weighting, the system can reduce the deviation caused by errors in detecting individual points, resulting in a more accurate coordinate offset.

[0051] Step S306: Update the coordinate offset of the current acupoint based on the weighted fusion result; Through this update process, the system can continuously optimize the positioning accuracy of acupoints. The updated coordinate offset reflects the actual location information of the current acupoint and takes into account the location information and coordination of all adjacent points to ensure accurate positioning under different postures.

[0052] Step S307: Calculate the coordinate position of each acupoint in the image coordinate system based on the updated coordinate offset, and output a coordinate set containing the coordinates of all acupoints as the current coordinates.

[0053] Specifically, the system can not only accurately locate the coordinates of each acupoint, but also transform these coordinates into specific locations in an image coordinate system, ultimately outputting a coordinate set containing the locations of all acupoints. This coordinate set will provide precise spatial information for subsequent laser projection, treatment positioning, or other medical applications.

[0054] In the above embodiments, the updated coordinate calculation yields a precise set of acupoint coordinates, providing high-precision spatial coordinates for subsequent treatment positioning or laser projection applications. This technical solution can be widely applied in fields such as smart healthcare and remote diagnosis, improving the automation and precision of the treatment process while reducing manual intervention and optimizing the patient's treatment experience.

[0055] In some embodiments, the loss function for training the keypoint detection model can satisfy the following formula: In the above formula, D KL (p,q) is the KL divergence value, typically used to measure the difference between two probability distributions, i.e., the loss. i ) and q(x i ) are distributions P and Q at point x, respectively. i The probability or probability density of the acupoints; λ, α, γ are hyperparameters, 0.1, 100, 50 respectively; N is the total number of samples; n is the total number of acupoints; x i p(x) is the distribution of the i-th sample; i ) represents the true distribution of the i-th sample, q(x) i ) represents the predicted distribution of the i-th sample. Φ(r) t ,r i The ) represents the normalized distance between the predicted value and the actual value. s It is the input image size multiplied by the segmentation factor. f represents the prediction accuracy, pred represents the predicted coordinate values, gt represents the true coordinate values, and bin represents the number of subpixel classifications on the vertical or horizontal axis. The f value reflects the accuracy of the prediction; the more accurate the prediction, the closer the f value is to 1; conversely, the larger the prediction error, the smaller the f value.

[0056] Reference Figure 4 As a further implementation of the automatic identification and projection method for human acupoints, the method also includes: Step S401: Obtain the next frame of real-time patient image; Each time, the system acquires a new patient image from a sensor (such as a camera, infrared device, or other image acquisition device). The frequency of image acquisition is usually related to the speed of patient movement and the system's response requirements, achieving real-time performance. The real-time acquired images provide data support for subsequent analysis and processing. In practical applications, the quality of this step directly affects the accuracy of subsequent target detection and acupoint recognition; therefore, the resolution and acquisition frequency of the image acquisition device must meet the needs of the medical scenario.

[0057] Step S402: Based on the target detection model, perform patient placement detection on the next frame of real-time patient image; Specifically, patient positioning detection is performed on the next frame of the real-time patient image based on an object detection model. This step utilizes the object detection model to detect the patient's position and posture in the current image. In practical applications, the goal of this step is to determine whether the patient is in position and whether they are in the correct treatment posture. The object detection model outputs a classification result of the positioning posture, such as "no patient," "incorrect posture," "supine," "lateral," or "prone," which provides crucial data for subsequent acupoint identification and coordinate calculation.

[0058] Step S403: Determine whether the positioning posture detection result has changed; if yes, proceed to step S404; if no, proceed to step S405. This includes monitoring the stability of the patient's posture. If the patient's posture remains unchanged, the system can directly use the previous coordinate template and detection results for acupoint identification, avoiding redundant calculations and improving system efficiency. If the posture changes, the updated coordinate template needs to be reloaded to ensure the accuracy and real-time nature of acupoint labeling.

[0059] Step S404: Terminate the current acupoint recognition process, update the acupoint coordinate template based on the changed positioning posture detection results, and continue to execute subsequent acupoint recognition steps; Since changes in patient posture can lead to relative shifts in acupoint locations, the acupoint coordinate template needs to be updated based on the new posture data. This step ensures accurate acupoint identification each time based on the patient's latest posture. By updating the coordinate template, the system can adapt to the acupoint distribution under different postures, thereby improving the accuracy of subsequent identification and projection.

[0060] Step S405: Based on the original acupoint coordinate template and key point detection model, perform acupoint recognition on the next frame of real-time patient image, update the coordinates of the patient's acupoints in the image coordinate system, and obtain the updated coordinates. The system continues to utilize the previously updated coordinate template and keypoint detection model (typically employing deep learning models such as convolutional neural networks and ResNet) to identify new acupoints. The model extracts the patient's joint and acupoint features from real-time images and, combined with the updated coordinate template, outputs the new coordinates of the patient's acupoints in the image coordinate system. The key to this process is accurately identifying the acupoints on the patient's body surface and updating their relative positions, which is crucial for subsequent laser projection and treatment localization.

[0061] Step S406: Calculate the absolute value of the difference between the updated coordinates and the current coordinates; Specifically, the system calculates the difference between the current coordinates and the acupoint coordinates in the previous frame. This difference reflects the magnitude of changes in patient posture or detection error. A large difference may indicate significant posture changes or system errors, requiring further correction. The purpose of calculating this difference is to determine whether the projection coordinates need adjustment, thereby ensuring the accuracy of laser projection.

[0062] In step S407, if the absolute value of the difference is greater than the preset error threshold, the updated coordinates are converted into new real position coordinates according to the projection correction amount, and the laser galvanometer is controlled to reproject the light spot.

[0063] The process of converting the updated coordinates into actual spatial coordinates is achieved through projection correction, which is typically based on factors such as camera calibration parameters, the scaling relationship between coordinate systems, and the relative position between the sensor and the patient. This correction ensures that the laser projection system accurately projects the light spot onto the patient's actual acupoint location. If the updated coordinate difference exceeds a preset error threshold, the system will recalculate the light spot's position and readjust the laser projection angle to achieve accurate acupoint marking.

[0064] It should be noted that after reprojection, step 401 can continue to be executed until the error between the projected light spot and the updated coordinates is less than or equal to the preset error threshold.

[0065] In the above embodiments, the acupoint labels are dynamically adjusted according to real-time patient images to ensure the accuracy of each labeling and avoid labeling deviations caused by changes in patient posture or detection errors. This can greatly improve the automation and accuracy of the medical treatment process, reduce manual intervention, optimize the patient's treatment experience, and has strong real-time response capabilities and adaptability.

[0066] Reference Figure 5In another example, as a preferred implementation for dynamic position correction, when more precise projection spot position information is needed, the system can input the next frame of real-time patient image into a single-stage target detection network for processing. This network is based on a VGG16 convolutional neural network architecture, selecting the feature map output from the third layer (Conv4_3) of the fourth convolutional block as the first detection layer to preserve high-resolution spatial details. Simultaneously, three core structural modifications are implemented: the original fully connected layer FC7 is replaced with an equivalent convolutional layer Conv7 to support full convolution operations, and four feature layers—Conv8, Conv9, Conv10, and Conv11—are added sequentially to form a cross-scale feature detection system. The network pre-determines the spatial distribution of initial prediction boxes at each feature layer and outputs detection results through a dual-branch parallel processing mechanism: the classification branch (cls) consists of six consecutive convolutional layers, outputting a tensor of dimension k×class_num (where k represents the number of prediction boxes in the current feature layer and class_num is the total number of acupoint categories), representing the probability of acupoint presence for each prediction box; the regression branch (loc) also processes six convolutional layers, outputting a displacement parameter tensor of dimension k×4, including the center point coordinate offset (Δx, Δy) and width and height scaling (Δw, Δh). The network iteratively corrects the initial prediction box coordinates based on the displacement parameters, gradually converging the prediction boxes to the actual acupoint locations, and finally outputting sub-pixel precision light spot coordinate update values.

[0067] This application also discloses an automatic human acupoint recognition and projection system.

[0068] An automatic acupoint recognition and projection system, comprising: The image acquisition module is used to acquire real-time patient images and perform preprocessing. The positioning posture detection module is used to perform patient positioning detection on pre-processed real-time patient images based on a pre-trained target detection model and output the patient's positioning posture detection results. The acupoint template loading module is used to load the acupoint coordinate template corresponding to the positioning posture detection result when the positioning posture detection result matches the preset treatment posture. The acupoint coordinate recognition module is used to recognize acupoints in preprocessed real-time patient images based on acupoint coordinate templates and pre-trained keypoint detection models, and output the current coordinates of the patient's acupoints in the image coordinate system. The real coordinate transformation module is used to convert the current coordinates into real position coordinates in the real coordinate system according to the preset projection correction amount; The laser projection module is used to control the laser galvanometer to project light spots onto the patient's body based on the real position coordinates in order to mark the location of acupoints.

[0069] Reference Figure 6The image acquisition module (e.g., a camera) can be positioned directly above the hospital bed, and together with the laser galvanometer in the laser projection module, it is fixed on a mounting frame and at the same horizontal level.

[0070] Reference Figure 7 It is also equipped with an optical scanning head, and the laser emitted by the laser galvanometer is projected onto the patient's acupoints after passing through the optical scanning head.

[0071] The automatic human acupoint recognition and projection system of this application embodiment can realize any of the above-mentioned automatic human acupoint recognition and projection methods, and the specific working process of each module in the automatic human acupoint recognition and projection system can refer to the corresponding process in the above-mentioned method embodiment.

[0072] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0073] This application also discloses a computer device.

[0074] Computer equipment, including memory, processor, and computer program stored in memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for automatic identification and projection of acupoints.

[0075] This application also discloses a computer-readable storage medium.

[0076] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as any of the above-described methods for automatic identification and projection of human acupoints.

[0077] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0078] It should be noted that the computer device and storage medium in the embodiments of this application are respectively electronic devices and storage media that apply the above-described automatic human acupoint recognition and projection method. Therefore, all embodiments of the above-described automatic human acupoint recognition and projection method are applicable to the computer device and storage medium, and can achieve the same or similar beneficial effects. For the computer device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple; relevant details can be found in the descriptions of the method embodiments.

[0079] In this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0080] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0081] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for automatic identification and projection of human acupoints, characterized in that, The method includes: Acquire real-time patient images and perform preprocessing; Based on a pre-trained target detection model, the pre-processed real-time patient image is used to detect patient positioning and output the patient's positioning posture detection result. When the positioning posture detection result matches the preset treatment posture, the acupoint coordinate template corresponding to the positioning posture detection result is loaded. Based on the acupoint coordinate template and the pre-trained keypoint detection model, acupoint recognition is performed on the pre-processed real-time patient image, and the current coordinates of the patient's acupoints in the image coordinate system are output. The current coordinates are converted into real position coordinates in the real coordinate system according to the preset projection correction amount; The laser galvanometer is controlled to project light spots onto the patient's body based on the actual position coordinates to mark the acupoint locations.

2. The method for automatic identification and projection of human acupoints according to claim 1, characterized in that, The target detection model includes a CSPDarknet backbone network, a feature pyramid network, and a decoupled detection head: The steps of performing patient placement detection on the pre-processed real-time patient image based on the pre-trained target detection model include: extracting multi-scale features of the real-time patient image through the CSPDarknet backbone network; The multi-scale features are fused using the feature pyramid network. The decoupled detection head outputs the pose category probability and predicted bounding box coordinates; wherein, the decoupled detection head uses anchor-free bounding box technology to compress the number of predicted bounding boxes to a single group; High-confidence prediction boxes are selected based on a global optimal matching strategy, and the in-place posture detection results are output.

3. The method for automatic identification and projection of human acupoints according to claim 1, characterized in that, The steps of identifying acupoints in the preprocessed real-time patient image based on the acupoint coordinate template and the pre-trained keypoint detection model, and outputting the current coordinates of the patient's acupoints in the image coordinate system, include: The real-time patient image is input into the ResNet50 backbone network of the key point detection model to extract image features and output a feature map. The target acupoint region is determined based on the acupoint coordinate template. On the feature map with the resolution reduced to a preset ratio, the initial confidence score and initial coordinate offset of each acupoint within the target acupoint area are generated simultaneously. For each acupoint, identify at least one adjacent acupoint and calculate the coordinate offset of each adjacent acupoint. According to the preset weighting rules, fuse the initial coordinate offset of the current acupoint with the coordinate offsets of all adjacent acupoints to obtain the weighted fusion result. The coordinate offset of the current acupoint is updated based on the weighted fusion result; The coordinates of each acupoint in the image coordinate system are calculated based on the updated coordinate offset, and the coordinate set containing the coordinates of all acupoints is output as the current coordinates.

4. The method for automatic identification and projection of human acupoints according to claim 1, characterized in that, The projection correction is a transformation matrix between the image coordinate system and the real coordinate system, and the transformation matrix is ​​defined by translation parameters (Δx, Δy) and rotation parameters θ. The projection correction amount satisfies the following formula: in, These are the correction values ​​for the x, y coordinates and angle θ in the projection correction, respectively. Δx, Δy, and Δθ are the x, y coordinate values ​​and angle values ​​in the translation and rotation parameters, respectively. i y i θ i The x, y coordinates and angle of the real-world coordinate point in the real-world coordinate system. d y d θ d These are the coordinates and angles of the real-world coordinates in the image coordinate system.

5. The method for automatic identification and projection of human acupoints according to any one of claims 1 to 4, characterized in that, The method further includes: Acquire the next frame of real-time patient image; Based on the target detection model, the patient's position is detected in the next frame of the real-time patient image, and it is determined whether the position posture detection result has changed. If so, terminate the current acupoint recognition process, update the acupoint coordinate template according to the changed positioning posture detection results, and continue to execute the subsequent acupoint recognition steps; If not, then based on the original acupoint coordinate template and the key point detection model, acupoint recognition is performed on the next frame of real-time patient image, and the coordinates of the patient's acupoints in the image coordinate system are updated to obtain the updated coordinates; Calculate the absolute value of the difference between the updated coordinates and the current coordinates; If the absolute value of the difference is greater than the preset error threshold, the updated coordinates are converted into new real position coordinates according to the projection correction amount, and the laser galvanometer is controlled to reproject the light spot.

6. The method for automatic identification and projection of human acupoints according to claim 5, characterized in that, The method further includes: When the positioning posture detection result is no patient or incorrect treatment posture, a posture error prompt instruction is generated, and the acupoint recognition and laser projection process is paused; Waiting to acquire new real-time patient images and re-performing the in-place detection step.

7. An automatic human acupoint recognition and projection system, characterized in that, The system includes: The image acquisition module is used to acquire real-time patient images and perform preprocessing. The positioning posture detection module is used to perform patient positioning detection on the pre-processed real-time patient image based on a pre-trained target detection model and output the patient positioning posture detection result. The acupoint template loading module is used to load the acupoint coordinate template corresponding to the positioning posture detection result when the positioning posture detection result matches the preset treatment posture. The acupoint coordinate recognition module is used to perform acupoint recognition on the preprocessed real-time patient image based on the acupoint coordinate template and the pre-trained key point detection model, and output the current coordinates of the patient's acupoints in the image coordinate system. The real coordinate transformation module is used to convert the current coordinates into real position coordinates in the real coordinate system according to a preset projection correction amount; The laser projection module is used to control the laser galvanometer to project light spots onto the patient's body based on the real position coordinates to mark the acupoint locations.

8. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images