Lymphedema identification method and device applied to postoperative breast cancer

Abstract

The embodiment of the invention discloses a lymphedema identification method and device applied to breast cancer postoperation. A specific embodiment of the method comprises the following steps: constructing an upper limb three-dimensional model according to an acquired upper limb image sequence; performing upper limb state recognition according to the upper limb image sequence and the upper limb three-dimensional model to generate upper limb state information; according to the upper limb state information, the collected chief complaint information, breast cancer operation treatment information corresponding to the to-be-recognized object and a pre-trained lymphedema recognition model, lymphedema recognition information is generated; and in response to the lymphedema identification information, representing that the to-be-identified object has lymphedema, and according to the lymphedema identification information, generating rehabilitation suggestion information corresponding to the to-be-identified object. By means of the implementation mode, the patient after the breast cancer operation can find out the lymphedema of the patient as soon as possible, and the patient can be prevented and treated as soon as possible when the lymphedema occurs.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and more particularly, to a method and apparatus for identifying lymphedema after breast cancer surgery. Background Art

[0002] Breast cancer-related lymphedema (BCRL) is a common chronic complication after breast cancer surgery. It is caused by the obstruction of the lymphatic fluid rich in protein from flowing back after breast cancer surgery, which then accumulates in the tissue space and causes edema, resulting in lymphatic system circulation disorders. It is manifested as limited range of limb function activities, pain, fat deposition, fibrosis and other upper limb changes in patients, and even severe skin infections. Since it is a chronic complication after surgery, how to enable patients to detect lymphedema in themselves as early as possible has positive significance for the treatment of lymphedema.

[0003] The above information disclosed in this background art section is only used to enhance the understanding of the background of the inventive concept, and thus, it may include information that does not form the prior art known to those of ordinary skill in the art. Summary of the Invention

[0004] This section of the present disclosure is used to introduce the inventive concept in a brief form, which will be described in detail in the following detailed implementation section. This section of the present disclosure is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0005] Some embodiments of the present disclosure propose a method and apparatus for identifying lymphedema after breast cancer surgery to solve the technical problems mentioned in the above background art section.

[0006] In a first aspect, some embodiments of the present disclosure provide a method for identifying lymphedema after breast cancer surgery. The method includes: constructing a three-dimensional model of the upper limb based on the acquired sequence of upper limb images, where the three-dimensional model of the upper limb is a local three-dimensional model corresponding to at least three upper limb measurement regions acquired from multiple angles for the object to be identified; identifying the state of the upper limb based on the sequence of upper limb images and the three-dimensional model of the upper limb to generate upper limb state information, where the upper limb state information includes: upper limb dimension features, upper limb skin features, the upper limb dimension features include: upper limb arm circumference, and the upper limb skin features include: upper limb color, allergy features, and scar features; generating lymphedema identification information based on the upper limb state information, the acquired chief complaint information, the breast cancer surgery treatment information corresponding to the object to be identified, and a pre-trained lymphedema identification model, where the chief complaint information is a description of the upper limb state provided by the object to be identified, and the breast cancer surgery treatment information includes: surgical site, surgical incision type, and radiotherapy information; in response to the lymphedema identification information indicating that the object to be identified has lymphedema, generating rehabilitation advice information corresponding to the object to be identified based on the lymphedema identification information.

[0007] In a second aspect, some embodiments of the present disclosure provide a device for identifying lymphedema after breast cancer surgery. The device includes: a construction unit configured to construct a three-dimensional model of the upper limb based on the acquired sequence of upper limb images, where the three-dimensional model of the upper limb is a local three-dimensional model corresponding to at least three upper limb measurement regions acquired from multiple angles for the object to be identified; an upper limb state identification unit configured to identify the state of the upper limb based on the sequence of upper limb images and the three-dimensional model of the upper limb to generate upper limb state information, where the upper limb state information includes: upper limb dimension features, upper limb skin features, the upper limb dimension features include: upper limb arm circumference, and the upper limb skin features include: upper limb color, allergy features, and scar features; a first generation unit configured to generate lymphedema identification information based on the upper limb state information, the acquired chief complaint information, the breast cancer surgery treatment information corresponding to the object to be identified, and a pre-trained lymphedema identification model, where the chief complaint information is a description of the upper limb state provided by the object to be identified, and the breast cancer surgery treatment information includes: surgical site, surgical incision type, and radiotherapy information; a second generation unit configured to, in response to the lymphedema identification information indicating that the object to be identified has lymphedema, generate rehabilitation advice information corresponding to the object to be identified based on the lymphedema identification information.

[0008] In a third aspect, some embodiments of the present disclosure provide an electronic device, including: one or more processors; a storage device storing one or more programs thereon, which, when executed by the one or more processors, cause the one or more processors to implement the method described in any implementation manner of the first aspect above.

[0009] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium storing a computer program thereon, wherein the program, when executed by a processor, implements the method described in any implementation manner of the first aspect above.

[0010] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: Through the application of some embodiments of the present disclosure to the method for identifying lymphedema after breast cancer surgery, patients after breast cancer surgery can detect the lymphedema that occurs to themselves as early as possible, so that they can carry out prevention and treatment as early as possible when lymphedema occurs. Specifically, since lymphedema is a common chronic complication after breast cancer surgery and may occur during the months to years after surgery, it is necessary for patients to spontaneously determine whether they have lymphedema. Based on this, in the method for identifying lymphedema after breast cancer surgery according to some embodiments of the present disclosure, first, an upper limb three-dimensional model is constructed according to the collected upper limb image sequence, wherein the above-mentioned upper limb three-dimensional model is a local three-dimensional model corresponding to at least three upper limb measurement regions collected from multiple angles for the object to be identified. By combining the upper limb images to construct a local three-dimensional model (upper limb three-dimensional model) corresponding to at least three upper limb measurement regions, the upper limb corresponding to the object to be identified can be effectively modeled. Secondly, the upper limb state is identified according to the above-mentioned upper limb image sequence and the above-mentioned upper limb three-dimensional model to generate upper limb state information, wherein the above-mentioned upper limb state information includes: upper limb dimension characteristics, upper limb skin characteristics, the above-mentioned upper limb dimension characteristics include: upper limb arm circumference, and the above-mentioned upper limb skin characteristics include: upper limb color, allergy characteristics and scar characteristics. In this way, by combining the upper limb images and the upper limb three-dimensional model, the upper limb dimension characteristics and upper limb skin characteristics of the object to be identified are identified. Then, according to the above-mentioned upper limb state information, the collected chief complaint information, the breast cancer surgery treatment information corresponding to the above-mentioned object to be identified, and the pre-trained lymphedema identification model, lymphedema identification information is generated, wherein the above-mentioned chief complaint information is a description of the upper limb state provided by the above-mentioned object to be identified, and the above-mentioned breast cancer surgery treatment information includes: surgical site, surgical incision type and radiotherapy information. Through the set lymphedema identification model, the lymphedema identification information is automatically generated by combining the upper limb state information, the chief complaint information and the breast cancer surgery treatment information. In this way, patients can quickly identify whether they have lymphedema by collecting the upper limb image sequence of their own upper limbs without going to the hospital. Finally, in response to the above-mentioned lymphedema identification information indicating that the above-mentioned object to be identified has lymphedema, according to the above-mentioned lymphedema identification information, rehabilitation advice information corresponding to the above-mentioned object to be identified is generated. In this way, rehabilitation advice that matches the type of lymphedema is generated, so that patients can relieve lymphedema at home by combining the rehabilitation advice when the symptoms are mild. In this way, patients after breast cancer surgery can detect the lymphedema that occurs to themselves as early as possible, so that they can carry out prevention and treatment as early as possible when lymphedema occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In conjunction with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages and aspects of the various embodiments of the present disclosure will become more apparent. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic and the elements and elements are not necessarily drawn to scale.

[0012] Figure 1 is a flowchart of some embodiments of a method for recognizing lymphedema after breast cancer surgery according to the present disclosure; Figure 2 is a schematic diagram of the position of the upper limb measurement area; Figure 3 is a comparison diagram of the upper limb image, the first channel image, the second channel image, and the third channel image; Figure 4 is a schematic diagram of the generation process of the set of image blocks corresponding to the first channel image; Figure 5 is a schematic structural diagram of some embodiments of a device for recognizing lymphedema after breast cancer surgery according to the present disclosure; Figure 6 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. Specific Embodiments

[0013] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0014] In addition, it should be noted that for the sake of convenience of description, only the parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0015] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or mutual dependence relationship of the functions performed by these devices, modules or units.

[0016] It should be noted that the modifications of "one" and "plural" mentioned in the present disclosure are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0017] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are for illustrative purposes only and are not used to limit the scope of these messages or information.

[0018] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0019] Reference Figure 1 , a flowchart 100 of some embodiments of a lymphoedema recognition method applied to breast cancer patients after surgery according to the present disclosure is shown. The lymphoedema recognition method applied to breast cancer patients after surgery includes the following steps: Step 101, constructing a three-dimensional model of the upper limb according to the acquired sequence of upper limb images.

[0020] In some embodiments, the execution subject (e.g., a computing device) of the lymphoedema recognition method applied to breast cancer patients after surgery can construct a three-dimensional model of the upper limb according to the acquired sequence of upper limb images. Among them, the sequence of upper limb images can be multiple images of the upper limb of the object to be recognized acquired by a mobile device (e.g., a mobile phone) with an image acquisition function. Specifically, during the upper limb image acquisition process, the object to be detected can place the arm horizontally on a horizontal plane in a relaxed state for upper limb image acquisition. Among them, the object to be detected flips the palm and rotates the upper limb to acquire upper limb images at multiple angles. In this way, there is no need to use special equipment for image acquisition, thereby reducing the difficulty of image acquisition for patients. The object to be recognized can be a patient with breast cancer who is to be recognized for lymphoedema after surgery. The three-dimensional model of the upper limb is a local three-dimensional model corresponding to at least three upper limb measurement regions acquired from multiple angles for the object to be recognized. In practice, the upper limb consists of six parts: the shoulder, upper arm, elbow, forearm, wrist, and hand. Among them, lymphoedema mainly occurs in the upper arm, forearm, and hand. Therefore, at least three measurement regions can be set to reduce the amount of data processing during three-dimensional model construction. In particular, for the upper arm and forearm, local regions in the upper arm and forearm (e.g., a region with a width of 5 cm in the middle of the upper arm and a region with a width of 5 cm in the middle of the forearm) can be used as upper limb measurement regions, thereby further reducing the amount of data processing during three-dimensional model construction.

[0021] As an example, refer to Figure 2 the schematic diagram of the positions of the upper limb measurement regions shown, where Figure 2 three upper limb measurement regions are shown, namely the upper limb measurement region corresponding to the hand, the upper limb measurement region corresponding to the local region in the forearm, and the upper limb measurement region corresponding to the local region in the upper arm. By performing three-dimensional modeling only on the upper limb measurement regions, the amount of data processing can be effectively reduced.

[0022] As another example, the upper limb images in the upper limb image sequence may include depth of field information. Therefore, first, the above-mentioned execution entity may crop a local image corresponding to the upper limb measurement area in the upper limb image. Secondly, perform edge feature point matching on at least three obtained local images. Then, then, through the RANSAC (RANdom SApmle Consensus) algorithm, according to the matching results of the edge feature points, estimate the relative position and orientation of the local image in at least one obtained local image relative to other local images. Next, in combination with the depth of field information included in the upper limb image, convert the pixel points in the local image from the two-dimensional image coordinates in the image coordinate system to the three-dimensional coordinates in the geodetic coordinate system. Further, unify the coordinate systems of the three-dimensional coordinates in the obtained three-dimensional coordinate set (for example, by means of three-dimensional point cloud registration). Immediately afterwards, perform surface reconstruction and texture mapping in combination with the three-dimensional coordinate set after coordinate system unification to obtain the above-mentioned local upper limb model.

[0023] It should be noted that the above-mentioned computing device may be hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or can be implemented as a single server or a single terminal device. When the computing device is embodied as software, it can be installed in the above-listed hardware devices. It can be implemented as, for example, multiple software or software modules for providing distributed services, or can be implemented as a single software or software module. No specific limitation is made here.

[0024] In some optional implementation manners of some embodiments, the above-mentioned execution entity constructs an upper limb three-dimensional model according to the acquired upper limb image sequence, including: In the first step, for each upper limb image in the above-mentioned upper limb image sequence, perform the following first processing steps: The first sub-step: Divide the above-mentioned upper limb image along the color channels to obtain a first channel image, a second channel image, and a third channel image.

[0025] Among them, the first channel image is the upper limb image under the blue channel. The above-mentioned second channel image is the upper limb image under the green channel. The above-mentioned third channel image is the upper limb image under the red channel.

[0026] As an example, the code for dividing the image along the color channels may be as follows: import cv2 image = cv2.imread("Storage path of the upper limb image") image_rgb = cv2.cvtColor(img,cv2.COLOR_BGR2RGB) r_channel, g_channel, b_channel = cv2.split(image_rgb) Among them, r_channel corresponds to the third-channel image. g_channel corresponds to the second-channel image. b_channel corresponds to the first-channel image.

[0027] As another example, refer to Figure 3 The comparison diagram of the upper limb image, the first-channel image, the second-channel image, and the third-channel image shown. Among them, it is analyzed that compared with the upper limb image, the first-channel image has a more obvious boundary between the upper limb and the background compared to the third-channel image and the second-channel image. Analyzing the reason, taking the yellow race as an example, their skin color presents yellow, red, pink, white, etc. When the background color of the upper limb image is similar to the skin color of the upper limb, it will cause the boundary between the upper limb and the background in the third-channel image to be unclear, which is not conducive to the subsequent extraction of key points on the upper limb boundary. From the perspective of the distribution range of colors on the color wheel, green is adjacent to yellow. Therefore, when the background color of the upper limb image is similar to the skin color of the upper limb, the boundary between the upper limb and the background in the second-channel image is relatively blurred compared to the boundary in the first-channel image and relatively clear compared to the boundary in the third-channel image. Therefore, the first-channel image is selected as the benchmark. At the same time, for the processing of single-channel images, compared with processing the upper limb image in the RGB (Red-Green-Blue) format, the data processing volume can also be effectively reduced.

[0028] Second sub-step: Divide the above first-channel image into a set of image patches.

[0029] Among them, the image patches in the above set of image patches have the same corresponding image patch size. In practice, the first-channel image can be divided into a set of image patches by using a uniform division method.

[0030] Third sub-step: For each image patch in the above set of image patches, perform the following second processing step: Step 1: Identify the type of the image patch for the above image patch to generate an image patch type.

[0031] Among them, the image patch type represents the part type of the part of the upper limb corresponding to the image patch.

[0032] In practice, a lightweight classification model can be constructed to identify the type of the image patch. The classification model can be composed of 5 serially connected convolutional layers and 1 fully connected layer to classify the image patch type. The classification model can use the image patches marked with the image patch type as training samples and sample labels and be trained by a supervised training method.

[0033] As an example, refer to Figure 4Schematic diagram of the generation process of the set of image patches corresponding to the first-channel image shown. The execution entity can divide the first-channel image into 2×5, a total of 10 image patches as the set of image patches. The relative positions among the 10 image patches are fixed. The advantage of block processing is that when the key points corresponding to the first key point information are not extracted within the image patches corresponding to the first-channel image, the key point extraction for the image patches corresponding to the second-channel image and the third-channel image can be skipped to reduce the data processing volume.

[0034] Step 2: In response to the type of the image patch being consistent with the part type corresponding to any one of the at least three upper limb measurement regions, perform key point recognition on the image patch to generate the first key point information corresponding to the image patch.

[0035] The first key point information includes: key point coordinates, image patch index, and key point feature vector. In practice, the first key point information can be obtained by edge detection, for example, by using the Canny operator to perform edge detection on the image patch. The key point coordinates represent the image coordinates of the edge key points corresponding to the first key point information. The key point feature vector represents the feature vector of the edge key points corresponding to the first key point information, such as being represented by a 256-dimensional feature vector. The image patch index represents the position of the image patch in the set of image patches.

[0036] Step 3: Filter the obtained first key point information group according to the second-channel image and the third-channel image to obtain the key point information group.

[0037] Optionally, filtering the obtained first key point information group according to the second-channel image and the third-channel image to obtain the key point information group includes: For each first key point information in the first key point information group, perform the following third processing step: S1: Perform key point recognition on the image patch in the second-channel image corresponding to the image patch index included in the first key point information to obtain the second key point information.

[0038] The second key point information includes: key point coordinates and key point feature vector.

[0039] In practice, edge detection can be performed, for example, using the Canny operator to detect the edges of the image block corresponding to the image block index included in the above second-channel image in the above first key point information, so as to obtain the second key point information. Since only the color channels of the first-channel image, the second-channel image, and the third-channel image are different, it is possible to locate the image blocks in different-channel images through the image block index for further key point recognition, without having to process the entire image, greatly reducing the amount of data processing.

[0040] S2: In response to the coordinate distance between the key point coordinates included in the above first key point information and the key point coordinates included in the above second key point information being less than a preset coordinate distance, determine the feature similarity between the key point feature vector included in the above first key point information and the key point feature vector included in the above second key point information, to obtain a first feature similarity.

[0041] In practice, since the key point feature vector is a high-dimensional feature vector, directly calculating the similarity between vectors involves a large amount of data processing. In particular, when there are multiple key points in the image block, the number of similarities to be calculated is even more. Therefore, comparing the distances between the coordinates first and then calculating the similarity can effectively reduce the amount of data processing. Specifically, the cosine similarity calculation method can be used to determine the feature similarity between the key point feature vector included in the above first key point information and the key point feature vector included in the above second key point information, to obtain a first feature similarity.

[0042] S3: In response to the above first feature similarity being greater than or equal to a preset feature similarity, determine the above first key point information as the key point information in the above key point information group.

[0043] S4: In response to the above first feature similarity being less than a preset feature similarity, perform key point recognition on the image block corresponding to the image block index included in the above first key point information in the above third-channel image, to obtain third key point information.

[0044] Among them, the third key point information includes: key point coordinates, key point feature vector.

[0045] In practice, the key point recognition method in "S2" can be adopted to perform key point recognition on the image block in the third-channel image corresponding to the image block index included in the first key point information, so as to obtain the third key point information. Details are not described herein again. Specifically, since the first-channel image and the second-channel image are clearer in boundary than the third-channel image, most of the matching key points can be determined by calculating the feature point similarity between the second-channel image and the third-channel image. For those that do not match, the third-channel image is combined for supplementation. Compared with the method of calculating the similarity twice for the key points corresponding to the first key point information, the key points corresponding to the second key point information, and the key points corresponding to the third key point information, when the key points corresponding to the first key point information match the key points corresponding to the second key point information, one feature matching can be reduced, further reducing the data processing volume.

[0046] S5: In response to the coordinate distance between the key point coordinates included in the first key point information and the key point coordinates included in the third key point information being less than the preset coordinate distance, determine the feature similarity between the key point feature vector included in the first key point information and the key point feature vector included in the third key point information, so as to obtain the second feature similarity.

[0047] Specifically, the cosine similarity calculation method can be adopted to determine the feature similarity between the key point feature vector included in the first key point information and the key point feature vector included in the third key point information, so as to obtain the second feature similarity.

[0048] S6: In response to the second feature similarity being greater than or equal to the preset feature similarity, determine the first key point information as the key point information in the key point information group.

[0049] Step 4: Construct an upper limb three-dimensional model according to the obtained set of key point information groups to obtain the upper limb three-dimensional model.

[0050] In practice, each key point information group in the set of key point information groups corresponds to an upper limb image. First, the key points that match in the upper limb images collected at different angles can be determined by means of feature point matching. Since the key point information includes key point feature vectors, the key points that match in different upper limb images can be determined by calculating vector similarity. Secondly, according to the obtained multiple pairs of matching key point information, the essential matrix is estimated. Specifically, the fundamental matrix can be estimated by combining multiple pairs of matching key point information, and the fundamental matrix can be converted into the essential matrix by combining the internal parameters of the camera. Then, according to the essential matrix, the rotation vector and the translation matrix are determined. Further, by combining the rotation vector and the translation matrix, the matching key points are mapped to the three-dimensional space as point cloud data through the triangulation method. Finally, a three-dimensional model of the upper limb is reconstructed according to the point cloud data to obtain a three-dimensional model of the upper limb.

[0051] Step 102: Perform upper limb state recognition based on the upper limb image sequence and the three-dimensional model of the upper limb to generate upper limb state information.

[0052] In some embodiments, the above-mentioned execution entity can perform upper limb state recognition based on the upper limb image sequence and the three-dimensional model of the upper limb to generate upper limb state information. Among them, the above-mentioned upper limb state information includes: upper limb dimension features and upper limb skin features. The above-mentioned upper limb dimension features include: upper limb arm circumference. The above-mentioned upper limb skin features include: upper limb color, allergy features, and scar features. The allergy features represent the type of allergy of the upper limb skin. The scar features represent the type of scar of the upper limb skin.

[0053] As an example, the above-mentioned execution entity can combine the three-dimensional model of the upper limb to determine the above-mentioned upper limb dimension features. Specifically, since the three-dimensional model of the upper limb is a local three-dimensional model corresponding to at least three upper limb measurement regions, the upper limb arm circumference can be obtained by combining the model circumference of the three-dimensional model of the upper limb in the upper limb measurement region. In particular, since there is a proportional relationship between the three-dimensional model of the upper limb and the actual upper limb size during modeling, it is necessary to map the upper limb arm circumference obtained through the three-dimensional model of the upper limb by combining the upper limb size of the object to be detected when normal as the upper limb arm circumference. Since it includes at least three measurement regions, at least three upper limb arm circumferences can be obtained.

[0054] As another example, the above-mentioned execution entity can combine the upper limb image sequence and obtain the above-mentioned upper limb skin features through image recognition. Specifically, a recurrent neural network model can be used to perform image recognition on the upper limb image to obtain upper limb skin features. In particular, since the upper limb skin features include upper limb color, allergy features, and scar features. Therefore, 3 classifiers need to be connected based on the recurrent neural network model to output the upper limb color, allergy features, and scar features.

[0055] In some alternative implementations of some embodiments, the above-mentioned execution entity performs upper limb state recognition based on the above-mentioned upper limb image sequence and the above-mentioned upper limb three-dimensional model to generate upper limb state information, including: First step, for each upper limb image in the above-mentioned upper limb image sequence, perform the following background segmentation steps: First sub-step, construct a segmentation region according to the key point information group corresponding to the above-mentioned upper limb image.

[0056] In practice, since the key points corresponding to the key point information group are the edge points of the upper limb, therefore, the closed region surrounded by the key points corresponding to the key point information group can be used as the segmentation region.

[0057] Second sub-step, perform image background stripping on the above-mentioned upper limb image according to the above-mentioned segmentation region to obtain the upper limb image after background stripping.

[0058] In practice, the above-mentioned execution entity can set the pixel values outside the segmentation region to 0 to obtain the upper limb image after background stripping. In this way, it can be avoided that the images of non-upper limb parts participate in subsequent image processing and the data processing amount can be reduced. At the same time, compared with the segmentation method, the method of setting to 0 can ensure that the image sizes of the upper limb images after background stripping are consistent.

[0059] Third sub-step, perform image feature extraction on the above-mentioned upper limb image after background stripping through the image feature extraction network included in the pre-trained upper limb skin feature extraction model to generate an image feature map.

[0060] In practice, the image feature extraction network can adopt the FPN (Feature Pyramid Network) network as the backbone network to perform image feature extraction on the upper limb image after background stripping from different receptive fields.

[0061] Fourth sub-step, generate a candidate upper limb color through the upper limb color classifier included in the above-mentioned upper limb skin feature extraction model and the above-mentioned image feature map.

[0062] The upper limb color classifier is a multi-classifier to adjust the feature dimension of the image feature map and finally output a 1×N feature vector. Among them, N represents the color categories that the upper limb color classifier can classify. In practice, in order to avoid the problem of difficult collection of training samples in refining skin colors, and since skin colors mainly focus on yellow, red, pink, white, and black, therefore, N can take the value of 6. Among them, "6" corresponds to yellow, red, pink, white, black, and other colors.

[0063] Fifth sub-step, generate a candidate allergy feature and a candidate scar feature through the allergy and scar position recognition network included in the above-mentioned upper limb skin feature extraction model and the above-mentioned image feature map.

[0064] In practice, in order to improve the recognition speed, the allergy and scar location recognition network can adopt ResNet-50 as the backbone network and connect two locators to generate candidate allergy features and candidate scar features respectively.

[0065] In the second step, vote screening is respectively performed on the obtained candidate upper limb color set, candidate allergy feature set, and candidate scar feature set to obtain the upper limb color included in the above-mentioned upper limb state information.

[0066] In the third step, according to the regional cross-sectional perimeters of the above-mentioned upper limb three-dimensional model in the above-mentioned at least three upper limb measurement regions, the upper limb arm circumference included in the above-mentioned upper limb state information is determined.

[0067] In practice, since there are at least three upper limb measurement regions, at least three upper limb arm circumferences can be obtained.

[0068] Step 103: Generate lymphedema recognition information according to the upper limb state information, the collected chief complaint information, the breast cancer surgery treatment information corresponding to the object to be recognized, and the pre-trained lymphedema recognition model.

[0069] In some embodiments, the above-mentioned execution subject can generate lymphedema recognition information according to the upper limb state information, the collected chief complaint information, the breast cancer surgery treatment information corresponding to the object to be recognized, and the pre-trained lymphedema recognition model. Among them, the chief complaint information is a description of the upper limb state provided by the above-mentioned object to be recognized. The breast cancer surgery treatment information includes the surgical site, the type of surgical incision, and radiotherapy information. In practice, since the object to be recognized is a patient who has undergone breast cancer surgery, there is a corresponding breast cancer treatment record, which is used as the breast cancer postoperative treatment information. Specifically, the surgical site includes: the lower inner chest, the upper inner chest, the areola, the lower outer chest, and the upper outer chest. The surgical incisions include: transverse incision, longitudinal incision, and oblique incision. The radiotherapy information includes: radiotherapy identification, radiotherapy location. The radiotherapy identification indicates whether the object to be recognized has undergone radiotherapy. The radiotherapy location indicates the location of radiotherapy. The radiotherapy location includes: breast, chest wall, lymph node region. The lymphedema recognition model can adopt a pre-trained model (for example, the GPT model) as the backbone network to generate lymphedema recognition information.

[0070] Optionally, the lymphedema recognition information includes: edema severity, edema type, and edema confidence. In practice, the edema severity characterizes the severity of the lymphedema of the object to be recognized. For example, the edema severity can include: Class A, Class B, and Class C. Among them, Class A, Class B, and Class C represent different severities of lymphedema. Class A corresponds to the highest severity. Class C corresponds to the lowest severity. The edema type characterizes the location where lymphedema occurs. The edema confidence characterizes the comprehensive confidence of the edema severity and the edema type. Specifically, the edema confidence is obtained by weighted summation of the confidence corresponding to the edema severity and the confidence corresponding to the edema type.

[0071] Optionally, the lymphedema recognition model includes: a chief complaint information encoder, a chief complaint information decoder, an input layer, K hidden layers, and an output layer. Among them, the input layer includes 8 neurons, and the output layer includes 2 neurons. The lymphedema recognition model and the upper limb skin feature extraction model as a whole are trained through a supervised training method. In practice, the chief complaint information encoder and the chief complaint information decoder adopt an encoder and a decoder based on the Transformer structure. The structures of the chief complaint information encoder and the chief complaint information decoder are symmetric. The input layer takes the output of the chief complaint information decoder, the upper limb arm circumference, the upper limb color, the allergy feature, the scar feature, the surgical site, the surgical incision type, and the radiotherapy information as 8 inputs. 8 ≥ K ≥ 5. This avoids the problem of excessive computational complexity caused by too deep hidden layers. The 2 neurons included in the output layer respectively output the edema severity and the edema type. Among them, the edema severity and the edema type respectively correspond to the corresponding confidences. The edema confidence is obtained by weighted summation of the confidence corresponding to the edema severity and the confidence corresponding to the edema type respectively.

[0072] In some optional implementation manners of some embodiments, the execution subject generates lymphedema recognition information according to the above-mentioned upper limb state information, the collected chief complaint information, the breast cancer surgery treatment information corresponding to the object to be recognized, and the pre-trained lymphedema recognition model, including: First step, encode the chief complaint information through the above-mentioned chief complaint information encoder to obtain chief complaint information features.

[0073] Second step, decode the chief complaint information features through the above-mentioned chief complaint information decoder to obtain chief complaint description features.

[0074] Third step, generate lymphedema recognition information through the above-mentioned input layer, the above-mentioned K hidden layers, and the above-mentioned output layer according to the above-mentioned chief complaint description features, the above-mentioned upper limb state information, and the above-mentioned breast cancer surgery treatment information.

[0075] Step 104, in response to the lymphedema recognition information indicating that the object to be recognized has lymphedema, generate rehabilitation advice information for the object to be recognized according to the lymphedema recognition information.

[0076] In some embodiments, the above-mentioned execution entity may, in response to the lymphedema recognition information indicating that the object to be recognized has lymphedema, generate rehabilitation advice information for the object to be recognized according to the lymphedema recognition information. Among them, the rehabilitation advice information is the rehabilitation advice for lymphedema of different severities. In practice, for mild lymphedema, since it is discovered in time, self-training rehabilitation can be used to relieve lymphedema. For severe lymphedema, medical treatment can be used to treat lymphedema.

[0077] In some optional implementation manners of some embodiments, the above-mentioned execution entity, in response to the above-mentioned lymphedema recognition information indicating that the above-mentioned object to be recognized has lymphedema, generates rehabilitation advice information for the above-mentioned object to be recognized according to the above-mentioned lymphedema recognition information, including: In response to the above-mentioned edema confidence level being greater than the preset edema confidence level, determine the above-mentioned rehabilitation advice information through a pre-constructed advice information decision tree according to the edema severity and edema type included in the above-mentioned lymphedema recognition information. In practice, an advice information decision tree can be constructed according to different edema severities and types, as well as the corresponding rehabilitation advice information, so as to quickly locate the rehabilitation advice information. For example, for severe lymphedema, medical treatment by surgery can be used, such as connecting lymphatic vessels to veins so that lymph fluid returns through the veins to relieve lymphedema. The method of inhibiting lymph nodes can also be used to stimulate the construction of new lymphatic pathways to relieve lymphedema. The method of removing redundant tissues or liposuction can also be used to relieve the swelling of the upper limb caused by lymphedema. For mild lymphedema, the methods of avoiding hot baths or excessive warming can be used to avoid lymph fluid accumulation. Sitting for a long time should also be avoided as much as possible to avoid lymph fluid accumulation. In addition, wearing a lymphedema compression sleeve can be recommended to control lymphedema.

[0078] In some optional implementation manners of some embodiments, the above method further includes: First step, generate rehabilitation training information for the object to be recognized according to the above-mentioned rehabilitation advice information and the object portrait corresponding to the object to be recognized.

[0079] Among them, the above rehabilitation training information includes: training action type, action frequency, and rehabilitation duration. In practice, different rehabilitation actions can be preset for different rehabilitation advice information to relieve lymphedema. At the same time, the basic information of different objects to be recognized, such as age, is different. Therefore, the action frequency, rehabilitation duration, and training action type can be adjusted in combination with the object portrait, so that the rehabilitation training information conforms to the physical state of the object to be recognized.

[0080] In the second step, in response to the above object to be recognized starting a rehabilitation training task, collect the real-time action information corresponding to the above object to be recognized.

[0081] In practice, the real-time action information corresponding to the above object to be recognized can be collected in real time through a mobile device (for example, a mobile phone). For example, by orienting the mobile device towards the object to be recognized, the action execution state of the object to be recognized can be collected and recognized in real time to obtain real-time action information.

[0082] In the third step, determine the action matching degree according to the above rehabilitation training information and the above real-time action information.

[0083] In practice, the above execution entity can extract the action type corresponding to the real-time action information by means of key point extraction, and count the execution frequency and duration of the actions of the object to be recognized, and compare them with the training action type, action frequency, and rehabilitation duration included in the rehabilitation training information, so as to determine the action matching degree.

[0084] In the fourth step, in response to the action matching degree being less than the preset matching degree, initiate an action error correction reminder to the above object to be recognized.

[0085] In practice, the above execution entity can initiate an action error correction reminder to the object to be recognized through a mobile device, so as to ensure that the object to be recognized executes the training actions corresponding to the rehabilitation training information at a relatively high matching degree, so as to achieve a better rehabilitation effect.

[0086] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: By applying the lymphoedema recognition method for breast cancer patients after surgery in some embodiments of the present disclosure, patients after breast cancer surgery can detect the lymphoedema that occurs to themselves as early as possible, so that they can carry out prevention and treatment as early as possible when lymphoedema occurs. Specifically, since lymphoedema is a common chronic complication after breast cancer surgery and may occur during the months to years after surgery, it is necessary for patients to spontaneously determine whether they have lymphoedema. Based on this, in the lymphoedema recognition method for breast cancer patients after surgery in some embodiments of the present disclosure, first, an upper limb three-dimensional model is constructed according to the acquired upper limb image sequence, wherein the above-mentioned upper limb three-dimensional model is a local three-dimensional model corresponding to at least three upper limb measurement regions collected from multiple angles for the object to be recognized. By combining the upper limb images to construct a local three-dimensional model (upper limb three-dimensional model) corresponding to at least three upper limb measurement regions, the upper limb corresponding to the object to be recognized can be effectively modeled. Secondly, the upper limb state is recognized according to the above-mentioned upper limb image sequence and the above-mentioned upper limb three-dimensional model to generate upper limb state information, wherein the above-mentioned upper limb state information includes: upper limb dimension characteristics, upper limb skin characteristics, the above-mentioned upper limb dimension characteristics include: upper limb arm circumference, and the above-mentioned upper limb skin characteristics include: upper limb color, allergy characteristics and scar characteristics. In this way, by combining the upper limb images and the upper limb three-dimensional model, the upper limb dimension characteristics and upper limb skin characteristics of the object to be recognized are recognized. Then, according to the above-mentioned upper limb state information, the collected chief complaint information, the breast cancer surgery treatment information corresponding to the above-mentioned object to be recognized, and a pre-trained lymphoedema recognition model, lymphoedema recognition information is generated, wherein the above-mentioned chief complaint information is a description of the upper limb state provided by the above-mentioned object to be recognized, and the above-mentioned breast cancer surgery treatment information includes: surgical site, surgical incision type and radiotherapy information. Through the set lymphoedema recognition model, the lymphoedema recognition information is automatically generated by combining the upper limb state information, the chief complaint information, and the breast cancer surgery treatment information. In this way, patients can quickly recognize whether they have lymphoedema by collecting the upper limb image sequence of their own upper limbs without going to the hospital. Finally, in response to the above-mentioned lymphoedema recognition information indicating that the above-mentioned object to be recognized has lymphoedema, according to the above-mentioned lymphoedema recognition information, rehabilitation advice information corresponding to the above-mentioned object to be recognized is generated. In this way, rehabilitation advice that matches the type of lymphoedema is generated, so that patients can relieve lymphoedema at home by combining the rehabilitation advice when the symptoms are mild. In this way, patients after breast cancer surgery can detect the lymphoedema that occurs to themselves as early as possible, so that they can carry out prevention and treatment as early as possible when lymphoedema occurs. Further referring to Figure 5 , as an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a lymphoedema recognition device for breast cancer patients after surgery, and these device embodiments are related to Figure 1The method embodiments shown correspond to a lymphedema recognition device applied after breast cancer surgery, which can be specifically applied to various electronic devices.

[0087] As Figure 5 shown, the lymphedema recognition device 500 applied after breast cancer surgery in some embodiments includes: a construction unit 501, an upper limb state recognition unit 502, a first generation unit 503, and a second generation unit 504. Among them, the construction unit 501 is configured to construct a three-dimensional model of the upper limb according to the acquired upper limb image sequence, where the three-dimensional model of the upper limb is a local three-dimensional model corresponding to at least three upper limb measurement regions acquired from multiple angles for the object to be recognized; the upper limb state recognition unit 502 is configured to perform upper limb state recognition according to the upper limb image sequence and the three-dimensional model of the upper limb to generate upper limb state information, where the upper limb state information includes: upper limb dimension features, upper limb skin features, the upper limb dimension features include: upper limb arm circumference, and the upper limb skin features include: upper limb color, allergy features, and scar features; the first generation unit 503 is configured to generate lymphedema recognition information according to the upper limb state information, the acquired chief complaint information, the breast cancer surgery treatment information corresponding to the object to be recognized, and a pre-trained lymphedema recognition model, where the chief complaint information is a description of the upper limb state provided by the object to be recognized, and the breast cancer surgery treatment information includes: surgical site, surgical incision type, and radiotherapy information; the second generation unit 504 is configured to, in response to the lymphedema recognition information indicating that the object to be recognized has lymphedema, generate rehabilitation advice information corresponding to the object to be recognized according to the lymphedema recognition information. It can be understood that the units described in the lymphedema recognition device 500 applied after breast cancer surgery correspond to Figure 1 the respective steps in the method described in the reference. Thus, the operations, features, and beneficial effects described above for the method also apply to the lymphedema recognition device 500 applied after breast cancer surgery and the units included therein, and will not be elaborated here. Next, refer to Figure 6 , which shows a schematic structural diagram of an electronic device (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure. As Figure 6As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory. The non-volatile storage medium can store an operating system and computer programs. The computer programs include program instructions, which when executed, can cause the processor to execute any one of the front-end page monitoring methods. The processor is used to provide computing and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the operation of the computer programs in the non-volatile storage medium. When the computer programs are executed by the processor, the processor can be caused to execute any one of the front-end page monitoring methods. The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 6 the structure shown in is only a block diagram of some structures related to the solution of the present disclosure, and does not constitute a limitation on the computer device to which the solution of the present disclosure is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0088] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0089] Among them, in one embodiment, the above-mentioned processor is used to run a computer program stored in the memory to implement the following steps: constructing a three-dimensional upper limb model based on the acquired upper limb image sequence, where the above-mentioned three-dimensional upper limb model is a local three-dimensional model corresponding to at least three upper limb measurement regions acquired from multiple angles for the object to be recognized; performing upper limb state recognition based on the above-mentioned upper limb image sequence and the above-mentioned three-dimensional upper limb model to generate upper limb state information, where the above-mentioned upper limb state information includes: upper limb dimension features and upper limb skin features, the above-mentioned upper limb dimension features include: upper limb arm circumference, and the above-mentioned upper limb skin features include: upper limb color, allergy features, and scar features; generating lymphedema recognition information based on the above-mentioned upper limb state information, the acquired chief complaint information, the breast cancer surgery treatment information corresponding to the above-mentioned object to be recognized, and a pre-trained lymphedema recognition model, where the above-mentioned chief complaint information is a description of the upper limb state provided by the above-mentioned object to be recognized, and the above-mentioned breast cancer surgery treatment information includes: surgical site, surgical incision type, and radiotherapy information; in response to the above-mentioned lymphedema recognition information indicating that the above-mentioned object to be recognized has lymphedema, generating rehabilitation advice information corresponding to the above-mentioned object to be recognized based on the above-mentioned lymphedema recognition information.

[0090] The embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored, and the computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of the lymphedema recognition method for breast cancer patients after surgery in the present disclosure.

[0091] Among them, the above-mentioned computer-readable storage medium can be an internal storage unit of the aforementioned computer device, such as the hard disk or memory of the above-mentioned computer device. The above-mentioned computer-readable storage medium can also be an external storage device of the above-mentioned computer device, such as a plug-in hard disk equipped on the above-mentioned computer device, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0092] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. Without further limitations, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0093] The above description is only some preferred embodiments of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the embodiments of the present disclosure that have similar functions.

Claims

1. A method for identifying lymphedema after breast cancer surgery, characterized in that: include: Constructing an upper limb three-dimensional model according to the acquired upper limb image sequence, wherein the upper limb three-dimensional model is a local three-dimensional model corresponding to at least three upper limb measurement areas acquired at multiple angles for the object to be identified; Perform upper limb state recognition according to the upper limb image sequence and the upper limb three-dimensional model to generate upper limb state information, wherein the upper limb state information includes: upper limb dimensional features and upper limb skin features, the upper limb dimensional features include: upper limb arm circumference, and the upper limb skin features include: upper limb color, allergy features, and scar features; Generate lymphedema recognition information according to the upper limb status information, the collected chief complaint information, the breast cancer surgical treatment information corresponding to the object to be identified, and the pre-trained lymphedema recognition model, wherein the chief complaint information is a description of the upper limb status provided by the object to be identified, and the breast cancer surgical treatment information includes: surgical site, surgical incision type, and radiotherapy information; In response to the lymphedema identification information indicating that the subject to be identified has lymphedema, rehabilitation suggestion information corresponding to the subject to be identified is generated according to the lymphedema identification information.

2. The method according to claim 1, characterized in that The lymphedema identification information includes: edema severity, edema type, and edema confidence; and In response to the lymphedema identification information indicating that the subject to be identified has lymphedema, generating rehabilitation suggestion information corresponding to the subject to be identified according to the lymphedema identification information, including: In response to the edema confidence being greater than a preset edema confidence, the rehabilitation advice information is determined according to the edema severity and edema type included in the lymphedema identification information through a pre-constructed advice information decision tree.

3. The method according to claim 2, characterized in that The method further comprises: Generate rehabilitation training information corresponding to the object to be identified according to the rehabilitation suggestion information and the object portrait corresponding to the object to be identified, wherein the rehabilitation training information includes: training action type, action frequency, and rehabilitation duration; In response to the subject to be identified starting a rehabilitation training task, collecting real-time action information corresponding to the subject to be identified; Determining a motion matching degree according to the rehabilitation training information and the real-time motion information; In response to the action matching degree being less than a preset matching degree, an action error correction reminder is initiated to the object to be identified.

4. The method according to claim 3, characterized in that The step of constructing the upper limb three-dimensional model according to the acquired upper limb image sequence comprises: For each upper limb image in the upper limb image sequence, the following first processing step is performed: Divide the upper limb image along the color channel to obtain a first channel image, a second channel image, and a third channel image, wherein the first channel image is an upper limb image under a blue channel, the second channel image is an upper limb image under a green channel, and the third channel image is an upper limb image under a red channel; Dividing the first channel image into an image block set, wherein image blocks in the image block set have the same image block size; For each image block in the set of image blocks, the following second processing step is performed: Performing image block type recognition on the image block to generate an image block type, wherein the image block type represents the part type of the upper limb part corresponding to the image block; In response to the image block type being consistent with the part type corresponding to any upper limb measurement area of ​​the at least three upper limb measurement areas, key point recognition is performed on the image block to generate first key point information corresponding to the image block, wherein the first key point information includes: key point coordinates, image block index, and key point feature vector; According to the second channel image and the third channel image, filtering the obtained first key point information group to obtain a key point information group; The three-dimensional model of the upper limb is constructed according to the obtained key point information group set to obtain the three-dimensional model of the upper limb.

5. The method according to claim 4, characterized in that The step of filtering the obtained first key point information group according to the second channel image and the third channel image to obtain the key point information group includes: For each first key point information in the first key point information group, the following third processing step is performed: Performing key point recognition on an image block in the second channel image that corresponds to the image block index included in the first key point information to obtain second key point information, wherein the second key point information includes: key point coordinates and key point feature vectors; In response to the coordinate distance between the key point coordinates included in the first key point information and the key point coordinates included in the second key point information being less than a preset coordinate distance, determining a feature similarity between a key point feature vector included in the first key point information and a key point feature vector included in the second key point information to obtain a first feature similarity; In response to the first feature similarity being greater than or equal to a preset feature similarity, determining the first key point information as key point information in the key point information group; In response to the first feature similarity being less than a preset feature similarity, performing key point recognition on an image block in the third channel image corresponding to the image block index included in the first key point information to obtain third key point information, wherein the third key point information includes: key point coordinates and key point feature vectors; In response to the coordinate distance between the key point coordinates included in the first key point information and the key point coordinates included in the third key point information being less than a preset coordinate distance, determining a feature similarity between a key point feature vector included in the first key point information and a key point feature vector included in the third key point information to obtain a second feature similarity; In response to the second feature similarity being greater than or equal to a preset feature similarity, the first key point information is determined as key point information in the key point information group.

6. The method according to claim 5, characterized in that The performing upper limb state recognition according to the upper limb image sequence and the upper limb three-dimensional model to generate upper limb state information includes: For each upper limb image in the upper limb image sequence, the following background segmentation steps are performed: Constructing a segmentation region according to the key point information group corresponding to the upper limb image; According to the segmented area, performing image background stripping on the upper limb image to obtain an upper limb image after background stripping; Performing image feature extraction on the upper limb image after background stripping by using an image feature extraction network included in a pre-trained upper limb skin feature extraction model to generate an image feature map; Generate candidate upper limb colors by using the upper limb color classifier included in the upper limb skin feature extraction model and the image feature map; Generate candidate allergy features and candidate scar features through the allergy and scar location recognition network and the image feature map included in the upper limb skin feature extraction model; Voting and screening the obtained candidate upper limb color set, candidate allergy feature set and candidate scar feature set respectively to obtain the upper limb color included in the upper limb state information; The upper limb arm circumference included in the upper limb status information is determined according to the regional cross-sectional perimeter of the upper limb three-dimensional model in the at least three upper limb measurement areas.

7. The method according to claim 6, characterized in that The lymphedema recognition model includes: a chief complaint information encoder, a chief complaint information decoder, an input layer, K hidden layers, and an output layer, wherein the input layer includes: 8 neurons, the output layer includes: 2 neurons, and the lymphedema recognition model and the upper limb skin feature extraction model are trained as a whole through a supervised training method; and The generating of lymphedema identification information according to the upper limb status information, the collected chief complaint information, the breast cancer treatment information corresponding to the object to be identified and the pre-trained lymphedema identification model includes: Performing text encoding on the chief complaint information by the chief complaint information encoder to obtain the chief complaint information features; Decoding the features of the chief complaint information by the chief complaint information decoder to obtain the features of the chief complaint description; According to the chief complaint description features, the upper limb status information and the breast cancer surgery treatment information, lymphedema identification information is generated through the input layer, the K hidden layers and the output layer.

8. A lymphedema identification device for breast cancer surgery, characterized in that: include: A construction unit is configured to construct an upper limb three-dimensional model according to the acquired upper limb image sequence, wherein the upper limb three-dimensional model is a local three-dimensional model corresponding to at least three upper limb measurement areas acquired at multiple angles for the object to be identified; an upper limb state recognition unit, configured to perform upper limb state recognition according to the upper limb image sequence and the upper limb three-dimensional model to generate upper limb state information, wherein the upper limb state information includes: upper limb dimension features and upper limb skin features, the upper limb dimension features include: upper limb arm circumference, and the upper limb skin features include: upper limb color, allergy features, and scar features; a first generating unit, configured to generate lymphedema identification information according to the upper limb status information, the collected chief complaint information, the breast cancer surgical treatment information corresponding to the object to be identified, and a pre-trained lymphedema identification model, wherein the chief complaint information is a description of the upper limb status provided by the object to be identified, and the breast cancer surgical treatment information includes: surgical site, surgical incision type, and radiotherapy information; The second generating unit is configured to generate rehabilitation suggestion information corresponding to the object to be identified in response to the lymphedema identification information indicating that the object to be identified has lymphedema, and according to the lymphedema identification information.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable medium, characterized in that A computer program is stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images