Axillary lymph node status feature detection method, device, equipment and storage medium

Abstract

The present application discloses a method, device, equipment and storage medium for detecting the state characteristics of axillary lymph nodes. The present application obtains multiple breast magnetic resonance images of the detection object, splices the multiple breast magnetic resonance images in the channel dimension to obtain a target detection image; divides the target detection image into multiple sub-images, detects the primary lesion area and the axillary area in each sub-image, and determines the primary lesion area image and the axillary area image in the target detection image according to each primary lesion area and each axillary area; according to the code corresponding to the primary lesion area image and the code corresponding to the axillary area image, jointly detects the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes. Through the above technical means, it is realized to identify the state characteristics of the patient's axillary lymph nodes before surgery by non-invasive methods, ensuring that the patient's life is not affected by the axillary lymph node examination.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a method, device, equipment and storage medium for detecting state characteristics of axillary lymph nodes. Background Art

[0002] Breast cancer is one of the most common malignant tumors in women worldwide. The status of axillary lymph nodes (ALN) is one of the most important prognostic factors for breast cancer patients. Therefore, accurate assessment of axillary lymph node status is crucial for the choice of treatment and prognosis of breast cancer patients.

[0003] In existing technologies, the sentinel lymph node (SLN) is the first stop for the primary tumor to metastasize to the axilla. Sentinel lymph node biopsy (SLNB) is generally used to evaluate the status of the sentinel lymph nodes and determine whether further traumatic axillary lymph node dissection (ALND) is needed. Both SLNB and ALND are invasive examinations, which can cause postoperative complications such as upper limb edema, neurosensory disorders, and limited movement, which seriously affect the patient's life. Therefore, there is an urgent need for a non-invasive method to identify a series of status characteristics of the patient's axillary lymph nodes before surgery. Summary of the invention

[0004] The present application provides a method, apparatus, device and storage medium for detecting the status characteristics of axillary lymph nodes, which detect the metastasis probability and number of sentinel lymph nodes in axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in axillary lymph nodes based on breast magnetic resonance images, thereby realizing the preoperative identification of the status characteristics of axillary lymph nodes of patients by non-invasive methods, solving the problem that the prior art requires invasive examination to detect the status characteristics of axillary lymph nodes, avoiding postoperative complications such as upper limb edema, neurosensory disorders and movement limitation caused by invasive examination, and ensuring that the patient's life is not affected by axillary lymph node examination.

[0005] In a first aspect, the present application provides a method for detecting state characteristics of axillary lymph nodes, comprising:

[0006] Acquire multiple breast nuclear magnetic resonance images of the detection object, and splice the multiple breast nuclear magnetic resonance images in the channel dimension to obtain a target detection image;

[0007] Dividing the target detection image into a plurality of sub-images, detecting a primary lesion region and an axillary region in each of the sub-images, and respectively determining a primary lesion region image and an axillary region image in the target detection image according to each of the primary lesion regions and each of the axillary regions;

[0008] According to the codes corresponding to the primary lesion region image and the codes corresponding to the axillary region image, the metastasis probability and the number of metastases of the sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of the non-sentinel lymph nodes in the axillary lymph nodes are jointly detected.

[0009] In a second aspect, the present application provides a device for detecting state characteristics of axillary lymph nodes, comprising:

[0010] An image stitching module is configured to acquire multiple breast nuclear magnetic resonance images of a detection object, and stitch the multiple breast nuclear magnetic resonance images in a channel dimension to obtain a target detection image;

[0011] a regional image detection module, configured to divide the target detection image into a plurality of sub-images, detect a primary lesion region and an axillary region in each of the sub-images, and respectively determine a primary lesion region image and an axillary region image in the target detection image according to each of the primary lesion regions and each of the axillary regions;

[0012] The state feature detection module is configured to jointly detect the metastasis probability and the number of metastases of the sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of the non-sentinel lymph nodes in the axillary lymph nodes according to the codes corresponding to the primary lesion area image and the codes corresponding to the axillary area image.

[0013] In a third aspect, the present application provides a device for detecting state characteristics of axillary lymph nodes, comprising:

[0014] One or more processors; a storage device storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method for detecting the status characteristics of axillary lymph nodes as described in the first aspect.

[0015] In a fourth aspect, the present application provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to perform the method for detecting status characteristics of axillary lymph nodes as described in the first aspect.

[0016] In the present application, multiple breast MRI images of the detection object are acquired, and the multiple breast MRI images are spliced ​​in the channel dimension to obtain a target detection image; the target detection image is divided into multiple sub-images, and the primary lesion area and the axillary area in each sub-image are detected, and the primary lesion area image and the axillary area image are respectively determined in the target detection image according to each primary lesion area and each axillary area; according to the code corresponding to the primary lesion area image and the code corresponding to the axillary area image, the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes are jointly detected. Through the above technical means, it is realized to identify the status characteristics of the patient's axillary lymph nodes before surgery by non-invasive methods, solve the problem that the prior art needs to detect the status characteristics of the axillary lymph nodes through invasive examination, avoid patients from having postoperative complications such as upper limb edema, neurosensory disorders and movement restriction due to invasive examination, and ensure that the patient's life is not affected by the axillary lymph node examination. The state characteristics of axillary lymph nodes are detected by using multiple breast MRI images of axillary lymph nodes, which improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by extracting the primary lesion area and axillary area in breast MRI images, which reduces the interference of invalid information and improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by jointly detecting the primary lesion area and axillary area, which enriches the information used to detect the state characteristics of axillary lymph nodes and improves the detection accuracy of state characteristics. By detecting the metastasis probability and number of sentinel lymph nodes in axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in axillary lymph nodes, more comprehensive state characteristics of axillary lymph nodes are obtained, which provides a more accurate and important theoretical basis for the diagnosis of benign and malignant axillary lymph nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a method for detecting state characteristics of axillary lymph nodes provided in an embodiment of the present application;

[0018] Figure 2 is a flow chart for determining the metastasis probability of a sentinel lymph node provided in an embodiment of the present application;

[0019] Figure 3 is a schematic diagram of a first code acquisition process provided in an embodiment of the present application;

[0020] Figure 4 is a schematic diagram of a second code acquisition process provided in an embodiment of the present application;

[0021] Figure 5 is a flow chart of determining the number of sentinel lymph node metastases adopted by an embodiment of the present application;

[0022] Figure 6 is a flow chart for determining the metastasis probability of non-sentinel lymph nodes provided in an embodiment of the present application;

[0023] Figure 7 is a schematic diagram of the structure of an end-to-end multi-task prediction model provided in an embodiment of the present application;

[0024] Figure 8 It is a structural schematic diagram of a device for detecting state characteristics of axillary lymph nodes provided in an embodiment of the present application;

[0025] Fig. 9 It is a structural schematic diagram of an axillary lymph node status characteristic detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. It is understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for the convenience of description, only part of the present application is shown in the accompanying drawings, but not all of the content. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The process can correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.

[0027] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0028] In the existing implementation methods, sentinel lymph node biopsy is used to evaluate the status of sentinel lymph nodes and determine whether further invasive axillary lymph node dissection is needed. Both sentinel lymph node biopsy and axillary lymph node dissection are invasive examinations, which can cause postoperative complications such as upper limb edema, neurosensory disorders, and movement restrictions, which seriously affect the patient's life.

[0029] To solve the above problems, the present embodiment provides a method for detecting the status characteristics of axillary lymph nodes, which detects the metastasis probability and number of sentinel lymph nodes in axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in axillary lymph nodes based on breast magnetic resonance images, thereby realizing the identification of the status characteristics of the patient's axillary lymph nodes before surgery through a non-invasive method, reducing unnecessary surgeries, avoiding postoperative complications such as upper limb edema, neurosensory disorders and movement limitation caused by invasive examinations, and ensuring that the patient's life is not affected by axillary lymph node examinations.

[0030] The state characteristic detection method of the axillary lymph nodes provided in this embodiment can be performed by a state characteristic detection device of the axillary lymph nodes, and the state characteristic detection device of the axillary lymph nodes can be implemented by software and / or hardware, and the state characteristic detection device of the axillary lymph nodes can be composed of two or more physical entities, or can be composed of one physical entity. For example, the state characteristic detection device of the axillary lymph nodes can be an intelligent terminal with strong processing power such as a computer and a server. Among them, the server can be implemented by an independent server or a server cluster composed of multiple servers.

[0031] The axillary lymph node status feature detection device is installed with at least one type of operating system, and the axillary lymph node status feature detection device can install at least one application based on the operating system, and the application can be an application that comes with the operating system, or an application downloaded from a third-party device or server. In this embodiment, the axillary lymph node status feature detection device is installed with at least an application that can execute the axillary lymph node status feature detection method.

[0032] For ease of understanding, this embodiment is described by taking a server as an example of a subject that executes the method for detecting the status characteristics of axillary lymph nodes.

[0033] Figure 1 A flow chart of a method for detecting the state characteristics of an axillary lymph node provided in an embodiment of the present application is given. Figure 1 The method for detecting the state characteristics of axillary lymph nodes specifically includes:

[0034] S110, acquiring multiple breast MRI images of the detection object, and splicing the multiple breast MRI images in the channel dimension to obtain a target detection image.

[0035] The detection object is a patient, and a breast MRI image is a medical image generated by using Magnetic Resonance Imaging (MRI) for breast detection. For example, a strong magnetic field and radio waves can be applied to the patient's breast and axillary area to generate multiple breast MRI images, and the status characteristics of the patient's axillary lymph nodes can be detected based on the multiple breast MRI images. Breast MRI can simultaneously perform multi-parameter and multi-directional imaging of primary breast cancer lesions and axillary lymph nodes. It can not only provide morphological information of axillary lymph nodes through T1-weighted imaging, T2-weighted imaging and T1-weighted enhanced scanning imaging, but also reflect the changes in tissue microstructure and microvascular blood perfusion characteristics at the molecular level through diffusion-weighted imaging (Diffusion Weighted Imaging, DWI), dynamic contrast enhanced-magnetic resonance imaging (Dynamic Contrast Enhanced-Magnetic Resonance Imaging, DCE-MRI) and apparent diffusion coefficient imaging (Apparent Diffusion Coefficient, ADC), thereby more objectively reflecting the microenvironment information of the lymph nodes, which is conducive to improving the detection accuracy of the characteristic status of axillary lymph nodes.

[0036] In this embodiment, a plurality of breast nuclear magnetic resonance images are generated by different imaging methods, including at least two of T1 images, T2 images, DWI images, DCE images and ADC images. Among them, T1 images, T2 images, DWI images, DCE images and ADC images are generated by T1 weighted imaging, T2 weighted imaging, diffusion weighted imaging, dynamic enhanced magnetic resonance imaging and apparent diffusion coefficient imaging, respectively. This embodiment is described by taking a plurality of breast nuclear magnetic resonance images including T2 images, DWI images, DCE images and ADC images as an example. T2 images, DWI images, DCE images and ADC images are all grayscale images, and the number of channels is one. The T2 images, DWI images, DCE images and ADC images can be spliced ​​in the channel dimension to obtain a target detection image with a channel number of four.

[0037] It should be noted that each breast MRI image includes depth information, that is, each breast MRI image is a 2.5D image. In one embodiment, the cross-section with the largest lesion can be determined based on the image information and depth information of each breast MRI image, and the corresponding cross-section slice image is obtained in each breast MRI image based on the cross-section with the largest lesion, and each cross-section slice image is spliced ​​in the channel dimension to obtain a target detection image. This embodiment obtains the cross-section slice image with the largest lesion in each breast MRI image. Since the cross-section slice image has no depth information, it can effectively reduce the amount of calculation of subsequent detection state features and improve the detection efficiency of state features.

[0038] In another embodiment, the target detection image subsequently generated loses the depth information of each breast MRI image due to the acquisition of the cross-sectional slice image. The lack of depth information will affect the detection accuracy of the state characteristics of the axillary lymph nodes. Compared with the detection efficiency, the detection accuracy is a more important indicator of the state characteristics of the axillary lymph nodes. In this regard, the present embodiment aims to splice multiple breast MRI images containing depth information in the channel dimension to obtain a target detection image containing depth information. Exemplarily, the dimensions of the T2 image, DWI image, DCE image and ADC image are H*W*D. After splicing the T2 image, DWI image, DCE image and ADC image in the channel dimension, a target detection image with a dimension of 4*H*W*D is obtained.

[0039] In this embodiment, there are situations where the same spatial point in multiple breast MRI images is not at the same pixel point, and there are situations where the intensity ranges of multiple breast MRI images are not uniform, which results in that the spatial points and intensities of multiple breast MRI images are not aligned and unified when multiple breast MRI images are spliced ​​in the channel dimension, so that the target detection image generated by splicing cannot accurately characterize the depth information and image information of the breast and axillary lymph nodes, affecting the accuracy of subsequent feature state detection. In this regard, before splicing multiple breast MRI images, multiple breast MRI images can be spatially normalized and aligned to obtain multiple breast MRI common frame images with spatial point alignment; multiple breast MRI common frame images with spatial point alignment are subjected to intensity normalization processing to obtain multiple breast MRI common frame images with intensities in the same range. Exemplarily, feature space points in multiple breast MRI images are extracted, feature space points in multiple breast MRI images are matched to determine matching feature space points, relative transformation parameters of multiple breast MRI images are determined based on multiple groups of matching feature space points, and multiple breast MRI images are transformed by relative transformation parameters to transform the same feature space point in multiple breast MRI images to the same pixel point, thereby achieving spatial point alignment of multiple breast MRI images. In the process of breast MRI image conversion, pixel values ​​at non-integer pixel points can be supplemented by an interpolation algorithm. Furthermore, according to the pixel value range of multiple breast MRI images after spatial point alignment, the pixel values ​​of each pixel point of multiple breast MRI images are adjusted to the same pixel value range, thereby achieving a unified intensity range of multiple breast MRI images. This embodiment performs spatial normalization alignment and intensity normalization processing on multiple breast MRI images to ensure that the spatial points and intensities of the multiple breast MRI images are aligned and unified, thereby ensuring that the target detection image generated by stitching multiple breast MRI images can accurately characterize the depth information and image information of the breast and axillary lymph nodes, thereby improving the accuracy of subsequent characteristic state detection of the axillary lymph nodes.

[0040] S120, dividing the target detection image into a plurality of sub-images, detecting the primary lesion region and the axillary region in each sub-image, and respectively determining the primary lesion region image and the axillary region image in the target detection image according to each primary lesion region and each axillary region.

[0041] For example, the size of the target detection image is large, and it is difficult for the subsequent encoder to directly process the target detection image. In this regard, the target detection image can be divided into multiple sub-images of uniform size, so that the sub-images can be used as inputs of the encoder. The size of the sub-image can be set according to actual needs.

[0042] In this embodiment, the target detection image can be slid and cropped by a slider of a preset size, and the image obtained by each slider cropping is used as a sub-image. When the slider traverses the entire target detection image, the target detection image can be obtained The multiple sub-images corresponding to the target detection image can be obtained. Among them, the size and step size of the slider can be set according to actual conditions. The size of the slider can be set to 64*64*64, and a slider based on this size can capture a sub-image with a dimension of 4*64*64*64. This embodiment can quickly divide the target detection image into multiple sub-images of equal size in the form of a slider, effectively improving the efficiency of obtaining sub-images.

[0043] After the target detection image is divided into a plurality of sub-images, each sub-image may be sequentially input into a preset region of interest detection model, and the primary lesion region and / or axillary region in the sub-image may be detected by the region of interest detection model. Among them, the primary lesion region and axillary region in the sub-image may be detected simultaneously by one region of interest detection model, and the region of interest detection model may be trained by sample images annotated with the primary lesion region and the axillary region. The primary lesion region in the sub-image may be detected by one region of interest detection model and the axillary region in the sub-image may be detected by another region of interest detection model, the region of interest detection model for detecting the primary lesion region may be obtained by training with sample images annotated with the primary lesion region, and the region of interest detection model for detecting the axillary region may be obtained by training with sample images annotated with the axillary region.

[0044] After the primary lesion region and the axillary region in each sub-image are obtained, the positions of the primary lesion region and the axillary region in the target detection image can be determined according to the positions of the sub-images in the target detection image. According to the positions of the primary lesion regions in the target detection image, multiple primary lesion regions that are connected in regions are spliced ​​into a mass image, and the multiple mass images are used as primary lesion region images. According to the positions of the axillary regions in the target detection image, multiple axillary regions that are connected in regions are spliced ​​into a unilateral axillary region image, the unilateral axillary region image located on the left is used as the left axillary region image, the unilateral axillary region image located on the right is used as the right axillary region image, and the left axillary region image and the right axillary region image are used as axillary region images.

[0045] S130, jointly detecting the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes according to the codes corresponding to the primary lesion region image and the codes corresponding to the axillary region image.

[0046] For example, the three indicators of whether the sentinel lymph nodes of the axillary lymph nodes are metastatic, the metastatic load of the sentinel lymph nodes, and whether the non-sentinel lymph nodes are metastatic are very important for surgical decisions. When the patient's axillary lymph nodes have sentinel lymph node metastasis but the metastatic load is low, SLNB and ALND can be exempted; when the patient's axillary lymph nodes have sentinel lymph node metastasis and the metastatic load is high, SLNB can be exempted and ALND can be performed directly; when the patient's axillary lymph nodes do not have sentinel lymph node metastasis, SLNB and ALND can be exempted; when the patient's axillary lymph nodes have sentinel lymph node metastasis but no non-sentinel lymph node metastasis, ALND can be exempted. Among them, when the number of sentinel lymph node metastasis is 1~2, the metastatic load is low, and when the number of sentinel lymph node metastasis is greater than or equal to 3, the metastatic load is high; the concept of non-sentinel lymph nodes is only when there is sentinel lymph node metastasis.

[0047] It can be understood that in order to obtain the three indicators of axillary lymph nodes, the present embodiment aims to determine whether the sentinel lymph nodes of the axillary lymph nodes are metastatic, the metastasis load, and whether the non-sentinel lymph nodes are metastatic based on the three state characteristics by detecting the metastasis probability and the number of metastases of the sentinel lymph nodes of the axillary lymph nodes and the metastasis probability of the non-sentinel lymph nodes.

[0048] Exemplarily, the code corresponding to the primary lesion region image and the code corresponding to the axillary region image can be extracted by a pre-trained encoder, and the code corresponding to the primary lesion region image and the code corresponding to the axillary region image can be spliced ​​to obtain a fused code. The fused code is respectively input into the pre-trained first deep learning model, the second deep learning model, and the third deep learning model, so as to predict the metastasis probability of the sentinel lymph nodes in the axillary lymph nodes based on the fused code by the first deep learning model, predict the number of metastases of the sentinel lymph nodes in the axillary lymph nodes based on the fused code by the second deep learning model, and predict the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes based on the fused code by the third deep learning model.

[0049] In one embodiment, Figure 2 : is a flow chart of determining the metastasis probability of sentinel lymph nodes provided in an embodiment of the present application. Figure 2 The step of determining the metastasis probability of the sentinel lymph node specifically includes S1301-S1303:

[0050] S1301 . Encode the primary lesion region image by using a preset first encoder to obtain a first code corresponding to the primary lesion region image.

[0051] The first code can be understood as feature information extracted by the first encoder in the primary lesion region image. The first encoder can be a deep learning model for extracting spatial features such as 3D CNN, Transformers, and Vision Mamba.

[0052] In this embodiment, the primary lesion region image includes multiple mass images, and the multiple mass images can be input into a first encoder with a shared weight to obtain codes corresponding to the multiple mass images output by the first encoder; the codes corresponding to the multiple mass images are input into a preset first attention model to obtain a first code output by the first attention model. Exemplarily, Figure 3 Schematic diagram of the first code acquisition process provided by the embodiment of the present application. Figure 3 As shown, the primary lesion region image includes mass image A, mass image B and mass image C. The mass image A, mass image B and mass image C are input into the first encoder with shared weights. The first encoder extracts features from the mass image A, mass image B and mass image C respectively, and outputs code A of the mass image A, code B of the mass image B and code C of the mass image C. Code A, code B and code C are input into the first attention model, and the first attention model performs weighted aggregation on code A, code B and code C to output the first code. Among them, the first attention model can adopt a multi-instance learning model based on the attention mechanism. After the first encoder extracts the codes of each mass image, the first attention model performs several rounds of aggregation on the codes of each mass image to generate a comprehensive first code that can accurately characterize the sentinel lymph node information in the primary lesion region. The first code comprehensively reflects the common features and differences of the sentinel lymph nodes in multiple mass images, thereby improving the detection accuracy of the metastasis probability of the sentinel lymph nodes.

[0053] S1302: Encode the armpit region image by using a preset second encoder to obtain a second code corresponding to the armpit region image.

[0054] The second code can be understood as feature information extracted by the second encoder in the axillary region image. The second encoder can be a deep learning model for extracting spatial features such as 3D CNN, Transformers, and Vision Mamba.

[0055] In this embodiment, the armpit region image includes a left armpit region image and a right armpit region image, and the left armpit region image and the right armpit region image can be input into a second encoder with a shared weight, so as to obtain a code corresponding to the left armpit region image and a code corresponding to the right armpit region image output by the second encoder; and the second code is determined according to the distance between the code corresponding to the left armpit region image and the code corresponding to the right armpit region image. Exemplarily, Figure 4Schematic diagram of the second code acquisition process provided by the embodiment of the present application. Figure 4 As shown, the left axillary region image and the right axillary region image are input into the second encoder with a shared weight, and the second encoder extracts features from the left axillary region image and the right axillary region image respectively, and outputs a code L of the left axillary region image and a code R of the right axillary region image. The distance between the code L and the code R is determined by Euclidean distance, cosine similarity or dot product, and the distance is used as the second code. In this embodiment, after extracting the code of the left axillary region image and the code of the right axillary region image by the second encoder, the distance between the code of the left axillary region image and the code of the right axillary region image is used as the second code, so as to accurately characterize the difference between the sentinel lymph nodes in the left and right axillaries by the second code, thereby improving the detection accuracy of the metastasis probability of the sentinel lymph nodes.

[0056] S1303: Concatenate the first code and the second code, input the concatenated code into a preset first multi-layer perceptron, and obtain the metastasis probability of the sentinel lymph node output by the first multi-layer perceptron.

[0057] Exemplarily, the first code and the second code are spliced ​​and input into the input layer of the first multilayer perceptron, the neurons of the hidden layer of the first multilayer perceptron receive the data transmitted by the input layer, the input of the neurons is calculated by weighted summation, and the calculation result is nonlinearly transformed by the activation function of the hidden layer to obtain the characteristics of the hidden layer output. The characteristics of the hidden layer output are input into the output layer of the first multilayer perceptron, and the output layer predicts and outputs the metastasis probability of the sentinel lymph node based on the characteristics. In this embodiment, the first encoder and the second encoder accurately extract the characteristic codes of the sentinel lymph nodes in the primary lesion area image and the axillary area image, and the first multilayer perceptron accurately predicts the metastasis probability of the sentinel lymph nodes based on the characteristic codes of the sentinel lymph nodes, thereby realizing non-invasive examination of the metastasis probability of the sentinel lymph nodes and ensuring the detection accuracy of the metastasis probability of the sentinel lymph nodes.

[0058] In this embodiment, during the training process, the untrained first encoder, the second encoder and the first multi-layer perceptron can be used as an end-to-end single-task prediction model. The end-to-end task prediction model is trained by using pre-collected sample images of the primary lesion area and sample images of the axillary area and the corresponding annotated information on whether the sentinel lymph nodes have metastasized, so as to simultaneously optimize the model parameters of the first encoder, the second encoder and the first multi-layer perceptron.

[0059] In one embodiment, Figure 5 : is a flow chart of determining the number of sentinel lymph node metastases according to an embodiment of the present application. Figure 5 The step of determining the number of metastases of the sentinel lymph nodes specifically includes S1304-S1305:

[0060] S1304: Obtain a third code of the hidden layer output in the first multi-layer perceptron.

[0061] Exemplarily, the features of the hidden layer output in the first multi-layer perceptron are used as the third code.

[0062] S1305: concatenate the first code, the second code, and the third code, input the concatenated code into a preset second multi-layer perceptron, and obtain the number of metastases of the sentinel lymph nodes output by the second multi-layer perceptron.

[0063] Exemplarily, the first code, the second code and the third code are spliced ​​and input into the input layer of the second multilayer perceptron, the neurons of the hidden layer of the second multilayer perceptron receive the data transmitted by the input layer, the input of the neurons is calculated by weighted summation, and the calculation result is nonlinearly transformed by the activation function of the hidden layer to obtain the characteristics of the hidden layer output. The characteristics of the hidden layer output are input into the output layer of the second multilayer perceptron, and the output layer predicts and outputs the metastasis probability of the sentinel lymph node based on the characteristics. In this embodiment, the third code further extracted from the first code and the second code is spliced ​​with the first code and the second code to obtain a feature code with richer and more comprehensive sentinel lymph node information, and the second multilayer perceptron accurately predicts the number of metastasis of the sentinel lymph node based on the feature code, thereby realizing non-invasive examination of the number of metastasis of the sentinel lymph node and ensuring the detection accuracy of the number of metastasis of the sentinel lymph node.

[0064] In this embodiment, during the training process, the trained first encoder and the second encoder and the untrained second multi-layer perceptron can be used as an end-to-end single-task prediction model, and the end-to-end single-task prediction model can be trained through the pre-collected sample images of the primary lesion area and the axillary area sample images and the corresponding annotated number of metastases of the sentinel lymph nodes to optimize the model parameters of the second multi-layer perceptron separately. In addition, the untrained first encoder, the second encoder, the first multi-layer perceptron and the second multi-layer perceptron can also be used as an end-to-end multi-task prediction model, and the end-to-end multi-task prediction model can be trained through the pre-collected sample images of the primary lesion area and the axillary area sample images and the corresponding annotated information on whether the sentinel lymph nodes are metastatic and the number of metastases to simultaneously optimize the model parameters of the first encoder, the second encoder, the first multi-layer perceptron and the second multi-layer perceptron.

[0065] In one embodiment, Figure 6 : is a flow chart of determining the metastasis probability of non-sentinel lymph nodes provided in an embodiment of the present application. Figure 6 As shown, the step of determining the metastasis probability of the non-sentinel lymph node specifically includes S1306-S1308:

[0066] S1306 , encoding the primary lesion region image by using a preset third encoder to obtain a fourth code corresponding to the primary lesion region image.

[0067] Among them, the fourth code can be understood as the feature information extracted by the third encoder in the primary lesion area image. The third encoder can be a deep learning model for extracting spatial features such as 3D CNN, Transformers and Vision Mamba. It should be noted that the first encoder and the third encoder can use the deep learning model with the same structure, but the specific model parameters are different.

[0068] In this embodiment, the primary lesion area image includes a plurality of mass images, and the plurality of mass images can be input into a third encoder with a shared weight to obtain the codes corresponding to the plurality of mass images output by the third encoder; the codes corresponding to the plurality of mass images are input into a preset second attention model to obtain the fourth code output by the second attention model. Exemplarily, the primary lesion area image includes mass image A, mass image B, and mass image C, and the mass image A, mass image B, and mass image C are input into a fourth encoder with a shared weight, and the fourth encoder extracts features from the mass image A, mass image B, and mass image C respectively, and outputs the code A' of the mass image A, the code B' of the mass image B, and the code C' of the mass image C. The codes A', B', and C' are input into the second attention model, and the codes A', B', and C' are weightedly aggregated by the second attention model to output the fourth code. Among them, the first attention model can adopt a multi-instance learning model based on the attention mechanism. After the third encoder extracts the codes of each mass image, this embodiment uses the second attention model to perform several rounds of aggregation on the codes of each mass image to generate a comprehensive fourth code that can accurately characterize the non-sentinel lymph node information in the primary lesion area. The fourth code comprehensively reflects the common characteristics and differences of the non-sentinel lymph nodes in multiple mass images, thereby improving the detection accuracy of the metastasis probability of the non-sentinel lymph nodes.

[0069] S1307: Encode the armpit area image by using a preset fourth encoder to obtain a fifth code corresponding to the armpit area image.

[0070] The fifth code can be understood as the feature information extracted by the fourth coder in the axillary region image. The fourth coder can be a deep learning model for extracting spatial features such as 3D CNN, Transformers and Vision Mamba. It should be noted that the second coder and the fourth coder can use the deep learning model with the same structure, but the specific model parameters are different.

[0071] In this embodiment, the axillary region image includes a left axillary region image and a right axillary region image. The left axillary region image and the right axillary region image can be input into a fourth encoder with a shared weight to obtain a code corresponding to the left axillary region image and a code corresponding to the right axillary region image output by the fourth encoder; the fifth code is determined according to the distance between the code corresponding to the left axillary region image and the code corresponding to the right axillary region image. Exemplarily, the left axillary region image and the right axillary region image are input into a fourth encoder with a shared weight, and the fourth encoder performs feature extraction on the left axillary region image and the right axillary region image, respectively, and outputs a code L' of the left axillary region image and a code R' of the right axillary region image. The distance between the code L' and the code R' is determined by Euclidean distance, cosine similarity or dot product, and the distance is used as the fifth code. In this embodiment, after extracting the code of the left axillary area image and the code of the right axillary area image through the fourth encoder, the distance between the code of the left axillary area image and the code of the right axillary area image is used as the second code, so as to accurately characterize the difference between the non-sentinel lymph nodes in the left and right axillaries through the second code, thereby improving the detection accuracy of the metastasis probability of the non-sentinel lymph nodes.

[0072] S1308: splice the first code, the second code, the fourth code and the fifth code, input the spliced ​​code into a preset third multi-layer perceptron, and obtain the transfer probability of the non-sentinel lymph node output by the third multi-layer perceptron.

[0073] Exemplarily, the first code, the second code, the fourth code and the fifth code are spliced ​​and input into the input layer of the third multilayer perceptron, the neurons of the hidden layer of the third multilayer perceptron receive the data transmitted by the input layer, the input of the neurons is calculated by weighted summation, and the calculation result is nonlinearly transformed by the activation function of the hidden layer to obtain the features of the hidden layer output. The features output by the hidden layer are input into the output layer of the third multilayer perceptron, and the output layer predicts and outputs the transfer probability of the non-sentinel lymph node according to the features. In this embodiment, the feature codes of the non-sentinel lymph nodes in the primary lesion area image and the axillary area image are accurately extracted by the third encoder and the fourth encoder, and the feature codes of the sentinel lymph nodes previously extracted are spliced ​​with the feature codes of the non-sentinel lymph nodes, so that the spliced ​​generated codes contain the feature information of the sentinel lymph nodes and the non-sentinel lymph nodes at the same time, enriching the feature information contained in the spliced ​​generated codes. The third multilayer perceptron accurately predicts the transfer probability of the non-sentinel lymph nodes based on the spliced ​​generated codes, and realizes the non-invasive examination of the transfer probability of the non-sentinel lymph nodes, while ensuring the detection accuracy of the transfer probability of the non-sentinel lymph nodes.

[0074] In this embodiment, during the training process, the trained first encoder and the second encoder and the untrained third encoder, the fourth encoder and the third multi-layer perceptron can be used as an end-to-end single-task prediction model, and the end-to-end single-task prediction model can be trained through pre-collected sample images of the primary lesion area and the axillary area and the corresponding labeled transfer probabilities of non-sentinel lymph nodes to simultaneously optimize the third encoder, the fourth encoder and the third multi-layer perceptron. In addition, the untrained first encoder, the second encoder, the third encoder, the fourth encoder, the first multi-layer perceptron, the second multi-layer perceptron and the third multi-layer perceptron can also be used as an end-to-end multi-task prediction model, and the end-to-end multi-task prediction model can be trained through pre-collected sample images of the lesion area and the axillary area and the corresponding labeled information on whether the sentinel lymph nodes have metastasized, the number of metastases and whether the non-sentinel lymph nodes have metastasized to simultaneously optimize the model parameters of the first encoder, the second encoder, the third encoder, the fourth encoder, the first multi-layer perceptron, the second multi-layer perceptron and the third multi-layer perceptron. Exemplarily, Figure 7 Schematic diagram of the structure of the end-to-end multi-task prediction model provided by the embodiment of the present application. Figure 7 As shown, the primary lesion area image and the axillary area image are input into the first encoder, the second encoder, the third encoder and the fourth encoder, and the first encoder, the second encoder, the third encoder and the fourth encoder respectively output the first code, the second code, the fourth code and the fifth code. The first code and the second code are spliced ​​and input into the first multilayer perceptron, and the first multilayer perceptron outputs the transfer probability of the sentinel lymph node. The first code, the second code and the third code output by the hidden layer of the first multilayer perceptron are spliced ​​and input into the second multilayer perceptron, and the second multilayer perceptron outputs the number of metastases of the sentinel lymph node. The first code, the second code, the fourth code and the fifth code are spliced ​​and input into the third multilayer perceptron, and the third multilayer perceptron outputs the transfer probability of the non-sentinel lymph node.

[0075] It should be noted that after obtaining the metastasis probability of the sentinel lymph node, the metastasis probability of the sentinel lymph node can be compared with a preset first probability threshold. When the metastasis probability of the sentinel lymph node is greater than the first probability threshold, the sentinel lymph node in the axillary lymph node is determined to be metastatic, and when the metastasis probability of the sentinel lymph node is less than or equal to the first probability threshold, the sentinel lymph node is determined to be non-metastatic. After obtaining the number of metastases of the sentinel lymph node, the metastasis load of the sentinel lymph node can be determined based on the number of metastases, that is, when the number of metastases is 1 to 2, the metastasis load of the sentinel lymph node is determined to be low, and when the number of metastases is greater than or equal to 3, the metastasis load of the sentinel lymph node is determined to be high. After obtaining the metastasis probability of the non-sentinel lymph nodes, the metastasis probability of the non-sentinel lymph nodes can be compared with a preset second probability threshold. When the metastasis probability of the non-sentinel lymph nodes is greater than the second probability threshold, it is determined that the non-sentinel lymph nodes in the axillary lymph nodes are metastatic. When the metastasis probability of the non-sentinel lymph nodes is less than or equal to the second probability threshold, it is determined that the non-sentinel lymph nodes are not metastatic. The first probability threshold and the second probability threshold can be set according to actual needs, and can generally be set to 50%.

[0076] In summary, the state characteristic detection method of axillary lymph nodes provided in the embodiment of the present application obtains multiple breast MRI images of the detection object, and splices the multiple breast MRI images in the channel dimension to obtain a target detection image; divides the target detection image into multiple sub-images, detects the primary lesion area and the axillary area in each sub-image, and determines the primary lesion area image and the axillary area image in the target detection image according to each primary lesion area and each axillary area; according to the code corresponding to the primary lesion area image and the code corresponding to the axillary area image, jointly detects the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes. Through the above-mentioned technical means, it is realized to identify the state characteristics of the patient's axillary lymph nodes before surgery by non-invasive methods, solves the problem that the prior art needs to detect the state characteristics of the axillary lymph nodes through invasive examination, avoids the postoperative complications such as upper limb edema, neurosensory disorders and movement limitation caused by invasive examination, and ensures that the patient's life is not affected by the axillary lymph node examination. The state characteristics of axillary lymph nodes are detected by using multiple breast MRI images of axillary lymph nodes, which improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by extracting the primary lesion area and axillary area in breast MRI images, which reduces the interference of invalid information and improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by jointly detecting the primary lesion area and axillary area, which enriches the information used to detect the state characteristics of axillary lymph nodes and improves the detection accuracy of state characteristics. By detecting the metastasis probability and number of sentinel lymph nodes in axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in axillary lymph nodes, more comprehensive state characteristics of axillary lymph nodes are obtained, which provides a more accurate and important theoretical basis for the diagnosis of benign and malignant axillary lymph nodes.

[0077] Based on the above embodiments, Figure 8 A schematic diagram of the structure of a device for detecting the state characteristics of an axillary lymph node provided in an embodiment of the present application. Figure 8 The state feature detection device for axillary lymph nodes provided in this embodiment specifically includes: an image stitching module 21, a regional image detection module 22 and a state feature detection module 23.

[0078] The image stitching module 21 is configured to obtain multiple breast MRI images of the detection object, and stitch the multiple breast MRI images in the channel dimension to obtain a target detection image;

[0079] The regional image detection module 22 is configured to divide the target detection image into a plurality of sub-images, detect the primary lesion region and the axillary region in each sub-image, and respectively determine the primary lesion region image and the axillary region image in the target detection image according to each primary lesion region and each axillary region;

[0080] The state feature detection module 23 is configured to jointly detect the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes according to the codes corresponding to the primary lesion area image and the codes corresponding to the axillary area image.

[0081] On the basis of the above embodiment, the state feature detection module 23 includes: a first coding acquisition unit, configured to encode the primary lesion area image through a preset first encoder to obtain a first code corresponding to the primary lesion area image; a second coding acquisition unit, configured to encode the axillary area image through a preset second encoder to obtain a second code corresponding to the axillary area image; a first transition probability determination unit, configured to splice the first code and the second code, input the spliced ​​code into a preset first multi-layer perceptron, and obtain the transfer probability of the sentinel lymph node output by the first multi-layer perceptron.

[0082] On the basis of the above embodiment, the primary lesion area image includes multiple mass images; accordingly, the first coding acquisition unit includes: a first coding subunit, configured to input the multiple mass images into the first encoder with shared weights, and obtain the codes corresponding to the multiple mass images output by the first encoder; the first coding acquisition subunit, configured to input the codes corresponding to the multiple mass images into a preset first attention model, and obtain the first code output by the first attention model.

[0083] On the basis of the above embodiment, the axillary area image includes a left axillary area image and a right axillary area image; accordingly, the second encoding acquisition unit includes: a second encoding unit, configured to input the left axillary area image and the right axillary area image into a second encoder with a shared weight, and obtain the code corresponding to the left axillary area image and the code corresponding to the right axillary area image output by the second encoder; a second encoding acquisition subunit, configured to determine the second code according to the distance between the code corresponding to the left axillary area image and the code corresponding to the right axillary area image.

[0084] On the basis of the above embodiment, the state feature detection module 23 includes: a third code acquisition unit, configured to obtain the third code output by the hidden layer in the first multi-layer perceptron; a transfer quantity determination unit, configured to splice the first code, the second code and the third code, and input the spliced ​​code into a preset second multi-layer perceptron to obtain the number of metastases of the sentinel lymph node output by the second multi-layer perceptron.

[0085] On the basis of the above embodiment, the state feature detection module 23 includes: a fourth code acquisition unit, configured to encode the primary lesion area image through a preset third encoder to obtain a fourth code corresponding to the primary lesion area image; a fifth code acquisition unit, configured to encode the axillary area image through a preset fourth encoder to obtain a fifth code corresponding to the axillary area image; a second transition probability determination unit, configured to splice the first code, the second code, the fourth code and the fifth code, and input the spliced ​​code into a preset third multi-layer perceptron to obtain the transition probability of the non-sentinel lymph node output by the third multi-layer perceptron.

[0086] Based on the above embodiment, the primary lesion area image includes multiple mass images; accordingly, the fourth encoding acquisition unit includes: a third encoding subunit, configured to input the multiple mass images into a third encoder with shared weights, and obtain the codes corresponding to the multiple mass images output by the third encoder; a fourth encoding acquisition subunit, configured to input the codes corresponding to the multiple mass images into a preset second attention model, and obtain the fourth code output by the second attention model.

[0087] On the basis of the above embodiment, the axillary area image includes a left axillary area image and a right axillary area image; accordingly, the fifth encoding acquisition unit includes: a fourth encoding subunit, configured to input the left axillary area image and the right axillary area image into a fourth encoder with a shared weight, and obtain the code corresponding to the left axillary area image and the code corresponding to the right axillary area image output by the fourth encoder; a fifth encoding acquisition subunit, configured to determine the fifth code according to the distance between the code corresponding to the left axillary area image and the code corresponding to the right axillary area image.

[0088] On the basis of the above embodiment, each breast MRI image includes depth information; the image stitching module includes: a spatial alignment unit, configured to perform spatial normalization alignment on multiple breast MRI images before stitching the multiple breast MRI images in the channel dimension to obtain a target detection image, so as to obtain multiple breast MRI common-frame images with spatial point alignment; an intensity processing unit, configured to perform intensity normalization processing on the multiple breast MRI common-frame images with spatial point alignment, so as to obtain multiple breast MRI common-frame images with intensities in the same range.

[0089] As mentioned above, the state characteristic detection device of axillary lymph nodes provided in the embodiment of the present application obtains multiple breast MRI images of the detection object, splices the multiple breast MRI images in the channel dimension to obtain a target detection image; divides the target detection image into multiple sub-images, detects the primary lesion area and the axillary area in each sub-image, and determines the primary lesion area image and the axillary area image in the target detection image according to each primary lesion area and each axillary area; according to the code corresponding to the primary lesion area image and the code corresponding to the axillary area image, jointly detects the metastasis probability and number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes. Through the above technical means, it is realized to identify the state characteristics of the patient's axillary lymph nodes before surgery by non-invasive methods, solves the problem that the prior art needs to detect the state characteristics of the axillary lymph nodes through invasive examination, avoids the postoperative complications such as upper limb edema, neurosensory disorders and movement limitation caused by invasive examination, and ensures that the patient's life is not affected by the axillary lymph node examination. The state characteristics of axillary lymph nodes are detected by using multiple breast MRI images of axillary lymph nodes, which improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by extracting the primary lesion area and axillary area in breast MRI images, which reduces the interference of invalid information and improves the detection accuracy of state characteristics. The state characteristics of axillary lymph nodes are detected by jointly detecting the primary lesion area and axillary area, which enriches the information used to detect the state characteristics of axillary lymph nodes and improves the detection accuracy of state characteristics. By detecting the metastasis probability and number of sentinel lymph nodes in axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in axillary lymph nodes, more comprehensive state characteristics of axillary lymph nodes are obtained, which provides a more accurate and important theoretical basis for the diagnosis of benign and malignant axillary lymph nodes.

[0090] The axillary lymph node status characteristic detection device provided in the embodiment of the present application can be used to execute the axillary lymph node status characteristic detection method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0091] Fig. 9 is a schematic diagram of a device for detecting the state characteristics of an axillary lymph node provided in an embodiment of the present application, with reference to Fig. 9 The state characteristic detection device of the axillary lymph node comprises: a processor 31, a memory 32, a communication device 33, an input device 34 and an output device 35. The number of the processors 31 in the state characteristic detection device of the axillary lymph node can be one or more, and the number of the memories 32 in the state characteristic detection device of the axillary lymph node can be one or more. The processor 31, the memory 32, the communication device 33, the input device 34 and the output device 35 of the state characteristic detection device of the axillary lymph node can be connected via a bus or other means.

[0092] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the state feature detection method of axillary lymph nodes in any embodiment of the present application (for example, the image stitching module 21, the regional image detection module 22 and the state feature detection module 23 in the state feature detection device). The memory 32 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the device, etc. In addition, the memory 32 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0093] The communication device 33 is used for data transmission.

[0094] The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 32, that is, implements the above-mentioned axillary lymph node status feature detection method.

[0095] The input device 34 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 35 may include a display device such as a display screen.

[0096] The axillary lymph node status characteristic detection device provided above can be used to execute the axillary lymph node status characteristic detection method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0097] An embodiment of the present application also provides a storage medium containing computer executable instructions, which, when executed by a computer processor, are used to execute a method for detecting the status characteristics of an axillary lymph node. The method for detecting the status characteristics of an axillary lymph node includes: acquiring multiple breast magnetic resonance images of a detection object, and splicing the multiple breast magnetic resonance images in a channel dimension to obtain a target detection image; dividing the target detection image into multiple sub-images, detecting a primary lesion area and an axillary area in each sub-image, and determining a primary lesion area image and an axillary area image in the target detection image according to each primary lesion area and each axillary area; and jointly detecting the metastasis probability and the number of metastases of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes according to the code corresponding to the primary lesion area image and the code corresponding to the axillary area image.

[0098] Storage medium - any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROM, floppy disk or tape device; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disk or optical storage); registers or other similar types of memory elements, etc. Storage media may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the first computer system in which the program is executed, or may be located in a different second computer system, which is connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media residing in different locations (for example, in different computer systems connected by a network). The storage medium may store program instructions (for example, embodied as a computer program) that can be executed by one or more processors.

[0099] Of course, the storage medium containing computer executable instructions provided in an embodiment of the present application, whose computer executable instructions are not limited to the above-mentioned method for detecting the status characteristics of axillary lymph nodes, can also execute related operations in the method for detecting the status characteristics of axillary lymph nodes provided in any embodiment of the present application.

[0100] The axillary lymph node status characteristic detection device, axillary lymph node status characteristic detection system, storage medium and axillary lymph node status characteristic detection equipment provided in the above embodiments can execute the axillary lymph node status characteristic detection method provided in any embodiment of the present application. For technical details not described in detail in the above embodiments, please refer to the axillary lymph node status characteristic detection method provided in any embodiment of the present application.

[0101] The above are only preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments herein, and various obvious changes, readjustments and substitutions that can be made by those skilled in the art will not deviate from the protection scope of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may also include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for detecting the state characteristics of axillary lymph nodes, characterized in that: include: Acquire multiple breast nuclear magnetic resonance images of the detection object, and splice the multiple breast nuclear magnetic resonance images in the channel dimension to obtain a target detection image; Dividing the target detection image into a plurality of sub-images, detecting a primary lesion region and an axillary region in each of the sub-images, and respectively determining a primary lesion region image and an axillary region image in the target detection image according to each of the primary lesion regions and each of the axillary regions; According to the codes corresponding to the primary lesion region image and the codes corresponding to the axillary region image, jointly detecting the metastasis probability and the number of metastases of the sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of the non-sentinel lymph nodes in the axillary lymph nodes; The method comprises: encoding the primary lesion region image by a preset first encoder to obtain a first code corresponding to the primary lesion region image; The axillary area image is encoded by a preset second encoder to obtain a second code corresponding to the axillary area image; the first code and the second code are spliced, and the spliced ​​code is input into a preset first multi-layer perceptron to obtain the transfer probability of the sentinel lymph node output by the first multi-layer perceptron; a third code output by the hidden layer in the first multi-layer perceptron is obtained; the first code, the second code and the third code are spliced, and the spliced ​​code is input into a preset second multi-layer perceptron to obtain the number of metastasis of the sentinel lymph nodes output by the second multi-layer perceptron.

2. The method for detecting the state characteristics of axillary lymph nodes according to claim 1, characterized in that: The primary lesion area image includes a plurality of mass images; Correspondingly, encoding the primary lesion region image by a preset first encoder to obtain a first code corresponding to the primary lesion region image includes: Inputting the plurality of mass images into a first encoder with a shared weight, and obtaining codes corresponding to the plurality of mass images output by the first encoder; The codes corresponding to the multiple mass images are input into a preset first attention model to obtain a first code output by the first attention model.

3. The method for detecting the state characteristics of axillary lymph nodes according to claim 1, characterized in that: The axillary region image includes a left axillary region image and a right axillary region image; Correspondingly, encoding the armpit area image by a preset second encoder to obtain a second code corresponding to the armpit area image includes: Inputting the left axillary region image and the right axillary region image into a second encoder with a shared weight, and obtaining a code corresponding to the left axillary region image and a code corresponding to the right axillary region image output by the second encoder; A second code is determined according to a distance between a code corresponding to the left axillary region image and a code corresponding to the right axillary region image.

4. The method for detecting the state characteristics of axillary lymph nodes according to claim 1, characterized in that: The step of jointly detecting the metastasis probability and the number of metastases of the sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of the non-sentinel lymph nodes in the axillary lymph nodes according to the code corresponding to the primary lesion region image and the code corresponding to the axillary region image comprises: Encoding the primary lesion region image by a preset third encoder to obtain a fourth code corresponding to the primary lesion region image; Encoding the axillary region image by using a preset fourth encoder to obtain a fifth code corresponding to the axillary region image; The first code, the second code, the fourth code and the fifth code are spliced ​​together, and the spliced ​​code is input into a preset third multi-layer perceptron to obtain the transfer probability of the non-sentinel lymph node output by the third multi-layer perceptron.

5. The method for detecting the state characteristics of axillary lymph nodes according to claim 4, wherein the primary lesion region image comprises a plurality of mass images; Correspondingly, encoding the primary lesion region image by a preset third encoder to obtain a fourth code corresponding to the primary lesion region image includes: Inputting the plurality of mass images into a third encoder with a shared weight, and obtaining codes corresponding to the plurality of mass images output by the third encoder; The codes corresponding to the multiple mass images are input into a preset second attention model to obtain a fourth code output by the second attention model.

6. The method for detecting the state characteristics of axillary lymph nodes according to claim 4, wherein the axillary region image comprises a left axillary region image and a right axillary region image; Correspondingly, encoding the armpit area image by a preset fourth encoder to obtain a fifth code corresponding to the armpit area image includes: Inputting the left axillary region image and the right axillary region image into a fourth encoder with a shared weight, and obtaining a code corresponding to the left axillary region image and a code corresponding to the right axillary region image output by the fourth encoder; A fifth code is determined according to a distance between a code corresponding to the left armpit area image and a code corresponding to the right armpit area image.

7. The method for detecting the state characteristics of axillary lymph nodes according to claim 1, characterized in that: Each of the breast MRI images includes depth information; before the plurality of breast MRI images are spliced ​​in the channel dimension to obtain the target detection image, the method further includes: Performing spatial normalization alignment on the multiple breast MRI images to obtain multiple breast MRI common frame images that are spatially point aligned; Intensity normalization processing is performed on the multiple breast MRI common-frame images aligned with the spatial points to obtain multiple breast MRI common-frame images with intensities in the same range.

8. A device for detecting the state characteristics of axillary lymph nodes, characterized in that: include: An image stitching module is configured to acquire multiple breast nuclear magnetic resonance images of a detection object, and stitch the multiple breast nuclear magnetic resonance images in a channel dimension to obtain a target detection image; a regional image detection module, configured to divide the target detection image into a plurality of sub-images, detect a primary lesion region and an axillary region in each of the sub-images, and respectively determine a primary lesion region image and an axillary region image in the target detection image according to each of the primary lesion regions and each of the axillary regions; The state feature detection module is configured to jointly detect the metastasis probability and the number of sentinel lymph nodes in the axillary lymph nodes and the metastasis probability of non-sentinel lymph nodes in the axillary lymph nodes according to the code corresponding to the primary lesion region image and the code corresponding to the axillary region image; wherein the state feature detection module comprises: a first code acquisition unit, configured to encode the primary lesion region image through a preset first encoder to obtain a first code corresponding to the primary lesion region image; a second code acquisition unit, configured to encode the axillary region image through a preset second encoder to obtain a first code corresponding to the axillary region image. a second code corresponding to the domain image; a first transfer probability determining unit, configured to splice the first code and the second code, input the spliced ​​code into a preset first multi-layer perceptron, and obtain the transfer probability of the sentinel lymph node output by the first multi-layer perceptron; a third code obtaining unit, configured to obtain the third code output by the hidden layer in the first multi-layer perceptron; a transfer quantity determining unit, configured to splice the first code, the second code and the third code, input the spliced ​​code into a preset second multi-layer perceptron, and obtain the transfer quantity of the sentinel lymph node output by the second multi-layer perceptron.

9. A device for detecting the state characteristics of axillary lymph nodes, characterized in that: include: one or more processors; A storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the state characteristic detection method of axillary lymph nodes as described in any one of claims 1-7.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the method for detecting the status characteristics of axillary lymph nodes as described in any one of claims 1-7.

Images