Methods, devices and media for predicting lymph node metastasis

Abstract

This invention discloses a method, apparatus, and medium for predicting lymph node metastasis. The method involves acquiring a three-dimensional image of a lymph node to be detected and determining the region of interest (ROI) within the image. Feature extraction is performed on the ROI to obtain local features corresponding to the three-dimensional lymph node image. The ROI data is then expanded to obtain pixel blocks to be processed, and feature extraction is performed on these pixel blocks to obtain global features corresponding to the three-dimensional lymph node image. Based on the local and global features, the lymph node metastasis prediction result for the three-dimensional lymph node image is determined. This invention achieves a comprehensive description of the features of the three-dimensional lymph node image through the complementarity of local and global features, solving the problem of low accuracy in lymph node metastasis prediction and improving its reliability. Furthermore, the local features only include the features of the lymph node itself, reducing redundant features.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method, apparatus, and medium for predicting lymph node metastasis results. Background Art

[0002] The status of cervical lymph nodes significantly affects the survival rate and determines the surgical plan for oral cancer. Especially in oral cancer, 20 - 40% of patients have occult cervical metastases. It is necessary to accurately detect these occult lymph node metastases before surgery.

[0003] Currently, the prior art usually uses an algorithm in 3DSlicer software to convert the lesion area in CT images into a region of interest, combines the extracted texture values and imports them into an imageomics feature extraction software to obtain a data set. Through the established artificial neural network model, the data set is dimensionally reduced, and the lymph node status of newly input oral cancer patients can be predicted based on the dimensional reduction result.

[0004] However, in the prior art, the dimensional reduction result is only based on imageomics features, the sample size available for training the artificial neural network model is limited, which has an adverse impact on the generalization of the prior art. Summary of the Invention

[0005] The present invention provides a method, apparatus, and medium for predicting lymph node metastasis results to solve the problem of low accuracy in predicting lymph node metastasis.

[0006] According to one aspect of the present invention, there is provided a method for predicting lymph node metastasis, including:

[0007] Obtain a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node;

[0008] Extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node;

[0009] Perform an expansion process on the region of interest data to obtain a pixel block to be processed, and extract features from the pixel block to be processed to obtain global features corresponding to the three-dimensional image of the lymph node;

[0010] Based on the local features and global features, determine the lymph node metastasis prediction result of the three-dimensional image of the lymph node.

[0011] According to another aspect of the present invention, there is provided a device for predicting lymph node metastasis, including:

[0012] A region of interest determination module, configured to obtain a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node;

[0013] The local feature extraction module is used to extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node;

[0014] The global feature extraction module is used to expand the data of the region of interest to obtain the pixel block to be processed, and to extract the features of the pixel block to obtain the global features corresponding to the three-dimensional image of the lymph node.

[0015] The lymph node metastasis prediction module is used to determine the lymph node metastasis prediction results of the three-dimensional lymph node image based on local and global features.

[0016] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0017] At least one processor; and

[0018] A memory that is communicatively connected to at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the lymph node metastasis prediction method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the lymph node metastasis prediction method of any embodiment of the present invention.

[0021] The technical solution of this invention involves acquiring a three-dimensional image of a lymph node to be detected and determining the region of interest (ROI) within the image; extracting features from the ROI to obtain local features corresponding to the three-dimensional lymph node image; expanding the ROI data to obtain pixel blocks to be processed, and extracting features from these pixel blocks to obtain global features corresponding to the three-dimensional lymph node image; and determining the lymph node metastasis prediction result based on the local and global features. By extracting local features corresponding to the three-dimensional lymph node image based on the ROI, the local features can include only the features of the lymph node itself, excluding features of the surrounding tissues, thereby reducing redundant features. By determining the lymph node metastasis prediction result based on both local and global features, where the local and global features are complementary, the prediction comprehensively describes the features of the three-dimensional lymph node image, solving the problem of low accuracy in lymph node metastasis prediction and improving its reliability.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 1 of the present invention;

[0026] Figure 3 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 2 of the present invention;

[0027] Figure 4 This is a schematic diagram of the structure of a lymph node metastasis prediction device provided in Embodiment 3 of the present invention;

[0028] Figure 5 This is a schematic diagram of the structure of an electronic device that implements the lymph node metastasis prediction method of this invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Example 1

[0032] Figure 1 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where lymph node metastasis prediction is performed based on three-dimensional lymph node images. The method can be executed by a lymph node metastasis prediction device, which can be implemented in hardware and / or software and can be configured in electronic devices such as computers and servers. Figure 1 As shown, the method includes:

[0033] In this embodiment, the three-dimensional lymph node image is a three-dimensional medical image of a target site (e.g., a human or animal body) with a histological diagnosis of oral squamous cell carcinoma; the target site is the neck, including lymph nodes; the three-dimensional medical image is a medical image with three-dimensional spatial information, such as a contrast-enhanced computed tomography (CECT) image and a magnetic resonance imaging (MRI) image. Lymph node metastasis prediction involves predicting the lymph node metastasis status contained in the three-dimensional lymph node image, where lymph node metastasis status refers to the metastasis of tumor cells from the primary tumor to lymph nodes in the lymphatic system.

[0034] S110. Obtain a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node.

[0035] In this embodiment, the three-dimensional image of the lymph node to be detected is the three-dimensional image of the lymph node for which lymph node metastasis prediction is required. It can be a three-dimensional image of the lymph node obtained by scanning and acquiring through an imaging system, or it can be a three-dimensional image of the lymph node obtained by transmitting through an internal or external storage system such as a storage system image archiving and communication system (PACS). This embodiment does not limit this.

[0036] Specifically, the method involves reading real-time acquired or stored 3D images of lymph nodes. The region of interest (ROI) is the local 3D region containing the lymph nodes within the 3D lymph node image. The ROI is obtained by extracting the ROI from the 3D lymph node image. For example, the ROI extraction can be manual, automatic, or semi-automatic; this embodiment does not impose any limitations on this method. Optionally, the contours of the lymph nodes are drawn layer by layer along the axial direction of the 3D lymph node image from multiple 2D lymph node images, and the ROI is determined based on these contours; the 3D lymph node image is composed of multiple 2D lymph node images stacked axially; and / or, the 3D lymph node image is segmented based on a lymph node segmentation model to obtain the ROI.

[0037] In some embodiments, the contours of lymph nodes are delineated layer by layer along the axis of the three-dimensional lymph node image, and the region of interest is determined based on the contours of the lymph nodes.

[0038] The three-dimensional image of the lymph node is composed of multiple two-dimensional images of the lymph node stacked axially.

[0039] In this embodiment, the axial direction of the three-dimensional lymph node image represents the vertical axis direction of the target object, and the two-dimensional lymph node image is a two-dimensional medical image obtained by a single scan of the target part of the target object. It can be understood that the two-dimensional lymph node image represents the cross section of the target part along the vertical axis direction of the target object.

[0040] It should be noted that during the acquisition of three-dimensional lymph node images, multiple two-dimensional lymph node images perpendicular to the vertical axis of the target object can be obtained by continuously and equally spaced scanning of the target area. These multiple two-dimensional lymph node images are stacked in the order of scanning to form a three-dimensional lymph node image. For example, assuming the pixel size (length × width × height) of the three-dimensional lymph node image is 512 × 512 × 100, then the pixel size of the two-dimensional lymph node image is 512 × 512, and the three-dimensional lymph node image is composed of 100 two-dimensional lymph node images stacked layer by layer.

[0041] Specifically, by importing 3D lymph node images into an image processing application or using a manual delineation toolkit, 2D lymph node images are sequentially presented on the display interface based on their axial order within the 3D lymph node image. The operator delineates the contours of the lymph nodes in each 2D image, and the delineated closed regions are defined as regions of interest (ROIs) in the presented 2D lymph node image. By stacking these ROIs layer by layer based on their axial order within the 3D lymph node image, the ROI of the 3D lymph node image is obtained. For example, the image processing application can be an open-source image processing application based on MATLAB or Python. By manually determining the ROI of the 3D lymph node image, personalized ROIs can be accurately identified, ensuring the accuracy of personalized lymph node metastasis prediction.

[0042] In some embodiments, based on a lymph node segmentation model, the three-dimensional image of lymph nodes is segmented to obtain the region of interest.

[0043] Specifically, the lymph node segmentation model is a segmentation model that uses lymph nodes as the segmentation object. By importing a 3D image of a lymph node into the lymph node segmentation model, the model automatically segments the lymph nodes in the 3D image. By extracting the lymph nodes and removing other background areas, the region of interest (ROI) is obtained. For example, the lymph node segmentation model can be one or more combinations of image segmentation algorithms based on thresholding, edge detection, region segmentation, and neural networks. The lymph node segmentation model can include a 3D segmentation model and / or a 2D segmentation model; this embodiment does not limit this. By automatically determining, rather than manually, the ROI of the 3D lymph node image based on the lymph node segmentation model, the speed of ROI determination can be accelerated, and the efficiency of lymph node metastasis prediction can be improved.

[0044] Due to segmentation errors in the lymph node segmentation model, the obtained regions of interest (ROIs) may contain errors. In some embodiments, after obtaining the ROIs of the three-dimensional lymph node image based on the lymph node segmentation model, the ROIs of multiple two-dimensional lymph node images are manually calibrated layer by layer along the axis of the three-dimensional lymph node image to improve the accuracy of the ROIs and ensure the reliability of subsequent lymph node metastasis prediction.

[0045] Due to the subjective factors of the operator, the manually drawn region of interest may include the surrounding tissue area. In some embodiments, after obtaining the region of interest of the three-dimensional image of the lymph node based on the manually drawn lymph node outline, the region of interest is automatically calibrated based on the lymph node segmentation model to reduce the error of the region of interest caused by the subjective factors of the operator and ensure the reliability of subsequent lymph node metastasis prediction.

[0046] S120. Extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node.

[0047] Specifically, local features refer to the features of the region of interest in a 3D lymph node image. By calling a feature extraction application or toolkit, feature extraction processing is performed on the region of interest in the 3D lymph node image to obtain the local features corresponding to the 3D lymph node image. For example, assuming the local features are radiomics features, the feature extraction application can be an open-source radiomics feature extraction application based on MATLAB or Python.

[0048] When the local features are radiomics features, the optional local features include at least one of the following: first-order features, shape features, gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level size region moment features, adjacent gray-level difference matrix features, and gray-level correlation matrix features.

[0049] First-order features characterize the voxel intensity distribution within the region of interest. These features include, but are not limited to, the mean, variance, and maximum value of the voxel intensity. First-order features help describe information such as the overall brightness, contrast, and texture of the region of interest.

[0050] Shape features are characteristics that characterize the shape and size of a region of interest. For example, shape features include, but are not limited to, the volume, surface area, and compactness of the region of interest. Shape features help describe the size, shape, and structure of the region of interest.

[0051] Gray-level co-occurrence matrix (GLCM) features characterize the spatial relationships and texture information between voxels in a region of interest. For example, GLCM features include, but are not limited to, contrast, roughness, and orientation. GLCM features help describe the texture structure of the region of interest.

[0052] Gray-level run-length matrix (GLRLM) features characterize the spatial distribution of continuous voxels with the same gray level in multiple directions. For example, GLRLM features include, but are not limited to, short-range low-gray-level repetition, long-range high-gray-level repetition, and gray-level non-uniformity. GLRLM features help represent the texture properties of regions of interest.

[0053] Gray Level Size Zone Matrix (GLSZM) features characterize the size distribution of consecutive regions with the same gray level value within a region of interest. For example, GLSZM features include, but are not limited to, small-area low-gray-level repetition, large-area high-gray-level repetition, and gray-level non-uniformity. GLSZM features help reflect the texture information of the region of interest.

[0054] The Neighborhood Gray-tone Difference Matrix (NGTDM) is a feature that represents the sum of differences between the gray levels of a voxel and the average gray levels of its neighboring voxels. NGTDM features help capture variations in gray levels within a region of interest, further revealing texture details within that region.

[0055] Gray-level Dependence Matrix (GLDM) features are used to characterize the differences between adjacent voxels based on voxel values. GLDM features help quantify the gray-level dependencies between adjacent pixels or voxels in an image, thereby revealing the texture characteristics of the image.

[0056] For example, local features include 1454 features in 7 categories: first-order features, shape features, gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level size region moment features, adjacent gray-level difference matrix features, and gray-level correlation matrix features.

[0057] S130. Expand the region of interest data to obtain the pixel block to be processed, and extract features from the pixel block to obtain the global features corresponding to the three-dimensional image of the lymph node.

[0058] In this embodiment, the pixel block to be processed is at least a local region of the three-dimensional image of the lymph node. It should be noted that the pixel block to be processed contains a region of interest.

[0059] Specifically, by expanding the region of interest outward, an expanded region is obtained. Pixel blocks containing the expanded region in the 3D image of the lymph node are then cropped to obtain pixel blocks to be processed. The region of interest can be located at the center of the pixel block to be processed or at any off-center position; this embodiment does not impose any restrictions on this. In some embodiments, the region of interest is uniformly expanded outward by a target distance so that the region of interest is located at the center of the pixel block to be processed. The target distance can be a preset distance or a distance determined based on the region of interest; this embodiment does not impose any restrictions on this. The preset distance is a pre-set fixed distance, for example, a preset distance of 3 millimeters.

[0060] The shape of the pixel block to be processed includes, but is not limited to, cubes, cuboids, spheres, and irregular cubes. In some embodiments, the pixel block to be processed is obtained by cropping the pixel block corresponding to the circumscribed cuboid of the extended region in the three-dimensional image of the lymph node.

[0061] It should be noted that the pixel spacing in the three dimensions of the lymph node 3D image may be inconsistent. For example, the pixel spacing in the length and width dimensions may both be 0.5 mm, while the pixel spacing in the height dimension may be 1 mm. This inconsistency in pixel spacing across the length, width, and height dimensions can adversely affect subsequent lymph node metastasis prediction. Optionally, the pixel block to be processed can be interpolated using a preset interpolation method to obtain a pixel block of preset size and shape.

[0062] The preset shapes include cubes.

[0063] In this embodiment, the preset size is the pre-set size of the pixel block to be processed, and the preset shape is the pre-set shape of the pixel block to be processed. It should be noted that when the preset shape is a cube, the preset size includes the dimensions of length, width, and height, and these three dimensions are identical. The preset size can be the actual size or the pixel size. The preset interpolation method is a pre-set interpolation method, which includes, but is not limited to, interpolation methods based on linear interpolation functions, bicubic interpolation functions, and 3D super-resolution reconstruction algorithms.

[0064] Specifically, by calling the interpolation function or algorithm corresponding to the preset interpolation method, and configuring the interpolation function or algorithm based on the preset size and preset shape, interpolation processing is performed on the outwardly expanded pixel block to be processed based on the configured interpolation function or algorithm to obtain the pixel block to be processed with the preset size and preset shape. In some embodiments, interpolation parameters are determined based on the preset size, preset shape, and pixel block to be processed. The interpolation parameters include, but are not limited to, one or more target interpolation dimensions and the interpolation ratio of each target interpolation dimension. The target interpolation dimension is the dimension to be interpolated, including at least one of the three dimensions: length, width, and height. It can be understood that, for the length, width, and height dimensions, if the size of any dimension of the pixel block to be processed is inconsistent with the size of that dimension in the preset size, that dimension is determined as the target interpolation dimension. The interpolation ratio of the target interpolation dimension is the change ratio of the size of the pixel block to be processed in the target interpolation dimension between before and after interpolation. By configuring the interpolation parameters for the interpolation function or algorithm, and based on the configured interpolation function or algorithm, a number of new voxels corresponding to the target interpolation dimension are inserted between adjacent voxels in each target interpolation dimension of the expanded pixel block to be processed, resulting in a pixel block of preset size and shape. The number of targets corresponding to the target interpolation dimension is set based on the interpolation ratio of the target interpolation dimension; that is, the number of targets corresponding to each target interpolation dimension is equal to the interpolation ratio of that target interpolation dimension.

[0065] For example, assuming the pixel size of the pixel block to be processed is 16×16×8, the preset shape is a cube, and the preset size is 32×32×32 pixels, then the interpolation factor for the long dimension is 2, the interpolation factor for the wide dimension is 2, and the interpolation factor for the high dimension is 4. In the interpolation process, two new voxels are inserted between any two adjacent voxels in the long dimension (or wide dimension), and four new voxels are inserted between any two adjacent voxels in the high dimension, thereby obtaining a cube-shaped pixel block to be processed with a pixel size of 32×32×32.

[0066] The technical solution of this embodiment, by performing interpolation processing on the pixel block to be processed, can improve the resolution of the pixel block to be processed and supplement details while correcting the consistency of pixel spacing in the three dimensions of length, width and height, which helps to improve the accuracy of lymph node metastasis prediction.

[0067] Specifically, the global features are the features of the pixel blocks corresponding to the region of interest in the 3D lymph node image. By calling feature extraction applications, toolkits, or algorithms, feature extraction is performed on the pixel blocks to be processed, obtaining the global features corresponding to the 3D lymph node image. For example, assuming the global features are deep learning features, the feature extraction algorithm can be a 3D deep learning feature extraction algorithm, including but not limited to 3D DenseNet, 3DResNet, and 3D VGG.

[0068] S140. Based on local and global features, determine the lymph node metastasis prediction results of the three-dimensional lymph node image.

[0069] In this embodiment, the lymph node metastasis prediction result is a result that characterizes the lymph node metastasis situation based on the three-dimensional image of the lymph nodes. The lymph node metastasis prediction result includes the lymph node metastasis situation.

[0070] It should be noted that a 3D lymph node image contains multiple lymph nodes, and each lymph node corresponds to a pair of local and global features. Both the local and global features are determined based on the region of interest containing that lymph node. In the following text, both local and global features will refer to the same lymph node.

[0071] Specifically, the lymph node metastasis status of each lymph node is determined by jointly analyzing each pair of local and global features corresponding to each lymph node. Based on the lymph node metastasis status of multiple lymph nodes in the 3D lymph node image, the lymph node metastasis status in the lymph node metastasis prediction result of the 3D lymph node image is determined. In some embodiments, the lymph node metastasis prediction result includes, but is not limited to, the density of lymph nodes contained in the region of interest, the region of interest identifier, the size parameters of the region of interest, and the lymph node metastasis status.

[0072] In some embodiments, the lymph node metastasis prediction result includes the lymph node metastasis status, which includes lymph node metastasis and lymph node non-metastasis. The lymph node metastasis status can be identified in the lymph node metastasis prediction result using preset identifiers. These preset identifiers include, but are not limited to, color identifiers, numerical logical identifiers, symbol identifiers, and shape identifiers. It should be noted that the preset identifiers for lymph node metastasis and lymph node non-metastasis are different. For example, the preset identifier for lymph node metastasis is "1", and the preset identifier for lymph node non-metastasis is "0". Optionally, based on an attention mechanism, local features and global features are adaptively weighted and fused to obtain fused features; the fused features are then used for prediction processing based on a multilayer perceptron network to obtain the lymph node metastasis status.

[0073] In this embodiment, the attention mechanism is a mechanism for adaptively adjusting the weights of local and global features. The attention mechanism includes, but is not limited to, self-attention mechanism, channel attention mechanism and spatial attention mechanism. The attention mechanism can be a single-head attention mechanism or a multi-head attention mechanism. This embodiment does not limit this.

[0074] Specifically, the attention algorithm corresponding to the attention mechanism is invoked, and local and global features are simultaneously input into the attention algorithm. Based on the attention algorithm, the weights of local and global features are adaptively adjusted, and a weighted sum is performed based on the local features, global features, the weights of local features, and the weights of global features to obtain a fused feature. The fused feature is the feature obtained by fusing each pair of local and global features, and each lymph node corresponds to one fused feature. The fused feature is then input into a multilayer perceptron network for prediction processing to obtain the lymph node transfer state corresponding to the fused feature. In some embodiments, the fused features corresponding to multiple nodules are simultaneously input into the multilayer perceptron network to obtain a lymph node transfer state; in some embodiments, the fused feature corresponding to each lymph node is input into the multilayer perceptron network individually to obtain the lymph node transfer state of that lymph node. If any lymph node transfer state among multiple lymph node transfer states indicates that the lymph node has transferred, the lymph node transfer state of the 3D image of that lymph node is determined to be that the lymph node has transferred. This embodiment does not impose any limitations on this approach.

[0075] The technical solution of this embodiment, by adaptively weighting and fusing local and global features based on an attention mechanism, can adaptively adjust the weights of local and global features, enhance effective features, suppress redundant features, and help improve the accuracy and generalization of lymph node metastasis status prediction.

[0076] For example, Figure 2 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 1 of the present invention. Figure 2 As shown, the contours of lymph nodes are delineated layer by layer along the axis of the three-dimensional lymph node image, and the region of interest (ROI) is determined based on the contours of the lymph nodes. Radiomics features are extracted from the ROI to obtain local features corresponding to the three-dimensional lymph node image. The ROI data is expanded to obtain pixel blocks to be processed. Interpolation is performed on the pixel blocks to be processed using a preset interpolation method to obtain pixel blocks of preset size and shape. Deep learning features are extracted from the pixel blocks to obtain global features corresponding to the three-dimensional lymph node image. Based on a self-attention mechanism, local and global features are adaptively weighted and fused to obtain fused features. Finally, the fused features are predicted using a multilayer perceptron network to obtain the lymph node transition state.

[0077] The technical solution of this embodiment involves acquiring a three-dimensional image of the lymph node to be detected and determining the region of interest (ROI) within the image; extracting features from the ROI to obtain local features corresponding to the three-dimensional image of the lymph node; expanding the ROI data to obtain pixel blocks to be processed, and extracting features from these pixel blocks to obtain global features corresponding to the three-dimensional image of the lymph node; and determining the lymph node metastasis prediction result based on the local and global features. By extracting local features corresponding to the three-dimensional image of the lymph node based on the ROI, the local features can include only the features of the lymph node itself, excluding features of the surrounding tissues, thereby reducing redundant features. By determining the lymph node metastasis prediction result based on both local and global features, where the local and global features are complementary, the features of the three-dimensional image of the lymph node are comprehensively described, solving the problem of low accuracy in lymph node metastasis prediction and helping to improve the reliability of lymph node metastasis prediction.

[0078] Example 2

[0079] Figure 3 This is a flowchart of a lymph node metastasis prediction method provided in Embodiment 2 of the present invention. The technical solution of this embodiment is an improvement upon the above embodiments. The explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here. Figure 3 As shown, the method includes:

[0080] S210. Obtain a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node.

[0081] S220. Extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node.

[0082] S230. Determine the density of lymph nodes contained in the region of interest. Based on the size parameters of the region of interest, the pixel pitch, and the density of lymph nodes, determine the outer expansion size.

[0083] In this embodiment, the expansion dimension is the size of the expanded region obtained by expanding the region of interest outward. The expansion dimension includes the expansion dimensions in three dimensions: length, width, and height. The expansion dimension corresponding to each region of interest is related to the size parameters of that region of interest, the pixel pitch, and the density of lymph nodes.

[0084] Specifically, the density of lymph nodes within the region of interest (ROI) is determined based on the average grayscale or average intensity values ​​of multiple voxels contained within the ROI. Taking a CECT scan of a 3D lymph node image as an example, the density of lymph nodes within the ROI is determined based on the average CT value of the ROI. The pixel spacing of the ROI is determined based on the 3D lymph node image; for example, the pixel spacing in both the length and width dimensions of the ROI is 0.5 mm, and the pixel spacing in the higher dimension is 1 mm. The size parameters of the ROI are determined based on its shape and size. Optionally, the maximum circumscribed cuboid of the ROI is determined, and the size parameters of the maximum circumscribed cuboid are used as the size parameters of the ROI.

[0085] In this embodiment, the maximum circumscribed cuboid is the smallest cuboid that completely contains the region of interest. The size parameters of the maximum circumscribed cuboid include the length, width, and height.

[0086] Specifically, by determining the maximum and minimum coordinate values ​​of the region of interest (ROI) in each dimension, and using the absolute difference between these values ​​as the size parameter of the maximum circumscribed cuboid in that dimension, the size parameters of the ROI in the length, width, and height dimensions are obtained. For example, assuming the ROI is the area enclosed by points (1,3,5), (2,4,6), and (5,4,3), the maximum coordinate value of the ROI in the length dimension is 5, and the minimum coordinate value is 1. Therefore, the size parameter of the maximum circumscribed cuboid in the length dimension is 4, and this 4 is set as the size parameter of the ROI in the length dimension.

[0087] In this embodiment, the maximum circumscribed cuboid of the region of interest (ROI) can concisely and intuitively represent the boundary of the ROI. By using the size parameters of the maximum circumscribed cuboid of the ROI, instead of other complex shapes, as the size parameters of the ROI, the calculation of the ROI size parameters can be simplified, computational resources can be saved, and the efficiency of determining the ROI size parameters can be improved.

[0088] The outer size mapping process is performed based on the size parameters of the region of interest, pixel spacing, and lymph node density to obtain the outer size corresponding to the region of interest. The outer size mapping process includes, but is not limited to, interval merging mapping and function mapping.

[0089] In some embodiments, a region of interest (ROI) data item is obtained by combining the size parameter of the ROI, pixel pitch, and lymph node density. The ROI data item is compared with multiple preset data item ranges. If the ROI data item is within any preset data item range, the preset expansion size corresponding to the preset data item range is determined as the expansion size of the ROI, thereby achieving interval merging mapping. The preset data item range is the range of the ROI data item used to determine the expansion size of the ROI. It can be understood that the preset data item range includes three dimensions: the size parameter of the ROI, pixel pitch, and lymph node density. Each preset data item range corresponds to an expansion size. For example, suppose there exists a preset data item range E consisting of: size parameters [A1,A2], pixel spacing [B1,B2], and lymph node density [C1,C2]. The outer dimension corresponding to this preset data item range is d, and the size parameters a∈[A1,A2], pixel spacing b∈[B1,B2], and lymph node density c∈[C1,C2] of the region of interest, then the data item {a,b,c} of the region of interest can be considered to be within the preset data item range E, and the outer dimension of the region of interest is determined to be d.

[0090] In some embodiments, the outward expansion size is calculated based on the size parameters of the region of interest (ROI), the pixel spacing, and the lymph node density to obtain the outward expansion size of the ROI in the three dimensions of length, width, and height, thereby achieving function mapping. For example, the outward expansion size of the ROI is obtained by calculating the size parameters of the ROI, the pixel spacing, and the lymph node density using the following formulas: Outward expansion size W_L = α × l × s_l + β × l × mean_hu in the length dimension; Outward expansion size W_W = α × w × s_w + β × w × mean_hu in the width dimension; Outward expansion size W_H = α × h × s_h + β × h × mean_hu in the height dimension, where the size parameters of the ROI l, w, and h are l, w, and h, respectively; the pixel spacings are s_l, s_w, and s_h, respectively; the lymph node density contained in the ROI is mean_hu; and α and β are adjustment factors ranging from 0 to 1.

[0091] S240. Based on the expanded size and the three-dimensional image of the lymph nodes, the region of interest is expanded along at least one preset direction to obtain the pixel block to be processed.

[0092] In this embodiment, the preset direction is a pre-set outward expansion direction, which can be the positive X-axis (long dimension), negative X-axis, positive Y-axis (wide dimension), negative Y-axis, positive Z-axis (high dimension), or negative Z-axis of the three-dimensional lymph node image.

[0093] Specifically, for each dimension, the size parameter of the region of interest in that dimension is compared with the expansion size of that dimension. If the expansion size of the region of interest in that dimension is greater than the size parameter, the region of interest is expanded along the preset direction corresponding to that dimension, so that the size parameter of the expanded region in that dimension is equal to the expansion size of that dimension.

[0094] For example, suppose the region of interest has three preset directions: reverse X-axis, reverse Y-axis, and reverse Z-axis. The dimensions of the region of interest and its expansion are 2 (length), 4 (width), 6 (height) and 2 (length), 4 (width), 7 (height), respectively. Then, by expanding the region of interest outward by 1 along the reverse Z-axis, we obtain an expanded region with dimensions of 2, 3, and 7 in length, width, and height. The pixel blocks containing the expanded region in the 3D lymph node image are then cropped to obtain the pixel blocks to be processed.

[0095] In some embodiments, the region of interest is set at the center of the pixel block to be processed. Optionally, the center point of the region of interest is determined. The three-dimensional image of the lymph node is cropped based on the outer expansion size, with the center point of the region of interest as the center, to obtain the pixel block to be processed.

[0096] In this embodiment, the center point is the point that represents the center of the region of interest. It can be understood that if the shape of the region of interest is irregular, the geometric center point of the regular-shaped region corresponding to the region of interest can be determined as the center point of the region of interest.

[0097] Specifically, based on regular shapes, a regular shape region corresponding to the region of interest (ROI) is constructed, and the geometric center point of the constructed inscribed or circumscribed regular shape is determined as the center point of the ROI. Regular shapes include, but are not limited to, cuboids and spheres. For example, if the regular shape is a cuboid, the regular shape region corresponding to the ROI is the maximum circumscribed cuboid, and the center point is the intersection of the two body diagonals of the maximum circumscribed cuboid. Based on the center point and expansion dimension of the ROI, an expansion region centered on the center point of the ROI and with the expansion dimension as the size parameter is determined in the 3D lymph node image. Pixel blocks containing the expansion region in the 3D lymph node image are cropped to obtain the pixel blocks to be processed. It can be understood that the preset direction corresponding to each dimension includes both the positive and negative directions of the corresponding axis. Taking the long dimension as an example, assuming the size parameter and expansion dimension of the ROI in the long dimension are 4 and 6 respectively, then the ROI is expanded outward by 1 ohm along the positive X-axis and outward by 1 ohm along the negative X-axis.

[0098] The technical solution of this embodiment, by cropping the three-dimensional image of the lymph node based on the outer expansion size with the center point of the region of interest as the center, obtains the pixel block to be processed. This ensures that the region of interest is located at the center of the pixel block to be processed, reduces the impact of edge effects and boundary conditions on the global features extracted based on the pixel block to be processed, and improves the reliability and stability of the global features.

[0099] S250. Extract features from the pixel blocks to be processed to obtain global features corresponding to the three-dimensional image of the lymph node.

[0100] S260. Based on local and global features, determine the lymph node metastasis prediction results of the three-dimensional lymph node image.

[0101] The technical solution of this embodiment determines the density of lymph nodes contained in the region of interest (ROI), and determines the expansion size based on the size parameters of the ROI, pixel spacing, and lymph node density. Based on the expansion size and the three-dimensional image of the lymph nodes, the ROI is expanded along at least one preset direction to obtain the pixel block to be processed. The expansion size is determined based on the size parameters of the ROI, pixel spacing, and lymph node density. Lymph nodes with different metastatic conditions have different morphologies and characteristics. By comprehensively considering the biological characteristics of the lymph nodes (lymph node density), image features (pixel spacing), and the physical scale of the image (size parameters), rather than relying solely on the physical scale of the image, the generalization of the expansion processing is improved. Adaptive expansion processing can be performed on lymph nodes with different metastatic conditions, and the size consistency between the expanded pixel block and the actual lymph node is enhanced, which helps to improve the robustness and generalization of lymph node metastasis prediction. Furthermore, by considering image features (pixel spacing), the expansion size can be made consistent in physical space, ensuring that the pixel block to be processed maintains the geometric shape and proportional relationship of the three-dimensional image of the lymph nodes.

[0102] Example 3

[0103] Figure 4 This is a schematic diagram of a lymph node metastasis prediction device provided in Embodiment 3 of the present invention. Figure 4 As shown, the device includes:

[0104] The region of interest determination module 310 is used to acquire a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node.

[0105] The local feature extraction module 320 is used to extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node.

[0106] The global feature extraction module 330 is used to expand the data of the region of interest to obtain the pixel block to be processed, and to extract features from the pixel block to obtain the global features corresponding to the three-dimensional image of the lymph node.

[0107] The lymph node metastasis prediction module 340 is used to determine the lymph node metastasis prediction results of the three-dimensional image of lymph nodes based on local and global features.

[0108] The technical solution of this embodiment involves acquiring a three-dimensional image of the lymph node to be detected and determining the region of interest (ROI) within the image; extracting features from the ROI to obtain local features corresponding to the three-dimensional image of the lymph node; expanding the ROI data to obtain pixel blocks to be processed, and extracting features from these pixel blocks to obtain global features corresponding to the three-dimensional image of the lymph node; and determining the lymph node metastasis prediction result based on the local and global features. By extracting local features corresponding to the three-dimensional image of the lymph node based on the ROI, the local features can include only the features of the lymph node itself, excluding features of the surrounding tissues, thereby reducing redundant features. By determining the lymph node metastasis prediction result based on both local and global features, where the local and global features are complementary, the features of the three-dimensional image of the lymph node are comprehensively described, solving the problem of low accuracy in lymph node metastasis prediction and helping to improve the reliability of lymph node metastasis prediction.

[0109] Based on the above embodiments, optionally, the lymph node metastasis prediction device further includes a pixel block interpolation module; wherein, the pixel block interpolation module is used to perform interpolation processing on the pixel block to be processed based on a preset interpolation method before the global feature extraction module 330, so as to obtain a pixel block to be processed with a preset size and a preset shape, wherein the preset shape includes a cube.

[0110] Based on the above embodiments, optionally, the global feature extraction module 330 includes: an outer expansion size determination submodule and an outer expansion processing submodule; wherein,

[0111] The outer dimension determination submodule is used to: determine the density of lymph nodes contained in the region of interest, and determine the outer dimension based on the size parameters of the region of interest, the pixel pitch, and the density of lymph nodes;

[0112] The outward expansion processing submodule is used to perform outward expansion processing on the region of interest along at least one preset direction based on the outward expansion size and the three-dimensional image of the lymph nodes, so as to obtain the pixel block to be processed.

[0113] Based on the above embodiments, optionally, the global feature extraction module 330 further includes: a size parameter determination submodule; wherein, the size parameter determination submodule is used to determine the largest circumscribed cuboid of the region of interest before the outer size determination submodule, and use the size parameter of the largest circumscribed cuboid as the size parameter of the region of interest.

[0114] Based on the above embodiments, optionally, the outward expansion processing submodule is specifically used for: determining the center point of the region of interest; and cropping the three-dimensional image of the lymph node based on the outward expansion size, with the center point of the region of interest as the center, to obtain the pixel block to be processed.

[0115] Based on the above embodiments, optionally, the lymph node metastasis prediction result includes the lymph node metastasis status, which includes lymph node metastasis and lymph node non-metastasis.

[0116] Based on the above embodiments, optionally, the lymph node metastasis prediction module 340 is specifically used to: adaptively weight and fuse local features and global features based on an attention mechanism to obtain fused features; and perform prediction processing on the fused features based on a multilayer perceptron network to obtain the lymph node metastasis status.

[0117] Based on the above embodiments, optionally, the region of interest determination module 310 is specifically used to: delineate the contours of multiple lymph node two-dimensional images layer by layer along the axis of the lymph node three-dimensional image, and determine the region of interest based on the contours of the lymph nodes; the lymph node three-dimensional image is composed of multiple lymph node two-dimensional images stacked axially; and / or, perform lymph node segmentation processing on the lymph node three-dimensional image based on the lymph node segmentation model to obtain the region of interest.

[0118] Based on the above embodiments, optionally, the local features include at least one of the following: first-order features, shape features, gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level size region moment features, adjacent gray-level difference matrix features, and gray-level correlation matrix features.

[0119] The lymph node metastasis prediction device provided in the embodiments of the present invention can execute the lymph node metastasis prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0120] Example 4

[0121] Figure 5This is a schematic diagram of an electronic device implementing the lymph node metastasis prediction method of this invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0122] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0123] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0124] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as lymph node metastasis prediction methods.

[0125] In some embodiments, the lymph node metastasis prediction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the lymph node metastasis prediction method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the lymph node metastasis prediction method by any other suitable means (e.g., by means of firmware).

[0126] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0127] Computer programs for implementing the lymph node metastasis prediction method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0128] Example 5

[0129] Embodiment 5 of the present invention also provides a computer-readable storage medium storing computer instructions for causing a processor to execute a lymph node metastasis prediction method, the method comprising:

[0130] A three-dimensional image of the lymph node to be detected is acquired, and the region of interest (ROI) in the three-dimensional image of the lymph node is determined. Feature extraction is performed on the ROI to obtain local features corresponding to the three-dimensional image of the lymph node. The ROI data is expanded to obtain pixel blocks to be processed, and feature extraction is performed on the pixel blocks to obtain global features corresponding to the three-dimensional image of the lymph node. Based on the local and global features, the lymph node metastasis prediction result of the three-dimensional image of the lymph node is determined.

[0131] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0132] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0133] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0134] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0135] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0136] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for predicting lymph node metastasis, characterized in that, include: Acquire a three-dimensional image of the lymph node to be detected, and determine the region of interest in the three-dimensional image of the lymph node; Feature extraction is performed on the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node; The region of interest data is expanded to obtain a pixel block to be processed, and features are extracted from the pixel block to obtain global features corresponding to the three-dimensional image of the lymph node, wherein the pixel block to be processed contains the region of interest; Based on the local features and the global features, the lymph node metastasis prediction result of the three-dimensional lymph node image is determined; The step of expanding the region of interest to obtain the pixel block to be processed includes: The density of lymph nodes contained in the region of interest is determined, and the outer expansion size is determined based on the size parameters of the region of interest, the pixel pitch, and the density of the lymph nodes; Based on the expanded dimensions and the three-dimensional image of the lymph node, the region of interest is expanded along at least one preset direction to obtain the pixel block to be processed, wherein the region of interest is located at the center of the pixel block to be processed.

2. The method according to claim 1, characterized in that, Before performing feature extraction on the pixel block to be processed, the method further includes: The pixel block to be processed is interpolated based on a preset interpolation method to obtain the pixel block to be processed with a preset size and a preset shape, wherein the preset shape includes a cube.

3. The method according to claim 1, characterized in that, Before determining the external expansion dimensions, the following is also included: Determine the largest circumscribed cuboid of the region of interest, and use the size parameters of the largest circumscribed cuboid as the size parameters of the region of interest.

4. The method according to claim 1, characterized in that, The process of expanding the region of interest based on the expanded size and the three-dimensional image of the lymph nodes to obtain the pixel block to be processed includes: Determine the center point of the region of interest; Centered on the center point of the region of interest, and based on the outward expansion size, the three-dimensional image of the lymph node is cropped to obtain the pixel block to be processed.

5. The method according to claim 1, characterized in that, The lymph node metastasis prediction result includes the lymph node metastasis status, which includes lymph node metastasis and lymph node non-metastasis; the step of determining the lymph node metastasis prediction result of the three-dimensional image of the lymph nodes based on the local features and the global features includes: Based on the attention mechanism, the local features and the global features are adaptively weighted and fused to obtain the fused features; The lymph node metastasis status is obtained by predicting the fused features based on a multilayer perceptron network.

6. The method according to claim 1, characterized in that, Determining the region of interest in the three-dimensional image of the lymph node includes: Along the axial direction of the three-dimensional lymph node image, the contours of multiple two-dimensional lymph node images are delineated layer by layer, and the region of interest is determined based on the contours of the lymph nodes; the three-dimensional lymph node image is formed by stacking multiple two-dimensional lymph node images axially; and / or, Based on the lymph node segmentation model, the three-dimensional image of the lymph nodes is segmented to obtain the region of interest.

7. The method according to claim 1, characterized in that, The local features include at least one of the following: first-order features, shape features, gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level size region moment features, adjacent gray-level difference matrix features, and gray-level correlation matrix features.

8. A lymph node metastasis prediction device, characterized in that, include: The region of interest determination module is used to acquire a three-dimensional image of the lymph node to be detected and determine the region of interest in the three-dimensional image of the lymph node. The local feature extraction module is used to extract features from the region of interest to obtain local features corresponding to the three-dimensional image of the lymph node; A global feature extraction module is used to expand the data of the region of interest to obtain a pixel block to be processed, and to extract features from the pixel block to obtain global features corresponding to the three-dimensional image of the lymph node, wherein the pixel block to be processed contains the region of interest; The lymph node metastasis prediction module is used to determine the lymph node metastasis prediction result of the three-dimensional image of the lymph nodes based on the local features and the global features; The global feature extraction module includes: an outer dimension determination submodule and an outer dimension processing submodule; wherein, The outer dimension determination submodule is used to: determine the density of lymph nodes contained in the region of interest, and determine the outer dimension based on the size parameters of the region of interest, the pixel pitch, and the density of lymph nodes; An expansion processing submodule is used to expand the region of interest along at least one preset direction based on the expansion size and the three-dimensional image of the lymph node to obtain a pixel block to be processed, wherein the region of interest is located at the center of the pixel block to be processed.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the lymph node metastasis prediction method according to any one of claims 1-7.

Images