Lymph node detection, model training methods, equipment and media

Abstract

The embodiments of the present application provide a lymph node detection and model training method, device and medium. In the lymph node detection method, the lymph node classification information of the scanned slice can be identified based on the image features of the scanned slice and the learned characteristic distribution of different lymph nodes. By combining the lymph node classification information, image features and the learned characteristic distribution of lymph nodes in different lymph nodes, lymph nodes can be detected and segmented from the scanned slice. Among them, the lymph node classification information can effectively guide the identification process of lymph nodes in different lymph nodes. Combining the lymph node classification information, image features and the characteristic distribution of lymph nodes in different lymph nodes to identify lymph nodes in the scanned slice is conducive to accurately distinguishing the subtle differences between lymph nodes and non-lymph nodes in different lymph nodes, reducing the interference of non-lymph nodes in different lymph nodes on the identification process, and thus accurately detecting and segmenting lymph nodes in different lymph nodes.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a lymph node detection and model training method, device, and medium. Background Art

[0002] Computer Aided Diagnosis (CAD) technology can help doctors detect lesions based on the powerful analytical and computing capabilities of computers, combined with imaging, medical image processing technology and other possible physiological and biochemical methods.

[0003] In the process of cancer grading and treatment, finding abnormal lymph nodes (LN) is a crucial task. Currently, computed tomography (CT) technology can be used to obtain images containing lymph nodes. Physicians can identify lymph nodes from CT images and analyze whether there are abnormalities in the lymph nodes based on the CT images of the lymph nodes. In existing CAD technology, there is a method for identifying lymph nodes from CT images, but this method cannot accurately distinguish between lymph nodes and surrounding soft tissues. Therefore, a new solution is needed. Summary of the Invention

[0004] Various aspects of the present application provide a lymph node detection and model training method, device, and medium to improve the accuracy of lymph node recognition in medical images.

[0005] An embodiment of the present application provides a method for detecting lymph nodes in medical images, comprising: obtaining a scanned slice containing lymph nodes; obtaining image features of the scanned slice using a neural network model; identifying lymph node classification information of the scanned slice based on the image features and learned characteristic distributions of different lymph nodes; and detecting lymph nodes from the scanned slice by combining the lymph node classification information, the image features, and the learned characteristic distributions of lymph nodes in different lymph nodes.

[0006] An embodiment of the present application also provides a method for detecting lymph nodes in medical images, including: responding to a call request from a client to a first interface, obtaining a scanned slice containing lymph nodes from the interface parameters of the first interface; identifying lymph node classification information of the scanned slice based on the image features and the learned characteristic distribution of different lymph nodes; detecting lymph nodes from the scanned slice by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes; and sending the location information of the lymph nodes to the client so that the client marks the lymph nodes in the scanned slice.

[0007] An embodiment of the present application also provides a neural network model training method, comprising: obtaining a training data set, the training data set comprising: a scanned slice sample containing lymph nodes; using a neural network model, obtaining image features of the scanned slice sample; identifying lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes; obtaining a lymph node detection result of the scanned slice sample based on the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes; and iteratively optimizing the neural network model based on the error between the lymph node detection result and the lymph node classification label of the scanned slice sample.

[0008] An embodiment of the present application also provides an electronic device, comprising: a memory and a processor; the memory is used to store one or more computer instructions; the processor is used to execute the one or more computer instructions to: execute the steps in the method provided in the embodiment of the present application.

[0009] The embodiment of the present application also provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps in the method provided in the embodiment of the present application.

[0010] In the lymph node detection method of medical imaging provided in the embodiment of the present application, after obtaining a scan slice containing lymph nodes, a neural network can be used to obtain the image features of the scan slice, and based on the image features and the learned characteristic distribution of different lymph nodes, the lymph node classification information of the scan slice can be identified. By combining the lymph node classification information, image features and the learned characteristic distribution of lymph nodes in different lymph nodes, lymph nodes can be detected from the scan slice. Among them, the lymph node classification information can effectively guide the identification process of lymph nodes in different lymph nodes. Combining the lymph node classification information, image features and the characteristic distribution of lymph nodes in different lymph nodes to identify lymph nodes in the scan slice is conducive to accurately distinguishing the subtle differences between lymph nodes and non-lymph nodes in different lymph nodes, reducing the interference of non-lymph nodes in different lymph nodes on the identification process, and thus accurately detecting lymph nodes in different lymph nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0012] Figure 1 A schematic flow chart of a method for detecting lymph nodes in medical images provided as an exemplary embodiment of the present application;

[0013] Figure 2A schematic diagram of the structure of a neural network model provided for an exemplary embodiment of the present application;

[0014] Figure 3 A schematic diagram of a lymph node segmentation result provided by an exemplary embodiment of the present application;

[0015] Figure 4 A flowchart of a neural network model training method provided by an exemplary embodiment of the present application;

[0016] Figure 5 A schematic diagram of an application scenario provided for an exemplary embodiment of the present application;

[0017] Figure 6 A schematic structural diagram of an electronic device provided as an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0018] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] Existing CAD technologies include a method for identifying lymph nodes from CT images. This method typically uses all lymph nodes within a lymphatic station as positive samples and other soft tissues as negative samples when training the classification model. However, this method cannot accurately distinguish between lymph nodes and surrounding soft tissue, and the accuracy of image detection needs to be improved.

[0020] In view of the above-mentioned technical problems existing in the prior art, a solution is provided in some embodiments of the present application. The technical solutions provided in each embodiment of the present application are described in detail below with reference to the accompanying drawings.

[0021] Figure 1 A schematic diagram of a flow chart of a method for detecting lymph nodes in medical images provided as an exemplary embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0022] Step 101: Obtain a scan slice containing lymph nodes.

[0023] Step 102: Utilize a neural network model to obtain image features of the scanned slice.

[0024] Step 103: Identify the lymph node classification information of the scanned slice based on the image features and the learned feature distribution of different lymph nodes.

[0025] Step 104 : Detect lymph nodes from the scan slice by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes.

[0026] The lymph node detection method provided in this embodiment can be used to distinguish lymph nodes from non-lymph nodes (such as soft tissue around lymph nodes) from medical images. Among them, the scan slice refers to an image obtained by performing axial tomographic imaging of a part that may contain lymph nodes based on CT technology. When CT scanning imaging is performed multiple times in succession, multiple consecutive axial scan slices can be obtained. After the lymph node is imaged, its characteristics can be reflected on the corresponding scan slice. Therefore, based on the characteristics shown in the scan slice, the lymph node structure contained therein can be identified.

[0027] In medicine, lymph nodes are divided into different lymph nodes, and different lymph nodes have multiple lymph nodes. For example, taking gastric cancer as an example, gastric cancer cells generally metastasize from near to far and from shallow to deep along the order of lymphatic drainage around the stomach. Sometimes, retrograde metastasis or jump metastasis may occur due to lymphatic obstruction. According to the different distribution of lymph nodes metastasized to gastric cancer in different parts of the body, the lymph nodes around the stomach are usually divided into three stations: the first station (N1), the second station (N2), and the third station (N3). Each station of lymph nodes is composed of several groups of lymph nodes. The first station is the lymph nodes around the tumor, the second station is the lymph nodes slightly farther away from the tumor, and the third station is the lymph nodes far away from the tumor.

[0028] Lymph nodes in different lymphatic stations are surrounded by different organs or soft tissues. Therefore, the appearance and background of lymph nodes in different lymphatic stations vary greatly after CT imaging. Based on this, the station information of lymph nodes can be combined to classify different objects in medical images to accurately detect lymph nodes and non-lymph nodes in different lymphatic stations.

[0029] In this embodiment, a neural network model can be pre-trained to learn the characteristic distributions of different lymph nodes. Thus, the neural network model can identify the lymph node classification information of the candidate region based on the image features of the candidate region and the learned characteristic distributions of different lymph nodes. Lymph node classification information describes the classification results of the scan slice at the lymph node level. For example, in the gastric cancer scenario, the lymph node classification information of the candidate region may be: the probability that the candidate region belongs to the first station is 80%, the probability that the candidate region belongs to the second station is 10%, and so on.

[0030] After obtaining the lymph node classification information for the scanned slice, lymph nodes can be detected from the scanned slice by combining this information, the image features of the scanned slice, and the learned characteristic distributions of lymph nodes within different lymph nodes. When there are multiple lymph nodes, the characteristic distributions of lymph nodes within each lymph node can be learned separately. Continuing with the gastric cancer example mentioned above, the characteristic distributions of lymph nodes within the first station can be learned, the characteristic distributions of lymph nodes within the second station can be learned, and the characteristic distributions of lymph nodes within the third station can be learned. This allows the common features of lymph nodes within a station to be abstracted, reducing the impact of differences in appearance and background between lymph nodes at different stations on detection results.

[0031] In some embodiments, based on the detection results of the lymph nodes, the lymph nodes may be segmented from the scan slice to facilitate lesion analysis.

[0032] In this embodiment, after obtaining a scan slice containing lymph nodes, a neural network can be used to obtain the image features of the scan slice, and based on the image features and the learned characteristic distribution of different lymph nodes, the lymph node classification information of the scan slice can be identified. By combining the lymph node classification information, image features, and the learned characteristic distribution of lymph nodes in different lymph nodes, lymph nodes can be detected from the scan slice. Among them, the lymph node classification information can effectively guide the identification process of lymph nodes in different lymph nodes. Combining the lymph node classification information, image features, and the characteristic distribution of lymph nodes in different lymph nodes to identify lymph nodes in the scan slice is conducive to accurately distinguishing the subtle differences between lymph nodes and non-lymph nodes in different lymph nodes, reducing the interference caused by non-lymph nodes in different lymph nodes to the identification process, and thus accurately detecting lymph nodes in different lymph nodes.

[0033] In the above and following embodiments of the present application, the process of identifying medical images can be implemented using a machine learning model based on neural networks (NN). The neural network may include, but is not limited to, one or more of a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), and a long short-term memory neural network (LSTM), or may be implemented based on a modified network of one or more of the above networks, which is not limited in this embodiment.

[0034] For ease of description, in the following embodiments of this application, the model used to identify medical images is described as a lymph node classification model. In some optional embodiments, the lymph node classification model can be implemented as Figure 2The "end-to-end" neural network model shown in Figure 2 As shown in the figure, the lymph node structure recognition model can be composed of a feature extraction network (i.e., backbone network), a candidate region selection network (Proposal Network), a lymph node classification network (Station branch), a multi-head lymph node classification network (Multi-head LN detection branch) and a fully connected layer. Figure 2 An exemplary description is given.

[0035] Among them, the feature extraction network is used to obtain the image features of the scanned slices.

[0036] In some optional embodiments, the image features of the scan slice are obtained by performing two-dimensional (2D) image feature extraction on the scan slice.

[0037] In other embodiments, in addition to appearance features, lymph nodes have certain spatial structural features. To improve the distinction between lymph nodes and other tubular organs (such as blood vessels and esophagus), the spatial structural features of lymph nodes can be obtained to improve the accuracy of lymph node recognition. Optionally, the spatial structural features can be obtained by scanning a slice and its adjacent slices.

[0038] The adjacent slices of a scan slice may include multiple preceding and / or subsequent scan slices of the scan slice, and this embodiment does not impose any limitation thereto. For example, in some embodiments, for a scan slice, the four preceding and four subsequent adjacent slices may be obtained, resulting in nine consecutive scan slices. These nine consecutive scan slices may be input as a group of slices into a neural network model.

[0039] In the neural network model, the feature extraction network can be used to extract 2D image features for each scan slice and its adjacent slices. The 2D image features of the scan slice reflect the appearance and background features of the scan slice. The 2D image features of the adjacent slices reflect the contextual appearance and background features of the scan slice. The 2D image features of the scan slice and its adjacent slices are fused to obtain a contextual fused feature of the scan slice, which serves as the image feature of the scan slice. This contextual fused feature includes both the 2D features of the scan slice and its 3D spatial structural features.

[0040] Optionally, the feature extraction network can be a 2.5D network. Compared to a 2D network, a 2.5D network can utilize the 3D information of the image. Compared to a 3D network, a 2.5D network is easier to converge and has higher accuracy, thus improving training efficiency and the performance of the trained neural network model.

[0041] After obtaining the image features of the scanned slice based on the above embodiment, in the candidate region selection network of the neural network model, the pre-learned model parameters can be used to extract at least one candidate region from the scanned slice according to the image features of the scanned slice. The candidate region refers to the region on the scanned slice where lymph nodes may exist. When the candidate region selection network selects any candidate region, it can obtain the location information of the candidate region. The candidate region selection network can learn the contextual image features of the lymph nodes, but it is difficult to distinguish the subtle structures in different lymph nodes. After determining the at least one candidate region, it can be further identified whether there are lymph nodes in the candidate region. Optionally, the candidate region selection network can be implemented based on a one-stage fully convolutional detector (FCOS).

[0042] The following will take any candidate region of the at least one candidate region as an example to exemplify an optional implementation of identifying whether a lymph node exists in the candidate region.

[0043] like Figure 2 As shown, the input of the lymph node classification network is connected to the feature extraction network and the candidate region selection network. For any of the at least one candidate region, the location information of the candidate region and the image features of the scanned slice can be input into the lymph node classification network. The lymph node classification network can then identify a first probability distribution indicating that the candidate region belongs to multiple different lymph nodes based on the image features of the scanned slice, the location information of the candidate region, and the learned feature distributions of different lymph nodes, and use this probability distribution as the lymph node classification information for the candidate region.

[0044] When there are multiple types of lymph nodes, the lymph node classification network can be a multi-classifier to classify candidate regions into different lymph nodes. During the training phase, the lymph node classification network learns the characteristic distribution of different lymph nodes based on the input scan slice samples of different lymph nodes and the supervision signal. The learned characteristic distribution of different lymph nodes is reflected in the model parameters of the lymph node classification network.

[0045] Continuing with the example of gastric cancer, lymph nodes are divided into the first, second, and third stations. In the lymph node classification model for gastric cancer, the lymph node classification network can be a three-category network. This three-category network learns the feature distributions of the first, second, and third stations, respectively, to obtain model parameters. Based on the learned model parameters, the lymph node classification network outputs the probability that a candidate region belongs to the first station, the second station, and the third station.

[0046] The characteristics of lymph nodes vary from station to station. For example, in gastric cancer, the appearance and structural characteristics of lymph nodes in the first station are different from those in the second and third stations; the structural characteristics of lymph nodes in the second station are also different from those in the third station.

[0047] Based on this, in this embodiment, the lymph node classification networks for each of the multiple lymph nodes can be trained separately, allowing each network to learn the characteristic distribution of lymph nodes within each lymph node. Thus, the neural network model can effectively distinguish between lymph nodes and non-lymph nodes (such as blood vessels, connective tissue, and esophageal tumors) within different lymph nodes based on the learned characteristics.

[0048] The neural network model includes a multi-head lymph node classification network, which corresponds to multiple lymph nodes. For example, when the lymph nodes to be classified are divided into c lymph nodes, c lymph node classification networks can be trained to accurately identify the lymph nodes in c lymph nodes, where c is a positive integer. In the c lymph node classification networks, each lymph node classification network is used to perform the classification task of the lymph nodes in the station. Figure 2 As shown, the input end of the multi-head lymph node classification network is connected to the feature extraction network and the candidate region selection network, which can distinguish lymph nodes from non-lymph nodes based on the image features of the input scan slices and the position information of the candidate regions.

[0049] Continuing to take any candidate region as an example, the lymph node classification process is exemplified.

[0050] In some optional Example A In the present invention, when detecting lymph nodes from the scan slice, the image features of the scan slice and the location information of the candidate region are input into a multi-head lymph node classification network, combining the lymph node classification information, the image features of the scan slice, and the location information of the candidate region. The multi-head lymph node classification network is used to calculate the lymph node classification information of the candidate region within the multiple different lymph nodes, combining the image features of the scan slice and the location information of the candidate region. The lymph node classification information of the candidate region within a particular lymph node is used to describe the probability that an object within the candidate region is classified as a lymph node and / or the probability that an object within the candidate region is classified as a non-lymph node within that lymph node.

[0051] For example, the jth lymph node classification network among c lymph node classification networks can calculate the classification information of the candidate region in the jth lymph station, where j = 1, 2, ..., c. The following uses the lymph node classification network corresponding to any lymph station in a multi-head lymph node classification network as an example for illustration.

[0052] Optionally, in a lymph node classification network corresponding to a lymph node station, the probability that the candidate region is a lymph node within the lymph node station can be calculated based on the input image features of the scan slice, the location information of the candidate region, and the characteristic distribution of lymph nodes within the lymph node station learned by the lymph node classification network. For example, the lymph node classification information output by the j-th lymph node classification network for a candidate region within the j-th lymph node station may be: the probability of being a lymph node is 80%, and the probability of not being a lymph node is 20%.

[0053] Based on the above embodiment, the lymph node detection results of the lymph node classification network corresponding to each lymph node station for the candidate region can be obtained. A second probability distribution is determined based on the probability that the candidate region output by the multi-head lymph node classification network is a lymph node in the multiple lymph nodes.

[0054] According to the lymph node classification information of the candidate region and the lymph node classification information of the candidate region in multiple different lymph nodes, the candidate region can be determined to be a target candidate region containing lymph nodes.

[0055] Optionally, a weighted calculation can be performed using the first probability distribution corresponding to the candidate region and the second probability distribution corresponding to the candidate region to obtain a weighted score. The weighted score calculation operation can be performed at the fully connected layer. The weighted score represents the likelihood that the candidate region contains a lymph node. If the weighted score meets a set score condition, the candidate region is determined to be a target candidate region containing a lymph node. This will be explained below with specific examples.

[0056] Assume that the candidate region selection network selects n candidate regions from the scan slice, where n is a positive integer. For the i-th candidate region, i∈[1,n], the probability of it being classified as the j-th lymph node station is t ij , the probability of being classified as a lymph node in the jth lymph station is s ij , then the weighted score s of the i-th candidate region final It can be expressed using the following formula 1:

[0057]

[0058] If the weighted score satisfies a set score condition, the candidate region is determined to be a target candidate region containing a lymph node. The set score condition may be: the weighted score of the i-th candidate region is greater than a specified score threshold; or, among the n candidate regions, the i-th candidate region has the highest weighted score.

[0059] Optionally, in this embodiment, after determining that the candidate region contains a lymph node, a detection frame of the lymph node can be determined based on the bounding box regression layer, and the detection frame of the lymph node can be output on the scan slice to indicate the location of the lymph node in the scan slice, such as Figure 3 As shown. Optionally, after determining that the candidate region is a target candidate region containing a lymph node, the target lymph node to which the lymph node belongs can be determined from the multiple different lymph nodes based on the first probability distribution, and the identifier of the target lymph node can be output. For example, if it is determined that the i-th candidate region contains a lymph node, and if the first probability distribution shows that the candidate region belongs to a certain lymph node a, the identifier of lymph node a can be output to indicate that the i-th candidate region contains a lymph node in lymph node a.

[0060] In this embodiment, lymph nodes can be distinguished from other soft tissues based on the lymph node classification information of the candidate area and the characteristic distribution of lymph nodes learned by multiple lymph node classification networks. The differences between different lymph nodes can be fully utilized to accurately classify lymph nodes and non-lymph nodes within a lymph node. At the same time, the lymph node detection results are determined based on the output results of the lymph node classification network and the c lymph node classification networks, which has higher stability and accuracy. Figure 3 As shown, the lymph node detection method provided in the embodiment of the present application can accurately detect lymph nodes and non-lymph nodes when there are slight differences in appearance between lymph nodes and non-lymph nodes, greatly improving the detection efficiency and accuracy of lymph nodes.

[0061] Continuing to take any candidate region as an example, the lymph node classification process is exemplified.

[0062] In some optional Example B In the present invention, when identifying lymph nodes from the scan slice, the lymph node classification information and the image features are combined, and a target lymph node that meets the preset probability conditions can be determined from the multiple different lymph nodes according to the first probability distribution, and the image features of the scan slice and the location information of the candidate region are input into the lymph node classification network corresponding to the target lymph node. Among them, the second probability condition can be: the probability that the candidate region belongs to the target lymph node is the highest, or the probability that the candidate region belongs to the target lymph node is greater than a preset probability threshold. The probability threshold can be 85% or 90%, etc., and this embodiment does not impose any restrictions.

[0063] In the lymph node classification network of the target lymph node station, the image features of the candidate area are determined based on the image features of the scan slice and the position information of the candidate area; and the third probability that the candidate area is a lymph node in the target lymph node station is calculated based on the image features of the candidate area and the characteristic distribution of the lymph nodes in the station learned by the lymph node classification network; if the third probability is greater than a preset third probability threshold, it is determined that the candidate area includes a lymph node in the target lymph node station.

[0064] In this embodiment, based on the lymph node classification information of the candidate area, some lymph node classification networks with higher possibilities can be selected from multiple lymph node classification networks, which can reduce the amount of calculation in the lymph node classification process and improve the classification processing efficiency.

[0065] In the above and following embodiments of the present application, after the lymph nodes in the candidate area are determined to be the lymph nodes in the target lymph station through the neural network model, the shape characteristics of the healthy lymph nodes in the target lymph station can be obtained, and the shape characteristics of the lymph nodes in the candidate area are compared with the shape characteristics of the healthy lymph nodes. If it is determined based on the results of the comparison that the lymph nodes in the candidate area are in an abnormal state, the candidate area is highlighted in the scanned slice. Among them, the method of highlighting may include: adding a mark to the candidate area, enlarging the candidate area for display, and highlighting the candidate area. At least one of the following is not limited in this embodiment. Based on this embodiment, after the lymph nodes are detected from the medical image, the types of lymph nodes can be automatically classified to more intuitively display the lymph node detection results.

[0066] In addition to the lymph node classification method provided in the aforementioned embodiment, the present embodiment further provides a method for training a neural network model for identifying lymph nodes from medical images. This method will be described below with reference to the accompanying drawings.

[0067] Figure 4 A flowchart of a neural network model training method provided by an exemplary embodiment of the present application is shown in FIG. Figure 4 As shown, the method includes:

[0068] Step 401: Acquire a training data set, where the training data set includes scanned slice samples of lymph nodes.

[0069] Step 402: Utilize a neural network model to obtain image features of the scanned slice sample; identify lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes; and obtain a lymph node detection result of the scanned slice sample by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes.

[0070] Step 403: Iteratively optimize the neural network model according to the error between the lymph node detection result and the lymph node classification label of the scanned slice sample.

[0071] Among them, the characteristic distribution of different lymph nodes and the characteristic distribution of lymph nodes in different lymph nodes learned by the neural network model can be expressed using model parameters, and the model parameters can be continuously optimized during the training process.

[0072] Optionally, the lymph node classification labels on the scanned slice samples may include: lymph node labels and lymph node labels, wherein samples marked with lymph node classification labels are positive samples, and samples not marked with such labels are negative samples.

[0073] Optionally, the training dataset may include multiple sets of training data, each of which may consist of a scanned slice sample and its neighboring slice samples. These neighboring slice samples may include multiple neighboring slice samples preceding and following the scanned slice sample. Based on these neighboring slice samples, spatial features of the object to be identified in the scanned slice can be obtained, enabling the neural network model to learn to distinguish between lymph nodes and other tubular tissues (e.g., blood vessels).

[0074] When extracting image features from the scanned slice sample using a neural network model, the scanned slice sample and its adjacent slice samples are input into the neural network model. The feature extraction network in the neural network model extracts two-dimensional image features of the scanned slice sample and the adjacent slice samples. The two-dimensional image features of the scanned slice sample and the adjacent slice samples are fused to obtain a contextual fusion feature of the scanned slice sample as the image feature of the scanned slice sample.

[0075] In subsequent embodiments, the image features of the scanned slice samples involved may be context fusion features of the scanned slice samples or two-dimensional image features of the scanned slices, which is not limited in this embodiment.

[0076] Optionally, when identifying the lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes, the candidate region selection network in the neural network model can be used to extract at least one candidate region from the scanned slice sample based on the image features of the scanned slice sample; for any candidate region of the at least one candidate region, the lymph node classification network in the neural network model can be used to identify the first probability distribution of the candidate region belonging to multiple different lymph nodes based on the image features of the scanned slice sample, the location information of the candidate region and the learned characteristic distribution of different lymph nodes, as the lymph node classification information of the candidate region.

[0077] Optionally, lymph nodes are identified from the scanned slice sample in combination with the lymph node classification information and the image features, including: inputting the image features of the scanned slice sample and the position information of the candidate area into a multi-head lymph node classification network in the neural network model; the multi-head lymph node classification network corresponds to multiple different lymph nodes; using the multi-head lymph node classification network, in combination with the image features of the scanned slice sample and the position information of the candidate area, the lymph node classification information of the candidate area in the multiple different lymph nodes is calculated; based on the lymph node classification information of the candidate area and the lymph node classification information of the candidate area in the multiple different lymph nodes, the candidate area is determined to be a target candidate area containing lymph nodes.

[0078] Optionally, the multi-head lymph node classification network is used, combined with the image features of the scanned slice sample and the position information of the candidate area, to calculate the lymph node classification information of the candidate area in the multiple different lymph stations, including: in the lymph node classification network corresponding to any lymph station, based on the image features of the scanned slice sample, the position information of the candidate area and the characteristic distribution of the lymph nodes in the station learned by the lymph node classification network, calculating the probability that the candidate area is a lymph node in the lymph station; and determining a second probability distribution based on the probability that the candidate area output by the multi-head lymph node classification network is a lymph node in the multiple different lymph stations.

[0079] Optionally, based on the lymph station classification information of the candidate area and the lymph node classification information of the candidate area in the multiple different lymph stations, the candidate area is determined to be a target candidate area containing lymph nodes, including: using the first probability distribution corresponding to the candidate area to perform a weighted calculation on the second probability distribution corresponding to the candidate area to obtain a weighted score of the candidate area; if the weighted score meets the set score condition, the candidate area is determined to be a target candidate area containing lymph nodes.

[0080] Optionally, the error between the lymph node detection result and the lymph node classification label of the scanned slice sample may include lymph node classification loss and lymph node classification loss. The lymph node classification loss L of the neural network model may be calculated based on the lymph node classification label. station , and according to the lymph node classification label, calculate the lymph node classification loss L of the neural network model cls .

[0081] Optionally, the lymph node classification loss and the lymph node classification loss can be calculated based on at least one of the cross entropy loss function, the logarithmic loss function and the square error loss function, which is not limited in this embodiment. Taking the cross entropy loss function as an example, the lymph node classification loss L cls This can be expressed as the following formula 2:

[0082]

[0083] Among them, n represents the number of candidate regions, σ represents the sigmoid function, and y i Represents the lymph node classification label. ij represents the probability that the i-th candidate region belongs to the j-th lymph node station, s ij The output of the lymph node classification network for the jth lymph station is obtained. During iterative optimization of this neural network model, convergence of the lymph node classification loss for that lymph station and that lymph node is used as the training objective. During this iterative optimization process, multiple lymph node classification networks learn the characteristic distribution of lymph nodes within their respective lymph stations and focus on mining subtle discriminant features to separate lymph nodes from non-lymph nodes within that station.

[0084] It should be noted that Figure 2 The direction indicated by the dotted arrow indicates that reverse gradient propagation is not performed. That is, the output error L of the multi-head lymph node classification network is not used. station The parameters of the lymph node classification network are reversely adjusted to reduce the interference of the classification performance of the multi-head lymph node classification network on the learning process of the lymph node classification network. Figure 2 As shown by another dotted arrow, the parameters of the feature extraction network may not be reversely adjusted according to the output error of the lymph node classification network, so as to reduce the influence of the classification performance of the lymph node classification network on the feature extraction network, thereby improving the learning efficiency of the neural network model.

[0085] After multiple rounds of iterative training, if L cls and L station If all converge to the specified value or range, the iteration stops and the trained neural network model is output.

[0086] In some scenarios, the lymph node detection method and neural network model training method of medical imaging provided by the aforementioned embodiments can be encapsulated as a software tool that can be used by a third party, such as a SaaS (Software-as-a-Service) tool. Among them, the SaaS tool can be implemented as a plug-in or an application. The plug-in or application can be deployed on a server and can open a specified interface to third-party users such as a client. For ease of description, in this embodiment, the specified interface is described as a first interface. Furthermore, third-party users such as a client can conveniently access and use the above-mentioned method provided by the server device by calling the first interface. Among them, the server can be a conventional server or a cloud server, and this embodiment does not impose any restrictions.

[0087] Taking the SaaS tool corresponding to the lymph node detection method of medical imaging as an example, the server can respond to the client's call request to the first interface, and obtain a scanned slice containing lymph nodes from the interface parameters of the first interface; identify the lymph node classification information of the scanned slice based on the image features and the learned characteristic distribution of different lymph nodes; detect the lymph nodes from the scanned slice by combining the lymph node classification information, the image features and the learned characteristic distribution of lymph nodes in different lymph nodes; send the location information of the lymph node to the client so that the client can mark the lymph node in the scanned slice.

[0088] Taking the SaaS tool corresponding to the neural network model training method as an example, the server can respond to the client's call request to the first interface and obtain a training data set from the interface parameters of the first interface. The training data set includes: a scanned slice sample containing lymph nodes; using the neural network model, obtaining the image features of the scanned slice sample; identifying the lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes; combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes to obtain the lymph node detection result of the scanned slice sample; iteratively optimizing the neural network model based on the error between the lymph node detection result and the lymph node classification label of the scanned slice sample. When the error converges to a specified value or a specified range, the neural network model can be output and sent to the client for use.

[0089] In this embodiment, the server can provide the client with lymph node detection services in medical images or neural network model training services based on the SaaS tools running on it, thereby reducing the client's computing pressure and computing costs.

[0090] It should be noted that the execution entity of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution entity of steps 101 to 104 can be device A; for another example, the execution entity of steps 101 and 102 can be device A, and the execution entity of step 103 can be device B; and so on.

[0091] In addition, some of the processes described in the above embodiments and the accompanying drawings include multiple operations that appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The sequence numbers of the operations, such as 101, 102, etc., are merely used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel.

[0092] It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to different types.

[0093] The following will be combined Figure 5 , a typical application scenario of the lymph node detection method of medical imaging provided in an embodiment of the present application is exemplified.

[0094] The lymph node detection method of medical imaging provided in the embodiments of the present application can be applied to the lymph node detection process of cancer diseases such as gastric cancer, lung cancer, and esophageal cancer. After the doctor obtains the CT image of the patient's lesion, he can input the CT image into an electronic device. The electronic device can execute the methods provided in the aforementioned embodiments to identify the lymph nodes on the CT image and output an image containing the lymph node detection results. The lymph node detection results include the marked locations of the lymph nodes, such as Figure 5 Lymph node detection box shown.

[0095] When detecting lymph nodes, the electronic device can combine lymph node classification information, image features, and learned characteristic distributions of lymph nodes within different lymph nodes to detect scanned sections containing lymph nodes. The lymph node classification information effectively guides the electronic device's identification of lymph nodes within different lymph nodes, allowing it to accurately distinguish subtle differences between lymph nodes and non-lymph nodes within different lymph nodes, thereby outputting more accurate lymph node detection and segmentation results.

[0096] Figure 6 is a structural diagram of an electronic device provided by an exemplary embodiment of the present application, such as Figure 6 As shown, the electronic device includes: a memory 601 and a processor 602.

[0097] The memory 601 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device.

[0098] The memory 601 may be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0099] Processor 602 is coupled to memory 601 and is configured to execute a computer program in memory 601 to: obtain a scan slice containing lymph nodes; obtain image features of the scan slice using a neural network model; identify lymph node classification information of the scan slice based on the image features and learned characteristic distributions of different lymph nodes; and detect lymph nodes from the scan slice based on the lymph node classification information, the image features, and the learned characteristic distributions of lymph nodes in different lymph nodes.

[0100] Optionally, when the processor 602 identifies the lymph node classification information of the scan slice based on the image features and the learned characteristic distribution of different lymph nodes, it is specifically used to: utilize the candidate region selection network in the neural network model to extract at least one candidate region from the scan slice based on the image features of the scan slice; for any candidate region of the at least one candidate region, utilize the lymph node classification network in the neural network model to identify, based on the image features of the scan slice, the location information of the candidate region and the learned characteristic distribution of different lymph nodes, a first probability distribution that the candidate region belongs to multiple different lymph nodes, as the lymph node classification information of the candidate region.

[0101] Optionally, when the processor 602 detects lymph nodes from the scan slice in combination with the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes, it is specifically used to: input the image features of the scan slice and the position information of the candidate area into the multi-head lymph node classification network in the neural network model; the multi-head lymph node classification network corresponds to multiple different lymph nodes; using the multi-head lymph node classification network, combined with the image features of the scan slice and the position information of the candidate area, calculate the lymph node classification information of the candidate area in the multiple different lymph nodes; based on the lymph node classification information of the candidate area and the lymph node classification information of the candidate area in the multiple different lymph nodes, determine that the candidate area is a target candidate area containing lymph nodes.

[0102] Optionally, when the processor 602 uses the multi-head lymph node classification network, combined with the image features of the scan slice and the position information of the candidate area, to calculate the lymph node classification information of the candidate area in the multiple different lymph stations, it is specifically used to: in the lymph node classification network corresponding to any lymph station, based on the image features of the scan slice, the position information of the candidate area and the characteristic distribution of the lymph nodes in the station learned by the lymph node classification network, calculate the probability that the candidate area is a lymph node in the lymph station; and determine the second probability distribution based on the probability that the candidate area output by the multi-head lymph node classification network is a lymph node in the multiple different lymph stations.

[0103] Optionally, when the processor 602 determines that the candidate area is a target candidate area containing lymph nodes based on the lymph station classification information of the candidate area and the lymph node classification information of the candidate area in the multiple different lymph stations, it is specifically used to: use the first probability distribution corresponding to the candidate area to perform a weighted calculation on the second probability distribution corresponding to the candidate area to obtain a weighted score of the candidate area; if the weighted score meets the set score condition, the candidate area is determined to be a target candidate area containing lymph nodes.

[0104] Optionally, after determining that the candidate region is a target candidate region containing a lymph node, the processor 602 is further configured to: determine, from the multiple lymph nodes according to the first probability distribution, a target lymph node to which the lymph node belongs; and output an identifier of the target lymph node.

[0105] Optionally, when the processor 602 detects lymph nodes from the scan slice in combination with the lymph station classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph stations, it is specifically used to: determine a target lymph station that meets a preset probability condition from the multiple different lymph stations based on the first probability distribution; input the image features of the scan slice and the position information of the candidate area into the lymph node classification network corresponding to the target lymph station; in the lymph node classification network of the target lymph station, determine the image features of the candidate area based on the image features of the scan slice and the position information of the candidate area; and, calculate the third probability that the candidate area is a lymph node in the target lymph station based on the image features of the candidate area and the characteristic distribution of lymph nodes in the station learned by the lymph node classification network; if the third probability is greater than the preset third probability threshold, it is determined that the candidate area contains lymph nodes in the target lymph station.

[0106] Optionally, when the processor 602 uses a neural network model to extract image features from the scan slice, it is specifically used to: input the scan slice and the adjacent slices of the scan slice into the neural network model; use the feature extraction network in the neural network model to extract the two-dimensional image features of the scan slice and the adjacent slice; fuse the two-dimensional image features of the scan slice and the adjacent slice to obtain the context fusion feature of the scan slice as the image feature of the scan slice.

[0107] Optionally, the processor 602 is further used to: obtain shape features of healthy lymph nodes in the target lymph node station; compare the shape features of the lymph nodes in the candidate area with the shape features of the healthy lymph nodes; if it is determined that the lymph nodes in the candidate area are in an abnormal state based on the result of the comparison, then highlight the candidate area in the scan slice.

[0108] Further, if Figure 6As shown, the electronic device also includes: a communication component 603, a display 604, a power supply component 605 and other components. Figure 6 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 6 Components shown.

[0109] Among them, the communication component 603 is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G or 5G, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component can be implemented based on near field communication (NFC) technology, radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0110] The display 604 includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, slides, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0111] The power supply assembly 605 provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0112] In this embodiment, the electronic device can identify the lymph node classification information of the scanned slice based on the image features of the scanned slice and the learned characteristic distribution of different lymph nodes. By combining the lymph node classification information, image features, and the learned characteristic distribution of lymph nodes in different lymph nodes, lymph nodes can be detected from the scanned slice. Among them, the lymph node classification information can effectively guide the identification process of lymph nodes in different lymph nodes. Combining the lymph node classification information, image features, and the characteristic distribution of lymph nodes in different lymph nodes to identify lymph nodes in the scanned slice is conducive to accurately distinguishing the subtle differences between lymph nodes and non-lymph nodes in different lymph nodes, reducing the interference caused by non-lymph nodes in different lymph nodes to the identification process, and thus accurately detecting lymph nodes in different lymph nodes.

[0113] It should be noted that Figure 6The illustrated electronic device, in addition to performing data processing operations according to the data processing logic described in the aforementioned embodiments, can also perform the following operations according to the lymph node detection method of medical imaging described below: the processor 602 is specifically used to: respond to the client's call request for the first interface, and obtain a scan slice containing lymph nodes from the interface parameters of the first interface; use a neural network model to obtain image features of the scan slice; identify lymph node classification information of the scan slice based on the image features and the learned characteristic distribution of different lymph nodes; detect lymph nodes from the scan slice in combination with the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes; and send the location information of the lymph node to the client so that the client can mark the lymph node in the scan slice.

[0114] The electronic device can also perform the following operations according to the neural network model training method described below: obtaining a training data set, which includes: a scanned slice sample containing lymph nodes; using a neural network model, obtaining image features of the scanned slice sample; identifying lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes; obtaining a lymph node detection result of the scanned slice sample based on the lymph node classification information, the image features and the learned characteristic distribution of lymph nodes in different lymph nodes; and iteratively optimizing the neural network model based on the error between the lymph node detection result and the lymph node classification label of the scanned slice sample.

[0115] Optionally, the lymph node classification label includes: a lymph node label and a lymph node label; when the processor 602 iteratively optimizes the neural network model based on the error between the lymph node detection result and the lymph node classification label of the scanned slice sample, it is specifically used to: calculate the lymph node classification loss of the neural network model based on the lymph node classification label; and calculate the lymph node classification loss of the neural network model based on the lymph node classification label; iteratively optimize the neural network model with the convergence of the lymph node classification loss and the lymph node classification loss as the training goal.

[0116] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which can implement the steps in the above method embodiment when the computer program is executed.

[0117] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable electronic device to produce a machine, so that the instructions executed by the processor of the computer or other programmable electronic device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable electronic device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable electronic device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0121] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0122] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0123] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0125] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for detecting lymph nodes in medical images, characterized in that: include: Obtain scans containing lymph nodes; Obtaining image features of the scan slice using a neural network model, including: inputting the scan slice and adjacent slices of the scan slice into the neural network model; extracting two-dimensional image features of the scan slice and the adjacent slices using a feature extraction network in the neural network model; fusing the two-dimensional image features of the scan slice and the adjacent slices to obtain context fusion features of the scan slice as the image features of the scan slice; identifying lymph node classification information of the scanned slice based on the image features and the learned feature distributions of different lymph nodes; Lymph nodes are detected from the scanned slices by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes.

2. The method according to claim 1, characterized in that Identifying lymph node classification information of the scanned slice based on the image features and the learned feature distribution of different lymph nodes, including: Utilizing a candidate region selection network in the neural network model, extracting at least one candidate region from the scan slice according to image features of the scan slice; For any candidate area among the at least one candidate area, the lymph node classification network in the neural network model is used to identify, based on the image features of the scanned slice, the location information of the candidate area, and the learned feature distributions of different lymph nodes, a first probability distribution that the candidate area belongs to multiple different lymph nodes, as the lymph node classification information of the candidate area.

3. The method according to claim 2, characterized in that Detecting lymph nodes from the scanned slices by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes includes: Inputting the image features of the scanned slice and the position information of the candidate area into the multi-headed lymph node classification network in the neural network model; the multi-headed lymph node classification network corresponds to a plurality of different lymph nodes; Calculating lymph node classification information of the candidate region within the multiple lymph nodes using the multi-head lymph node classification network in combination with image features of the scanned slice and position information of the candidate region; The candidate region is determined to be a target candidate region containing lymph nodes according to the lymph node classification information of the candidate region and the lymph node classification information of the candidate region in the multiple different lymph nodes.

4. The method according to claim 3, characterized in that Utilizing the multi-head lymph node classification network, combined with the image features of the scan slice and the position information of the candidate region, calculating the lymph node classification information of the candidate region in the multiple different lymph nodes, including: In a lymph node classification network corresponding to any lymph node station, the probability that the candidate region is a lymph node within the lymph node station is calculated based on the image features of the scan slice, the location information of the candidate region, and the characteristic distribution of the lymph nodes within the station learned by the lymph node classification network; A second probability distribution is determined according to the probabilities that the candidate regions output by the multi-head lymph node classification network are lymph nodes in the multiple different lymph stations.

5. The method according to claim 4, characterized in that Determining, based on the lymph node classification information of the candidate region and the lymph node classification information of the candidate region in the multiple different lymph nodes, that the candidate region is a target candidate region containing lymph nodes includes: Performing a weighted calculation on the second probability distribution corresponding to the candidate region using the first probability distribution corresponding to the candidate region to obtain a weighted score for the candidate region; If the weighted score satisfies a set score condition, the candidate region is determined to be a target candidate region containing a lymph node.

6. The method according to claim 5, characterized in that After determining that the candidate region is a target candidate region containing a lymph node, the method further includes: determining, from the plurality of different lymph nodes according to the first probability distribution, a target lymph node to which the lymph node belongs; The identification of the target lymphatic station is output.

7. The method according to claim 2, characterized in that Detecting lymph nodes from the scanned slices by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes includes: determining, according to the first probability distribution, a target lymph node that meets a preset probability condition from the multiple different lymph nodes; Inputting the image features of the scanned slice and the position information of the candidate region into the lymph node classification network corresponding to the target lymph node station; In the lymph node classification network of the target lymph node station, the image features of the candidate area are determined based on the image features of the scan slice and the position information of the candidate area; and the third probability that the candidate area is a lymph node in the target lymph node station is calculated based on the image features of the candidate area and the characteristic distribution of the lymph nodes in the station learned by the lymph node classification network; if the third probability is greater than a preset third probability threshold, it is determined that the candidate area includes the lymph nodes in the target lymph node station.

8. The method according to claim 6 or 7, characterized in that Also includes: Acquiring shape characteristics of healthy lymph nodes in the target lymph node station; comparing the shape characteristics of the lymph nodes in the candidate region with the shape characteristics of the healthy lymph nodes; If it is determined according to the comparison result that the lymph nodes in the candidate region are in an abnormal state, the candidate region is highlighted in the scan slice.

9. A method for detecting lymph nodes in medical imaging, characterized in that: include: In response to a call request to the first interface from the client, obtaining a scan slice containing a lymph node from interface parameters of the first interface; Obtaining image features of the scan slice using a neural network model, including: inputting the scan slice and adjacent slices of the scan slice into the neural network model; extracting two-dimensional image features of the scan slice and the adjacent slices using a feature extraction network in the neural network model; fusing the two-dimensional image features of the scan slice and the adjacent slices to obtain context fusion features of the scan slice as the image features of the scan slice; identifying lymph node classification information of the scanned slice based on the image features and the learned characteristic distribution of different lymph nodes; and detecting lymph nodes from the scanned slice based on the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes; The location information of the lymph node is sent to the client, so that the client marks the lymph node in the scan slice.

10. A neural network model training method, characterized in that: include: Acquire a training data set, the training data set comprising: scan slice samples containing lymph nodes; Obtaining image features of the scanned slice sample using a neural network model, including: inputting the scanned slice sample and adjacent slices of the scanned slice sample into the neural network model; extracting two-dimensional image features of the scanned slice sample and the adjacent slices using a feature extraction network in the neural network model; fusing the two-dimensional image features of the scanned slice sample and the adjacent slices to obtain contextual fusion features of the scanned slice sample as image features of the scanned slice sample; identifying lymph node classification information of the scanned slice sample based on the image features and the learned characteristic distribution of different lymph nodes; and obtaining a lymph node segmentation result of the scanned slice sample by combining the lymph node classification information, the image features, and the learned characteristic distribution of lymph nodes in different lymph nodes; The neural network model is iteratively optimized according to the error between the lymph node segmentation result and the lymph node classification label of the scanned slice sample.

11. The method according to claim 10, characterized in that The lymph node classification labels include: lymph node station labels and lymph node labels; according to the error between the lymph node segmentation result and the lymph node classification labels of the scanned slice sample, the neural network model is iteratively optimized, including: Calculating the lymph node classification loss of the neural network model based on the lymph node classification labels; and calculating the lymph node classification loss of the neural network model based on the lymph node classification labels; The neural network model is iteratively optimized with the convergence of the lymph station classification loss and the lymph node classification loss as a training goal.

12. An electronic device, characterized in that: include: memory and processor; The memory is used to store one or more computer instructions; The processor is configured to execute the one or more computer instructions to perform the steps of the method according to any one of claims 1 to 11.

13. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 can be implemented.

Images