SPECT-oriented bone metastasis identification method, equipment, medium and product
Through the global-local recognition model combined with Swin Transformer and EfficientNet model, the global and local features of SPECT images are extracted and fused, which solves the problem of low accuracy in the prior art and improves the accuracy of bone metastasis recognition.
Patent Information
- Application Number
- CN202411112322.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-12-06
AI Technical Summary
The existing SPECT bone imaging image analysis method relies on manual feature extraction or a single CNN-based method, making it difficult to effectively extract global information, resulting in low accuracy of bone metastasis recognition, and neglecting the coupling between local lesions and global image information. effect.
The global-local recognition model is adopted, and the global and local features of the SPECT image are extracted through the global feature extraction module and the local feature extraction module, and feature fusion is performed. Finally, the multi-layer perceptron module is used for bone metastasis recognition, combining the Swin Transformer and EfficientNet model. , realize effective integration of different view features.
The accuracy of bone metastasis recognition is improved, the coupling effect between local information of the lesion and global image information is fully considered, and the ability to identify bone metastasis lesions is enhanced.
Smart Images

Figure CN119091198A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of medical image analysis, and in particular to a method, device, medium and product for identifying bone metastasis based on global-local information fusion of SPECT bone imaging. Background Art
[0002] Bone metastasis is one of the common complications of advanced cancer, which is the spread of cancer cells from the primary tumor to other parts of the bones. Early detection of bone metastasis is crucial for the treatment and management of patients.
[0003] Single-photon emission computed tomography (SPECT) is a nuclear medicine imaging technique used to observe the function, structure and metabolic activities of human internal organs, especially in cancer, cardiovascular disease, nervous system diseases, etc. Therefore, SPECT technology can be applied to the identification of bone metastasis.
[0004] At present, the processing methods of SPECT bone imaging are mainly divided into two categories: one is based on traditional machine learning technology, and the other is based on deep learning technology. In the field of traditional machine learning, previous studies mainly used artificial neural networks to classify hot spot features extracted from SPECT images. For example, CADBOSS, as a computer-aided diagnosis system, extracts the features of the metastatic area by gridding the image, and classifies it through an artificial neural network classifier to determine whether there is metastasis. Mac et al. used K-nearest neighbor and support vector machine classifiers to identify objects extracted from SPECT images, and adopted edge segmentation methods and full Lambda scheduling algorithms. However, the common limitation of these studies is that they mainly rely on manually extracted hot spot features for detection, and manually extracted features may have problems such as insufficient capabilities and unstable performance.
[0005] With the development of deep learning technology, especially convolutional neural network (CNN), it has achieved great success in the field of computer vision. Image classification and object segmentation based on CNN have been widely used in medical image analysis. A large number of studies have been conducted on the automatic detection of bone metastases based on CNN for bone scan images. For example, a master's thesis from Lund University first attempted to use CNN to identify hot spots in SPECT images to determine whether they represent bone metastases caused by prostate cancer or other physiological processes. They developed a CNN model consisting of five branches to classify hot spot image blocks found in the spine, achieving an accuracy of 89%. In another study, a CNN model consisting of three sub-networks was proposed to detect the presence of bone metastases with an accuracy of 95%. Liu et al. proposed a medical image fusion method based on convolutional neural network. First, two source images are passed through a conjoined convolutional network to generate a weight map that integrates pixel activity information from the two source images. Then, each source image is decomposed into a Laplacian pyramid, and the weight map obtained from the network is decomposed into a Gaussian pyramid. The fusion process is performed at each decomposition level. The fusion process is performed in a multi-scale manner through an image pyramid, and the fusion mode of the decomposition coefficients is adaptively adjusted using a local similarity-based strategy. Lin et al. proposed a network for automatically detecting different diseases in whole-body SPECT scan images, which can determine whether a given image contains a disease and which diseases it contains by classifying these images (i.e., normal, metastasis, arthritis, and metastasis & arthritis). In view of the problem of limited SPECT image samples, different data augmentation methods are used to expand the dataset. The CNN-based dSPIC network first extracts the optimal features from whole-body SPECT images, and then classifies these images into the focus class. A dataset consisting of a set of real whole-body SPECT images is constructed to evaluate the customized end-to-end classification network.
[0006] It can be seen that the current feature extraction method in the view fusion task is mainly based on CNN. However, the internal working mechanism of CNN is difficult to control and explain, which will lead to insufficient feature extraction. Moreover, CNN only extracts local information in a relatively small receptive field, and it is difficult to extract global information to generate high-quality fused images. The coupling effect between the local information of the lesion and the global information of the image is ignored, resulting in limited feature representation of the image and low accuracy of the recognition result. Summary of the invention
[0007] The purpose of this application is to provide a SPECT bone metastasis identification method, device, medium and product, which can improve the accuracy of bone metastasis identification.
[0008] To achieve the above objectives, this application provides the following solutions:
[0009] In a first aspect, the present application provides a method for identifying bone metastasis using SPECT, comprising:
[0010] Obtaining a SPECT whole-body bone image of a human body to be identified, wherein the SPECT whole-body bone image of the human body to be identified includes a whole-body anterior view and a whole-body posterior view, the whole-body anterior view is a face-side SPECT whole-body bone image, and the whole-body posterior view is a back-of-head-side SPECT whole-body bone image;
[0011] Extracting regions of interest from the whole-body anterior view and the whole-body posterior view respectively, and obtaining an anterior view of interest and a posterior view of interest correspondingly;
[0012] The anterior view of interest and the posterior view of interest are input into a trained global-local recognition model, and a bone metastasis recognition result is output, wherein the trained global-local recognition model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a multi-layer perceptron module, the global feature extraction module and the local feature extraction module are respectively connected to the feature fusion module, and the feature fusion module is connected to the multi-layer perceptron module; the global feature extraction module is used to extract and fuse the global features of the anterior view of interest and the posterior view of interest to obtain fused global features; the local feature extraction module is used to extract and fuse the local features of the anterior view of interest and the posterior view of interest to obtain fused local features; the feature fusion module is used to fuse the fused global features and the fused local features to obtain final fused features; the multi-layer perceptron module is used to identify the final fused features and output the bone metastasis recognition result.
[0013] Optionally, respectively extracting regions of interest from the whole-body anterior view image and the whole-body posterior view image to correspondingly obtain an anterior view of interest and a posterior view of interest specifically includes:
[0014] Preprocessing the whole-body anterior view image and the whole-body posterior view image respectively to obtain a chest region anterior view and a chest region posterior view, wherein the preprocessing includes data denoising, effective region extraction, cropping and removal processing, and background filling, the effective region is a human body contour region, and the cropping and removal processing is to remove the head region, the leg region, the pelvis region, and the arm region;
[0015] The anterior view of the thoracic region is used as the anterior view of interest, and the posterior view of the thoracic region is used as the posterior view of interest.
[0016] Optionally, during the training process, the trained global-local recognition model uses an image geometric transformation method to expand data in the SPECT chest bone imaging data set.
[0017] Optionally, the global feature extraction module adopts a Swin Transformer model, and the Swin Transformer model includes a Patch Partition layer, a Stage 1 layer, a Stage 2 layer, a Stage 3 layer and a Stage 4 layer connected in sequence, the Stage 1 layer includes a Linear Embedding layer and a Swin TransformerBlock layer connected in sequence, and the Stage 2 layer, the Stage 3 layer and the Stage 4 layer all include a Patch Merging layer and a Swin TransformerBlock layer connected in sequence.
[0018] Optionally, the local feature extraction module adopts an EfficientNet model, which includes 1 Conv convolutional layer, 7 MBConv convolutional layers and 1 combination layer connected in sequence, and the combination layer includes 1 Conv convolutional layer, 1 average pooling layer and 1 fully connected layer connected in sequence.
[0019] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a SPECT bone metastasis identification method as described in any one of the first aspects above.
[0020] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a SPECT bone metastasis identification method as described in any one of the first aspects above.
[0021] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements a SPECT bone metastasis identification method as described in any one of the first aspects above.
[0022] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0023] The present application provides a method, device, medium and product for SPECT bone metastasis identification. First, the global features of the anteroposterior view in the SPECT image are mined through a global feature extraction module, and the global features of the two are fused to obtain the fused global features; the local features of the anteroposterior view in the SPECT image are mined through a local feature extraction module, and the local features of the two are fused to obtain the fused local features; the fused global features and the fused local features are then fused to obtain the final fused features; finally, whether bone metastasis exists is identified based on the final fused features. The present application focuses on mining the global and local features in the anteroposterior view in the SPECT image and performing full-full and local-local feature fusion, taking into account the complementarity of information between different views under the same modality, and realizing the effective integration of features extracted from different views; the fused global-local features fully consider the coupling effect between the local information of the lesion and the global information of the image, and improve the accuracy of bone metastasis identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0025] Figure 1 This is an application environment diagram of a bone metastasis identification method in Example 1 of the present application;
[0026] Figure 2 This is a schematic diagram of the process of extracting a region of interest in Example 1 of the present application;
[0027] Figure 3 This is a model overview diagram of the global-local recognition model in Example 1 of the present application;
[0028] Figure 4 This is a schematic diagram of the structure of the SwinTransformer model in Example 1 of the present application;
[0029] Figure 5 (a) is a schematic diagram of the structure of the Conv convolutional layer in the EfficientNet model of Example 1 of the present application, Figure 5 (b) is a schematic diagram of the structure of the MBConv convolutional layer in the EfficientNet model of Example 1 of the present application;
[0030] Figure 6 This is a schematic diagram of the structure of the feature fusion module in Example 1 of the present application;
[0031] Figure 7A schematic diagram of the structure of a computer device provided in Example 2 of the present application. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0033] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0034] Example 1
[0035] The SPECT bone metastasis identification method provided in the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the SPECT whole-body bone imaging of the human body to be identified to the server 104. After the server 104 receives the SPECT whole-body bone imaging of the human body to be identified, the region of interest of the whole-body anterior view image and the whole-body posterior view image in the SPECT whole-body bone imaging of the human body to be identified is extracted, and the anterior view of interest and the posterior view of interest are obtained correspondingly; the global feature extraction module in the trained global-local recognition model is used to extract the global features of the anterior view of interest and the posterior view of interest and fuse them to obtain the fused global features; the local feature extraction module extracts the local features of the anterior view of interest and the posterior view of interest and fuses them to obtain the fused local features; the feature fusion module is used to fuse the fused global features and the fused local features to obtain the final fused features; the multi-layer perceptron module is used to identify the final fused features, and the output is whether there is a bone metastasis recognition result. The server 104 can feed back the obtained bone metastasis identification result to the terminal 102. In addition, in some embodiments, the SPECT bone metastasis identification method can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform bone metastasis identification processing on the SPECT whole-body bone image of the human body to be identified, or the server 104 can obtain the SPECT whole-body bone image of the human body to be identified from the data storage system, and perform bone metastasis identification processing on the SPECT whole-body bone image of the human body to be identified.
[0036] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.
[0037] The inventors have found that in the current bone metastasis recognition method, the feature extraction method is mainly based on CNN during the view fusion task, and the internal working mechanism of CNN is difficult to control and explain, resulting in insufficient feature extraction. Moreover, CNN only extracts local information in a relatively small receptive field, and it is difficult to extract global information to generate high-quality fused images. The coupling effect between the local information of the lesion and the global information of the image is ignored, resulting in limited feature representation of the image and reduced accuracy of model recognition. This single-dependency method fails to fully utilize the correlation between the local information of the lesion and the global information of the image, resulting in limitations in the recognition results. In addition, current research on view fusion mainly focuses on fusing views between different modalities, while the anterior and posterior views are not fully utilized. Figure 1 Generally, it is simply processed as the input of the model without performing feature fusion between views, ignoring the complementary information between different views in the same modality.
[0038] Based on this background, in an exemplary embodiment, the present application proposes a method for identifying bone metastasis by fusing global and local features in anterior-posterior SPECT images. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 201 to 203. Among them:
[0039] Step 201, obtaining a SPECT whole-body bone image of a human body to be identified, wherein the SPECT whole-body bone image of the human body to be identified includes a whole-body anterior view image and a whole-body posterior view image, the whole-body anterior view is a face-side SPECT whole-body bone image, and the whole-body posterior view is a back-of-the-head-side SPECT whole-body bone image.
[0040] Step 202 , respectively extracting regions of interest from the whole-body anterior view image and the whole-body posterior view image, and correspondingly obtaining an anterior view of interest and a posterior view of interest.
[0041] Step 203, inputting the anterior view of interest and the posterior view of interest into the trained global-local recognition model, and outputting a bone metastasis recognition result, wherein the trained global-local recognition model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a multi-layer perceptron module, the global feature extraction module and the local feature extraction module are respectively connected to the feature fusion module, and the feature fusion module is connected to the multi-layer perceptron module; the global feature extraction module is used to extract the global features of the anterior view of interest and the posterior view of interest and fuse them to obtain fused global features; the local feature extraction module is used to extract the local features of the anterior view of interest and the posterior view of interest and fuse them to obtain fused local features; the feature fusion module is used to fuse the fused global features and the fused local features to obtain the final fused features; the multi-layer perceptron module is used to identify the final fused features and output the bone metastasis recognition result.
[0042] As an optional implementation, the specific training process of the global-local recognition model is as follows, including:
[0043] Step 301: constructing a SPECT chest bone imaging data set, wherein the SPECT chest bone imaging data set includes a plurality of chest region anterior views and corresponding chest region posterior views.
[0044] 1) Region of interest extraction
[0045] Since the spine and ribs are the most common metastatic sites of various cancers and are located in the chest area, and the chest area is mostly a high-risk area for lung cancer bone metastasis, it is necessary to extract the chest area as the region of interest from human SPECT whole-body bone imaging.
[0046] This embodiment uses a threshold segmentation method to crop the chest area from the whole body anterior view and the whole body posterior view of the human body SPECT whole body bone imaging. Taking the cropping of the chest area from the whole body anterior view as an example, Figure 2 As shown, specifically including:
[0047] The whole body front view is preprocessed to obtain the chest area, where the preprocessing includes data denoising, effective area extraction, cropping and removal, and background filling. The effective area is the human body contour area (i.e. Figure 3 The cropping and removal process is to remove the head area, leg area, pelvis area and arm area.
[0048] 2) Data expansion
[0049] Deep learning models can often only express good results on massive data sets, and medical imaging data is often difficult to collect due to the confidentiality of patient information. Based on this, this embodiment uses an image geometric transformation method to expand the data of a certain amount of chest area images. Image geometric transformation achieves data expansion by mapping the original coordinate position in a given image to another position through a certain transformation. Image geometric transformation includes image flipping, image translation, and image rotation:
[0050] (1) Image flipping, also known as image mirroring, refers to the axial symmetric transformation of an image along its axis.
[0051] (2) Image translation is to translate all pixels in the image according to a given displacement amount.
[0052] (3) Image rotation means that the image will randomly rotate to the left or right around its geometric center by a certain angle.
[0053] Step 302: Construct a global-local recognition model.
[0054] like Figure 3 As shown, this embodiment uses the anteroposterior view of SPECT chest bone imaging of lung cancer bone metastasis (i.e., the anterior view of the chest region and the posterior view of the chest region in the data set) as input, and uses a feature extractor based on Transformer-CNN (i.e., the global-local recognition model) to effectively extract the global and local features in the anteroposterior view of the SPECT image. The global-local recognition model mainly includes a global feature extraction module (using the Swin-Transformer model), a local feature extraction module (using the EfficientNet model), a feature fusion module (FusionLayer) and a multilayer perceptron module (MultilayerPerceptron, MLP).
[0055] like Figure 4 As shown, the SwinTransformer model is used to extract the global feature information of the anteroposterior views of SPECT chest bone imaging. The SwinTransformer model adopts a hierarchical design, including a PatchPartition layer, a Stage1 layer, a Stage2 layer, a Stage3 layer, and a Stage4 layer connected in sequence. The Stage1 layer includes a LinearEmbedding layer and a SwinTransformerBlock layer connected in sequence, and the Stage2 layer, the Stage3 layer, and the Stage4 layer all include a PatchMerging layer and a Swin TransformerBlock layer connected in sequence.
[0056] Except for Stage1, Stage2, Stage3, and Stage4 layers will first reduce the resolution of the input feature map through the PatchMerging layer and perform downsampling operations, expanding the receptive field layer by layer like CNN in order to obtain global information. At the beginning of the input, a PatchPartition is performed, which is the Patch Embedding operation in ViT. The image is cut into patches through the convolution layer and embedded into the Embedding. The W-MSA module calculates self-attention in non-overlapping local windows, replacing the global self-attention. Although W-MSA reduces the computational complexity from quadratic to linear, the lack of communication and connection between windows will limit its modeling and representation capabilities. In order to introduce cross-window connections while maintaining the computational efficiency of non-overlapping windows, the SW-MSA module is used, which alternates between two partitioning / partitioning configurations in consecutive SwinTransformerBlocks.
[0057] like Figure 5 As shown in Figure 2, the EfficientNet model is used to extract local feature information of the anteroposterior view of SPECT chest bone imaging, including a sequentially connected Conv convolutional layer (such as Figure 5 (a)), 7 MBConv convolutional layers ((as shown Figure 5 (b)) and a combination layer, wherein the combination layer includes a Conv convolution layer and an average pooling layer connected in sequence.
[0058] The EfficientNet model is divided into 9 stages. Stage 1 is the first Conv convolution layer, which is a common convolution layer with a convolution kernel size of 3x3 and a stride of 2 (including BN and activation function Swish); Stage 2 to Stage 8 are repeated stacked MBConv structures, that is, the corresponding 7 MBConv convolution layers (the Layers in the last column indicates how many times the MBConv structure is repeated in this stage); Stage 9 is the combination layer, which consists of an ordinary 1x1 convolution layer (including BN and activation function Swish) and an average pooling layer.
[0059] The MBConv structure mainly consists of a 1x1 ordinary convolution, a kxk Depthwise Conv convolution, a SE module, a 1x1 ordinary convolution, and a Dropout layer.
[0060] like Figure 6 As shown in the figure, the feature fusion module consists of 1×1 convolution, ReLU function and residual connection.
[0061] Step 303: Using the SPECT chest bone imaging data set to train a global-local recognition model, the specific process includes:
[0062] (1) The anterior view of the thoracic region and the posterior view of the thoracic region are respectively input into the global feature extraction module to obtain corresponding anterior image global features and posterior image global features.
[0063] (2) Fusing the front bitmap global features and the back bitmap global features to obtain fused global features.
[0064] (3) The anterior view of the thoracic region and the posterior view of the thoracic region are respectively input into the local feature extraction module to obtain corresponding local features of the anterior image and local features of the posterior image.
[0065] (4) Fusing the local features of the front bitmap and the local features of the back bitmap to obtain fused local features.
[0066] (5) The fused global features and the fused local features are input into the fused global features to obtain the final fused features.
[0067] (6) The final fusion feature is input into the multi-layer perceptron module, and a bone metastasis identification result is output.
[0068] The present embodiment provides a method for identifying bone metastasis by fusing global and local features in an anteroposterior SPECT image. The method aims at automatically identifying bone metastasis, takes an anteroposterior view SPECT chest bone imaging as input, and utilizes a feature extractor based on Transformer-CNN to effectively extract global and local features in the anteroposterior view of the SPECT image. First, the global information of the SPECT anteroposterior view is extracted by the attention mechanism of SwinTransformer, and the local information of the SPECT anteroposterior view is extracted by EffcientNet, and full-full and local-local information fusion is performed to realize feature fusion of different views; secondly, the fused global features and the fused local features are further fused; finally, the final fused features are classified by the MLP module to realize automatic identification of bone metastasis.
[0069] Compared with other traditional methods, this application focuses on mining the global and local features between the anteroposterior views in SPECT images and performing full-to-full and local-to-local information fusion, thereby achieving effective integration of features extracted from different views. This fusion strategy makes full use of the respective characteristics of EfficientNe and Swin Transformer, so that the model can better understand the relationship between multi-view information, thereby improving the ability to identify bone metastasis lesions. In addition, the bone metastasis identification method that fuses global and local features in anteroposterior SPECT images proposed in this embodiment has higher robustness and can reliably assist in further diagnosis of lung cancer bone metastasis. In general, this application improves the automated and accurate identification of lung cancer bone metastasis through efficient network structure and full use of multi-view information.
[0070] advantage:
[0071] (1) Based on the data characteristics, an intelligent and specific SPECT chest bone imaging dataset was constructed;
[0072] (2) Aiming at the intelligent identification of bone metastasis, a global-local feature extraction structure based on Transformer-CNN was proposed to realize the extraction of global-local features in SPECT.
[0073] (3) A global-local feature fusion module was constructed. Through the global-global and local-local feature fusion strategies, the model can better explore the relationship between different frequencies and achieve effective integration of global-local features, thereby improving the ability to identify bone metastasis lesions.
[0074] The present application also provides an application scenario, which applies the above-mentioned SPECT bone metastasis identification method. Specifically: The SPECT bone metastasis identification method provided in this embodiment can be applied in the bone metastasis identification result publishing scenario. The bone metastasis identification result publishing process includes an image acquisition link (collecting SPECT whole-body bone imaging), an image recognition result determination link (using the bone metastasis identification method provided in this embodiment to obtain image recognition results and determine whether bone metastasis exists) and an image recognition result publishing link (display and publication of image recognition results). Since the above-mentioned SPECT bone metastasis identification method in this embodiment can improve the recognition ability of bone metastasis lesions, in actual application, doctors can more accurately diagnose the patient's disease progression based on the displayed and published image recognition results. The bone metastasis identification method provided in this embodiment provides clinicians with a more reliable basis for diagnosis.
[0075] Example 2
[0076] This embodiment provides a computer device, which may be a server or a terminal. Its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Wherein, the processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. Wherein, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data in the bone metastasis identification method. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a SPECT bone metastasis identification method provided in Example 1 is implemented.
[0077] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0078] Example 3
[0079] This embodiment further provides a computer device, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the SPECT bone metastasis identification method provided in the above embodiment 1 is implemented.
[0080] Example 4
[0081] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the SPECT-oriented bone metastasis identification method provided in the above-mentioned embodiment 1 is implemented.
[0082] Example 5
[0083] This embodiment provides a computer program product, including a computer program. When the computer program is executed by a processor, the SPECT-oriented bone metastasis identification method provided in the above-mentioned embodiment 1 is implemented.
[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0085] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0086] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.
[0087] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0088] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for identifying bone metastasis based on SPECT, characterized in that: The SPECT-oriented bone metastasis identification method comprises: Acquire a SPECT whole-body bone image of a human body to be identified, wherein the SPECT whole-body bone image of the human body to be identified includes a whole-body anterior view image and a whole-body posterior view image, the whole-body anterior view is a face-side SPECT whole-body bone image, and the whole-body posterior view is a back-of-head-side SPECT whole-body bone image; Respectively extracting regions of interest from the whole-body anterior view image and the whole-body posterior view image, and correspondingly obtaining an anterior view of interest and a posterior view of interest; The anterior view of interest and the posterior view of interest are input into a trained global-local recognition model, and a bone metastasis recognition result is output, wherein the trained global-local recognition model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a multi-layer perceptron module, the global feature extraction module and the local feature extraction module are respectively connected to the feature fusion module, and the feature fusion module is connected to the multi-layer perceptron module; the global feature extraction module is used to extract the global features of the anterior view of interest and the posterior view of interest and fuse them to obtain fused global features; the local feature extraction module is used to extract the local features of the anterior view of interest and the posterior view of interest and fuse them to obtain fused local features; the feature fusion module is used to fuse the fused global features and the fused local features to obtain final fused features; the multi-layer perceptron module is used to identify the final fused features and output the bone metastasis recognition result.
2. A method for identifying bone metastasis based on SPECT according to claim 1, characterized in that: Respectively extracting regions of interest from the whole-body anterior view image and the whole-body posterior view image, and correspondingly obtaining an anterior view of interest and a posterior view of interest, specifically includes: Preprocessing the whole-body anterior view image and the whole-body posterior view image respectively to obtain a chest region anterior view and a chest region posterior view, wherein the preprocessing includes data denoising, effective region extraction, cropping and removal processing, and background filling, the effective region is a human body contour region, and the cropping and removal processing is to remove the head region, the leg region, the pelvis region, and the arm region; The anterior view of the thoracic region is used as the anterior view of interest, and the posterior view of the thoracic region is used as the posterior view of interest.
3. A method for identifying bone metastasis based on SPECT according to claim 1, characterized in that: During the training process, the trained global-local recognition model uses an image geometric transformation method to expand the data in the SPECT chest bone imaging data set.
4. The method for identifying bone metastasis based on SPECT according to claim 1, characterized in that: The global feature extraction module adopts the Swin Transformer model, which includes a Patch Partition layer, a Stage 1 layer, a Stage 2 layer, a Stage 3 layer and a Stage 4 layer connected in sequence. The Stage 1 layer includes a Linear Embedding layer and a Swin Transformer Block layer connected in sequence. The Stage 2 layer, the Stage 3 layer and the Stage 4 layer all include a Patch Merging layer and a Swin Transformer Block layer connected in sequence.
5. The method for identifying bone metastasis based on SPECT according to claim 1, characterized in that: The local feature extraction module adopts the EfficientNet model, which includes 1 Conv convolution layer, 7 MBConv convolution layers and 1 combination layer connected in sequence, and the combination layer includes 1 Conv convolution layer and 1 average pooling layer connected in sequence.
6. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a SPECT bone metastasis identification method according to any one of claims 1 to 5.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the SPECT-oriented bone metastasis identification method described in any one of claims 1 to 5 is implemented.
8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the SPECT-oriented bone metastasis identification method described in any one of claims 1 to 5 is implemented.
Citation Information
Cited By
Rheumatism muscle and bone injury intelligent evaluation system and method based on image processing
CN120125576A
Bone lesion classification method and device based on whole-body bone scanning image and clinical information, equipment and medium
CN120808043A
Bone metastasis detecting and positioning method and system based on SPECT bone imaging
CN121213539A
