Cancer risk prediction model training method, and cancer risk prediction method and device
By segmenting the sample image data and extracting the feature of sliced image, and training the cancer risk prediction model with clinical data, the problem of low accuracy of cancer risk prediction in the existing technology is solved, and the accuracy and reliability of lung cancer recurrence risk prediction is improved. .
Patent Information
- Application Number
- CN202411228342.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-11-15
AI Technical Summary
The existing cancer risk prediction methods are based on a single characteristic dimension and cannot guarantee the accuracy of cancer risk prediction, especially in the prediction of recurrence risk after lung cancer treatment.
By image segmentation of the sample image data, only the tumor image area is retained, slice images are collected, imagingomics characteristics and clinical data are extracted, and the cancer risk prediction model is comprehensively trained until the cross entropy loss is less than the preset threshold.
It improves the accuracy and reliability of cancer risk prediction, provides richer prediction information, and enhances the training accuracy of the model.
Smart Images

Figure CN118968226A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of biomedical engineering technology, and in particular to a cancer risk prediction model training method, a cancer risk prediction method and a cancer risk prediction device. Background Art
[0002] Lung cancer is one of the malignant tumors with the highest morbidity and mortality in the world. Particle implantation therapy, as a local treatment method, plays an important role in the treatment of lung cancer. However, due to individual differences, there are significant differences in the treatment effect. Therefore, how to accurately predict the risk of recurrence after treatment before treatment is of great significance to improving the treatment effect.
[0003] At present, cancer risk prediction for patients is often based on extracting tumor image features from imaging data, and cancer risk prediction is performed based on tumor image features. However, this method of predicting cancer risk based on a single feature dimension cannot guarantee the accuracy of cancer risk prediction. Summary of the invention
[0004] The present invention provides a cancer risk prediction model training method, a cancer risk prediction method and a cancer risk prediction device, which can improve the accuracy of cancer risk prediction.
[0005] In a first aspect, a method for training a cancer risk prediction model is provided, comprising:
[0006] Acquire sample data of the sample object, where the sample data includes first sample image data, sample clinical data, and a corresponding cancer risk label;
[0007] Performing image segmentation on the first sample image data to obtain second sample image data that only retains the tumor image area;
[0008] Collecting a portion of sample slice images in the second sample image data, and extracting a first imaging omics feature based on the portion of sample slice images;
[0009] Taking some sample slice images, the first radiomics features and sample clinical data as input features, and the cancer risk label as the training label, an iterative training process of the cancer risk prediction model is performed until the cross entropy loss of the cancer risk prediction model is less than the preset threshold, and the cancer risk prediction model training is judged to be completed.
[0010] In a second aspect, a method for predicting cancer risk is provided, comprising:
[0011] Acquiring first target image data and target clinical data of a target object;
[0012] Performing image segmentation on the first target image data to obtain second target image data that only retains the tumor image area;
[0013] Acquiring a partial target slice image in the second target image data, and extracting target imaging omics features based on the partial target slice image;
[0014] Part of the target slice images, target imaging features and target clinical data are input into the above-mentioned cancer risk prediction model to obtain the cancer risk prediction results of the target object.
[0015] In a third aspect, a training device for a cancer risk prediction model is provided, comprising:
[0016] A first acquisition module, used to acquire sample data of a sample object, the sample data including first sample image data, sample clinical data and a corresponding cancer risk label;
[0017] A first segmentation module is used to perform image segmentation on the first sample image data to obtain second sample image data that only retains the tumor image area;
[0018] A first extraction module, used for collecting a portion of sample slice images in the second sample image data, and extracting a first radiomics feature based on the portion of sample slice images;
[0019] The training module is used to use part of the sample slice images, the first imaging genomics features and the sample clinical data as input features, and the cancer risk label as the training label, to perform an iterative training process on the cancer risk prediction model until the cross entropy loss of the cancer risk prediction model is less than a preset threshold, and it is judged that the cancer risk prediction model training is completed.
[0020] In a fourth aspect, a cancer risk prediction device is provided, comprising:
[0021] A second acquisition module, used to acquire first target image data and target clinical data of a target object;
[0022] A second segmentation module is used to perform image segmentation on the first target image data to obtain second target image data that only retains the tumor image area;
[0023] A second extraction module is used to collect a portion of the target slice images in the second target image data, and extract target imaging omics features based on the portion of the target slice images;
[0024] The prediction module is used to input part of the target slice image, the target imaging features and the target clinical data into the above-mentioned cancer risk prediction model to obtain the cancer risk prediction result of the target object.
[0025] In a fifth aspect, an electronic device is provided, comprising: a processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory to execute the method in the first aspect or the second aspect.
[0026] In a sixth aspect, a computer-readable storage medium is provided for storing a computer program, wherein the computer program enables a computer to execute the method in the first aspect or the second aspect.
[0027] Through the technical solution provided by the present invention, after obtaining the sample data corresponding to the target object including the first sample image data, the sample clinical data and the corresponding cancer risk label, the first sample image data can be segmented to obtain the second sample image data that only retains the tumor image area, so as to avoid the influence of the remaining tissues outside the tumor on the feature extraction; then, the difficulty of feature extraction can be reduced by collecting part of the sample slice images in the second sample image data, and the first imaging genomics feature can be extracted based on the part of the sample slice images, and the part of the sample slice images, the first imaging genomics feature and the sample clinical data are used as input features, and the cancer risk label is used as a training label to perform an iterative training process for the cancer risk prediction model until the cross entropy loss of the cancer risk prediction model is less than the preset threshold, and it is judged that the cancer risk prediction model training is completed. In view of the fact that the sample slice image only contains the tumor image area, the imaging genomics features are also extracted based on the segmented second sample image data, so the reliability of the corresponding input features of the cancer risk prediction model can be improved. On this basis, the cancer risk prediction model is comprehensively trained using three data dimensions: slice images, imaging genomics features, and clinical data. Multiple data dimensions can provide richer information for the cancer risk prediction model, improve the training accuracy of the cancer risk prediction model, and thus improve the accuracy and reliability of cancer risk prediction.
[0028] It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present application. Other features and advantages of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of 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 creative work.
[0030] Figure 1 An application scenario diagram provided for an embodiment of the present application;
[0031] Figure 2 A flowchart of a method for training a cancer risk prediction model provided in an embodiment of the present application;
[0032] Figure 3 A model architecture diagram of a cancer risk prediction model provided in an embodiment of the present application;
[0033] Figure 4 A schematic diagram of a cancer risk prediction method provided in an embodiment of the present application;
[0034] Figure 5 A schematic diagram of the structure of a cancer risk prediction model training device provided in an embodiment of the present application;
[0035] Figure 6 A schematic diagram of the structure of a cancer risk prediction device provided in an embodiment of the present application;
[0036] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0037] 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 making creative work are within the scope of protection of this application.
[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0039] Lung cancer is one of the malignant tumors with the highest morbidity and mortality in the world. Particle implantation therapy, as a local treatment method, plays an important role in the treatment of lung cancer. However, due to individual differences, there are significant differences in the treatment effect. Therefore, how to accurately predict the risk of recurrence after treatment before treatment is of great significance to improving the treatment effect.
[0040] At present, cancer risk prediction for patients is often based on extracting tumor image features from imaging data, and cancer risk prediction is performed based on tumor image features. However, this method of predicting cancer risk based on a single feature dimension cannot guarantee the accuracy of cancer risk prediction.
[0041] It should be noted that the technical solution in this application can be applied to cancer risk prediction for non-therapeutic purposes, and the predicted cancer type can be changed according to the actual application scenario. The predicted cancer types include but are not limited to lung cancer, breast cancer, cervical cancer, gastric cancer, intestinal cancer, pancreatic cancer, prostate cancer, etc. In the following embodiment steps of this application, the recurrence risk prediction of lung cancer after treatment is taken as an example to illustrate the technical solution in this application, but it does not constitute a specific limitation.
[0042] In order to solve the above technical problems, the inventive concept of the present invention is that after obtaining sample data corresponding to the target object including the first sample image data, sample clinical data and the corresponding cancer risk label, the first sample image data is segmented to obtain the second sample image data that only retains the tumor image area to avoid the influence of the remaining tissues outside the tumor on the feature extraction; then, the difficulty of feature extraction can be reduced by collecting part of the sample slice images in the second sample image data, and the first imaging omics features are subsequently extracted based on the part of the sample slice images, and the cancer risk prediction model is trained using the part of the sample slice images, the first imaging omics features, the sample clinical data and the cancer risk label. The trained cancer risk prediction model is used to predict cancer risk in specific application scenarios.
[0043] It should be understood that the technical solution of the present application can be applied to the following scenarios, but is not limited to:
[0044] In some possible implementations, Figure 1 An application scenario diagram provided for an embodiment of the present application, such as Figure 1 As shown, the application scenario may include an electronic device 110 and a network device 120. The electronic device 110 may establish a connection with the network device 120 via a wired network or a wireless network.
[0045] Exemplarily, the electronic device 110 may be a desktop computer, a laptop computer, a tablet computer, etc., but is not limited thereto. The network device 120 may be a terminal device or a server, but is not limited thereto. In one embodiment of the present application, the electronic device 110 may send a request message to the network device 120, and the request message may be used to request to obtain a sample data set. Further, the electronic device 110 may receive a response message sent by the network device 120, and the response message includes the sample data set.
[0046] also, Figure 1An electronic device 110 and a network device 120 are provided as an example, but other numbers of electronic devices and network devices may actually be included, and the present application does not impose any limitation on this.
[0047] In other possible implementations, the technical solution of the present application may also be executed by the above-mentioned electronic device 110, or the technical solution of the present application may also be executed by the above-mentioned network device 120, and the present application does not impose any limitation on this.
[0048] After introducing the application scenarios of the embodiments of the present application, the technical solutions of the present application will be described in detail below:
[0049] Figure 2 A flowchart of a method for training a cancer risk prediction model provided in an embodiment of the present application, the method can be performed as follows: Figure 1 The electronic device 110 shown in the figure performs, but is not limited to this. Figure 2 As shown, the method may include the following steps:
[0050] Step 210: Obtain sample data of the sample object, where the sample data includes first sample image data, sample clinical data, and a corresponding cancer risk label.
[0051] Among them, the sample data is training data for training the cancer risk prediction model, and the first sample image data is data generated by a chest computed tomography (CT) sample object. When the first sample image data is lung image data, these data usually contain grayscale values of slice images at multiple levels, and each slice image shows a cross-section of the lung structure; the lung image data mainly includes the lung area, which usually refers to the part of the lung tissue visible in the image, including structures such as lung parenchyma, bronchi and blood vessels. These areas appear as specific grayscale or density patterns on the image and have certain morphological and texture characteristics; the sample clinical data may include basic information of the sample object (such as gender, age, etc.), tumor type (such as lung squamous cell carcinoma, lung adenocarcinoma, etc.), medical history and lifestyle, and evaluation indicators of treatment effect, such as overall survival (OS), progression-free survival (PFS), objective response rate (ORR) and duration of response (DOR). Cancer risk labels usually refer to training labels that are manually or automatically labeled based on specific areas or structures in lung images, which are used to characterize the probability of cancer recurrence after treatment or the probability of successful recovery after treatment. These labeled data can be used to train machine learning models, perform image segmentation, quantitative analysis, or assist doctors in diagnosis.
[0052] Step 220: perform image segmentation on the first sample image data to obtain second sample image data that only retains the tumor image area.
[0053] In a specific application scenario, the first sample image data may contain other tissue regions in addition to the tumor image region. In order to avoid the adverse effects of other tissue regions on subsequent feature extraction and cancer risk prediction, the first sample image data may be segmented to obtain the second sample image data that only retains the tumor image region.
[0054] When performing image segmentation on the first sample image data:
[0055] As a possible implementation method, the first sample image data may be segmented based on a threshold segmentation method, which divides the pixels of the image into different categories (such as target and background) by setting one or more thresholds. In tumor image region segmentation, a threshold may be set based on the grayscale characteristics of the tumor, thereby distinguishing the tumor region from normal tissue.
[0056] As another possible implementation, the first sample image data may be segmented based on a region segmentation method, which clusters pixels with common properties into regions by analyzing the similarity of pixels (such as color, texture, etc.). In tumor image region segmentation, lesion regions with similar characteristics may be clustered as a whole.
[0057] As another possible implementation, the first sample image data may be segmented based on a boundary segmentation method, which mainly distinguishes different regions based on edges. In tumor segmentation, segmentation may be achieved by detecting the boundary between the tumor and normal tissue.
[0058] As another possible implementation method, the first sample image data can be segmented based on machine learning and deep learning methods. Specifically, the features and boundaries of the tumor can be learned through training models, so that more accurate tumor segmentation can be achieved. Commonly used deep learning models include convolutional neural networks (CNN), U-Net, etc., which can automatically extract image features and perform segmentation with high accuracy and robustness.
[0059] In this embodiment, the specific image segmentation method is not specifically limited, and multiple methods can be adaptively selected or integrated according to the actual application scenario to improve the accuracy and robustness of segmentation. For example, a threshold-based or region-based method can be used to perform rough segmentation on the image, and then a boundary-based or deep learning-based method can be used for fine segmentation.
[0060] Accordingly, for the embodiment of the present disclosure, performing image segmentation on the first sample image data in step 220 to obtain second sample image data retaining only the tumor image area may include the following steps:
[0061] Step 220 - 1 , performing image preprocessing on the first sample image data.
[0062] Among them, preprocessing of medical images aims to improve image quality, enhance image features, and improve diagnostic accuracy and consistency, which may include operations such as image denoising and standardization. Specifically, it can be done through Gaussian filter, median filter, wavelet denoising, and non-local means denoising: based on the principle of similarity, similar blocks in the image are found and weighted average is performed to remove noise. Medical images may be interfered by various noises during acquisition and transmission, such as equipment noise, electromagnetic interference, etc. These noises will reduce the quality of the image and affect the doctor's diagnosis. Through image denoising technology, these noises can be effectively eliminated or weakened to improve the clarity and signal-to-noise ratio of the image. After removing the noise, the useful information in the image (such as tissue boundaries, lesion areas, etc.) will become more prominent, which will help doctors to more accurately identify and analyze the key features in the image. When performing standardization, specific methods such as gray level normalization, histogram equalization, Z-score normalization, and minimum-maximum normalization can be used. Medical imaging devices produced by different manufacturers may have differences in imaging quality, brightness, contrast, etc. Through standardization, these differences can be eliminated to ensure that images generated by different devices are visually consistent and comparable. Standardized medical imaging processing procedures can reduce diagnostic differences between different doctors and different devices, and improve the consistency and reliability of diagnostic results. This is of great significance for improving medical quality and reducing medical costs.
[0063] Step 220 - 2 : Determine the regional position information of the tumor image region in the first sample medical image after image preprocessing.
[0064] For the disclosed embodiment, any of the above optional image segmentation methods can be used to determine the regional position information of the tumor image region, which is not specifically limited here. The regional position information may include but is not limited to the regional boundary coordinate position of the tumor image region.
[0065] Step 220 - 3 : based on the region position information, segment the first sample image data to obtain second sample image data that only retains the tumor image region.
[0066] Step 230 : Acquire a portion of the sample slice images in the second sample image data, and extract a first radiomics feature based on the portion of the sample slice images.
[0067] In a specific application scenario, the second sample image data may include multiple sample slice images. In order to reduce the difficulty of extraction while ensuring the effectiveness of feature extraction, the most representative and meaningful sample slice images may be selected from multiple sample slice images to reduce data dimensions and increase computing speed. Subsequent feature extraction and analysis processes may then be performed based on some sample slice images. By selecting some slice images, the impact of redundant information can be reduced, relevant information can be extracted more specifically, and the accuracy and efficiency of analysis can be improved.
[0068] Among them, when selecting to extract a part of the sample slice images from multiple sample slice images, as a possible implementation method, the slice images can be uniformly sampled in the second sample image data based on a preset sampling frequency to obtain a part of the slice images of the second sample image data; as a possible implementation method, the sample slice images can be screened according to a defined feature selection criterion (such as relevance, importance, etc.). Statistical methods (such as chi-square test, mutual information), machine learning algorithms (such as decision trees, random forests) or other feature selection techniques can be used to determine the optimal part of the sample slice images from multiple sample slice images.
[0069] Step 240: Using part of the sample slice images, the first radiomics feature, and the sample clinical data as input features, and the cancer risk label as a training label, an iterative training process of the cancer risk prediction model is performed until the cross entropy loss of the cancer risk prediction model is less than a preset threshold, and it is determined that the cancer risk prediction model training is completed.
[0070] Among them, Figure 3 As shown, the cancer risk prediction model includes a ResNet3D network module 31, a first feature processing module 32, a second feature processing module 33, a feature fusion module 34 and a risk prediction module 35.
[0071] Accordingly, the iterative training process of the cancer risk prediction model can include the following steps:
[0072] Step 240 - 1 , input part of the sample slice image into the ResNet3D network module for feature extraction, and obtain the spatial structural features of the tumor image area corresponding to the preset feature dimension.
[0073] ResNet is a deep convolutional neural network architecture that has made significant breakthroughs in computer vision tasks with its depth and performance. The main innovation of ResNet is the introduction of residual blocks. Incremental changes to the input are learned through jump connections instead of directly learning the complete mapping. This helps solve the problems of gradient vanishing and gradient exploding in deep neural networks, making it possible to train very deep networks. Since 2D convolution can only process slice information and cannot fully reflect three-dimensional structural features, 3D convolution is usually used when extracting tumor features to capture spatial information in the tumor volume. Therefore, the present invention replaces the 2D convolution in ResNet18 with 3D convolution to obtain the final ResNet3D model.
[0074] For the disclosed embodiment, part of the sample slice image (1×20×224×224) can be input into the ResNet3D network module for feature extraction to obtain the spatial structure features (120×1) of the tumor image area corresponding to the preset feature dimension.
[0075] Step 240 - 2 : Input the first radiomics feature into the first feature processing module for feature transformation processing to obtain a second radiomics feature under a preset feature dimension, wherein the second radiomics feature includes at least a shape feature, a texture feature, and an intensity feature.
[0076] Among them, Figure 3 As shown, the first feature processing module includes a first network layer, and the first network layer includes a linear layer (Linear), a normalization layer (BN), and an activation layer (SELU) connected in sequence.
[0077] Accordingly, for the disclosed embodiment, the embodiment steps may include: inputting the first imaging omics feature into the first feature processing module, using the first network layer to sequentially perform linear transformation processing, normalization processing, and activation processing on the first imaging omics feature, and obtain the second imaging omics feature under the preset feature dimension. That is, for the disclosed embodiment, the first imaging omics feature (228×1) can be input into the first feature processing module, and the linear layer (Linear), normalization layer (BN), and activation layer (SELU) in the first feature processing module can sequentially perform linear transformation processing, normalization processing, and activation processing on the first imaging omics feature, and obtain the second imaging omics feature (120×1) under the preset feature dimension.
[0078] Step 240 - 3: Input the sample clinical data into the second feature processing module for feature extraction to obtain clinical features under preset feature dimensions.
[0079] Among them, Figure 3As shown, the second feature processing module includes a second network layer, the second network layer is composed of two first network layers, and the first network layer includes a linear layer (Linear), a normalization layer (BN) and an activation layer (SELU) connected in sequence.
[0080] Accordingly, for the embodiment of the present disclosure, the embodiment steps may include: inputting the sample clinical data into the second feature processing module, using the second network layer to perform linear transformation processing, normalization processing and activation processing on the sample clinical data twice in succession, and obtaining clinical features under the preset feature dimension. That is, for the embodiment of the present disclosure, the sample clinical data (29×1) may be input into the second feature processing module, and the linear layer (Linear), normalization layer (BN) and activation layer (SELU) in the second feature processing module may be used to perform linear transformation processing, normalization processing and activation processing on the sample clinical data twice in succession, and obtain clinical features under the preset feature dimension (120×1).
[0081] Step 240 - 4: Use the feature fusion module to perform feature splicing on the spatial structure feature, the second radiomics feature, and the clinical feature to obtain a high-dimensional feature.
[0082] For the embodiments of the present disclosure, Figure 3 As shown, the feature fusion module (Concat) can be used to fuse the two spatial structure features (120×1), the second radiomics features (120×1) and the clinical features (120×1) to obtain high-dimensional features (360×1).
[0083] Step 240 - 5: Input the high-dimensional features into the risk prediction module to obtain the cancer risk prediction result of the sample object.
[0084] Among them, Figure 3 As shown, the risk prediction module includes a fully connected layer, which includes a linear layer (Linear) and a Sigmoid function.
[0085] Correspondingly, for the embodiments of the present disclosure, the embodiment steps may include: inputting high-dimensional features into a fully connected layer for linear transformation processing to obtain linear transformation features; and using a Sigmoid function to determine a cancer risk prediction value of the sample object based on the linear transformation features.
[0086] Step 240 - 6: Calculate the cross entropy loss of the cancer risk prediction model based on the cancer risk prediction result and the cancer risk label, and optimize the model parameters of the cancer risk prediction model based on the cross entropy loss.
[0087] In specific application scenarios, the cross entropy loss function can be used to calculate the cross entropy between the cancer risk predicted by the cancer risk prediction model and the cancer risk label. This loss value represents the accuracy of the model prediction. The smaller the loss value, the more accurate the model prediction. When optimizing the model parameters of the cancer risk prediction model based on the cross entropy loss, the gradient of the loss function with respect to the model parameters can be calculated through the back propagation algorithm. This gradient indicates how to adjust the model parameters to reduce the loss value. Then, an optimization algorithm (such as gradient descent, Adam, etc.) is used to update the model parameters according to the calculated gradient. The purpose of parameter updating is to minimize the loss function, thereby improving the prediction accuracy of the model.
[0088] In summary, according to the training method of the cancer risk prediction model provided by the present invention, after obtaining the sample data corresponding to the target object including the first sample image data, the sample clinical data and the corresponding cancer risk label, the first sample image data can be segmented to obtain the second sample image data that only retains the tumor image area, so as to avoid the influence of the remaining tissues outside the tumor on the feature extraction; then, the difficulty of feature extraction can be reduced by collecting part of the sample slice images in the second sample image data, and the first imaging genomics feature can be extracted based on the part of the sample slice images, and the part of the sample slice images, the first imaging genomics feature and the sample clinical data are used as input features, and the cancer risk label is used as a training label to perform an iterative training process for the cancer risk prediction model until the cross entropy loss of the cancer risk prediction model is less than the preset threshold, and it is judged that the cancer risk prediction model training is completed. In view of the fact that the slice image only contains the tumor image area, the imaging genomics features are also extracted based on the segmented second sample image data, so the reliability of the corresponding input features of the cancer risk prediction model can be improved. On this basis, the cancer risk prediction model is comprehensively trained using three data dimensions: slice images, imaging genomics features, and clinical data. Multiple data dimensions can provide richer information for the cancer risk prediction model, improve the training accuracy of the cancer risk prediction model, and thus improve the accuracy and reliability of cancer risk prediction.
[0089] Figure 4 A flowchart of a cancer risk prediction method according to an embodiment of the present invention is provided. The method can be performed as follows: Figure 1 The electronic device 110 shown in the figure performs, but is not limited to this. Figure 4 As shown, the following steps may be included:
[0090] Step 410: Acquire first target image data and target clinical data of the target object.
[0091] Among them, the target object is the object to be predicted for cancer risk; the first target image data is the data generated by the chest computed tomography (CT) target object. When the first target image data is lung image data, these data usually contain grayscale values of slice images at multiple levels, and each slice image shows a cross-section of the lung structure; the lung image data mainly includes the lung area, which usually refers to the part of the lung tissue visible in the image, including structures such as lung parenchyma, bronchi and blood vessels. These areas appear as specific grayscale or density patterns on the image and have certain morphological and texture characteristics; the target clinical data may include basic information of the target object (such as gender, age, etc.), tumor type (such as lung squamous cell carcinoma, lung adenocarcinoma, etc.), medical history and lifestyle, and evaluation indicators of treatment effect, such as overall survival (OS), progression-free survival (PFS), objective response rate (ORR) and duration of response (DOR).
[0092] Step 420: perform image segmentation on the first target image data to obtain second target image data that only retains the tumor image area.
[0093] In a specific application scenario, the first target image data may contain other tissue regions in addition to the tumor image region. In order to avoid the adverse effects of other tissue regions on subsequent feature extraction and cancer risk prediction, the first target image data may be segmented to obtain the second target image data that only retains the tumor image region.
[0094] When performing image segmentation on the first target image data, the specific implementation steps may refer to the relevant description in step 220 of the embodiment, which will not be repeated here.
[0095] Step 430 : Acquire a portion of the target slice images in the second target image data, and extract target imaging omics features based on the portion of the target slice images.
[0096] In a specific application scenario, the second target image data may include multiple target slice images. In order to reduce the difficulty of extraction while ensuring the effectiveness of feature extraction, the most representative and meaningful partial target slice images may be selected from multiple target slice images to reduce data dimensions and increase computing speed. Subsequent feature extraction and analysis processes may then be performed based on partial target slice images. By selecting partial target slice images, the impact of redundant information may be reduced, relevant information may be extracted more specifically, and the accuracy and efficiency of analysis may be improved. When selecting to extract partial target slice images from multiple target slice images, the specific implementation process may refer to the relevant description in step 230 of the embodiment, which will not be repeated here.
[0097] Step 440: Input part of the target slice image, the target imaging genomics features, and the target clinical data into the trained cancer risk prediction model to obtain a cancer risk prediction result for the target subject.
[0098] Among them, the cancer risk prediction model includes a ResNet3D network module, a first feature processing module, a second feature processing module, a feature fusion module and a risk prediction module.
[0099] For the embodiment of the present disclosure, when a portion of the target slice image, the target radiomics features, and the target clinical data are input into the trained cancer risk prediction model to obtain the cancer risk prediction result of the target object, step 440 may include:
[0100] Step 440 - 1 , input part of the target slice image into the ResNet3D network module for feature extraction, and obtain the target spatial structure features under the preset feature dimension corresponding to the tumor image area.
[0101] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 240-1 of the embodiment, which will not be repeated here.
[0102] Step 440 - 2 : Input the target radiomics features into the first feature processing module for feature transformation processing to obtain the target radiomics features under the preset feature dimensions, where the target radiomics features include at least shape features, texture features and intensity features.
[0103] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 240-2 of the embodiment, which will not be repeated here.
[0104] Step 440 - 3 : Input the target clinical data into the second feature processing module for feature extraction to obtain the target clinical features under the preset feature dimensions.
[0105] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 240-3 of the embodiment, which will not be repeated here.
[0106] Step 440 - 4 , using a feature fusion module to perform feature splicing on the target spatial structure features, target radiomics features, and target clinical features to obtain target high-dimensional features.
[0107] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 240-4 of the embodiment, which will not be repeated here.
[0108] Step 440 - 5 : Input the target high-dimensional features into the risk prediction module to obtain the cancer risk prediction result of the target object.
[0109] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 240-5 of the embodiment, which will not be repeated here.
[0110] In summary, after determining the first target image data and target clinical data of the target object, the present application can perform image segmentation on the first target image data to obtain the second target image data that only retains the tumor image area, so as to avoid the influence of the remaining tissues outside the tumor on the feature extraction; then, by collecting part of the target slice images in the second target image data, the difficulty of feature extraction can be reduced, and the target imaging genomics features can be subsequently extracted based on part of the sample slice images. Subsequently, part of the target slice images, the target imaging genomics features and the target clinical data are input into the trained cancer risk prediction model to obtain the cancer risk prediction result of the target object. In view of the fact that the target slice image only contains the tumor image area, the target imaging genomics features are also extracted based on the segmented second target image data, so the reliability of the corresponding input features of the cancer risk prediction model can be improved. In addition, in view of the fact that the cancer risk prediction model is obtained by comprehensive training using three data dimensions: slice images, imaging genomics features and clinical data, the accuracy and reliability of cancer risk prediction can be improved.
[0111] Based on the above Figure 2 A detailed description of the training method of the cancer risk prediction model provided, such as Figure 5 As shown, Figure 5 FIG. 1 is a block diagram of a training device for a cancer risk prediction model according to an exemplary embodiment. Figure 5 As shown, the device comprises:
[0112] A first acquisition module 51, which can be used to acquire sample data of a sample object, where the sample data includes first sample image data, sample clinical data, and a corresponding cancer risk label;
[0113] The first segmentation module 52 may be used to perform image segmentation on the first sample image data to obtain second sample image data that only retains the tumor image area;
[0114] A first extraction module 53, which may be used to collect a portion of sample slice images in the second sample image data, and extract a first radiomics feature based on the portion of sample slice images;
[0115] The training module 54 can be used to use part of the sample slice images, the first imaging genomics features and the sample clinical data as input features, and the cancer risk label as a training label to perform an iterative training process on the cancer risk prediction model until the cross entropy loss of the cancer risk prediction model is less than a preset threshold, and it is determined that the cancer risk prediction model training is completed.
[0116] In some embodiments of the present application, the first segmentation module 52 can be specifically used to perform image preprocessing on the first sample image data; determine the regional position information of the tumor image area in the first sample medical image after image preprocessing; and based on the regional position information, segment the first sample image data to obtain the second sample image data that only retains the tumor image area.
[0117] In some embodiments of the present application, the cancer risk prediction model includes a ResNet3D network module, a first feature processing module, a second feature processing module, a feature fusion module and a risk prediction module; when the training module 54 executes the iterative training process of the cancer risk prediction model, it includes: inputting part of the sample slice image into the ResNet3D network module for feature extraction to obtain the spatial structure feature of the tumor image area corresponding to the preset feature dimension; inputting the first imaging genomics feature into the first feature processing module for feature transformation processing to obtain the second imaging genomics feature under the preset feature dimension, and the second imaging genomics feature at least includes shape feature, texture feature and intensity feature; inputting the sample clinical data into the second feature processing module for feature extraction to obtain the clinical feature under the preset feature dimension; using the feature fusion module to perform feature splicing on the spatial structure feature, the second imaging genomics feature and the clinical feature to obtain high-dimensional features; inputting the high-dimensional features into the risk prediction module to obtain the cancer risk prediction result of the sample object; calculating the cross entropy loss of the cancer risk prediction model based on the cancer risk prediction result and the cancer risk label, and optimizing the model parameters of the cancer risk prediction model based on the cross entropy loss.
[0118] In some embodiments of the present application, the first feature processing module includes a first network layer, the second feature processing module includes a second network layer, the second network layer is composed of two first network layers, and the first network layer includes a linear layer, a normalization layer and an activation layer connected in sequence; when the training module 54 executes the input of the first imaging genomics feature into the first feature processing module for feature transformation processing to obtain the second imaging genomics feature under the preset feature dimension, it includes: inputting the first imaging genomics feature into the first feature processing module, and using the first network layer to perform linear transformation processing, normalization processing and activation processing on the first imaging genomics feature in sequence to obtain the second imaging genomics feature under the preset feature dimension; when the training module 54 executes the input of the sample clinical data into the second feature processing module for feature extraction to obtain the clinical features under the preset feature dimension, it includes: inputting the sample clinical data into the second feature processing module, and using the second network layer to perform linear transformation processing, normalization processing and activation processing on the sample clinical data twice in succession to obtain the clinical features under the preset feature dimension.
[0119] In some embodiments of the present application, the risk prediction module includes a fully connected layer, which includes a linear layer and a Sigmoid function; when the training module 54 inputs high-dimensional features into the risk prediction module to obtain the cancer risk prediction result of the sample object, it includes: inputting the high-dimensional features into the fully connected layer for linear transformation processing to obtain linear transformation features; using the Sigmoid function to determine the cancer risk prediction value of the sample object based on the linear transformation features.
[0120] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0121] In the embodiment of the present application, since the sample slice image only contains the tumor image area, the imaging genomics features are also extracted based on the segmented second sample image data, so the reliability of the corresponding input features of the cancer risk prediction model can be improved. On this basis, the cancer risk prediction model is comprehensively trained using the three data dimensions of slice images, imaging genomics features, and clinical data. Multiple data dimensions can provide richer information for the cancer risk prediction model, which can improve the training accuracy of the cancer risk prediction model, thereby improving the accuracy and reliability of cancer risk prediction.
[0122] Based on the above Figure 4 A specific description of the cancer risk prediction method provided is provided. After the cancer risk prediction model is obtained through training, the use of the cancer risk prediction model can refer to the use of a normal deep learning model. Specifically, a cancer risk prediction device is provided, such as Figure 6 As shown, the device comprises:
[0123] The second acquisition module 61 may be used to acquire first target image data and target clinical data of a target object;
[0124] The second segmentation module 62 may be used to segment the first target image data to obtain second target image data that only retains the tumor image area;
[0125] A second extraction module 63, which can be used to collect a portion of the target slice images in the second target image data, and extract target imaging features based on the portion of the target slice images;
[0126] The prediction module 64 can be used to input part of the target slice image, the target imaging features and the target clinical data into the above-mentioned cancer risk prediction model to obtain the cancer risk prediction result of the target object.
[0127] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0128] In the embodiment of the present application, since the target slice image only contains the tumor image area, the target imaging genomics features are also extracted based on the segmented second target image data, so the reliability of the corresponding input features of the cancer risk prediction model can be improved. In addition, since the cancer risk prediction model is obtained by comprehensive training using three data dimensions: slice images, imaging genomics features, and clinical data, the accuracy and reliability of cancer risk prediction can be improved.
[0129] In the above, the training device of the cancer risk prediction model and the cancer risk prediction device of the embodiment of the present invention are described from the perspective of the functional module in combination with the accompanying drawings. It should be understood that the functional module can be implemented in hardware form, can be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, the training method of the cancer risk prediction model and the steps of the cancer risk prediction method embodiment in the embodiment of the present invention can be completed by the hardware integrated logic circuit and / or software instructions in the processor, and the training method of the cancer risk prediction model and the steps of the cancer risk prediction method applied in combination with the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory, and combines its hardware to complete the steps in the above-mentioned cancer risk prediction model training method and cancer risk prediction method embodiment.
[0130] Figure 7It is a schematic block diagram of an electronic device 700 according to an embodiment of the present invention.
[0131] like Figure 7 As shown, the electronic device 700 may include:
[0132] The memory 710 and the processor 720, the memory 710 is used to store the computer program and transmit the program code to the processor 720. In other words, the processor 720 can call and run the computer program from the memory 710 to implement the method in the embodiment of the present invention.
[0133] For example, the processor 720 may be configured to execute the above method embodiments according to instructions in the computer program.
[0134] In some embodiments of the present invention, the electronic device 720 may include but is not limited to:
[0135] General-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.
[0136] In some embodiments of the present invention, the memory 710 includes but is not limited to:
[0137] Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM).
[0138] In some embodiments of the present invention, the computer program may be divided into one or more modules, which are stored in the memory 710 and executed by the processor 720 to complete the method provided by the present invention. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the controller.
[0139] like Figure 7 As shown, the electronic device 700 may further include:
[0140] The transceiver 730 may be connected to the processor 720 or the memory 710 .
[0141] The processor 720 may control the transceiver 730 to communicate with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. The transceiver 730 may include a transmitter and a receiver. The transceiver 730 may further include an antenna, and the number of antennas may be one or more.
[0142] It should be understood that the various components in the electronic device are connected via a bus system, wherein the bus system includes not only a data bus but also a power bus, a control bus and a status signal bus.
[0143] The present invention also provides a computer storage medium having a computer program stored thereon, which, when executed by a computer, enables the computer to perform the method of the above method embodiment. In other words, an embodiment of the present invention also provides a computer program product containing instructions, which, when executed by a computer, enables the computer to perform the method of the above method embodiment.
[0144] When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (Digital Subscriber Line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integration. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a digital video disc (Digital Video Disc, DVD)), or a semiconductor medium (e.g., a solid-state drive (Solid State Disk, SSD)), etc.
[0145] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments applied herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0146] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the module is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0147] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. For example, each functional module in each embodiment of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0148] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in this application, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A method for training a cancer risk prediction model, characterized in that: include: Acquire sample data of a sample object, wherein the sample data includes first sample image data, sample clinical data, and a corresponding cancer risk label; Performing image segmentation on the first sample image data to obtain second sample image data retaining only the tumor image area; Acquiring a portion of sample slice images in the second sample image data, and extracting a first imaging omics feature based on the portion of sample slice images; The partial sample slice image, the first imaging genomics feature and the sample clinical data are used as input features, and the cancer risk label is used as a training label. An iterative training process of the cancer risk prediction model is performed until the cross entropy loss of the cancer risk prediction model is less than a preset threshold, and it is determined that the cancer risk prediction model training is completed.
2. The method according to claim 1, characterized in that The performing image segmentation on the first sample image data to obtain second sample image data retaining only the tumor image area includes: performing image preprocessing on the first sample image data; Determining the regional position information of the tumor image region in the first sample medical image after the image preprocessing; Based on the region position information, the first sample image data is segmented to obtain second sample image data that only retains the tumor image region.
3. The method according to claim 1, characterized in that The cancer risk prediction model includes a ResNet3D network module, a first feature processing module, a second feature processing module, a feature fusion module and a risk prediction module; The iterative training process of the cancer risk prediction model includes: Inputting the partial sample slice images into the ResNet3D network module for feature extraction to obtain spatial structural features of the tumor image region corresponding to a preset feature dimension; Inputting the first radiomics feature into the first feature processing module for feature transformation processing to obtain a second radiomics feature under the preset feature dimension, wherein the second radiomics feature at least includes a shape feature, a texture feature, and an intensity feature; Inputting the sample clinical data into the second feature processing module for feature extraction to obtain clinical features under the preset feature dimension; Using the feature fusion module to perform feature splicing on the spatial structure feature, the second radiomics feature, and the clinical feature to obtain a high-dimensional feature; Inputting the high-dimensional features into the risk prediction module to obtain a cancer risk prediction result for the sample object; The cross entropy loss of the cancer risk prediction model is calculated based on the cancer risk prediction result and the cancer risk label, and the model parameters of the cancer risk prediction model are optimized based on the cross entropy loss.
4. The method according to claim 3, characterized in that The first feature processing module includes a first network layer, the second feature processing module includes a second network layer, the second network layer is composed of two of the first network layers, and the first network layer includes a linear layer, a normalization layer, and an activation layer connected in sequence; Inputting the first radiomics feature into the first feature processing module for feature transformation processing to obtain a second radiomics feature under the preset feature dimension, including: Inputting the first radiomics feature into the first feature processing module, and using the first network layer to sequentially perform linear transformation processing, normalization processing, and activation processing on the first radiomics feature to obtain a second radiomics feature under the preset feature dimension; Inputting the sample clinical data into the second feature processing module for feature extraction to obtain clinical features under the preset feature dimension, including: The sample clinical data is input into the second feature processing module, and the second network layer is used to perform linear transformation processing, normalization processing and activation processing on the sample clinical data twice in succession to obtain the clinical features under the preset feature dimension.
5. The method according to claim 3, characterized in that: The risk prediction module includes a fully connected layer, and the fully connected layer includes a linear layer and a Sigmoid function; Inputting the high-dimensional features into the risk prediction module to obtain the cancer risk prediction result of the sample object includes: Inputting the high-dimensional features into the fully connected layer for linear transformation processing to obtain linear transformation features; The Sigmoid function is used to determine the cancer risk prediction value of the sample object based on the linear transformation feature.
6. A method for predicting cancer risk, characterized in that: include: Acquiring first target image data and target clinical data of a target object; Performing image segmentation on the first target image data to obtain second target image data retaining only the tumor image area; Acquiring a portion of the target slice images in the second target image data, and extracting target imaging omics features based on the portion of the target slice images; The partial target slice image, the target imaging features and the target clinical data are input into the cancer risk prediction model described in any one of claims 1 to 5 to obtain a cancer risk prediction result for the target object.
7. A training device for a cancer risk prediction model, characterized in that: include: A first acquisition module, used to acquire sample data of a sample object, wherein the sample data includes first sample image data, sample clinical data, and a corresponding cancer risk label; A first segmentation module, used to perform image segmentation on the first sample image data to obtain second sample image data that only retains the tumor image area; a first extraction module, configured to collect a portion of sample slice images in the second sample image data, and extract a first radiomics feature based on the portion of sample slice images; A training module is used to use the partial sample slice image, the first imaging genomics feature and the sample clinical data as input features, and the cancer risk label as a training label to perform an iterative training process on the cancer risk prediction model until the cross entropy loss of the cancer risk prediction model is less than a preset threshold, and it is determined that the cancer risk prediction model training is completed.
8. A cancer risk prediction device, characterized in that: include: A second acquisition module, used to acquire first target image data and target clinical data of a target object; A second segmentation module is used to perform image segmentation on the first target image data to obtain second target image data that only retains the tumor image area; a second extraction module, configured to collect a portion of target slice images in the second target image data, and extract target imaging omics features based on the portion of target slice images; A prediction module is used to input the partial target slice image, the target imaging features and the target clinical data into the cancer risk prediction model according to any one of claims 1 to 5 to obtain a cancer risk prediction result for the target object.
9. An electronic device, characterized in that: include: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory to execute the training method for the cancer risk prediction model as described in any one of claims 1 to 5 or the cancer risk prediction method as described in claim 6.
10. A computer-readable storage medium, characterized in that: Used to store a computer program to execute the training method of the cancer risk prediction model as described in any one of claims 1 to 5 or the cancer risk prediction method as described in claim 6.
Citation Information
Patent Citations
Image omics based lesion tissue auxiliary prognosis system and method
CN105653858A
Method and system for predicting liver cancer recurrence through combination of radiomics and deep learning, and medium
CN113780274A
Multi-modal prediction model construction method and system for analyzing liver cancer recurrence data
CN117612711A
Cited By
Method for training prediction model for bleeding risk and related product
CN121416083A
Methods for training a predictive model for bleeding risk and related products
CN121416083B
