Method, System, Device and Storage Medium for Predicting Relative Survival Risk of Breast Cancer

Abstract

The present invention belongs to the field of biotechnology, and discloses a method, system, computer device and storage medium for predicting the relative survival risk of breast cancer by combining histological whole-slide images and gene mutation tags. The method includes: for each of a pair of patients, obtaining tumor histological whole-slide image data and gene mutation data; obtaining image features according to the obtained image data; selecting genes that have a significant impact on survival from the gene mutation data to obtain genomic features; processing the image features and the genomic features through a siamese network, including: using a recurrent neural network to process the image features and using a fully connected network to process the genomic features; splicing the processed image features and genomic features to obtain the fusion features of the patient; and for the pair of patients, predicting the relative survival risk of the pair of patients using the output linear layer based on the difference of the fusion features of the pair of patients.

Description

Technical Field

[0001] The present invention relates to the field of biomedicine, and in particular to a method, system, computer device and non-transitory computer-readable storage medium for predicting the relative survival risk of breast cancer by combining histological whole-slide images and gene mutation tags. Background Art

[0002] According to World Health Organization cancer data (https: / / www.who.int / en / news-room / fact-sheets / detail / cancer)[1], breast cancer is one of the leading causes of death in women. Breast cancer is a very complex disease, and the outcomes of different patients often vary greatly. Currently, standard breast cancer treatment regimens include surgery (mastectomy), chemotherapy, radiotherapy, and possibly hormone therapy or targeted therapy. Existing treatment regimens aim to remove tumors and kill any remaining tumor cells, and often need to be adjusted according to the patient's tumor grade and overall health status. Therefore, if the survival risk of patients can be predicted more accurately, it can help doctors better adjust treatment regimens.

[0003] Existing breast cancer survival analysis methods can be classified into three categories according to the data they use: methods that only use imaging data, such as "Assessing risk of breast cancer recurrence" by Barnes et al. (US Patent Application No. 10489904) [2] and "Assessment of nodalinvolvement and survival analysis in breast cancer patients using imagecytometric data: statistical, neural network and fuzzy approaches" by Seker, Huseyin et al. (Anticancerresearch 22.1A (2002): pp. 433-438 [3]; methods that only use genomics data, such as "SALMON: survival analysis learning with multi-omics neural networks on breast cancer" by Huang, Zhi, et al. (Frontiers in genetics 10 (2019): 166) [4], "Deep learning-based feature-level integration of multi-omics data for breast cancer patients survival analysis" by Li, et al. (BMC Medical Informatics and Decision Making, (2020) 20:225) [5], "12-gene prognostic signature for breast cancer survival" by Snijders, et al. (U.S. Patent Application No. 10876767) [6], "Methods for determining a prognosis for survival for a patient with breast cancer" by Pendino, et al. (U.S. Patent Application No. 9512484) [7], and "Breast cancer prognosis assessment model and method for establishing the same" by Wang, Xin, et al. (Chinese Patent Application Publication No. CN110656173A) [8]; and methods that only use clinical data, such as "Construction the model on the breast cancer survival analysis using support vector machine, logistic regression and decision tree" by Chao, Cheng-Min, et al. (Journal of medical systems 38.10 (2014): pp. 1-7) [9] and "A method for predicting breast cancer prognosis survival rate based on dynamic Cox model" by Teng, Jing, et al. (Chinese Patent Application Publication No. CN108922628A)

[10] .

[0004] Specifically, reference [2] uses whole-slide histopathological images to predict the recurrence risk (high risk / low risk) of breast cancer patients. Reference [3] uses the data of cell counts in images to predict the 5-year survival status of breast cancer patients. Reference [4] is based on multi-omics data (mRNA sequencing data, miRNA sequencing data, copy number burden, tumor mutation burden, estrogen and progesterone receptor status), and uses deep learning methods to predict the survival risk of breast cancer patients. Reference [5] also uses multi-omics data (gene expression, DNA methylation, miRNA expression, copy number variation), and uses deep learning methods to predict the survival risk of breast cancer patients. Reference [6] proposes a prognostic index based on 12 genes. Reference [7] uses the expression level of CXXC5 mRNA to predict the survival risk of breast cancer patients and monitor the effectiveness of breast cancer treatment. Reference [8] extracts 190 genes based on the analysis of the expression levels of RNA sequence data and uses a support vector machine to predict whether breast cancer patients will relapse. Reference [9] is based on clinical data (pathological grade, whether chemotherapy is received, whether radiotherapy is received, age, tumor size, number of lymph nodes examined, number of lymph nodes attacked), and uses a support vector machine, logistic regression or decision tree to predict whether breast cancer patients will survive. Reference

[10] is based on clinical data (tumor size and location, number of lymph nodes examined, number of lymph nodes attacked), and uses a dynamic Cox model to predict the survival risk of breast cancer patients.

[0005] All these techniques only use a single data source, and the information contained therein is limited. Summary of the Invention

[0006] The present invention proposes a method for predicting the relative survival risk of breast cancer patients by combining whole-slide histopathological images and gene mutation tags. This method uses a siamese network to predict the relative survival risk of patients. First, image features and genomic features are extracted separately, and then the above siamese network is used to fuse them. Finally, the fused features are used for the prediction of relative risk.

[0007] Specifically, according to the first aspect of the present invention, the present invention provides a method for predicting the relative survival risk of breast cancer by combining whole-slide histopathological images and gene mutation tags, and the method includes the following steps:

[0008] (a) For each of a pair of patients, obtain whole-slide histopathological image data of the tumor tissue and gene mutation data;

[0009] (b) Obtain image features according to the obtained image data, preferably including: dividing the whole-slide histopathological image into image patches, screening out non-tumor image patches and clustering the remaining image patches, and using the sorted class centers as the image features;

[0010] (c) Select genes with significant survival impact from the gene mutation data to obtain genomic features;

[0011] (d) Process the image features and the genomic features through a siamese network, the siamese network including a recurrent neural network (RNN) for processing image features, a fully connected network (FCN) for processing genomic features, and an output linear layer for outputting results, including: processing the image features using the recurrent neural network (RNN), and processing the genomic features using the fully connected network (FCN);

[0012] (e) Concatenate the processed image features and genomic features to obtain the fusion features of the patient; and

[0013] (f) For the pair of patients, predict the relative survival risk of the pair of patients using the output linear layer based on the difference of the fusion features of the pair of patients (e.g., directly subtracting the corresponding elements).

[0014] In one embodiment, in step (b), slicing the histological whole slide image into the image patches includes: slicing side by side from the magnified histological whole slide image to obtain the image patches.

[0015] In one embodiment, step (b) further includes: performing color normalization processing on the image patches after slicing the histological whole slide image into the image patches.

[0016] In one embodiment, step (b) further includes: performing feature extraction on the image patches, using a pre-trained neural network to perform feature extraction on the image patches, preferably further including: the pre-trained neural network is a PNASNet neural network pre-trained on an image classification database such as ImageNet.

[0017] In one embodiment, screening out non-tumor image patches in step (b) includes: filtering the extracted features using a Gaussian mixture model (GMM) trained on the features of non-tumor region image patches to screen out non-tumor image patches, preferably using the obtained GMM model to rank the class centers in step (b).

[0018] In one embodiment, the gene selection in step (c) is achieved through a log-rank test of survival information on the mutation set and non-mutation set of genes, preferably with a p-value threshold of 0.05.

[0019] In one embodiment, the RNN in step (d) is an independent two-layer network, with 1024 hidden nodes in each layer.

[0020] In one embodiment, the FCN in step (d) is a three-layer network, where the number of nodes in each layer is 1024, 512, and 256 respectively.

[0021] In one embodiment, the output linear layer in step (f) has no bias parameter.

[0022] In one embodiment, the siamese network is trained by using the data of multiple patients with breast cancer survival risk data as the training set data. Preferably, the steps of the training include:

[0023] (a) For each of the multiple patients, obtain the whole-slide image data of tumor histology and gene mutation data;

[0024] (b) Obtain image features according to the obtained image data, including: dividing the whole-slide histology image into image patches, extracting features from the image patches, screening out non-tumor image patches and clustering the remaining image patches, and taking the sorted class centers as the image features;

[0025] (c) Select genes that have a significant impact on survival from the gene mutation data to obtain genomic features;

[0026] (d) Process the image features and the genomic features through the siamese network, including: using a recurrent neural network (RNN) to process the image features and using a fully connected network (FCN) to process the genomic features;

[0027] (e) Concatenate the processed image features and genomic features to obtain the fused features of the patient;

[0028] (f) For each pair of the multiple patients, train the siamese network based on the difference between the fused features of the pair of patients and their relative survival risks, including: using the cross-entropy loss function as supervision according to the predicted relative survival risk for the pair of patients and the actual survival data of the pair of patients to train the network parameters of the siamese network;

[0029] (g) The training process of the siamese network is divided into three stages: in the first stage, only use the image feature part and train the RNN part alone; in the second stage, add the genomic features, fix the parameters of the RNN, and only train the FCN part; in the third stage, release the fixation of the RNN parameters and perform further joint optimization training on the parameters of the RNN and FCN.

[0030] In one embodiment, the performance evaluation index for evaluating the prediction accuracy is the coincidence index.

[0031] In one embodiment, the method is applicable to cancers having the histological whole slide image and the gene mutation label.

[0032] In a second aspect, the present invention provides a system for predicting survival risk by combining a histological whole slide image and a gene mutation label, including a processor configured to execute computer instructions to cause the method described in the first aspect of the present invention to be executed.

[0033] In a third aspect, the present invention provides a computer device, including a memory and a processor, where computer instructions are stored on the memory, and when the computer instructions are executed by the processor, the method described in the first aspect of the present invention is caused to be executed.

[0034] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the method described in the first aspect of the present invention is caused to be executed.

[0035] Using the solution of the present invention, a histological whole slide image and a gene mutation label can be combined to predict survival risk, which helps to improve the accuracy of predicting survival risk, thereby assisting doctors in better adjusting treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention will now be described only by way of non-limiting examples with reference to the accompanying drawings, where:

[0037] Figure 1 Schematically shows the process of obtaining image features from a histological whole slide of a patient according to an embodiment of the present invention.

[0038] Figure 2 Schematically shows a siamese network designed for predicting relative survival risk according to an embodiment of the present invention.

[0039] Figure 3 Schematically shows the concordance index (c-index) of a 5-fold cross-validation experiment on the TCGA-BRCA dataset using three inputs: only image features, only genomic features, and image features and genomic features, respectively, according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The inventors of the present invention have experimentally confirmed that the combination of imaging data and genomic data helps to improve the accuracy of predicting breast cancer survival risk. The information contained in imaging data and genomic data is different, and there is no relevant research on the combination of the two.

[0041] Regarding the selection of imaging data features, Figure 1Shows the process of obtaining image features from a patient's histological whole slide according to an embodiment of the present invention. The specific details are described as follows.

[0042] For a patient, there may often be multiple histological whole slide images. Therefore, the inventor assumes that patient p i has N i histological whole slide images of size . Usually, histological whole slide images are very large, and it is impractical to directly use them as input to train a network. In addition, the tumor area in the image is less than the normal area, and the imbalance of this data will also make training more difficult. To alleviate these difficulties, the inventor divides the N i images into image patches, extracts features from the image patches, clusters the image patches into several categories, and uses the mean value of the image features of each category as the image feature of patient p i , as shown in Figure 1 . Specifically, the size of the image patch is 256×256×3, which is obtained by slicing side by side on a histological whole slide image with a 20-fold physical magnification. Patient p i has a total of M i image patches. During the generation of the image patches, image patches with a particularly large brightness mean (background patches) are directly discarded. Next, the color of the image patches is normalized (refer to Marc Macenko et al., “A method for normalizing histology slides for quantitative analysis”, 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro (ISBI)

[11] ) to reduce the color difference of the image patches sliced from different histological whole slides. Then, a pre-trained neural network called PNASNet (which is trained with the general image classification large database ImageNet, refer to Chenxi Liu et al., “Progressive neural architecture search”, Proceedings of the European Conference on Computer Vision (ECCV). 2018

[12] ) is used to extract features from the color-normalized image patches. When extracting features, the inventor uses average pooling and max pooling respectively, and combines the obtained features together to improve robustness. In this way, the feature dimension of patient p i is M i×8640. Next, the inventor uses a Gaussian mixture model (GMM) trained on the features of non-tumor region image patches (roughly delimited by doctors, and usually the data of 10 whole histological sections is sufficient for training) to filter the above-extracted features, screening out the first one-third of the image patches with higher scores to filter out non-tumor region image patches. Finally, using the K-means method, the obtained features are clustered into 128 classes, and the 128 class centers are sorted in ascending order of the above GMM scores. These 128 sorted class centers are used as the final image features, with a dimension of 128×8640.

[0043] The following describes the selection of genomic features. There are many gene mutations in breast cancer patients. For example, for the TCGA-BRCA dataset (refer to "Radiology Data from The Cancer Genome Atlas Breast Invasive Carcinoma [TCGA BRCA] collection" by Lingle W et al., The Cancer Imaging Archive, 2016

[13] ), there are a total of 21,057 gene mutation records. The number of these gene mutations is too large and some are redundant. Therefore, the inventor needs to select important gene mutations from these alternative gene mutations. When selecting genes, the inventor only uses the data of the training set to avoid information leakage. Specifically, for each alternative gene, the training set is divided into two parts according to whether the gene has a mutation. Then the inventor uses the log-rank test to calculate the significance level of the two subsets in terms of survival time. Finally, those alternative genes with p-values less than 0.05 are selected as genomic features. Typically, only hundreds of genes will be retained after screening. This screening reduces the feature dimension while enhancing the discriminative power of the features.

[0044] Figure 2 Shows a designed twin network for predicting relative survival risk according to an embodiment of the present invention.

[0045] For the survival risk prediction problem, the absolute value of the risk is usually meaningless. Only the relative magnitude of the survival risks of two patients is meaningful. To this end, the solution of the present invention designs a twin network (refer to "Learning a similarity metric discriminatively, with application to face verification" by S. Chopra et al., In Computer Vision and Pattern Recognition, pages 539–546, IEEE, 2005

[14] ) to meet this requirement, as Figure 2as shown

[0046] The image features are processed by an independent recurrent neural network (IndRNN) (refer to "Independently recurrent neural network (IndRNN): Building a longer and deeper RNN" by Shuai Li et al., Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR). 2018

[15] ), and the genomic features are processed by a fully connected network (FCN). The IndRNN model used has 2 layers, with 1024 hidden nodes in each layer. The FCN used consists of 3 layers, with the number of nodes in each layer being 1024, 512, and 256 respectively. Next, the processed features (with dimensions of 1024 and 256 respectively) are concatenated as joint features (with a dimension of 1280). For a pair of patients (p i , p j ), after their features are processed by a network with shared weights respectively, the difference of their joint features (the corresponding elements of the p i features and the p j features are directly subtracted) is fed into the output linear layer to predict the relative risk. If patient p i has a higher survival risk, it is labeled as 1, otherwise it is labeled as 0. This prediction problem becomes a binary classification problem, so the cross-entropy loss function can be used to guide the training process.

[0047] Existing methods usually directly estimate the absolute survival risk of patients. For example, in "Predicting cancer outcomes from histology and genomics using convolutional networks" by P. Mobadersany et al. (Proceedings of the National Academy of Sciences, vol. 115, no. 13, pp E2970 - E2979, 2018

[16] ) and "Whole slide images based cancer survival prediction using attention guided deep multiple instance learning networks" by Yao, Jiawen et al. (Medical Image Analysis 65 (2020) 101789

[17] ), and use the partial negative log - likelihood as the loss function. Compared with existing methods, the advantages of using the above - mentioned siamese network are as follows: (a) Only relative survival risk is meaningful, and the siamese network of the present invention directly processes relative survival risk. (b) Pairs of patients are used as the input to the network, rather than individual patients. Therefore, assuming the number of patients is P, the number of training samples is of the order of O(P^2). Furthermore, using pairs of patients as the input greatly increases the amount of training samples, making the training process easier. (c) The proposed method is not affected by the batch size in training. When the batch size is limited due to the graphics card memory size, the training samples cannot be loaded into memory simultaneously. For existing methods, when the batch size is limited, not all pairs of patient combinations are visible during one training epoch, so the training process may not be stable. Correspondingly, in the method of the present invention, for one training epoch, all pairs of patients will be traversed regardless of whether the batch size is large or small. Therefore, the method of the present invention is not affected by the batch size and its training process is stable.

[0048] Figure 3 The concordance index (c - index) of the 5 - fold cross - validation experiment on the TCGA - BRCA dataset using three types of inputs: only image features, only genomic features, and image features and genomic features respectively, according to an embodiment of the present invention, is shown. The specific details are described as follows.

[0049] In the TCGA-BRCA dataset, a total of 1,026 patients had both histological whole-slide images and gene mutation labels. Among these patients, 882 patients were right-censored and 144 patients experienced events (i.e., died of breast cancer). Considering this data imbalance, when dividing the training set and the test set, the solution of the present invention used a stratified sampling method. The ratio of the training set to the test set is 4 / 1. In the experiment, the solution of the present invention adopted 5-fold cross-validation, and the evaluation index used was the coincidence index. The solution of the present invention tested the prediction accuracy under different input conditions: using only image features, using only genomic features, and using both features simultaneously. The coincidence index on the test set is as Figure 3 shown, and higher prediction accuracy can be obtained by using both features than using a single feature.

[0050] Those of ordinary skill in the art should understand that the schematic diagram of the Siamese network shown in the drawings is only an exemplary explanatory block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the computer device, processor, or computer program embodying the solution of the present invention. The specific computer device, processor, or computer program may include more or fewer components or modules than those shown in the figure, or combine or split some components or modules, or may have a different arrangement of components or modules.

[0051] It should be understood that each unit of the system of the present invention can be implemented in whole or in part by software, hardware, firmware, or a combination thereof. Each of the units can be embedded in the processor of the computer device in the form of hardware or firmware or independent of the processor, or can be stored in the memory of the computer device in the form of software for the processor to call to execute the operations of each unit. Each of the units can be implemented as an independent component or module, or two or more units can be implemented as a single component or module.

[0052] In one embodiment, a computer device is provided, which includes a memory and a processor. Computer instructions executable by the processor are stored on the memory. When executed by the processor, the computer instructions instruct the processor to execute the steps of the method of the present invention. The computer device can generally be a server, a vehicle-mounted terminal, or any other electronic device with necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, a memory, a network interface, a communication interface, etc. connected through a system bus. The processor of the computer device can be used to provide necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and an internal memory. An operating system, computer programs, etc. may be stored in or on the non-volatile storage medium. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices through a network. When the computer program is executed by the processor, it executes the steps of the auxiliary method of the present invention.

[0053] The present invention can be implemented as a computer-readable storage medium, on which a computer program is stored. When executed by a processor, the computer program causes the steps of the method of the present invention to be executed. In one embodiment, the computer program is distributed among a plurality of network-coupled computer devices or processors, so that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, can be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations can be executed by one or more computer devices or processors, and one or more other method steps / operations can be executed by one or more other computer devices or processors. One or more computer devices or processors can execute a single method step / operation, or execute two or more method steps / operations.

[0054] Those of ordinary skill in the art will understand that all or part of the steps of the method of the present invention can be completed by a computer program instructing relevant hardware such as a computer device or a processor. The computer program can be stored in a non-transitory computer-readable storage medium. When the computer program is executed, the steps of the method of the present invention are caused to be executed. Depending on the circumstances, any reference herein to a memory, storage, database, or other medium may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0055] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification as long as such a combination is not contradictory.

[0056] Although the present invention has been described in conjunction with embodiments, those skilled in the art should understand that the above description and the accompanying drawings are merely exemplary and not restrictive, and the present invention is not limited to the disclosed embodiments. Various modifications and variations are possible without departing from the spirit of the present invention.

Claims

1. A method for predicting the relative survival risk of breast cancer by combining histological whole-slide images and gene mutation tags, the method comprising the following steps: (a) For each of a pair of patients, obtain tumor histological whole-slide image data and gene mutation data; (b) Obtain image features according to the obtained image data, including: dividing the histological whole-slide image into image patches, extracting features from the image patches, screening out non-tumor image patches and clustering the remaining image patches, and taking the sorted class centers as the image features; (c) Select genes with significant effects on survival from the gene mutation data to obtain genomic features; (d) Process the image features and the genomic features through a siamese network, the siamese network including a recurrent neural network for processing image features, a fully-connected network for processing genomic features, and an output linear layer for outputting results, including: using the recurrent neural network to process the image features and using the fully-connected network to process the genomic features; (e) Concatenate the processed image features and genomic features to obtain the fused features of the patient; and (f) For the pair of patients, predict the relative survival risk of the pair of patients using the output linear layer based on the difference in the fused features of the pair of patients.

2. The method according to claim 1, wherein the splitting of the whole histological section image into the image patches in step (b) comprises: The image patches are obtained by cutting side by side from the magnified histological whole-slide image.

3. The method according to claim 1, wherein step (b) further comprises: After dividing the histological whole-slide image into image patches, perform color normalization processing on the image patches.

4. The method according to claim 1, wherein step (b) further comprises: Use a pre-trained neural network to extract features from the image patches.

5. The method according to claim 4 further comprises: The pre-trained neural network is the PNASNet neural network pre-trained on an image classification database.

6. According to the method of claim 5, the image classification database is the ImageNet database.

7. The method according to claim 1, wherein the screening of non-tumor image patches in step (b) comprises: Use a Gaussian mixture model trained on the features of non-tumor region image patches to filter the extracted features to screen out non-tumor image patches.

8. According to the method of claim 7, use the obtained Gaussian mixture model to sort the class centers in step (b).

9. According to the method of claim 1, the gene selection in step (c) is achieved by the log-rank test of survival information on the mutation set and non-mutation set of genes, where the p-value threshold is 0.

05.

10. According to the method of claim 1, the recurrent neural network in step (d) is an independent two-layer network.

11. According to the method of claim 1, the fully-connected network in step (d) is a three-layer network.

12. According to the method of claim 1, the output linear layer in step (f) has no bias parameter.

13. According to the method of claim 1, the siamese network is trained by using the data of multiple patients with breast cancer survival risk data as training set data, and the training steps include: (a) For each of the multiple patients, obtain tumor histological whole-slide image data and gene mutation data; (b) Obtain image features based on the acquired image data, including: dividing the histological whole-slide image into image patches, extracting features from the image patches, screening out non-tumor image patches and clustering the remaining image patches, and taking the sorted class centers as the image features; (c) Select genes with significant impact on survival from the gene mutation data to obtain genomic features; (d) Process the image features and the genomic features through a siamese network, including: processing the image features using a recurrent neural network and processing the genomic features using a fully-connected network; (e) Concatenate the processed image features and genomic features to obtain the fusion features of the patient; (f) For each pair of the multiple patients, train the siamese network based on the difference between the fusion features of the pair of patients and their relative survival risks, including: training the network parameters of the siamese network using the cross-entropy loss function as supervision according to the predicted relative survival risk for the pair of patients and the actual survival data of the pair of patients.

14. The method according to claim 13, wherein the performance evaluation metric for evaluating the prediction accuracy is the coincidence index.

15. The method according to claim 1, wherein the method is applicable to cancers with the histological whole-slide image and the gene mutation labels.

16. A system for predicting survival risk by combining histological whole-slide images and gene mutation labels, comprising a processor configured to execute computer instructions to cause the method according to any one of claims 1-15 to be executed.

17. A computer device, comprising a memory and a processor, wherein computer instructions are stored on the memory, and when executed by the processor, cause the method according to any one of claims 1-15 to be executed.

18. A non-transitory computer-readable storage medium, having computer instructions stored thereon, which when executed by a processor cause the method according to any one of claims 1-15 to be executed.

Images