Method and system for classifying RAS wild type metastatic colorectal cancer patients
Through the fusion model of fusing CT imaging omics data and genetic data, it is predicted that the sensitivity of RAS wild-type metastatic colorectal cancer patients to anti-EGFR targeted drugs is solved, and the problems of poor treatment effect and high economic costs in the prior art are achieved, and precise medication and cost reduction are achieved.
Patent Information
- Application Number
- CN202311377224.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-23
- Publication Date
- 2025-04-25
Smart Images
Figure CN119885051A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent medical technology, and in particular to a classification method and system for patients with RAS wild-type metastatic colorectal cancer. Background Art
[0002] Colorectal cancer (CRC) is a common digestive tract malignancy with a high incidence and mortality rate worldwide. Metastatic colorectal cancer (mCRC) has also become a difficult and important point in clinical diagnosis and treatment.
[0003] Anti-epidermal growth factor receptor (anti-EGFR) monoclonal antibodies combined with chemotherapy are the first-line standard treatment for patients with RAS wild-type metastatic colorectal cancer (RAS wild-type mCRC).
[0004] However, the diversity of drug target genes has a certain impact on drug efficacy. Different genotypes have different sensitivities (effectiveness) to different anti-EGFR targeted drugs. In order to ensure the accuracy of drug administration for conversion therapy of metastatic colorectal cancer and reduce unnecessary economic costs, it is crucial to screen out sensitive populations for conversion therapy with anti-EGFR targeted drugs.
[0005] Based on this, the present application proposes a classification method and system for patients with RAS wild-type metastatic colorectal cancer. Summary of the invention
[0006] This specification provides a classification method and system for RAS wild-type metastatic colorectal cancer patients. By inputting the region of interest and the target gene characteristics into a fusion model, the sensitivity prediction results of the RAS wild-type metastatic colorectal cancer patients about the drug to be tested are obtained to assist in guiding the user's attention.
[0007] The present application provides a method for classifying patients with RAS wild-type metastatic colorectal cancer using the following technical solution, including:
[0008] Obtaining basic information of patients with RAS wild-type metastatic colorectal cancer, the basic information including CT radiomics data and gene data;
[0009] Extracting a region of interest from the CT radiomics data;
[0010] Screening the gene data to obtain target gene features associated with information of the drug to be tested;
[0011] The region of interest and the target gene features are input into the fusion model to obtain the sensitivity prediction results of the RAS wild-type metastatic colorectal cancer patients with respect to the drug to be tested.
[0012] Optionally, extracting the region of interest in the CT radiomics data includes:
[0013] Resampling the CT radiomics data to adjust the voxels of the CT radiomics data;
[0014] Standardizing the resampled CT imaging omics data to obtain the preprocessed CT imaging omics data;
[0015] The preprocessed CT radiomics data are outlined, and the outlined area is identified and extracted as a region of interest.
[0016] Optionally, the inputting the region of interest and the target gene feature into a fusion model to obtain a prediction result of the sensitivity of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested includes:
[0017] Extracting features from the region of interest using a first classification model to obtain a first classification probability;
[0018] Predicting the target gene feature by a second classification model to obtain a second classification probability;
[0019] The first classification probability and the second classification probability are integrated to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0020] Optionally, extracting features from the region of interest to obtain a first classification probability includes:
[0021] Extract local features from the region of interest to obtain microenvironment features;
[0022] Performing global average pooling on the microenvironment features to convert the microenvironment features into a one-dimensional feature vector;
[0023] The one-dimensional feature vector is mapped to a classification label space to obtain the first classification probability.
[0024] Optionally, screening and classifying the gene data to obtain a second classification probability includes:
[0025] Inputting the target gene features into the random forest model to obtain the prediction results of each decision tree;
[0026] The prediction result with the largest number of occurrences is used as the second classification probability.
[0027] Optionally, the fusing of the first classification probability and the second classification probability to obtain a prediction result of the sensitivity of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested includes:
[0028] Taking a weighted average of the first classification probability and the second classification probability;
[0029] The sensitivity label is determined according to the weighted average result to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0030] Optionally, the weight ratio of the first classification probability to the second classification probability is 3:7.
[0031] The present application provides a classification system for patients with RAS wild-type metastatic colorectal cancer using the following technical solutions, including:
[0032] An acquisition module, used to acquire basic information of patients with RAS wild-type metastatic colorectal cancer, wherein the basic information includes CT imaging data and gene data;
[0033] An extraction module, used to extract a region of interest in the CT radiomics data;
[0034] A screening module, used to screen the gene data to obtain target gene features associated with information of the drug to be tested;
[0035] A fusion module is used to input the region of interest and the target gene characteristics into a fusion model to obtain a sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient with respect to the drug to be tested.
[0036] Optionally, the extraction module includes:
[0037] An adjustment submodule, used for resampling the CT radiomics data and adjusting the voxels of the CT radiomics data;
[0038] A standardization submodule is used to perform standardization processing on the resampled CT imaging omics data to obtain the preprocessed CT imaging omics data;
[0039] The identification submodule is used to outline the preprocessed CT imaging data, identify and extract the outlined area as the region of interest.
[0040] Optionally, the fusion module includes:
[0041] A feature extraction submodule, used for extracting features from the region of interest using a first classification model to obtain a first classification probability;
[0042] A classification prediction submodule is used to predict the target gene feature through a second classification model to obtain a second classification probability;
[0043] The fusion submodule is used to fuse the first classification probability and the second classification probability to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0044] Optionally, the feature extraction submodule includes:
[0045] A local extraction unit, used for extracting local features from the region of interest to obtain microenvironment features;
[0046] A pooling unit, used to perform global average pooling on the microenvironment features to convert the microenvironment features into a one-dimensional feature vector;
[0047] A mapping unit is used to map the one-dimensional feature vector to a classification label space to obtain the first classification probability.
[0048] Optionally, the classification prediction submodule includes:
[0049] An input unit, used to input the target gene feature into the random forest model to obtain the prediction result of each decision tree;
[0050] A summarizing unit is used to take the prediction result with the largest number of occurrences as the second classification probability.
[0051] Optionally, the fusion submodule includes:
[0052] A first fusion unit, configured to perform weighted averaging of the first classification probability and the second classification probability;
[0053] The second fusion unit is used to determine the sensitivity label according to the weighted average result, and obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0054] Optionally, the weight ratio of the first classification probability to the second classification probability is 3:7.
[0055] This specification also provides an electronic device, wherein the electronic device includes:
[0056] processor; and,
[0057] A memory storing computer executable instructions, which when executed cause the processor to perform any of the above methods.
[0058] The present specification also provides a computer-readable storage medium, wherein the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, any of the above methods is implemented.
[0059] In the present application, basic information of patients with RAS wild-type metastatic colorectal cancer is obtained, wherein the basic information includes CT imaging genomics data and gene data; regions of interest in the CT imaging genomics data are extracted; the gene data are screened to obtain target gene features associated with information of the drug to be tested; the regions of interest and the target gene features are input into a fusion model to obtain prediction results of the sensitivity of the patients with RAS wild-type metastatic colorectal cancer to the drug to be tested, and people who are sensitive to conversion therapy of the drug to be tested are screened out to assist doctors in making decisions and achieve precise medication. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 A schematic diagram of the principle of a classification method for RAS wild-type metastatic colorectal cancer patients provided in the embodiments of this specification;
[0061] Figure 2 A schematic diagram of the structure of a classification system for patients with RAS wild-type metastatic colorectal cancer provided in an embodiment of this specification;
[0062] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification;
[0063] Figure 4 A schematic diagram of a computer-readable medium provided for an embodiment of this specification. DETAILED DESCRIPTION
[0064] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.
[0065] Exemplary embodiments of the present invention will now be described more fully with reference to the accompanying drawings. However, exemplary embodiments can be implemented in a variety of forms, and should not be construed as limiting the present invention to the embodiments set forth herein. On the contrary, providing these exemplary embodiments enables the present invention to be more comprehensive and complete, and is more convenient for fully conveying the inventive concept to those skilled in the art. The same reference numerals in the figures represent the same or similar elements, components or parts, and thus their repeated description will be omitted.
[0066] Under the premise of being consistent with the technical concept of the present invention, the features, structures, characteristics or other details described in a specific embodiment do not exclude that they can be combined in one or more other embodiments in a suitable manner.
[0067] In the description of specific embodiments, the features, structures, characteristics or other details described in the present invention are intended to enable those skilled in the art to fully understand the embodiments. However, it does not exclude that those skilled in the art can practice the technical solutions of the present invention without one or more of the specific features, structures, characteristics or other details.
[0068] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0069] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0070] The term "and / or" or "and / or" includes all combinations of any one or more of the associated listed items.
[0071] Figure 1 A schematic diagram of the principle of a classification method for patients with RAS wild-type metastatic colorectal cancer provided in an embodiment of this specification, the method comprising:
[0072] S1 obtains basic information of a patient with RAS wild-type metastatic colorectal cancer, wherein the basic information includes CT imaging data and gene data;
[0073] S2 extracts a region of interest in the CT radiomics data;
[0074] S3 screens the gene data to obtain target gene features associated with information of the drug to be tested;
[0075] S4 inputs the region of interest and the target gene features into the fusion model to obtain the sensitivity prediction results of the RAS wild-type metastatic colorectal cancer patient with respect to the drug to be tested.
[0076] Colorectal cancer (CRC) is a common digestive tract malignancy with a high incidence and mortality rate worldwide. CRC mainly spreads through the lymphatic and hematogenous routes, as well as the peritoneal route. The most common sites of metastasis are regional lymph nodes, liver, lungs, and peritoneum. In most cases, metastatic colorectal cancer (mCRC) remains an incurable disease, becoming a difficulty and focus in clinical diagnosis and treatment.
[0077] Taking colorectal cancer liver metastasis (CRLM) as an example, currently, more than half of CRC patients will develop colorectal cancer liver metastasis (CRLM), which is the main cause of death in CRC patients. Although liver resection is currently the best treatment for patients with colorectal cancer liver metastasis (CRLM patients), which can increase their 5-year survival rate from 5% to more than 50%, only about 20% of CRLM patients have the opportunity to undergo radical surgery to remove liver metastases at the time of initial diagnosis, and about 80% of CRLM patients are initially considered incapable of radical surgery. Therefore, how to convert it into resectable colorectal cancer for radical surgery through preoperative chemoradiotherapy is currently a major clinical pain point in colorectal surgery.
[0078] With the development and application of surgical techniques and drugs, conversion therapy is the most effective treatment method for CRLM patients with unresectable liver metastases to achieve long-term survival. Through conversion therapy, about 85% of the initial "CRLM patients with unresectable liver metastases" are expected to become "patients with resectable liver metastases". Therefore, by actively receiving conversion therapy, the tumor of most patients can be regressed and the purpose of radical resection can be achieved, so that the patient has a chance of long-term survival. This is the value of conversion therapy.
[0079] There are two main types of targeted drugs suitable for colorectal cancer. One type is drugs that target epidermal growth factor receptor (EGFR); the other type is drugs that target vascular endothelial growth factor (VEGF) and its receptor (VEGFR).
[0080] Among them, drugs targeting epidermal growth factor receptor (EGFR) mainly include small molecule tyrosine kinase inhibitors, such as gefitinib, erlotinib, lapatinib, etc.; large molecule monoclonal antibodies, such as cetuximab, panitumumab, pertuzumab, etc.
[0081] However, for the same anti-EGFR target drug, such as cetuximab, nearly half of RAS wild-type metastatic colorectal cancer patients are insensitive to conversion therapy and do not respond to cetuximab. If we can find these cetuximab conversion therapy insensitive people and choose another drug, such as anti-vascular endothelial growth factor therapy (such as bevacizumab), it may increase the response rate, reduce adverse reactions, and thus reduce the economic burden.
[0082] There is a lack of gold standard for the administration of cetuximab conversion therapy for patients with RAS wild-type metastatic colorectal cancer (RAS wild-type mCRC patients). When predicting the efficacy and clinical benefits of RAS wild-type mCRC patients, a convenient and efficient method for screening sensitive populations is undoubtedly the focus of research. Based on this, the present application proposes a classification method for patients with RAS wild-type metastatic colorectal cancer (RAS wild-type mCRC patients) to determine the category of anti-EGFR targeted drug conversion therapy for RAS wild-type mCRC patients, and to screen out sensitive populations for anti-EGFR targeted drug conversion therapy, so as to assist doctors in decision-making and achieve precision medication. Specifically, the technical solution of the present application includes:
[0083] S1 obtains basic information of a patient with RAS wild-type metastatic colorectal cancer, wherein the basic information includes CT imaging data and gene data;
[0084] The drug to be tested in this specification refers to the anti-EGFR targeted drug, and the information of the drug to be tested includes the drug name of the anti-EGFR targeted drug. That is, this application is used to screen out sensitive and insensitive populations for conversion therapy with anti-EGFR targeted drugs.
[0085] Anti-EGFR targeted drugs include cetuximab, panitumumab, etc., which mainly inhibit the growth and proliferation of tumor cells by inhibiting the signal transduction of EGFR (epidermal growth factor receptor). However, RAS / RAF mutations can cause mutations in the RAS / RAF part of the EGFR signal transduction pathway, making anti-EGFR targeted drugs unable to effectively inhibit the growth and proliferation of tumor cells. Therefore, for patients with RAS / RAF mutant mCRC, the therapeutic effect of anti-EGFR targeted drugs may be limited. Therefore, anti-EGFR targeted drugs are not recommended for patients with RAS / RAF mutant mCRC.
[0086] Based on its special circumstances, the classification method of this application does not involve patients with RAS mutant mCRC. This application only classifies the sensitivity of patients with RAS wild-type mCRC undergoing conversion therapy with anti-EGFR targeted drugs.
[0087] First, based on the initial diagnosis results, RAS wild-type mCRC patients are identified. In this manual, the initial diagnosis results include colonoscopic biopsy, which is a small specimen obtained from the patient's tumor tissue using biopsy forceps during colonoscopy for pathological diagnosis. The doctor will perform molecular biological testing on the biopsy specimen to determine the mutation status of the RAS gene. If the test results show that the RAS gene has not mutated, the patient can be considered a RAS wild-type mCRC patient.
[0088] Considering that high-throughput information mining of CT images by computer neural networks can fully extract morphological and functional information related to the tumor microenvironment, it has an important prompting effect on analyzing drug sensitivity and patient prognosis. Therefore, after determining that the patient is a RAS wild-type mCRC patient, the CT imaging group data of the RAS wild-type mCRC patient is obtained. Among them, the CT imaging group data includes CT images of metastases, which are CT image data obtained by routine examination of RAS wild-type mCRC patients before treatment. Preferably, the CT imaging group data is an enhanced CT image of metastases; in one embodiment of the present specification, if the RAS wild-type mCRC patient is a colorectal cancer liver metastasis, the CT imaging group data includes a liver metastasis CT image. Preferably, the CT imaging group data is in DICOM format.
[0089] Due to differences in imaging equipment, CT radiomics data often suffer from poor external validation efficacy and low classification accuracy.
[0090] In clinical applications, targeted drugs are highly effective and low toxic during conversion therapy, which can reduce adverse reactions in patients with RAS wild-type mCRC. However, if patients with RAS wild-type mCRC are not suitable for a certain targeted drug, the treatment effect is poor. For example, if the patient with RAS wild-type mCRC does not have a KRAS gene mutation, the use of panitumumab to treat colon cancer is ineffective. Therefore, genetic testing can help doctors avoid unnecessary treatment and relieve patients' pain. Therefore, in order to improve classification accuracy, it is also necessary to collect genetic data of patients with RAS wild-type mCRC.
[0091] In one embodiment of the present specification, the genetic data includes all genetic data of RAS wild-type mCRC patients.
[0092] Considering that the RAS gene is located downstream of the epidermal growth factor receptor (EGFR), when the RAS gene mutates, the EGFR gene and the epidermal growth factor bind to some oncogene molecules in the cell nucleus, causing cancer cells to divide and proliferate. When accumulated to a certain extent, the quantitative change turns into qualitative change, eventually leading to cancer.
[0093] In another embodiment of the present specification, the gene data includes RAS gene data of RAS wild-type mCRC patients. When detecting the RAS gene, PCR (polymerase chain reaction) technology, Southern Blotting technology, gene sequencing, etc. can be used, which will not be elaborated in detail here. Among them, high-throughput sequencing can be performed using whole exon sequencing technology to detect all exon regions in the individual genome, so as to discover gene mutations related to the disease.
[0094] In this specification, the acquisition of CT imaging genomics data and genetic data of the RAS wild-type mCRC patient does not require additional invasive examinations of the RAS wild-type mCRC patient, thereby achieving the purpose of being non-invasive, convenient, safe, and having high patient compliance.
[0095] It also needs to be made clear that different stages of cancer may require different drug treatments. Therefore, genomics must be combined with the patient's actual situation to make more accurate recommendations.
[0096] This application takes into account the lack of research on the value of CT imaging genomics classification models in predicting the efficacy of transformative treatment of CRLM patients. Although the relevant CT imaging genomics classification models have developed rapidly due to their non-invasive, convenient and low-cost characteristics, and the efficacy of drugs can be predicted only by imaging and clinical information, previous imaging genomics model research has problems such as the inability to guarantee sample quality and the simple model building method with poor reproducibility.
[0097] Genomics is a bioinformatics technology that has gradually developed and matured in recent years. It can reveal the intrinsic connection between genomic changes and the efficacy of targeted drugs. Genomics technology is mature, highly stable, and has good external consistency. However, the genetic classification model used to screen the population needs to be combined with actual conditions to obtain more accurate prediction results.
[0098] Therefore, in order to improve the accuracy of sensitivity classification and thereby alleviate the difficulty in medication decision-making for RAS wild-type mCRC patients, sensitivity classification of RAS wild-type mCRC patients to tested drugs was performed based on enhanced CT imaging and whole exome sequencing.
[0099] S2 extracts a region of interest in the CT radiomics data;
[0100] Considering that different CT devices and different parameter settings will affect the repeatability and reliability of feature extraction, first, S21 preprocesses the CT impactomics data;
[0101] In one embodiment of the present specification, S211 resamples the CT radiomics data to adjust the voxels of the CT radiomics data. Preferably, the CT radiomics data is resampled to 3×3×3 mm 3 , to make it isotropic, so as to reduce the variation caused by differences in scanning equipment and schemes, different lesion sizes of patients, etc. The image scan obtains the voxel spacing in the X, Y, and Z directions. The voxel spacing in the X and Y directions is small, and the voxel spacing in the Z direction is slightly larger. We call this situation anisotropy, and resample to 1×1×1mm 3 Finally, the anisotropic image becomes isotropic, the voxel spacing is standardized, the dependence of imaging omics features on the image voxel size is minimized, the actual physical size represented by each voxel is ensured to be consistent, and the impact of individual differences is reduced.
[0102] S212 performs standardization processing on the resampled CT imaging omics data to obtain the preprocessed CT imaging omics data;
[0103] In one embodiment of the present specification, the resampled CT imaging genomics data is first subjected to denoising to reduce image noise in order to achieve stable intensity resolution. Preferably, the voxel intensity value is discretized by using a fixed bin width of 25HU. In order to reduce the difference in different acquisitions and their impact on image signal intensity, the image is normalized to 1-500HU after discretization to reduce the impact of different acquisitions and their impact on image signal intensity. In order to reduce the impact of inconsistent imaging parameters on imaging genomics feature variation, the image grayscale value is subjected to Z-score normalization after normalization to obtain preprocessed CT imaging genomics data.
[0104] S22 outlining the preprocessed CT radiomics data, identifying and extracting the outlined area as a region of interest;
[0105] In this specification, the region of interest includes the abnormal structural region of RAS wild-type mCRC patients. By determining the region of interest of RAS wild-type mCRC patients, the region of interest is further analyzed to obtain lesion characteristics and morphological information. Among them, the CT images of metastatic lesions outside the region of interest are discarded to reduce the amount of calculation and improve the speed of feature extraction and classification in the later stage.
[0106] Specifically, in one embodiment of the present specification, professionals can also manually outline the CT imaging data to obtain the manually outlined region of interest. Preferably, two experienced radiologists outline the region of interest of liver metastases, check and reach a consensus;
[0107] In another embodiment of the present specification, a training set for regional delineation is created, specifically, historical personal information of several patients with RAS wild-type mCRC is collected, the historical personal information including sensitivity to targeted drugs and historical metastatic lesion CT images. Professionals manually delineate the above-obtained historical metastatic lesion CT images, mark the regions of interest, and obtain the marked historical metastatic lesion CT images.
[0108] The historical metastatic CT images and the marked historical metastatic CT images of the same patient are used as a set of regional labeling data samples, and the regional labeling data samples are collected to form a regional labeling training set and a regional labeling verification set. The regional labeling data samples of the regional labeling training set and the regional labeling verification set can be determined by random sampling or in chronological order.
[0109] A region labeling model is constructed, and the region labeling data samples in the region labeling training set are input into the region labeling model for training. After the training is completed, the region labeling model is verified and optimized using the region labeling data samples in the region labeling verification set.
[0110] When the CT radiomics data are delineated using the regional labeling model, in order to ensure its accuracy, after the region of interest of the CT radiomics data is obtained, it is checked and corrected by professionals.
[0111] After the region of interest is determined, the region of interest is extracted based on the outlined edges.
[0112] S3 screens the gene data to obtain target gene features associated with information of the drug to be tested;
[0113] Specifically, the genetic data of several historical patients with RAS wild-type mCRC and the information of the drugs to be tested were obtained to construct a genetic screening training set and a genetic screening test set.
[0114] The training set is fitted using the Lasso regression model, and the effect of feature selection is controlled by adjusting the penalty coefficient. Specifically, the feature coefficients are compressed by constructing the penalty coefficients, so that the feature coefficients of some features become zero. Features with zero feature coefficients are considered to be unimportant features. Features with non-zero coefficients are considered to be important features that have been screened out, namely, target gene features, which are associated with the information of the drug to be tested.
[0115] That is, the gene data is screened by the Lasso regression model to obtain the target gene features. In this specification, all the gene data of the colorectal cancer liver metastasis patients were screened by the Lasso regression model, and the optimal penalty coefficient was finally obtained by 5-fold cross validation of the hyperparameters, which was 1.893×10 -2 .
[0116] In one embodiment of the present specification, for the prediction of sensitivity to cetuximab conversion therapy, the following genes need to be screened: RET, GNA11, MDC1, NOBOX, NOTCH1, TP53, TCF3, COL1A1, APC.
[0117] S4 inputs the region of interest and the target gene features into the fusion model to obtain the sensitivity prediction results of the RAS wild-type metastatic colorectal cancer patient with respect to the drug to be tested.
[0118] S41 extracts features from the region of interest to obtain a first classification probability;
[0119] Specifically, extracting features from the region of interest according to the first classification model to obtain a first classification probability;
[0120] Among them, the first classification model is a neural network model, and further, the neural network model is preferably a ResNet18 model.
[0121] In one embodiment of the present specification, the learning rate used is determined to be 7×10 -4 , the loss function is the Cross-Entropy loss function. By minimizing the loss function, the gap between the model's predicted value and the label is reduced and evaluated. The optimizer is preferably the Adam optimizer, and the weight decay is 1×10 -4 In addition, the training set batch size is set to 16 and the training epoch is set to 17. The random seed is set to 12 to ensure optimization efficiency.
[0122] Among them, the ResNet18 model includes input layer, convolution layer, residual block, global average pooling layer, fully connected layer and output layer.
[0123] Specifically, an image of a region of interest is received through an input layer.
[0124] Then, local features are extracted from the region of interest through a convolutional layer to obtain microenvironment features.
[0125] In this specification, high-throughput feature extraction is performed on the region of interest to obtain microenvironmental features of the region of interest, and microenvironmental features refer to the physical, chemical and biological properties that describe the environment surrounding an organ, tissue or cell.
[0126] There are 8 residual blocks in the ResNet18 model. Each residual block consists of two convolutional layers and a skip connection to solve the problems of gradient disappearance and gradient explosion in deep convolutional neural networks.
[0127] Perform global average pooling on the microenvironment features to convert the microenvironment features into a one-dimensional feature vector; map the one-dimensional feature vector to a classification label space to obtain a first classification probability.
[0128] In one embodiment of the present specification, all channels of the input region of interest are average pooled to obtain a single output channel, and the region of interest is converted into a one-dimensional vector. This process can reduce the number of parameters of the model and provide more global statistical information.
[0129] In one embodiment of the present specification, M categories are provided. The fully connected layer is a weight matrix of size 1xM, where M is the number of categories. The output vector after average pooling is multiplied by this weight matrix, and then passed through an activation function (such as a sigmoid function), an output vector of size 1xM can be obtained, and the output vector contains the probability that the image belongs to each category. This output vector is input into the softmax function to obtain the final first classification probability. In one embodiment of the present specification, the category can be a decimal between [0, 1] and 0, 1. Preferably, M is 11, including 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1. In other embodiments of the present specification, the category can be a decimal between [0, 1] and 0, 1; of course, the category can be a decimal of other digits between [0, 1] and 0, 1. The larger the value, the higher the sensitivity of the patient to the conversion therapy of the drug to be tested, and the more effective the conversion therapy.
[0130] S42 predicts the target gene feature by using a second classification model to obtain a second classification probability;
[0131] The target gene characteristics and sensitivity results of several RAS wild-type mCRC patients were obtained to construct the second classification training set and the second classification test set.
[0132] Construct a second classification model. In one embodiment of the present specification, a random forest algorithm is used for classification modeling, the second classification model is trained by a second classification training set, and the model effect is evaluated by a second classification test set. The number of classifiers in the second classification model is set to 20, and the random seed is set to 2.
[0133] The random forest algorithm combines multiple decision trees. The second classification model randomly selects training data from the second classification training set with replacement each time and randomly selects some features as input. For many input data, the importance of input features is evaluated by random forest, while producing highly accurate classification results.
[0134] The random forest algorithm can perform parallel calculations, which improves the training speed. In addition, the random forest can balance the number of samples of different categories through its bagging and random feature selection process, thereby improving the classification accuracy of unbalanced data and balancing the errors of unbalanced samples.
[0135] The target gene feature is input into the random forest model to obtain the prediction result of each decision tree, and the prediction result with the most occurrences is used as the second classification probability according to the prediction result of each decision tree for the target gene feature. Among them, the prediction results include the patient being a sensitive population for conversion therapy of the drug to be tested and the patient being an insensitive population for conversion therapy of the drug to be tested. Preferably, the prediction results include 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.
[0136] In one embodiment of the present specification, the prediction result includes one decimal place between [0, 1] and 0, 1. Preferably, the prediction result includes 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1. In other embodiments of the present specification, the prediction result includes two decimal places between [0, 1] and 0, 1; of course, the prediction result may also be a decimal place of other digits between [0, 1] and 0, 1. The larger the value, the higher the sensitivity of the patient to the conversion therapy of the drug to be tested, and the more effective the conversion therapy is.
[0137] Preferably, the prediction result is consistent with the number of decimal places in the category.
[0138] If multiple decision trees give the same prediction, then the result is more likely to be correct. But if multiple decision trees give different predictions, more evidence and further analysis are needed to draw a final conclusion.
[0139] S43 performs weighted averaging of the first classification probability and the second classification probability; determines a sensitivity label according to the weighted average result, and obtains a sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient with respect to the drug to be tested.
[0140] Among them, the weight ratios of 1:9 and 2:8 are quite different from the actual classification probabilities. The weight w1 of the first classification probability is 0.3, and the weight w2 of the second classification probability is 0.7, that is, w1:w2=3:7.
[0141] In determining the sensitivity label, in one embodiment of the present specification, the weighted average result is rounded to determine the sensitivity label, wherein the number of sensitivity labels is two, and the sensitivity labels include 0 and 1.
[0142] If the sensitivity label is 0, the corresponding sensitivity prediction result is: the patient is an insensitive population for the drug conversion treatment to be tested. If the sensitivity label is 1, the corresponding sensitivity prediction result is: the patient is a sensitive population for the drug conversion treatment to be tested.
[0143] Of course, in another embodiment of the present specification, the result between [0, 0.5) corresponds to the sensitivity label 0; the result between [0.5, 1] corresponds to the sensitivity label 1.
[0144] The present invention has been experimentally verified, wherein the experimental data and experimental results are as follows:
[0145] (I) Test data:
[0146] CT radiomics and genetic data of 15 RAS wild-type mCRC patients who were “sensitive to cetuximab conversion therapy”;
[0147] CT radiomics and genetic data of 15 RAS wild-type mCRC patients who were "refractory to cetuximab treatment".
[0148] (II) Description
[0149] The prediction accuracy in “sensitive to cetuximab treatment” was taken as sensitivity, and the prediction accuracy in “insensitive to cetuximab” was taken as specificity.
[0150] The label is 0, indicating "not sensitive to cetuximab treatment"; the label is 1, indicating "sensitive to cetuximab conversion therapy"
[0151] (III) Data Verification:
[0152] 1. Classification prediction based only on CT radiomics data
[0153] The regions of interest (liver metastases) of 15 RAS wild-type mCRC patients who were “sensitive to cetuximab conversion therapy” were annotated, and the regions of interest (liver metastases) of 15 RAS wild-type mCRC patients who were “insensitive to cetuximab conversion therapy” were annotated;
[0154] Then, it was input into the CT radiomics classification model to determine the sensitivity of each patient to cetuximab conversion therapy. Among them, the output labels of 15 RAS wild-type mCRC patients who were "sensitive to cetuximab treatment" were as follows: (Table 1)
[0155] Serial number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Label 1 1 1 1 1 1 0 1 0 1 1 1 0 1 1
[0156] (Table 1)
[0157] The output labels of 15 patients with RAS wild-type mCRC who were “insensitive to cetuximab treatment” are as follows: (Table 2)
[0158] Serial number 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Label 0 1 0 1 0 1 1 0 1 0 1 0 1 1 0
[0159] (Table 2)
[0160] The results showed that in the judgment of "sensitive to cetuximab treatment", 12 cases were correctly judged and 3 cases were wrong; in the judgment of "insensitive to cetuximab", 7 cases were correctly judged and 8 cases were wrong. The sensitivity and specificity of classification prediction based only on CT imaging data were 80% and 46.6%, respectively.
[0161] 2. Classification prediction based only on genetic data
[0162] The genetic data of 15 patients with RAS wild-type mCRC who were “sensitive to cetuximab conversion therapy” were screened to obtain the target gene characteristics, and the genetic data of 15 patients with RAS wild-type mCRC who were “insensitive to cetuximab conversion therapy” were screened to obtain the target gene characteristics;
[0163] Then, it was input into the gene classification model to determine the sensitivity of each patient to cetuximab conversion therapy. Among them, the output labels of 15 RAS wild-type mCRC patients who were "sensitive to cetuximab treatment" were as follows: (Table 3)
[0164] Serial number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Label 1 1 0 1 1 1 0 1 1 1 1 1 0 1 1
[0165] (Table 3)
[0166] The output labels of 15 patients with RAS wild-type mCRC who were “insensitive to cetuximab treatment” are as follows: (Table 4)
[0167] Serial number 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Label 1 1 0 0 0 1 1 0 1 0 0 0 1 0 0
[0168] (Table 4)
[0169] The results showed that in the judgment of "sensitive to cetuximab treatment", 12 cases were correctly judged and 3 cases were wrong; in the judgment of "insensitive to cetuximab", 9 cases were correctly judged and 6 cases were wrong. The sensitivity and specificity of classification prediction based on genetic data alone were 80% and 60.0%, respectively.
[0170] 3. Inputting the region of interest and the target gene features into the fusion model in this specification
[0171] The first classification probability and the second classification probability of 15 RAS wild-type mCRC patients who were "sensitive to cetuximab treatment" were fused, and the final prediction results were as follows: (Table 5)
[0172] Serial number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Label 1 1 1 1 1 1 0 1 1 1 1 1 0 1 1
[0173] (Table 5)
[0174] The results showed that in the judgment of "sensitive to cetuximab treatment", 12 cases were correctly judged and 3 cases were wrong;
[0175] The first classification probability and the second classification probability of 15 RAS wild-type mCRC patients who were "insensitive to cetuximab treatment" were fused to obtain the final prediction results as follows: (Table 6)
[0176]
[0177]
[0178] (Table 6)
[0179] The results showed that in the judgment of "sensitive to cetuximab treatment", 13 cases were correctly judged and 2 cases were wrongly judged; in the judgment of "insensitive to cetuximab", 11 cases were correctly judged and 4 cases were wrongly judged. The sensitivity and specificity of the prediction of the fusion model of the present invention were 86.7% and 73.3% respectively.
[0180] In summary, the sensitivity and specificity of classification prediction based only on CT radiomics data were 80% and 46.6%, respectively; the sensitivity and specificity of classification prediction based only on genetic data were 80% and 60.0%, respectively.
[0181] The sensitivity and specificity predicted by the fusion model of the present invention are 86.7% and 73.3% respectively, and the area under the AUC curve of its external validation is 0.86 (the sensitivity and specificity values are 86.7% and 73.3% respectively). The sensitivity and specificity of the fusion model of the present invention are both high.
[0182] After obtaining the sensitivity prediction results of the tested drugs for patients with RAS wild-type metastatic colorectal cancer, professionals can conduct further drug screening and prediction of the efficacy of conversion therapy based on the sensitivity prediction results.
[0183] It should be noted that the drugs to be tested in this specification include but are not limited to cetuximab and panitumumab, and may also be other anti-EGFR targeted drugs.
[0184] Since common metastatic sites of colorectal cancer include regional lymph nodes, liver, lungs and peritoneum, although approximately more than half of CRC patients will develop colorectal cancer liver metastasis (CRLM), the RAS wild-type mCRC patients in this application include but are not limited to RAS wild-type colorectal cancer liver metastasis patients (CRLM patients).
[0185] In other words, the practical application scenarios of the present invention include but are not limited to the prediction of sensitivity to cetuximab conversion therapy for patients with RAS wild-type colorectal cancer liver metastases (RAS wild-type CRLM patients).
[0186] In this specification, based on CT imaging genomics data and whole exome sequencing gene data, a fusion model is used to predict the sensitivity of anti-EGFR targeted drug conversion therapy for patients with RAS wild-type metastatic colorectal cancer, which can be applied to the first-line formulation of clinical medication plans.
[0187] Figure 2 A schematic diagram of a classification system for patients with RAS wild-type metastatic colorectal cancer provided in an embodiment of this specification, the system comprising:
[0188] An acquisition module 210 is used to acquire basic information of patients with RAS wild-type metastatic colorectal cancer, wherein the basic information includes CT imaging data and gene data;
[0189] An extraction module 220, for extracting a region of interest from the CT radiomics data;
[0190] A screening module 230 is used to screen the gene data to obtain target gene features associated with information of the drug to be tested;
[0191] The fusion module 240 is used to input the region of interest and the target gene feature into a fusion model to obtain a sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0192] Optionally, the extraction module 220 includes:
[0193] An adjustment submodule, used for resampling the CT radiomics data and adjusting the voxels of the CT radiomics data;
[0194] A standardization submodule is used to perform standardization processing on the resampled CT imaging omics data to obtain the preprocessed CT imaging omics data;
[0195] The identification submodule is used to outline the preprocessed CT imaging data, identify and extract the outlined area as the region of interest.
[0196] Optionally, the fusion module 240 includes:
[0197] A feature extraction submodule, used for extracting features from the region of interest using a first classification model to obtain a first classification probability;
[0198] A classification prediction submodule is used to predict the target gene feature through a second classification model to obtain a second classification probability;
[0199] The fusion submodule is used to fuse the first classification probability and the second classification probability to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0200] Optionally, the feature extraction submodule includes:
[0201] A local extraction unit, used for extracting local features from the region of interest to obtain microenvironment features;
[0202] A pooling unit, used to perform global average pooling on the microenvironment features to convert the microenvironment features into a one-dimensional feature vector;
[0203] A mapping unit is used to map the one-dimensional feature vector to a classification label space to obtain the first classification probability.
[0204] Optionally, the classification prediction submodule includes:
[0205] An input unit, used to input the target gene feature into the random forest model to obtain the prediction result of each decision tree;
[0206] A summarizing unit is used to take the prediction result with the largest number of occurrences as the second classification probability.
[0207] Optionally, the fusion submodule includes:
[0208] A first fusion unit, configured to perform weighted averaging of the first classification probability and the second classification probability;
[0209] The second fusion unit is used to determine the sensitivity label according to the weighted average result, and obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
[0210] Optionally, the weight ratio of the first classification probability to the second classification probability is 3:7.
[0211] The functions of the system of the embodiment of the present invention have been described in the above method embodiment, so for details not fully described in this embodiment, please refer to the relevant description in the above embodiment, and no further description will be given here.
[0212] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0213] The electronic device embodiment of the present invention is described below, and the electronic device can be regarded as a physical implementation of the method and device embodiments of the present invention described above. The details described in the electronic device embodiment of the present invention should be regarded as a supplement to the above method or device embodiments; details not disclosed in the electronic device embodiment of the present invention can be implemented with reference to the above method or device embodiments.
[0214] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. Figure 3 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0215] like Figure 3 As shown, the computer device 300 of this exemplary embodiment is in the form of a general data processing device. The components of the computer device 300 may include but are not limited to: at least one processor 310, at least one memory 320, a network interface 330, a display unit 340, an input component 350, etc.
[0216] The memory 320 stores a computer-readable program, which may be a source program or a code of a read-only program. The program may be executed by the processor 310, so that the processor 310 performs the steps of various embodiments of the present invention. For example, the processor 310 may perform the following steps: Figure 1 Steps shown.
[0217] The memory 320 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) and / or a cache memory unit, and may further include a read-only memory unit (ROM). The memory 320 may also include a program / utility having a set (at least one) of program modules, such program modules including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include the implementation of a network environment.
[0218] Also included is a bus (not shown) which may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.
[0219] The computer device 300 may also communicate with one or more external devices (e.g., keyboard, display, network device, Bluetooth device, etc.) so that a user can interact with the computer device 300 via these external devices, and / or so that the computer device 300 can communicate with one or more other data processing devices (e.g., routers, modems, etc.). Such communication may be performed through the network interface 330, and may also be performed through a network adapter with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public network, such as the Internet). The network adapter may communicate with other modules of the computer device 300 via a bus. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in the computer device 300, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.
[0220] Figure 4 Schematic diagram of a computer readable medium embodiment of the present invention. Figure 4 As shown, the computer program can be stored on one or more computer-readable media. The computer-readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. When the computer program is executed by one or more data processing devices, the computer-readable medium is enabled to implement the above method of the present invention.
[0221] Through the description of the above implementation modes, it is easy for those skilled in the art to understand that the exemplary embodiments described in the present invention can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation mode of the present invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a data processing device (which can be a personal computer, a server, or a network device, etc.) to perform the above method according to the present invention.
[0222] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by an instruction execution system, an apparatus, or a device or used in combination with it. The program code contained on the readable storage medium may be transmitted with any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.
[0223] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).
[0224] In summary, the present invention can be implemented by a method, apparatus, electronic device or computer-readable medium that executes a computer program. In practice, a general data processing device such as a microprocessor or a digital signal processor (DSP) can be used to implement some or all functions of the present invention.
[0225] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the present invention is not inherently related to any specific computer, virtual device or electronic device, and various general devices can also implement the present invention. The above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for classifying patients with RAS wild-type metastatic colorectal cancer, characterized in that: include: Obtaining basic information of patients with RAS wild-type metastatic colorectal cancer, the basic information including CT radiomics data and gene data; Extracting a region of interest from the CT radiomics data; Screening the gene data to obtain target gene features associated with information of the drug to be tested; The region of interest and the target gene features are input into the fusion model to obtain the sensitivity prediction results of the RAS wild-type metastatic colorectal cancer patients with respect to the drug to be tested.
2. A method for classifying patients with RAS wild-type metastatic colorectal cancer according to claim 1, characterized in that: The extracting the region of interest in the CT radiomics data comprises: Resampling the CT radiomics data to adjust the voxels of the CT radiomics data; Standardizing the resampled CT imaging omics data to obtain the preprocessed CT imaging omics data; The preprocessed CT radiomics data are outlined, and the outlined area is identified and extracted as a region of interest.
3. A method for classifying patients with RAS wild-type metastatic colorectal cancer according to claim 1, characterized in that: The step of inputting the region of interest and the target gene feature into the fusion model to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested includes: Extracting features from the region of interest using a first classification model to obtain a first classification probability; Predicting the target gene feature by a second classification model to obtain a second classification probability; The first classification probability and the second classification probability are integrated to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
4. A method for classifying patients with RAS wild-type metastatic colorectal cancer as claimed in claim 3, characterized in that: The extracting features from the region of interest to obtain a first classification probability includes: Extract local features from the region of interest to obtain microenvironment features; Performing global average pooling on the microenvironment features to convert the microenvironment features into a one-dimensional feature vector; The one-dimensional feature vector is mapped to a classification label space to obtain the first classification probability.
5. The method for classifying patients with RAS wild-type metastatic colorectal cancer according to claim 3, characterized in that: The screening and classification of the gene data to obtain a second classification probability includes: Inputting the target gene features into the random forest model to obtain the prediction results of each decision tree; The prediction result with the largest number of occurrences is used as the second classification probability.
6. A method for classifying patients with RAS wild-type metastatic colorectal cancer as claimed in claim 3, characterized in that: The fusing of the first classification probability and the second classification probability to obtain a prediction result of the sensitivity of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested includes: Taking a weighted average of the first classification probability and the second classification probability; The sensitivity label is determined according to the weighted average result to obtain the sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient to the drug to be tested.
7. A method for classifying patients with RAS wild-type metastatic colorectal cancer according to claim 6, characterized in that: The weight ratio of the first classification probability to the second classification probability is 3:
7.
8. A classification system for patients with RAS wild-type metastatic colorectal cancer, characterized in that: include: An acquisition module, used to acquire basic information of patients with RAS wild-type metastatic colorectal cancer, wherein the basic information includes CT imaging data and gene data; An extraction module, used to extract a region of interest in the CT radiomics data; A screening module, used to screen the gene data to obtain target gene features associated with information of the drug to be tested; A fusion module is used to input the region of interest and the target gene characteristics into a fusion model to obtain a sensitivity prediction result of the RAS wild-type metastatic colorectal cancer patient with respect to the drug to be tested.
9. An electronic device, wherein: The electronic device includes: processor; and, A memory storing computer executable instructions which, when executed, cause the processor to perform a method according to any one of claims 1-7.
10. A computer-readable storage medium, wherein: The computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, the method of any one of claims 1 to 7 is implemented.