Lung cancer risk prediction method and device and storage medium

Abstract

The invention relates to the field of medical intelligent analysis, and discloses a lung cancer risk prediction method and device and a storage medium. The method comprises the following steps: performing combination processing on clinical features, a GRIK2 methylation value, an HOXA9 methylation value, a PTGER4 methylation value, an SHOX2 methylation value and a PITX2 methylation value to generate a feature vector; based on a preset machine learning algorithm, performing classification processing on the feature vector to obtain a classification risk value; when the classification risk value is greater than a preset high-risk threshold value, generating a lung cancer high-risk prediction result; when the classification risk value is smaller than a preset low-risk threshold value, generating a lung cancer low-risk prediction result; and when the classification risk value is not greater than a preset high-risk threshold value and not less than a preset low-risk threshold value, generating a risk prediction result in the lung cancer. In the embodiment of the invention, the consistency of lung cancer risk prediction is realized, and the accuracy of lung cancer risk prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of medical intelligent analysis, and in particular to a method, device and storage medium for predicting lung cancer risk. Background Technology

[0002] The epidemiological status of lung cancer is characterized by high incidence and mortality rates, making early detection and treatment of lung cancer extremely important clinically. The current clinical situation and challenges of lung cancer lie in the fact that the existing screening technology is low-dose CT (LDCT). The widespread adoption of this screening technology has led to a surge in the detection rate of lung nodules, hindering broader early intervention for lung cancer.

[0003] However, existing lung cancer screening technologies suffer from several drawbacks. LDCT has a high false-positive rate, stemming from inherent technological limitations. Furthermore, LDCT involves radiation exposure, leading to poor patient compliance, high equipment costs, and limited clinical accessibility. The lack of standardized interpretation of LDCT results in reliance on radiologists' experience and subjective judgment, resulting in high inter-observer inconsistencies. Other clinical risk assessment models (such as the Brock and Mayo models) are primarily based on demographic and imaging characteristics, but their predictive performance is insufficient due to the lack of integration of molecular-level information. Therefore, a new technological solution is needed to address the current technical challenges of inconsistent and inaccurate predictions in lung cancer screening programs. Summary of the Invention

[0004] The main objective of this invention is to address the technical problems of insufficient consistency and predictive accuracy in current lung cancer screening programs.

[0005] The first aspect of this invention provides a method for predicting lung cancer risk, comprising the steps of: Receive the clinical characteristics of the target user and obtain the GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of the target user's plasma cfDNA; The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value are combined to generate a feature vector. Based on a preset machine learning algorithm, the feature vector is classified to obtain a classification risk value; When the classification risk value is greater than the preset high-risk threshold, a high-risk lung cancer prediction result is generated for the target user. When the classification risk value is less than the preset low-risk threshold, a low-risk lung cancer prediction result is generated for the target user. When the classification risk value is not greater than the preset high-risk threshold and not less than the preset low-risk threshold, a lung cancer intermediate risk prediction result corresponding to the target user is generated.

[0006] Optionally, in a first implementation of the first aspect of the present invention, before the step of classifying the feature vector based on a preset machine learning algorithm to obtain a classification risk value, the method further includes: The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of N training samples are combined to generate N training feature vectors, where N is a positive integer; The N training feature vectors are labeled to obtain the labeled results corresponding to the N training feature vectors; Based on a preset machine learning algorithm, the N training feature vectors are classified and processed to generate N predicted values. Based on the preset first hyperparameter and the preset second hyperparameter, the N predicted values ​​are processed to obtain N predicted results. The first hyperparameter includes a parameter for a high-risk threshold, and the second hyperparameter includes a parameter for a low-risk threshold. Based on the annotation results corresponding to the N training feature vectors, the accuracy of the N prediction results is calculated to obtain the test accuracy. Based on the test accuracy, the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained until the test accuracy converges.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the step of classifying the N training feature vectors based on a preset machine learning algorithm to generate N predicted values ​​includes: Based on a preset logistic regression algorithm, logistic regression classification is performed on the N training feature vectors to obtain N first predicted values. Based on a preset support vector machine, the N training feature vectors are subjected to kernel function classification processing to obtain N second predicted values; Based on the preset random forest algorithm, decision classification processing is performed on the N training feature vectors to obtain N third prediction values; Based on the preset XGBoost algorithm, regression mapping and classification are performed on the N training feature vectors to obtain N fourth prediction values.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of performing result determination processing on the N predicted values ​​according to preset first hyperparameters and preset second hyperparameters to obtain N predicted results includes: Based on the preset first hyperparameter and the preset second hyperparameter, the results of the N first predicted values, the N second predicted values, the N third predicted values ​​and the N fourth predicted values ​​are divided to obtain N first predicted results, N second predicted results, N third predicted results and N fourth predicted results.

[0009] Optionally, in the fourth implementation of the first aspect of the present invention, the step of calculating the accuracy of the N prediction results based on the annotation results corresponding to the N training feature vectors to obtain the test accuracy includes: Based on the annotation results corresponding to the N training feature vectors, the accuracy of the N first prediction results, the N second prediction results, the N third prediction results, and the N fourth prediction results is calculated to obtain the first accuracy, the second accuracy, the third accuracy, and the fourth accuracy.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the step of adjusting the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter based on the test accuracy and iteratively training until the test accuracy converges includes: Based on the first accuracy, the parameters of the logistic regression algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the first convergence accuracy. Based on the second accuracy, the parameters of the support vector machine, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the second convergence accuracy. Based on the third accuracy, the parameters of the random forest algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the third convergence accuracy. Based on the fourth accuracy, the parameters of the XGBoost algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the fourth convergence accuracy. The type of machine learning algorithm is determined based on the maximum value of the first convergence accuracy, the second convergence accuracy, the third convergence accuracy, and the fourth convergence accuracy.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, the step of combining the clinical features, the GRIK2 methylation value, the HOXA9 methylation value, the PTGER4 methylation value, the SHOX2 methylation value, and the PITX2 methylation value to generate a feature vector includes: The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value were filled with missing values ​​and normalized to obtain preprocessed clinical features, preprocessed GRIK2 methylation value, preprocessed HOXA9 methylation value, preprocessed PTGER4 methylation value, preprocessed SHOX2 methylation value, and preprocessed PITX2 methylation value. The pre-processed clinical features, pre-processed GRIK2 methylation value, pre-processed HOXA9 methylation value, pre-processed PTGER4 methylation value, pre-processed SHOX2 methylation value, and pre-processed PITX2 methylation value are successively filled into the preset feature framework to generate feature vectors.

[0012] Optionally, in a seventh implementation of the first aspect of the present invention, the clinical characteristics include: age, sex, smoking history category, and family history of cancer category.

[0013] A second aspect of the present invention provides a lung cancer risk prediction device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the lung cancer risk prediction device to perform the lung cancer risk prediction method described above.

[0014] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the lung cancer risk prediction method described above.

[0015] In this embodiment of the invention, feature vectors are generated by combining multi-gene methylation with clinical features. These feature vectors are then input into a pre-trained machine learning model for classification, generating well-trained risk prediction values. Finally, lung cancer risk is classified using trained dual thresholds, achieving consistency in lung cancer risk prediction. Combining gene methylation parameters with clinical features in medicine improves the accuracy of lung cancer prediction and solves the technical problems of insufficient consistency and prediction accuracy in current lung cancer screening programs. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of one embodiment of the lung cancer risk prediction method in this invention; Figure 2 This is a schematic diagram of a model training embodiment prior to step 103 of the lung cancer risk prediction method in this invention. Figure 3aThis is a schematic diagram of the ROC curve of the random forest algorithm in the lung cancer risk prediction method of this invention on the training dataset. Figure 3b This is a schematic diagram of the ROC curve of the random forest algorithm for lung cancer risk prediction method in the embodiment of the present invention on the validation dataset; Figure 4a This is a schematic diagram of the results of the random forest algorithm in the lung cancer risk prediction method of the present invention using the double threshold partitioning of the training dataset. Figure 4b As described in the embodiments of the present invention Figure 4a A schematic diagram of the sample partitioning and distribution; Figure 4c This is a schematic diagram of the random forest algorithm for lung cancer risk prediction in an embodiment of the present invention, using the dual-threshold partitioning results of the validation dataset. Figure 4d As described in the embodiments of the present invention Figure 4c A schematic diagram of the sample partitioning and distribution; Figure 5 This is a schematic diagram of one embodiment of the lung cancer risk prediction device in this invention. Detailed Implementation

[0017] This invention provides a method, device, and storage medium for predicting lung cancer risk.

[0018] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the lung cancer risk prediction method in this invention includes: 101. Receive the clinical characteristics of the target user and obtain the GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of the target user's plasma cfDNA; In this embodiment, the selection of high-risk epidemiological characteristics for lung cancer is based on clinical guidelines. High-risk factors for lung cancer include at least the following clinical characteristics: age (greater than 40 years old) and at least one of the following risk factors: 1. Cumulative smoking index ≥20 pack / year, including those who have smoked ≥20 pack / year but quit smoking less than 15 years ago; 2. Environmental or occupational exposure (such as long-term exposure to radon, silicon, cadmium, arsenic, beryllium, chromium, nickel, asbestos, diesel fumes, coal smoke, radioactive elements, etc.); 3. Family history of lung cancer in first-degree relatives; 4. Comorbid COPD, diffuse pulmonary fibrosis, or old pulmonary tuberculosis; 5. History of malignant tumors; 6. Long-term exposure to secondhand smoke (home or indoor workplace >2h / d, at least 10 years) or long-term exposure to kitchen fumes (such as cooking environments such as stir-frying, pan-frying, deep-frying, etc.). Considering relevance and ease of information access, the following four features were selected as input variables: age (years, continuous), smoking history (yes / no, categorical), family history of cancer (yes / no, categorical), and gender (male / female, categorical).

[0021] In other words, clinical characteristics include: age, gender (1 for male, 0 for female), smoking history category (1 for yes, 0 for no), and family history of cancer category (1 for present, 0 for absent).

[0022] Based on a lung cancer methylation detection kit, or by selecting plasma cfDNA as the sample, we used qPCR (quantitative polymerase chain reaction) equipment to collect methylation data of lung cancer-related genes in the user's blood. The genes involved were GRIK2, HOXA9, PTGER4, SHOX2, and PITX2, and the dCt values ​​of these five genes were calculated as methylation values. The calculation method is as follows: dCtGRIK2= Ct(GRIK2)–Ct(ACTB); dCtHOXA9= Ct(HOXA9)–Ct(ACTB); dCtPTGER4= Ct(PTGER4)–Ct(ACTB); dCtSHOX2= Ct(SHOX2)–Ct(ACTB); dCtPITX2= Ct(PITX2)–Ct(ACTB); Ct value is the cycle threshold for each gene in quantitative real-time PCR. The internal reference gene ACTB refers to the β-actin gene. The dCt value reflects the methylation level of the target gene relative to the internal reference gene ACTB. This step eliminates the influence of RNA quantity and technical error between samples, allowing for a relatively quantitative comparison of the expression level of the target gene.

[0023] Therefore, we selected the following five methylation features as input variables: dCtGRIK2, dCtHOXA9, dCtPTGER4, dCtSHOX2, and dCtPITX2.

[0024] 102. Combine the clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value to generate a feature vector; In this embodiment, the clinical feature [36,1,0,0] and methylation values ​​such as dCtGRIK2, dCtHOXA9, dCtPTGER4, dCtSHOX2 and dCtPITX2 are combined to generate the feature vector [36,1,0,0,dCtGRIK2,dCtHOXA9,dCtPTGER4,dCtSHOX2,dCtPITX2].

[0025] Furthermore, step 102 includes the following specific implementation methods: 1021. The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value are subjected to missing value filling and normalization to obtain preprocessed clinical features, preprocessed GRIK2 methylation value, preprocessed HOXA9 methylation value, preprocessed PTGER4 methylation value, preprocessed SHOX2 methylation value, and preprocessed PITX2 methylation value; 1022. Fill the pre-processed clinical features, pre-processed GRIK2 methylation value, pre-processed HOXA9 methylation value, pre-processed PTGER4 methylation value, pre-processed SHOX2 methylation value, and pre-processed PITX2 methylation value into the preset feature framework one by one to generate the feature vector.

[0026] In steps 1021-1022, the clinical feature [36,1,0,0] is analyzed for missing values. If missing values ​​are found, median imputation (continuous type) or mode imputation (categorical type) is used. Min-Max normalization is performed to map to the 0-1 interval to obtain the preprocessed clinical feature [1.0000,0.0278,0.0000,0.0000].

[0027] Analyze whether there are missing values ​​in dCtGRIK2, dCtHOXA9, dCtPTGER4, dCtSHOX2, and dCtPITX2. If missing values ​​are found, imput them using the median. Then, normalize the five dCt values ​​of the imputed data dCtGRIK2, dCtHOXA9, dCtPTGER4, dCtSHOX2, and dCtPITX2 to generate five preprocessed dCt values.

[0028] The feature framework can be configured with relevant positions according to requirements. For example, the feature framework can be designed as [dCtPITX2,dCtSHOX2,dCtHOXA9,dCtPTGER4,dCtGRIK2, Gender, Cancer Family History Category, Smoking History Category, Age]. The pre-processed data is then filled into the prepared framework to generate feature vectors.

[0029] 103. Based on a preset machine learning algorithm, the feature vector is classified to obtain a classification risk value; In this embodiment, the machine learning algorithm selected from logistic regression, vector machine, random forest, and XGBoost algorithms based on performance screening is used to classify the feature vectors and obtain a classification risk value. A classification risk value falling within the range of 0 to 1 is considered a "probability of malignancy".

[0030] For details, please refer to Figure 2 , Figure 2 This is a schematic diagram of a model training embodiment prior to step 103 of the lung cancer risk prediction method in this invention. Prior to step 103, the following specific implementation methods are included: 1031. Combine the clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of N training samples to generate N training feature vectors, where N is a positive integer; 1032. Perform result annotation processing on the N training feature vectors to obtain the annotation results corresponding to the N training feature vectors; 1033. Based on a preset machine learning algorithm, classify the N training feature vectors respectively to generate N predicted values; 1034. Based on the preset first hyperparameter and the preset second hyperparameter, the N predicted values ​​are processed to obtain N predicted results, wherein the first hyperparameter includes a parameter for a high-risk threshold, and the second hyperparameter includes a parameter for a low-risk threshold. 1035. Based on the annotation results corresponding to the N training feature vectors, calculate the accuracy of the N prediction results to obtain the test accuracy. 1036. Based on the test accuracy, adjust the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter, and iteratively train until the test accuracy converges.

[0031] In steps 1031-1036, the data from multiple training samples are first vectorized to obtain multiple training feature vectors. After vectorization, each training feature vector is labeled as high-risk, medium-risk, or low-risk.

[0032] Using a machine learning algorithm, N training samples are classified to generate N predicted values. Then, based on parameters with high-risk and low-risk thresholds, the N predicted values ​​are divided into N predicted results. These N predicted results are then compared with the original labeled results to generate the test accuracy.

[0033] Based on the test accuracy, the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained until the test accuracy converges. The test accuracy mainly observes the performance of accuracy, true positive rate, and true negative rate. Here, the true positive rate represents the ability to identify how many people are at risk, and the true negative rate represents the ability to ignore how many people are at risk.

[0034] Specifically, step 1033 includes the following specific implementation methods: 10331. Based on a preset logistic regression algorithm, perform logistic regression classification on the N training feature vectors to obtain N first predicted values; 10332. Based on a preset support vector machine, perform kernel function classification on the N training feature vectors to obtain N second predicted values; 10333. Based on the preset random forest algorithm, the N training feature vectors are subjected to decision classification processing to obtain N third prediction values; 10334. Based on the preset XGBoost algorithm, regression mapping classification is performed on the N training feature vectors to obtain N fourth prediction values.

[0035] In steps 10331-10334, the logistic regression algorithm, support vector machine, random forest algorithm, and XGBoost algorithm are used in parallel to classify the N training feature vectors. Logistic regression, kernel function classification, decision classification, and regression mapping classification are performed according to the rules of the algorithm, and N first predicted values, second predicted values, third predicted values, and fourth predicted values ​​are obtained in parallel.

[0036] Specifically, step 1034 includes the following specific implementation methods: 10341. Based on the preset first hyperparameter and the preset second hyperparameter, the results of the N first predicted values, the N second predicted values, the N third predicted values ​​and the N fourth predicted values ​​are divided to obtain N first predicted results, N second predicted results, N third predicted results and N fourth predicted results.

[0037] In step 10341, the first hyperparameter is uniformly given as the high-risk threshold and the second hyperparameter as the low-risk threshold. The results are divided in parallel for N first predicted values, N second predicted values, N third predicted values, and N fourth predicted values. However, the first and second hyperparameters of the first to fourth predicted values ​​will be adjusted independently until the accuracy of subsequent calculations converges to the minimum point of gradient descent.

[0038] Specifically, step 1035 includes the following specific implementation methods: 10352. Based on the annotation results corresponding to the N training feature vectors, calculate the accuracy of the N first prediction results, the N second prediction results, the N third prediction results, and the N fourth prediction results respectively to obtain the first accuracy, the second accuracy, the third accuracy, and the fourth accuracy.

[0039] In step 10352, the accuracy of the prediction results from the first to the fourth is calculated based on the annotation results corresponding to the training feature vectors. Here, the accuracy can be calculated using the true positive rate or the true negative rate, or the weights of the true positive rate and the true negative rate can be set according to the requirements to calculate the corresponding accuracy.

[0040] Specifically, step 1036 includes the following specific implementation methods: 10361. Based on the first accuracy, adjust the parameters of the logistic regression algorithm, the first hyperparameter, and the second hyperparameter for iterative training to obtain the first convergence accuracy; 10362. Based on the second accuracy, adjust the parameters of the support vector machine, the first hyperparameter, and the second hyperparameter and iteratively train to obtain the second convergence accuracy; 10363. Based on the third accuracy, adjust the parameters of the random forest algorithm, the first hyperparameter, and the second hyperparameter for iterative training to obtain the third convergence accuracy; 10364. Based on the fourth accuracy, adjust the parameters of the XGBoost algorithm, the first hyperparameter, and the second hyperparameter for iterative training to obtain the fourth convergence accuracy; 10365. Based on the maximum value of the first convergence accuracy, the second convergence accuracy, the third convergence accuracy, and the fourth convergence accuracy, determine the type of machine learning algorithm.

[0041] In steps 10361-10365, the first, second, third, and fourth accuracies are used as results to readjust the parameters of the original algorithm. Simultaneously, the first and second hyperparameters assigned to the corresponding parallel processing algorithm are also adjusted. GridSearchCV can be used for hyperparameter tuning, and the final selected parameters are as follows: n_estimators=200, max_depth=10, min_samples_split=5, min_samples_leaf=2, random_state=42. The hyperparameters are then adjusted. Using loss functions such as cross-entropy as the basis for backpropagation during training, the convergence accuracies of the corresponding algorithms are compared to obtain the first to fourth convergence accuracies. The machine learning algorithm corresponding to the maximum convergence accuracies from the first to the fourth is selected as the machine learning algorithm used for actual classification.

[0042] Furthermore, the N training feature vectors can be used as the training set, and another M feature vectors can be used as the validation set. The convergence accuracy of each machine learning algorithm can be verified using the validation set, and then the category corresponding to the highest convergence accuracy can be selected as the machine learning algorithm for actual classification.

[0043] In a preferred embodiment, the random forest prediction model is the best-performing model after screening; please refer to [link / reference]. Figure 3a , Figure 3a This is a schematic diagram of the ROC curve of the random forest algorithm for lung cancer risk prediction in this embodiment of the invention on the training dataset. The horizontal axis of the ROC curve is the true negative rate, and the vertical axis is the true positive rate. The performance of the random forest model built based on epidemiological and methylation features on the independent test dataset is as follows: AUC=0.960 (0.917-0.992); true positive rate=0.831 (0.700-0.941); true negative rate=0.934 (0.851-1.000). AUC is the area under the ROC curve.

[0044] Please see Figure 3b , Figure 3b This is a schematic diagram of the ROC curve of the random forest algorithm for lung cancer risk prediction in the embodiment of the present invention on the validation dataset. The horizontal axis of the ROC curve is the true negative rate, and the vertical axis is the true positive rate. In the validation dataset, AUC=0.93, which still maintains excellent prediction performance.

[0045] Please see Figure 4a , Figure 4aThis diagram illustrates the results of the random forest algorithm for lung cancer risk prediction in this embodiment of the invention, using a dual-threshold partitioning of the training dataset. Red represents lung cancer positive samples, and blue represents healthy samples. The dual thresholds for prediction probabilities are set to a high-risk threshold of 0.75 and a low-risk threshold of 0.35. The verified performance is as follows: the positive prediction value for the high-risk group is 0.947, and the negative prediction value for the low-risk group is 0.952.

[0046] Please see Figure 4b , Figure 4b As described in the embodiments of the present invention Figure 4a The sample distribution diagram shows that 51.2% of the sample is low-risk, with the clinical response strategy being routine follow-up; 25.6% of the sample is medium-risk, with the clinical response strategy being enhanced monitoring; and 23.2% of the sample is high-risk, with the clinical response strategy being immediate intervention.

[0047] Please see Figure 4c , Figure 4c This diagram illustrates the results of dual-threshold partitioning of the validation dataset using the random forest algorithm for lung cancer risk prediction in this embodiment of the invention. Red represents lung cancer positive samples, and blue represents healthy samples. The dual thresholds for prediction probabilities are set as follows: a high-risk threshold of 0.75 and a low-risk threshold of 0.35. The validation performance is as follows: a positive prediction value of 1.0 for the high-risk group and a negative prediction value of 1.0 for the low-risk group.

[0048] Please see Figure 4d , Figure 4d As described in the embodiments of the present invention Figure 4c The sample distribution diagram shows that 17.5% of the sample is low-risk, with the clinical response strategy being routine follow-up; 37.5% of the sample is medium-risk, with the clinical response strategy being enhanced monitoring; and 45.0% of the sample is high-risk, with the clinical response strategy being immediate intervention.

[0049] 104. When the classification risk value is greater than the preset high-risk threshold, a high-risk lung cancer prediction result corresponding to the target user is generated; In this embodiment, the high-risk threshold is set to 0.75. When the analyzed risk value is greater than 0.75, a high-risk prediction result for lung cancer of the target user is generated.

[0050] 105. When the classification risk value is less than the preset low-risk threshold, a low-risk lung cancer prediction result corresponding to the target user is generated. In this embodiment, the low-risk threshold is set to 0.35. When the classification risk value is less than 0.35, a low-risk prediction result for lung cancer of the target user is generated.

[0051] 106. When the classification risk value is not greater than the preset high-risk threshold and not less than the preset low-risk threshold, a lung cancer intermediate risk prediction result corresponding to the target user is generated.

[0052] In this embodiment, the high-risk threshold is set to 0.75 and the low-risk threshold is set to 0.35. When the classification risk value is between 0.75 and 0.35, the lung cancer risk prediction result for the target user is generated.

[0053] In this embodiment of the invention, feature vectors are generated by combining multi-gene methylation with clinical features. These feature vectors are then input into a pre-trained machine learning model for classification, generating well-trained risk prediction values. Finally, lung cancer risk is classified using trained dual thresholds, achieving consistency in lung cancer risk prediction. Combining gene methylation parameters with clinical features in medicine improves the accuracy of lung cancer prediction and solves the technical problems of insufficient consistency and prediction accuracy in current lung cancer screening programs.

[0054] Figure 5 This is a schematic diagram of a lung cancer risk prediction device 500 provided in an embodiment of the present invention. The lung cancer risk prediction device 500 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 510 (e.g., one or more processors) and a memory 520, and one or more storage media 530 (e.g., one or more mass storage devices) for storing application programs 533 or data 532. The memory 520 and storage media 530 can be temporary or persistent storage. The program stored in the storage media 530 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the lung cancer risk prediction device 500. Furthermore, the processor 510 may be configured to communicate with the storage media 530 and execute the series of instruction operations in the storage media 530 on the lung cancer risk prediction device 500.

[0055] The lung cancer risk prediction device 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input / output interfaces 560, and / or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, Linux, Free BSD, etc. Those skilled in the art will understand that... Figure 5 The lung cancer risk prediction device structure shown does not constitute a limitation on lung cancer risk prediction devices, which may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0056] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the lung cancer risk prediction method.

[0057] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0058] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0059] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for predicting lung cancer risk, characterized in that, Including the following steps: Receive the clinical characteristics of the target user and obtain the GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of the target user's plasma cfDNA; The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value are combined to generate a feature vector. Based on a preset machine learning algorithm, the feature vector is classified to obtain a classification risk value; When the classification risk value is greater than the preset high-risk threshold, a high-risk lung cancer prediction result is generated for the target user. When the classification risk value is less than the preset low-risk threshold, a low-risk lung cancer prediction result is generated for the target user. When the classification risk value is not greater than the preset high-risk threshold and not less than the preset low-risk threshold, a lung cancer intermediate risk prediction result corresponding to the target user is generated.

2. The lung cancer risk prediction method according to claim 1, characterized in that, Before the step of classifying the feature vector based on a preset machine learning algorithm to obtain a classification risk value, the method further includes: The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value of N training samples are combined to generate N training feature vectors, where N is a positive integer; The N training feature vectors are labeled to obtain the labeled results corresponding to the N training feature vectors; Based on a preset machine learning algorithm, the N training feature vectors are classified and processed to generate N predicted values. Based on the preset first hyperparameter and the preset second hyperparameter, the N predicted values ​​are processed to obtain N predicted results. The first hyperparameter includes a parameter for a high-risk threshold, and the second hyperparameter includes a parameter for a low-risk threshold. Based on the annotation results corresponding to the N training feature vectors, the accuracy of the N prediction results is calculated to obtain the test accuracy. Based on the test accuracy, the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained until the test accuracy converges.

3. The lung cancer risk prediction method according to claim 2, characterized in that, The step of classifying the N training feature vectors based on a preset machine learning algorithm to generate N predicted values ​​includes: Based on a preset logistic regression algorithm, logistic regression classification is performed on the N training feature vectors to obtain N first predicted values. Based on a preset support vector machine, the N training feature vectors are subjected to kernel function classification processing to obtain N second predicted values; Based on the preset random forest algorithm, decision classification processing is performed on the N training feature vectors to obtain N third prediction values; Based on the preset XGBoost algorithm, regression mapping and classification are performed on the N training feature vectors to obtain N fourth prediction values.

4. The lung cancer risk prediction method according to claim 3, characterized in that, The step of performing result determination processing on the N predicted values ​​according to the preset first hyperparameter and the preset second hyperparameter to obtain N prediction results includes: Based on the preset first hyperparameter and the preset second hyperparameter, the results of the N first predicted values, the N second predicted values, the N third predicted values ​​and the N fourth predicted values ​​are divided to obtain N first predicted results, N second predicted results, N third predicted results and N fourth predicted results.

5. The lung cancer risk prediction method according to claim 4, characterized in that, The accuracy calculation based on the annotation results corresponding to the N training feature vectors, yielding the test accuracy, includes: Based on the annotation results corresponding to the N training feature vectors, the accuracy of the N first prediction results, the N second prediction results, the N third prediction results, and the N fourth prediction results is calculated to obtain the first accuracy, the second accuracy, the third accuracy, and the fourth accuracy.

6. The lung cancer risk prediction method according to claim 5, characterized in that, The step of adjusting the parameters of the machine learning algorithm, the first hyperparameter, and the second hyperparameter based on the test accuracy and iteratively training until the test accuracy converges includes: Based on the first accuracy, the parameters of the logistic regression algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the first convergence accuracy. Based on the second accuracy, the parameters of the support vector machine, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the second convergence accuracy. Based on the third accuracy, the parameters of the random forest algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the third convergence accuracy. Based on the fourth accuracy, the parameters of the XGBoost algorithm, the first hyperparameter, and the second hyperparameter are adjusted and iteratively trained to obtain the fourth convergence accuracy. The type of machine learning algorithm is determined based on the maximum value of the first convergence accuracy, the second convergence accuracy, the third convergence accuracy, and the fourth convergence accuracy.

7. The lung cancer risk prediction method according to claim 1, characterized in that, The process of combining the clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value to generate a feature vector includes: The clinical features, GRIK2 methylation value, HOXA9 methylation value, PTGER4 methylation value, SHOX2 methylation value, and PITX2 methylation value were filled with missing values ​​and normalized to obtain preprocessed clinical features, preprocessed GRIK2 methylation value, preprocessed HOXA9 methylation value, preprocessed PTGER4 methylation value, preprocessed SHOX2 methylation value, and preprocessed PITX2 methylation value. The pre-processed clinical features, pre-processed GRIK2 methylation value, pre-processed HOXA9 methylation value, pre-processed PTGER4 methylation value, pre-processed SHOX2 methylation value, and pre-processed PITX2 methylation value are successively filled into the preset feature framework to generate feature vectors.

8. The lung cancer risk prediction method according to claim 1, characterized in that, The clinical characteristics include: age, sex, smoking history category, and family history of cancer category.

9. A lung cancer risk prediction device, characterized in that, The lung cancer risk prediction device includes: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; The at least one processor invokes the instructions in the memory to cause the lung cancer risk prediction device to perform the lung cancer risk prediction method as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a processor, it implements the lung cancer risk prediction method as described in any one of claims 1-8.

Images