TWI630501B - Establishment of a cancer prediction model and a method for analyzing cancer detection results in combination with a tumor marker set - Google Patents

Abstract

A method for establishing a cancer prediction model and a method for analyzing cancer detection results in combination with a tumor marker set: (A) inputting a plurality of tumor marker sets for detecting a protein of a subject and the corresponding cancer disease state to a machine learning In the module; (B) use the variable selection method in the module, select the variables, and select the variables with the best classification efficiency; (C) use the selected variables, values and cancer disease status, by supervision The method of machine learning is used to establish a cancer prediction model, and then the cancer detection result can be analyzed by using the cancer prediction model established above in combination with the test result of the tumor marker set.

Description

Establishment of a cancer prediction model and a method for analyzing cancer detection results in combination with a tumor marker set

A method for predicting cancer, in particular, a method for establishing a cancer prediction machine learning model and a method for analyzing cancer detection results in combination with a tumor marker set, capable of establishing a machine learning model through cancer prediction, and combining multiple tumors in a tumor marker set The flag test results are predicted.

In the related art, the cancer screening of the general healthy group is mainly based on the four cancer screenings currently promoted in China: oral mucosa examination, mammography, fecal occult blood test, and Pap smear. Four cancer screening tests for oral cancer, breast cancer, colorectal cancer, and cervical cancer, respectively, can screen out potential cancers without obvious symptoms. The four cancer screening tests have achieved good results. However, the above-mentioned examination methods are only for a single cancer that is common to four people. This type of examination cannot detect cancers of a type other than four cancers.

In addition to the four cancer screenings implemented by the state, there are many methods for screening for potential cancers, including: low-dose chest computed tomography screening for lung cancer, colonoscopy for colorectal cancer screening, and abdominal ultrasound for screening liver cancer. and many more. However, these methods can only be screened for a specific single cancer. If multiple types of cancer screening are to be performed, the subject must perform multiple different tests or examinations. This is not only inconvenient, but also expensive. It may also expose the subject to potential iatrogenic damage and radiation.

According to the above missing, the prior art still has room for improvement.

The present invention provides a method for establishing a cancer prediction model, comprising at least the following steps: (A) inputting a plurality of tumor marker set test results for detecting a test subject protein and corresponding cancer disease states to a machine learning module (B) use the variable selection method in the module, select the variables, and select the variables with the best classification efficiency; and (C) use the selected variables, values and cancer disease status, by supervised machine The method of learning is to establish a cancer prediction model.

The supervised machine learning method is a logical regression, a K-neighbor method, a support vector machine, a neural network learning, a decision tree, a Bayesian decision method, or any combination thereof.

Wherein, the cancer disease state is a state classification with cancer/no cancer, a state classification of early cancer/advance cancer (such as a tumor state classification of a TNM tumor staging system), or a classification of cancer species (such as liver cancer/lung cancer/colorectal cancer) ...Wait).

Wherein, the date of determination of the cancer disease state and the date of inspection of the tumor marker set are between 1 day and 3 years.

Among them, the established machine learning model can calculate Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV) according to cancer markers of different modules. ), Accuracy, Area under ROC Curve (AUC), and Youden index.

The invention further provides a cancer prediction model combined with a tumor marker kit The method for analyzing the results of cancer detection comprises at least the following steps: (A) providing a sample of a new subject; and (B) simultaneously performing a plurality of the above-mentioned samples through a plurality of tumor markers containing the detection protein in the tumor marker set. Tumor marker detection; (C) Entering the tumor marker set test results to the above cancer prediction model, the cancer test results can be compared and analyzed; (D) making a risk prediction for cancer, thereby reminding Further follow-up actions can be taken by the examiner or the medical terminal.

The sample may be human blood, urine, saliva, sweat, feces, pleural effusion, ascites, and cerebral ridge.

Wherein, the plurality of tumor markers are Alpha Fetal Protein (AFP), Carcinoembryonic Antigen (CEA), Carbohydrate Antigen 19-9 (CA19-9), cytokeratin Antigen (Cytokeratin Fragment 21-1, CYFRA21-1), Squamous Cell Carcinoma Antigen (SCC), Prostate Specific Antigen (PSA), Carbohydrate Antigen (CA15) -3), Carbohydrate Antigen 125 (CA125), Human Epstein-Barr Virus IgA (EBV IgA), Carbohydrate Antigen (CA27-29), Beta 2Beta-2-microglobulin, Beta-human Chorionic Gonadotropin (Beta-hCG), Cluster of Differentiation 177 (CD 177), Differentiation group antigen 20 (Cluster) Of Differentiation 20, CD 20), Chromogranin A (CgA), Human Epididymis Secretory Protein 4 (HE 4), Lactate Dehydrogenase (LDH), Thyroglobulin (Thyroglobulin), nerve NSE (Neuron-specific Enolase, NSE), nuclear matrix protein 22 (Nuclear Matrix Protein 22), program cell death ligand 1 (Programmed Death Ligand 1, PD-L1).

The advantage of the machine learning model is that the present invention adopts a supervised machine learning operation method, and can establish a cancer prediction model by combining a large amount of clinical data in the past, and can simultaneously perform auxiliary reading and analysis of multiple tumor markers, so as to maximize the In the existing data, the differences in the distribution of tumor markers in different disease state cases are analyzed, and the classification basis of different disease states is found out from the overall data distribution appearance, which can improve the accuracy of prediction, timeliness, economic efficiency and reproducibility. .

The advantages of using a variety of tumor markers in tumor marker kits are: First, the specimens are easy to pick, have no contact radiation, have low discomfort, and have no risk of anesthesia. They can also be sampled with minimally invasive surgery that is currently promoted. The willingness to investigate and the possibility of large-scale screening can also facilitate the screening of all types of cancer in the asymptomatic general population.

Second, tumor markers can be detected by automated systems. Through automated and strict quality control, the accuracy and accuracy of tumor markers can be maintained at a certain level.

Third, the interpretation of tumor marker test results is objective and can prevent possible inconsistencies in interpretation.

Finally, various tumor markers can be obtained and quantified from a single sample of the sample, and different types of cancer can be screened by different combinations of tumor markers. In other words, cancer screening for tumor markers is safe, objective, and cost-effective. And can also screen for various types of cancer.

Figure 1 is a flow chart showing the steps of the method for establishing a cancer prediction machine learning model of the present invention.

2 is a schematic flow chart of a method for analyzing cancer detection results by combining a cancer prediction machine learning model of the present invention with a tumor marker set

Figure 3A is the receiver operating curve (ROC curve) for each single tumor marker (male)

Figure 3B is a receiver operating characteristic curve (ROC curve) of a different type of supervised machine learning method incorporating multiple tumor markers of the present invention (male)

Figure 3C shows the receiver operating curve (ROC curve) for each single tumor marker (female)

Figure 3D is a receiver operating characteristic curve (ROC curve) of a different type of supervised machine learning method combining multiple tumor markers of the present invention (female)

Referring to Fig. 1, the present invention provides a method for establishing a cancer prediction machine learning model, comprising at least the following steps: (A) inputting a plurality of tumor marker sets of test proteins for detecting a tester and corresponding cancer disease states To the machine learning module; (B) use the variable selection method in the module, select the variables, and select the variables with the best classification efficiency; and (C) use the selected variables, values and cancer diseases State, the establishment of a cancer prediction model by supervised machine learning.

The supervised machine learning method is a logical regression, a K-neighbor method, a support vector machine, a neural network learning, a decision tree, a Bayesian decision method, or any combination thereof.

Wherein, the cancer disease state is a state classification with cancer/no cancer, a state classification of early cancer/advance cancer (such as a tumor state classification of a TNM tumor staging system), or a classification of cancer species (such as liver cancer/lung cancer/colorectal cancer) ...Wait).

Wherein, the date of determination of the cancer disease state and the date of inspection of the tumor marker set are between 1 day and 3 years.

Among them, the established machine learning model can calculate Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV) according to cancer markers of different modules. ), Accuracy, Area under ROC curve (AUC), and Youden index.

Referring to FIG. 2, the present invention further provides a method for analyzing cancer detection results using a cancer prediction machine learning model in combination with a tumor marker set, comprising at least the following steps: (A) providing a sample of a new subject; (B) The tumor marker kit contains a plurality of tumor markers for detecting proteins, and simultaneously performs multiple tumor marker detection on the above-mentioned specimens; (C) inputting the tumor marker set test results to the cancer prediction machine learning model described above, and performing cancer detection results Comparison and analysis; (D) make a risk prediction for cancer, which can remind the subject or the medical end to further follow up.

The sample may be human blood, urine, saliva, sweat, feces, pleural effusion, ascites, and cerebral ridge.

Wherein, the plurality of tumor markers are Alpha Fetal Protein (AFP), Carcinoembryonic Antigen (CEA), Carbohydrate Antigen 19-9 (CA19-9), cytokeratin Antigen (Cytokeratin Fragment 21-1, CYFRA21-1), Squamous Cell Carcinoma Antigen (SCC), Prostate Specific Antigen (PSA), Carbohydrate Antigen (CA15) -3), Carbohydrate Antigen 125 (CA125), human type IV Epstein-Barr Virus IgA (EBV IgA), Carbohydrate Antigen (CA27-29), Beta-2-microglobulin, Beta Human Human Chorionic Hormone ( Beta-human Chorionic Gonadotropin, Beta-hCG), Cluster of Differentiation 177 (CD 177), Cluster of Differentiation 20 (CD 20), Chromogranin A (CgA), Human Epididymis Secretory Protein 4 (HE 4), Lactate Dehydrogenase (LDH), Thyroglobulin, Neuron-specific Enolase (NSE) , Nuclear Matrix Protein 22, Programmed Death Ligand 1, PD-L1.

Referring to Figures 3A to 3D, in this embodiment, the sample used is a blood sample, and in the cancer detection, 8 tumor markers are analyzed by data exploration methods, respectively AFP, CEA, CA19-9, CYFRA21- 1, SCC, PSA, CA15-3, CA125, the following is the operation process:

1. Subject Condition Setting, including Conditions and Numbers of Inclusion and Exclusion: The subject of the present invention is an adult of a 20-year-old adult who selects a cancer marker screening kit for self-funded health testing.

2. Design and method: The main measurement value is the test value of eight cancer markers. The follow-up survey subjects have new cancers and their types within one year after the test. After the data is completed, the embodiment establishes several supervised learning models, including: logical regression, K-nearest neighbor, and support vector machine.

3. During the retrospective period of the data in this example: the data is traced back from January 1, 1999 to December 31, 2013.

4. Evaluation of results and statistical methods: This example calculates various tumor markers The distribution of the target data, and the selection of variables before the model is established, to select the number of variables with the best classification efficiency. In the present embodiment, the variable effectiveness calculates the area under the curve (AUC) by the ROC curve to evaluate its effectiveness, and selects the combination of variables with the best classification ability. In addition, in this embodiment, the predictive ability of the model is verified by the internally verified group, and the statistical indicators such as the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of the classifier model are calculated accordingly.

Figures 3A and 3B individually show the cancer screening efficacy (male) of various single different tumor markers and different machine learning algorithms, where LR is logistic regression and KNN is k nearest neighbor (K nearest Neighbor), SVM is a support vector machine (Support vector machine), and then from Table 1 can prove the tumor marker set and use the machine learning algorithm, compared to the use of a single tumor marker to significantly improve the effectiveness of cancer screening, The tumor markers in a cancer test use AFP, CEA, CA19-9, CYFRA21-1, SCC, PSA.

Figures 3C and 3D show the cancer screening efficacy (female) of various single different tumor markers and different machine learning algorithms. Among them, LR is logistic regression, KNN is k nearest neighbor (K nearest Neighbors, SVM is a support vector machine. Then, from Table 2, the use of the tumor marker set combined with the machine learning algorithm can be seen. Compared with the use of a single tumor marker, the cancer screening performance is significantly improved. Table 2 shows the tumor markers in cancer detection using AFP, CEA, CA19-9, CYFRA21-1, SCC, CA15-3, CA125.

Further, see Tables 3 and 4 below to compare the performance differences between machine learning algorithms and traditional interpretation methods.

Table 3 reveals the effectiveness of machine learning algorithms and traditional interpretation methods (male): the effectiveness of machine learning algorithms such as support vector machines and k-nearest neighbor methods is significantly better than traditional interpretation methods.

Table 4 reveals the effectiveness of machine learning algorithms and traditional interpretation methods (female): support vector machines, k-nearest neighbors, and logical regression provide better performance than traditional interpretation methods.

The results of Tables 3 and 4 show that the use of machine learning algorithms for the analysis and study of tumor marker sets data can improve the effectiveness of cancer screening, both for men and women.

In summary, the embodiments of the present invention can simultaneously improve the convenience, economy, and correctness of general population cancer screening, and the medical personnel can learn from more aspects through the single-multiple tumor marker test of the tumor marker set. The physical condition and possible potential cancer, the subject does not have to carry out multiple, different items of testing. Tumor marker sets can improve the timeliness of testing, while also reducing many possible iatrogenic injuries and radiation exposure, and because the tumor marker set contains a considerable amount of information, a predictive model for cancer formation in combination with machine learning algorithms The ability to interpret and interpret the results of multiple data can provide considerable improvement in correctness, timeliness and reproducibility of interpretation results.

It is to be noted that the above-described embodiments are merely illustrative of the principles of the invention and its advantages, and are not intended to limit the scope of the invention. Modifications and variations of the embodiments can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore the right of the present invention The scope of protection shall be as described in the scope of the patent application described later.

Claims (8)

A method for establishing a cancer prediction machine learning model comprises at least the following steps: (A) inputting a plurality of tumor marker set test results for detecting a subject protein and a corresponding cancer disease state to a machine learning module; B) using the variable selection method in the module, selecting the variables, and selecting the variables with the best classification efficiency; and (C) using the selected variables, values, and cancer disease states, by supervised machine learning Methods The establishment of a cancer prediction model was performed. The method for establishing a cancer prediction machine learning model according to claim 1, wherein the supervised machine learning method is a logical regression, a K-neighbor method, a support vector machine, a neural network learning, a decision tree. , Bayesian decision making or any combination of the above. The method for establishing a cancer prediction machine learning model according to claim 1, wherein the cancer disease state is a cancer/no cancer state classification, an early cancer/advance cancer state classification, or a cancer species classification. The method for establishing a cancer prediction machine learning model according to claim 1, wherein the date of determination of the cancer disease state and the date of inspection of the tumor marker set are 1 day to 3 years apart. . The method for establishing a cancer prediction machine learning model according to claim 1, wherein the established machine learning model can calculate sensitivity (Sensitivity), specificity (Specificity), and positive according to cancer markers of different modules. Positive Predictive Value (PPV), Negative Predictive Value (NPV), Accuracy, Area Under ROC Curve (AUC), Yoden (Yonden) Index) The statistical indicator. A method for performing cancer screening using a cancer prediction machine learning model as described in claim 1 in combination with a tumor marker kit, comprising at least the following steps: (A) providing a sample of a new subject; (B) The tumor marker kit contains a plurality of tumor markers for detecting proteins, and simultaneously performs multiple tumor marker detection on the above-mentioned specimens; (C) inputting the tumor marker set test results to the cancer prediction machine learning model described above, and performing cancer detection results Comparison and analysis; (D) make a risk prediction for cancer. The method for analyzing a cancer detection result according to the cancer prediction machine learning model described in claim 6 of the patent scope, wherein the sample may be blood, urine, saliva, sweat, and feces , pleural effusion, ascites, brain ridge with fluid. The method for analyzing cancer detection results by combining a cancer prediction machine learning model according to claim 6 of the patent application scope, wherein the plurality of tumor markers are Alpha Fetal Protein (AFP), cancer Embryo antigen (Carinoembryonic Antigen, CEA), Carbohydrate Antigen 19-9 (CA19-9), Cytokeratin Fragment 21-1 (CYFRA21-1), Squamous Cell Carcinoma Antigen, SCC), Prostate Specific Antigen (PSA), Carbohydrate Antigen (CA15-3), Carbohydrate Antigen 125 (CA125), Human Type IV Herpes Virus Antibody (Epstein-Barr Virus IgA, EBV IgA), Carbohydrate Antigen (CA27-29), Beta-2-microglobulin, Beta Human Human Chorionic Hormone (Beta-human) Chorionic Gonadotropin, Beta-hCG), Cluster of Differentiation 177 (CD 177), Cluster of Differentiation 20 (CD 20), Chromogranin A (CgA), Human Paracrine Secretion Protein 4 (Human Epididymis Secretory Protein 4, HE 4), Lactate Dehydrogenase (LDH), Thyroglobulin, Neuron-specific Enolase (NSE), Nuclear Matrix Protein 22 (Nuclear Matrix Protein 22), Programmed Death Ligand 1, PD-L1.