阅读说明：本技术 抗血管生成药物联合免疫检查点抑制剂治疗疗效预测方法 (Prediction method for treatment efficacy of anti-angiogenesis drug combined with immune checkpoint inhibitor ) 是由 孙惠川 饶圣祥 徐彬 朱小东 董三源 黄成� 沈英皓 曾蒙苏 周俭 樊嘉 于 2021-08-05 设计创作，主要内容包括：本发明涉及一种抗血管生成药物联合免疫检查点抑制剂治疗疗效预测方法,准备接受抗血管生成药物联合免疫检查点抑制剂治疗患者的疗效评价数据；利用3D医学图像处理软件读取治疗前患者的增强磁共振成像资料,然后由数个相关医生对同一成像资料进行病灶范围或病灶区域勾画,增强磁共振成像序列和每个医生的勾画信息形成数个特征勾画文件；利用pyradiomics包对数份勾画文件分别提取病灶的影像组学特征参数；利用获得的数份影像组学特征参数文件,进行筛选获得有效特征,根据筛选得到的特征建立预测模型,并训练和验证预测模型用于对未接受治疗前病患进行治疗后疗效预测,以辅助临床决策、筛选有效人群,提高整体疗效,减轻患者经济负担,避免不良反应。(The invention relates to a prediction method of curative effect of anti-angiogenesis drug combined immune checkpoint inhibitor treatment, which prepares curative effect evaluation data of patients receiving anti-angiogenesis drug combined immune checkpoint inhibitor treatment; reading enhanced magnetic resonance imaging data of a patient before treatment by using 3D medical image processing software, then carrying out focus range or focus area delineation on the same imaging data by a plurality of related doctors, and forming a plurality of characteristic delineation files by an enhanced magnetic resonance imaging sequence and delineation information of each doctor; respectively extracting image omics characteristic parameters of the focus by using a pyradiomics package to the number of sketching files; screening by using the obtained several image omics characteristic parameter files to obtain effective characteristics, establishing a prediction model according to the screened characteristics, and training and verifying the prediction model to predict the curative effect of the patient before treatment, so as to assist clinical decision, screen effective people, improve the overall curative effect, reduce the economic burden of the patient and avoid adverse reactions.)

1. A method for predicting the treatment effect of an anti-angiogenesis drug combined with an immune checkpoint inhibitor is characterized by comprising the following steps:

1) prepare data for efficacy assessment of patients receiving anti-angiogenic drugs in combination with immune checkpoint inhibitors: evaluating previous case data, collecting data before and after treatment, evaluating the examination result of each stage of a treated subject by using an internationally prevailing curative effect evaluation standard, and reserving the optimal primary result as a final evaluation result;

2) reading enhanced magnetic resonance imaging data of the patient before treatment in the step 1) by using 3D medical image processing software, then performing focus range or region delineation on the same imaging data by a plurality of related doctors, and enhancing a magnetic resonance imaging sequence and delineating information of each doctor to form a plurality of feature delineation files;

3) extracting the image omics characteristic parameters of the focus from the plurality of drawing files in the step 2) by using a radiopacimics package, and storing the extracted image omics characteristic parameters as text files, namely a plurality of image omics characteristic parameter files;

4) performing characteristic parameter correlation calculation by using the plurality of image omics characteristic parameter files obtained in the step 3), screening out characteristics with high correlation coefficient, obtaining effective characteristics through algorithm screening, establishing a prediction model according to the characteristics obtained through screening, and training and verifying the prediction model by using the data obtained in the step 1);

5) after the prediction model is established, a clinician inputs basic information before receiving the combination therapy and enhanced magnetic resonance imaging data after lesion range or region delineation into the prediction model, and predicted curative effect evaluation data can be obtained.

2. The method for predicting the curative effect of the anti-angiogenesis drug combined with the immune checkpoint inhibitor according to claim 1, wherein the specific method realized in the step 4) is as follows:

reading all the curative effect evaluation data of the patients in the step 1) and the characteristic parameters of the omics obtained in the step 3) in Python or R, and respectively standardizing the characteristic parameters of the several groups of the omics;

screening for the first time: using a plurality of image omics characteristic parameter files, taking files with the same characteristic parameter as a group, calculating the intra-group relevance of the characteristic parameter, reserving the characteristic parameter with the number of relevant systems being more than or equal to 0.8, and calculating and screening the intra-group relevance of all the characteristic parameters by using the method to obtain the characteristic parameter with high relevance;

and (3) screening for the second time: the characteristic parameters which are reserved after the first screening are further screened through a Lasso algorithm to obtain final characteristics;

establishing a prediction model according to the screened characteristics, using the case data in the step 1), and performing prediction model training by adopting one or more machine learning algorithms of a neural network, a support vector machine and a logistic regression;

and (3) verifying the trained prediction model by using part of case data, and comparing the verification result with the evaluation result obtained in the step 1) to obtain the evaluation of the model.

3. The method for predicting the therapeutic effect of the anti-angiogenic drug in combination with the immune checkpoint inhibitor according to claim 2, wherein the strategy of training and verifying the prediction model is as follows:

if the case data is enough, randomly grouping the data into a training group and an internal verification group, training the model by using the training group, and then verifying the trained model by using the verification group;

if the number of cases is small, performing model training and verification by using all data by adopting a cross verification or bootstrap method;

in the process of training and verifying the model, the model can be evaluated in the aspects of discrimination and calibration.

4. The method for predicting the treatment efficacy of the anti-angiogenic drug in combination with immune checkpoint inhibitor according to claim 3, wherein after the prediction model trained and validated by the screening characteristics of the iconomics is established, each patient obtains a parameter by calculation according to the prediction model, and the parameter is used as an independent index to establish a new prediction model in combination with the basic information, the baseline tumor information and the baseline laboratory examination result index of the patient in step 1) before the patient receives the combination therapy.

5. The method of predicting the therapeutic efficacy of an anti-angiogenic drug in combination with an immune checkpoint inhibitor as in claim 4, wherein the evaluation of the model is obtained by comparing the predictive power of the simple imaging omics predictive model and the predictive power of the new predictive model using the index.

6. The method of predicting therapeutic efficacy of an anti-angiogenic drug in combination with an immune checkpoint inhibitor according to claim 5, wherein the index is any one of an area under an operational characteristic curve of the subject, a net reclassification index and a comprehensive discriminant improvement index.

7. The method for predicting the curative effect of the anti-angiogenic drug in combination with immune checkpoint inhibitor according to any one of claims 1 to 6, wherein the trained and verified prediction model is established as a visual prediction model, or a webpage tool based on the prediction model is developed, and the finally established prediction model is visualized in a risk scoring and nomogram mode.

Technical Field

The invention relates to an image analysis technology, in particular to a prediction method for treatment effect of an anti-angiogenesis drug combined immune checkpoint inhibitor.

Background

Taking liver cancer as an example, the first-line anti-angiogenic drugs currently used for treating hepatocellular carcinoma include sorafenib, ranvatinib and doraninib, and the second-line anti-angiogenic drugs include regorafenib and apatinib. The 5 medicines are all multi-target tyrosine kinase inhibitors, and no effective therapeutic effect prediction molecular marker is found at present.

Although immune checkpoint inhibitors have been used in the treatment of a variety of solid malignancies, the efficacy predictors vary, including tumor cell expression levels of programmed death protein ligand-1 (PD-L1), Tumor Mutation Burden (TMB), Combined Positive Score (CPS), and Microsatellite Instability (MSI). However, for hepatocellular carcinoma, no reliable molecular marker for predicting its efficacy was found in any of the three phase III clinical trials involving treatment of hepatocellular carcinoma with immune checkpoint inhibitors.

For the combined treatment (anti-angiogenesis drug combined with immune checkpoint inhibitor), the use of molecular markers to predict the efficacy of the combination of two drugs is more difficult, the clinical application is also more difficult, and no marker is available to effectively predict the efficacy of the combined treatment. At present, some researches are exploring to predict the curative effect by using the change condition of the tumor marker before and after the combination treatment, but the method cannot predict the curative effect before the patient receives the combination treatment, also brings certain economic burden to the patient and is not an optimal curative effect prediction method; still other studies are exploring the use of peripheral blood to predict efficacy before patients receive combination therapy, which not only increases the clinical workload, but also makes the method of peripheral blood detection more complex and costly, and is not suitable for large-scale clinical applications.

Therefore, for diseases which need to be treated in a certain period of time for disease control, the prediction of the treatment efficacy is particularly important before treatment, patients with high efficacy (effective patients) can be selected, the incidence rate of adverse reactions is reduced, and the operation cost of treatment is reduced.

Disclosure of Invention

Aiming at the problem of difficult prediction of the curative effect of the disease, a prediction method of the curative effect of the combination of the anti-angiogenesis drug and the immune checkpoint inhibitor is provided, and a calculation method is established by utilizing enhanced magnetic resonance imaging data generated by a patient in the conventional diagnosis and treatment process so as to predict the curative effect of the combination of the anti-angiogenesis drug and the immune checkpoint inhibitor.

The technical scheme of the invention is as follows: a prediction method for treatment efficacy of an anti-angiogenesis drug combined with an immune checkpoint inhibitor specifically comprises the following steps:

1) prepare data for efficacy assessment of patients receiving anti-angiogenic drugs in combination with immune checkpoint inhibitors: evaluating previous case data, collecting data before and after treatment, evaluating the examination result of each stage of a treated subject by using an internationally prevailing curative effect evaluation standard, and reserving the optimal primary result as a final evaluation result;

2) reading enhanced magnetic resonance imaging data of the patient before treatment in the step 1) by using 3D medical image processing software, then performing focus range or region delineation on the same imaging data by a plurality of related doctors, and enhancing a magnetic resonance imaging sequence and delineating information of each doctor to form a plurality of feature delineation files;

3) extracting the image omics characteristic parameters of the focus from the plurality of drawing files in the step 2) by using a radiopacimics package, and storing the extracted image omics characteristic parameters as text files, namely a plurality of image omics characteristic parameter files;

4) performing characteristic parameter correlation calculation by using the plurality of image omics characteristic parameter files obtained in the step 3), screening out characteristics with high correlation coefficient, obtaining effective characteristics through algorithm screening, establishing a prediction model according to the characteristics obtained through screening, and training and verifying the prediction model by using the data obtained in the step 1);

5) after the prediction model is established, a clinician inputs basic information before receiving the combination therapy and enhanced magnetic resonance imaging data after lesion range or region delineation into the prediction model, and predicted curative effect evaluation data can be obtained.

Further, the specific method implemented in the step 4) is as follows:

reading all the curative effect evaluation data of the patients in the step 1) and the characteristic parameters of the omics obtained in the step 3) in Python or R, and respectively standardizing the characteristic parameters of the several groups of the omics;

screening for the first time: using a plurality of image omics characteristic parameter files, taking files with the same characteristic parameter as a group, calculating the intra-group relevance of the characteristic parameter, reserving the characteristic parameter with the number of relevant systems being more than or equal to 0.8, and calculating and screening the intra-group relevance of all the characteristic parameters by using the method to obtain the characteristic parameter with high relevance;

and (3) screening for the second time: the characteristic parameters which are reserved after the first screening are further screened through a Lasso algorithm to obtain final characteristics;

establishing a prediction model according to the screened characteristics, using the case data in the step 1), and performing prediction model training by adopting one or more machine learning algorithms of a neural network, a support vector machine and a logistic regression;

and (3) verifying the trained prediction model by using part of case data, and comparing the verification result with the evaluation result obtained in the step 1) to obtain the evaluation of the model.

Further, the strategy of the prediction model training and verification is as follows:

if the case data is enough, randomly grouping the data into a training group and an internal verification group, training the model by using the training group, and then verifying the trained model by using the verification group;

if the number of cases is small, performing model training and verification by using all data by adopting a cross verification or bootstrap method;

in the process of training and verifying the model, the model can be evaluated in the aspects of discrimination and calibration.

Further, after the prediction model trained and verified by the screening characteristics of the image omics is established, each patient calculates a parameter according to the prediction model, the parameter is used as an independent index, and a new prediction model is established by combining the parameter with the basic information, the baseline tumor information and the baseline laboratory examination result index of the patient in the step 1) before the patient receives the combined treatment.

And further, the prediction capabilities of the simple image omics prediction model and the new prediction model are compared by adopting indexes to obtain the evaluation of the models.

Further, the index is any one of an area under an operation characteristic curve of the subject, a net reclassification index and a comprehensive judgment improvement index.

Further, the prediction model after training and verification is established as a visual prediction model, or a webpage tool based on the prediction model is developed, and the finally established prediction model is visualized in a risk scoring and nomogram mode.

The invention has the beneficial effects that: the method for predicting the treatment effect of the anti-angiogenesis drug and immune checkpoint inhibitor can predict the treatment effect of the anti-angiogenesis drug and immune checkpoint inhibitor on liver cancer, and assist clinical decision and screen effective people, so that the overall treatment effect is improved, the economic burden of patients is reduced, and adverse reactions are avoided.

Detailed Description

The present invention will be described in detail below with reference to specific examples. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.

In clinical diagnosis and treatment of hepatocellular carcinoma, enhanced magnetic resonance imaging of the liver is a routine examination item. Doctors often rely on enhanced magnetic resonance imaging data of the liver to assess the size, extent, nature, etc. of tumors, to determine whether patients can undergo surgical treatment, and to approximate the survival of patients.

But doctors can only distinguish the outline of the tumor by naked eyes in the process of reading magnetic resonance imaging data, and the detailed characteristics of the tumor on the magnetic resonance imaging are difficult to quantitatively evaluate; furthermore, different physicians may have discrepancies in the evaluation of the same mri data. The imaging technology (radiopomics package) can extract high-flux imaging characteristic parameters on magnetic resonance imaging data, and the characteristic parameters can reflect image details which cannot be distinguished by naked eyes and have good repeatability; and then through statistical analysis, the correlation between the characteristic parameters of the image omics and the curative effect of the combined treatment can be established, so that the curative effect of the combined treatment can be predicted.

Therefore, the method aims to predict the curative effect of the patient before the patient receives the combined treatment by utilizing the enhanced magnetic resonance imaging data generated by the patient in the conventional diagnosis and treatment and utilizing the imaging technology.

The invention relates to a prediction method of curative effect of anti-angiogenesis drug combined immune checkpoint inhibitor treatment, which comprises the following steps:

1. prepare data for efficacy assessment of patients receiving anti-angiogenic drugs in combination with immune checkpoint inhibitors: evaluating previous case data, storing data before treatment, acquiring data after treatment of a treatment plan subject, evaluating the examination result of each stage of the treatment subject by using the corresponding curative effect standard of the existing case, and reserving the best one-time result as the final evaluation result.

For example: evaluating the tumor focus with the longest diameter in the liver by adopting mRECIST standard, taking the best one of all evaluation results as the final curative effect evaluation result of the patient, and saving the identification code of the patient, the basic information before receiving the combined treatment, the baseline tumor information, the baseline laboratory examination result and the curative effect evaluation result as a text file (for example, txt, csv, xlsx);

2. reading enhanced magnetic resonance imaging data of the patient before treatment in the step 1 by using 3D Slicer or ITK-SNAP software (3D medical image processing software), then performing lesion range or region delineation on the same imaging data by a plurality of related doctors, and forming a plurality of characteristic delineation files by the enhanced magnetic resonance imaging sequence and the delineation information of each doctor;

for example: for liver tumor enhancement magnetic resonance imaging data, two radiologists or imaging physicians independently delineate the tumor focus range with the longest liver diameter layer by layer on the sequences of the arterial phase, the portal phase, the venous phase, the delay phase and the like, after the delineation is completed, the corresponding enhancement magnetic resonance imaging sequence (for example, nrrd, ni.gz format) and delineation information (for example, seg.nrrd, ni.gz format) are saved as a delineation file, and the delineation files of the two radiologists or imaging physicians are respectively saved and saved as two delineation files.

3. Extracting the image omics characteristic parameters of the focus from the plurality of drawing files in the step 2 by utilizing a radiophysics package (used for image omics characteristic extraction of medical images), and storing the extracted image omics characteristic parameters as text files (such as txt, csv and xlsx), namely a plurality of image omics characteristic parameter files;

4. performing characteristic parameter correlation calculation by using the plurality of image omics characteristic parameter files obtained in the step 3, screening out characteristics with high correlation coefficient, obtaining effective characteristics through algorithm screening, establishing a prediction model according to the characteristics obtained through screening, and training and verifying the prediction model by using the existing data;

1) reading the curative effect evaluation data of all patients in the step 1 and the characteristic parameters of the proteomics obtained in the step 3 in Python or R, and respectively standardizing the characteristic parameters of the several proteomics; performing feature screening (2 steps of feature screening are performed, namely all features → retaining the features of which the intra-group correlation coefficient is more than or equal to 0.8 → retaining the features obtained by the Lasso algorithm screening), on the features by adopting algorithms such as minimum absolute value convergence and selection algorithm (Lasso), establishing a prediction model by utilizing the finally screened image omics features, and training and verifying the image omics prediction model by using case data; the method for training the model can adopt one or more machine learning methods of neural network, support vector machine and logic regression, or deep learning algorithm; the strategy for model training and validation is as follows: if the number of cases is large, the data can be randomly grouped into a training group and an internal verification group, and after the model is trained by using the training group, the model obtained by training is verified by using the verification group; if the number of cases is less, a cross validation or bootstrap method can be adopted to carry out model training and validation by using all data; during the training and verification of the model, the model can be evaluated in terms of discrimination (such as indexes of area under the operation characteristic curve of the subject, C statistic and the like), calibration degree (such as indexes of a calibration curve, goodness-of-fit test and the like);

wherein, the calculation process of the intra-group correlation coefficient is as follows:

for example, n sets of image omics feature parameter files are shared, k feature parameters are extracted from each image omics feature parameter file, the intra-group correlation coefficient of the feature parameter 1 is calculated by using the parameters (1,1), (2,1), (3,1), … and (n,1) as a group, and similarly, the intra-group correlation coefficient … … of the feature parameter 2 is calculated by using the parameters (1,2), (2,2), (3,2), … and (n,2) as a group, that is, files with the same feature parameter as a group, and the intra-group correlation degree of the feature parameter is calculated to obtain the intra-group correlation coefficient.

And respectively calculating the intra-group correlation coefficients of the k characteristic parameters according to the method, namely calculating each characteristic parameter to obtain one intra-group correlation coefficient.

The Lasso algorithm: each time the result of the feature parameter combination obtained by the Lasso algorithm has randomness, so that the Lasso algorithm is repeated for a sufficient number of times (for example, 500 times and 1000 times), and the feature parameter combination with the largest occurrence number in the repeated number is used as the finally obtained feature. The number of the characteristic parameters can be controlled within a reasonable range.

2) Optional steps are as follows: after the imaging omics prediction model is established, each patient can calculate to obtain a parameter according to the imaging omics prediction model, the parameter is used as an independent index, a new prediction model is established with the indexes of basic information, baseline tumor information, baseline laboratory examination results and the like of the patient before the patient receives combined treatment in the step 1, and the training, verification and evaluation of the new prediction model can refer to 1);

in addition, indexes such as the area under the operation characteristic curve of the testee, the net weight reclassification index, the comprehensive discrimination improvement index and the like can be adopted to compare the prediction capacities of the simple image omics prediction model and the new prediction model;

for example, the proteomics prediction model is: y is the imaging group parameter 1+ the imaging group parameter 2, then each patient can calculate y according to the imaging group parameter 1 and the imaging group parameter 2, and y is used as a new independent index of the patient, and the index can be named as an "imaging index";

the imaging index can be parallel to indexes such as alpha-fetoprotein concentration, abnormal prothrombin concentration and the like, and a new prediction model is established together: y 'is the imaging index + alpha-fetoprotein concentration + abnormal prothrombin concentration, and the new model y' is also used to predict the efficacy of the combination therapy.

3) The model is externally validated and evaluated using external independent data, if possible, to assess the generalization of the predictive model.

5. After the prediction model is established, basic information and enhanced magnetic resonance imaging of a patient before the patient receives combined treatment are input into the prediction model, and predicted curative effect evaluation data can be obtained.

Furthermore, a visual prediction model can be established, or a webpage tool based on the prediction model is developed, and the finally established prediction model is visualized in a risk scoring, nomogram and other modes, or the webpage tool based on the prediction model is developed, so that the prediction model can be directly operated and applied in actual clinical work conveniently.

6. With the increase of the number of cases, the subsequent increased cases can be used as new data to supplement, correct and update the existing prediction model so as to continuously improve the accuracy of model prediction. The steps of supplementing, revising and updating the model refer to 1-5.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.