AI-based lung cancer immune phenotype prediction and analysis method

Abstract

The invention relates to the technical field of medical artificial intelligence, and discloses an AI-based lung cancer immune phenotype prediction and analysis method, which comprises the following steps: collecting multi-modal medical data of a plurality of historical lung cancer patients, and determining effective data elements influencing a lung cancer immune phenotype prediction effect on the basis of the multi-modal medical data; dividing the effective data elements into first data and second data, setting a reference threshold value of the first data, collecting a first biological sample and a second biological sample of the target patient at a first time point, and predicting the immune response of the target patient based on a first difference value between the first biological sample and the second biological sample; collecting a third biological sample and a fourth biological sample of the target patient at a second time point after the first time point; and based on the second difference value and the third difference value, predicting the progression-free lifetime and the total lifetime of the target patient through a trained prediction model. The application can accurately predict the response of the lung cancer patient to immunotherapy and the progression-free lifetime and the total lifetime of the lung cancer patient.

Description

Technical Field

[0001] The present application relates to the field of medical artificial intelligence technology, and in particular to an AI-based method for predicting and analyzing the immune phenotype of lung cancer. Background Art

[0002] Lung cancer is one of the leading causes of cancer-related deaths worldwide, and early diagnosis and treatment are crucial for improving patient survival. In recent years, the successful clinical application of immune checkpoint inhibitors has brought new hope for cancer treatment. However, only a subset of patients respond to immune checkpoint inhibitor therapy. Therefore, reliable biospecimen combinations that can predict response to immune checkpoint blockade are essential for estimating whether a patient will respond to immune blockade alone and whether combination therapy or other treatments are warranted.

[0003] Similar prior art includes a Chinese patent application with publication number CN117954101A, which discloses a method and system for building an artificial intelligence-based lung cancer survival prediction model. The method mainly relates to the technical field of prediction model building, and is used to solve the problem that existing lung cancer models are usually based on multiple types of data sets during training, resulting in the model giving relatively broad results when faced with specific data. The method includes: obtaining initial sample data; training a preset deep learning algorithm with the initial sample data to obtain an initial deep learning model; determining the key factors affecting lung cancer survival rate and the basic information of the current patient through a preset feature determination interface; then extracting processed data corresponding to the key factors from the initial training data; extracting final training data from the processed data based on the basic information; importing the final training data into the initial deep learning model for secondary learning to obtain a trained deep learning model, and then obtaining the predicted data corresponding to the current patient through the trained deep learning model.

[0004] A similar prior art includes a Chinese patent application with publication number CN119323694A, which provides an intelligent judgment method and system for lung cancer tissue pathology typing, including: obtaining pathology slice image data; preprocessing the pathology slice image data to extract feature information in the pathology slice image; analyzing the feature information using a trained deep learning model, where the deep learning model is trained based on a large number of labeled lung cancer pathology images; intelligently judging the pathology typing category to which the pathology slice image belongs based on the output results of the deep learning model and the lung cancer pathology typing standards in the expert knowledge base; and outputting the pathology typing category and corresponding confidence information to the user.

[0005] However, neither of the above two technical solutions determines whether lung cancer patients are suitable for treatment with immune checkpoint inhibitors, nor does it involve the progression-free survival and overall survival of lung cancer patients after they receive treatment with immune checkpoint inhibitors. Summary of the Invention

[0006] To solve the above technical problems, the present application provides an AI-based lung cancer immune phenotype prediction and analysis method that can accurately predict the response of lung cancer patients to immunotherapy and the progression-free survival and overall survival of lung cancer patients.

[0007] In a first aspect, the present application provides an AI-based method for predicting and analyzing lung cancer immune phenotypes, the method comprising:

[0008] Step S1: Collect and determine effective data elements that affect the prediction effect of lung cancer immune phenotype based on multimodal medical data of multiple historical lung cancer patients;

[0009] Step S2: dividing the valid data elements into first data and second data, and setting a reference threshold for the first data; collecting a first biological sample and a second biological sample from a target patient at a first time point; and predicting an immune response of the target patient based on a first difference between a measured value of the first biological sample and the reference threshold for the first data;

[0010] Step S3: Based on the immune response of the target patient, the target patient is treated with an immune checkpoint inhibitor, and a third biological sample and a fourth biological sample of the target patient are collected at a second time point after the first time point;

[0011] Step S4: Based on the second difference between the first biological sample and the third biological sample and the third difference between the second biological sample and the fourth biological sample, predict the progression-free survival and overall survival of the target patient using the trained prediction model.

[0012] In conjunction with the first aspect, in a first implementation of the first aspect of the present application, in step S1, determining valid data elements that affect the prediction effect of lung cancer immune phenotype includes:

[0013] Step S11: performing a single element linear regression analysis on each data element in the multimodal medical data, and screening out data elements related to the prediction effect of lung cancer immune phenotype as a first screening result based on a preset first value (significance level);

[0014] Step S12: adding the data elements in the first screening result to a multi-element linear regression model, and again screening out data elements related to the lung cancer immune phenotype prediction effect based on the first value as a second screening result;

[0015] Step S13: using the data elements in the second screening result as data elements that affect the prediction effect of the lung cancer immune phenotype.

[0016] In combination with the first aspect, in a second implementation of the first aspect of the present application, step S11 further includes:

[0017] Perform univariate linear regression analysis on each data element in the multimodal medical data, calculate the first change, first estimated range and first probability value of each data element, and determine the data element whose first probability value is less than the first value as a data element related to the prediction effect of the lung cancer immune phenotype, and use it as the first screening result.

[0018] In combination with the first aspect, in a third implementation of the first aspect of the present application, step S12 further includes:

[0019] The data elements in the first screening result are added to a multi-element linear regression model, and the second change amount, second estimation range and second probability value of the first screening result are calculated. The data elements whose second probability value is less than the first value are determined to be data elements related to the prediction effect of the lung cancer immune phenotype and are used as the second screening result.

[0020] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, the valid data element is divided into first data and second data, including:

[0021] The samples in which the difference between the measurement values of the patients before and after treatment with immune checkpoint inhibitors in the valid data elements is less than or equal to a first threshold are taken as the first data, and the samples in which the difference between the measurement values is greater than the first threshold are taken as the second data.

[0022] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, setting a reference threshold for the first data includes:

[0023] Collecting first data of M lung cancer patients and first data of M healthy people as test data, selecting the value of each first data in the test data, starting from the minimum value, as the second threshold, taking the ratio of the number of first data with a value greater than or equal to the second threshold to M as the first ratio of lung cancer patients, and taking the ratio of the number of first data with a value less than the second threshold to M as the second ratio of healthy people;

[0024] Based on the first ratio and the second ratio, the y value under each second threshold is calculated, and the second threshold that maximizes the y value is selected as the reference threshold of the first data.

[0025] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, the y value at each second threshold is calculated:

[0026] Based on the first ratio and the second ratio, the y value in the formula y=x1+x2-1 is calculated at each second threshold, where y refers to an indicator for evaluating the performance of the classification model for lung cancer patients and healthy people, x1 represents the first ratio, and x2 represents the second ratio.

[0027] In combination with the first aspect, in the seventh implementation of the first aspect of the present application, training the prediction model includes: obtaining biological samples collected from historical lung cancer patients at the first time point and the second time point, respectively, and generating training data based on the difference between the biological samples collected at the first time point and the second time point and the annotated progression-free survival and overall survival of each historical lung cancer patient, creating a prediction model, and training, validating and testing the prediction model based on the training data.

[0028] In combination with the first aspect, in an eighth implementation manner of the first aspect of the present application, the first data and the second data respectively refer to one or more biological samples, and the first data and the second data are different biological samples.

[0029] In combination with the first aspect, in a ninth implementation of the first aspect of the present application, the multimodal medical data includes: clinical data, imaging data, and blood data;

[0030] The clinical data includes the age, gender, smoking history, tumor stage and previous treatment history of the lung cancer patient; the imaging data includes high-resolution CT scans and PET-CT scans, and the morphological characteristics of the tumor, including the size, shape, texture and density of the tumor, are extracted from the imaging data; blood samples are obtained based on venous blood drawing from lung cancer patients, and the blood data is then extracted from the blood samples, and the blood samples contain multiple biomarkers.

[0031] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0032] In the technical solution provided by this application, by collecting and basing on the multimodal medical data of multiple historical lung cancer patients, effective data elements that affect the prediction effect of lung cancer immune phenotype are determined to reduce the dimension of data processing and improve prediction efficiency. The effective data elements are divided into first data and second data, and a reference threshold of the first data is set. At a first time point, a first biological sample and a second biological sample of the target patient are collected, and the immune response of the target patient is predicted based on the first difference between the measured value of the first biological sample and the reference threshold of the first data; so as to identify the patient group that may respond well to immunotherapy as early as possible and reduce the side effects and economic burden of ineffective treatment.

[0033] Based on the immune response of the target patient, the target patient is treated with an immune checkpoint inhibitor, and at a second time point after the first time point, a third biological sample and a fourth biological sample of the target patient are collected to provide a basis for predicting the progression-free survival and overall survival of the target patient. Based on the second difference between the first biological sample and the third biological sample and the third difference between the second biological sample and the fourth biological sample, the progression-free survival and overall survival of the target patient are predicted by a trained prediction model. The present application can accurately predict the response of lung cancer patients to immunotherapy and the progression-free survival and overall survival of lung cancer patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 Schematic diagram of the AI-based lung cancer immune phenotype prediction and analysis method in the embodiments of this application;

[0036] Figure 2 Schematic diagram of the first biological sample and the second biological sample of the target patient before and after immunotherapy in the embodiment of the present application;

[0037] Figure 3 This is a schematic diagram of determining valid data elements that affect the prediction effect of lung cancer immune phenotype in the embodiment of this application;

[0038] Figure 4 A schematic diagram of setting a reference threshold for first data in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The embodiment of the present application provides an AI-based lung cancer immune phenotype prediction and analysis method, which predicts the response of lung cancer patients to immunotherapy by integrating multimodal medical data, including clinical data, imaging data, and blood data. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0040] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 As shown, an embodiment of the AI-based lung cancer immunophenotype prediction and analysis method in the embodiments of the present application includes:

[0041] Step S1: Collect and determine the effective data elements that affect the prediction effect of lung cancer immune phenotype based on the multimodal medical data of multiple historical lung cancer patients. The multimodal medical data includes clinical data, imaging data, and blood data. The specific multimodal medical data will be described in detail below.

[0042] Specifically, hospitals or medical data centers can collect a large amount of multimodal medical data on historical lung cancer patients. However, some data elements in this collected multimodal medical data may have no effect on predicting the immune response of lung cancer patients or on predicting their progression-free survival and overall survival. Therefore, AI algorithms are needed to mine and analyze this multimodal medical data to accurately identify the valid data elements that influence the prediction of lung cancer immune phenotypes. For example, if the patient's age and gender have no effect on the immune response, then when screening for valid data elements that influence the prediction of lung cancer immune phenotypes, the age and gender of lung cancer patients can be removed from the multimodal medical data to reduce the dimensionality of data processing and improve prediction efficiency. The specific method for determining the valid data elements that influence the prediction of lung cancer immune phenotypes will be described in detail below.

[0043] Step S2: Divide the valid data elements into first data and second data, and set a reference threshold for the first data. At a first time point, collect a first biological sample and a second biological sample from the target patient, and predict the immune response of the target patient based on a first difference between the measured value of the first biological sample and the reference threshold for the first data.

[0044] Specifically, assuming that valid data elements include biological samples such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 in the blood, as well as tumor size and morphology from imaging data, the first data can be defined as a reference biological sample of lung cancer patients before receiving immune checkpoint inhibitor treatment. The first data can include one or more biological samples and is used to predict the potential degree of autoimmune system activation in lung cancer patients after treatment with immune checkpoint inhibitors. In other words, it can be used to predict whether lung cancer patients will respond to immune checkpoint inhibitor treatment in advance, thereby identifying patients who are likely to respond well to immunotherapy as early as possible and reducing the side effects and economic burden of ineffective treatment. The second data is used to compare the reference biological samples before and after treatment with immune checkpoint inhibitors. The second data can include one or more biological samples to assess the therapeutic effect after treatment with immune checkpoint inhibitors, providing a basis for predicting progression-free survival and overall survival in lung cancer patients. A reference threshold for the first data is also set to address the lack of a unified standard for assessing the immune response of lung cancer patients before treatment in traditional methods.

[0045] Collect a first biological sample and a second biological sample from the target patient, i.e., two types of biological samples before treatment. The first biological sample is used to predict in advance whether the patient is suitable for treatment with immune checkpoint inhibitors. The first biological sample can be one or more biological samples, including GM-CSF and CHI3L1 from a collected blood sample. The second biological sample is used to compare changes before and after receiving the immune checkpoint inhibitor. The second biological sample can also be one or more biological samples, for example, including CXCL2, VEGF, IFNα2, and MMP2 from a collected blood sample. See Figure 2As shown in the figure, suppose that before treatment, the GM-CSF level in the first biological sample of the target patient is 60pg / ml, which is higher than the reference threshold of 50pg / ml. This indicates that the lung cancer patient's immune system may be activated, indicating a better treatment effect after receiving immune checkpoint inhibitors. The CHI3L1 level is 90ng / ml, which is lower than the reference threshold of 100ng / ml, indicating a decrease in inflammatory chemokines in the tumor microenvironment and a better treatment effect after receiving immune checkpoint inhibitors. Finally, this indicates that the lung cancer patient is suitable for treatment with immune checkpoint inhibitors. Suppose that before treatment, the GM-CSF level in the first biological sample of another lung cancer patient is 40pg / ml, which is lower than the reference threshold of 50pg / ml, and the CHI3L1 level is between 100ng / ml and 125ng / ml, which is higher than the reference threshold of 100ng / ml. This indicates that the prognosis of immune checkpoint inhibitors is not good and the lung cancer patient is not suitable for treatment with immune checkpoint inhibitors.

[0046] Step S3: Based on the immune response of the target patient, the target patient is treated with an immune checkpoint inhibitor, and a third biological sample and a fourth biological sample of the target patient are collected at a second time point after the first time point.

[0047] Specifically, if the target patient has a good immune response, the target patient will start to receive immune checkpoint inhibitors, such as anti-PD-1 or anti-PD-L1 antibodies, and after receiving immune checkpoint inhibitor treatment, that is, at the second time point, usually about 7 weeks after the start of treatment, the third and fourth biological samples will be collected to predict the target patient's progression-free survival and overall survival. The first time point refers to the time point before receiving immune checkpoint inhibitor treatment. Among them, the second time point is set at about 7 weeks after receiving immune checkpoint inhibitor treatment because this provides enough time to observe the effect of immunotherapy on the level of biological samples, while also avoiding the delayed evaluation of treatment effects that may result from too late monitoring.

[0048] Step S4: Based on the second difference between the first biological sample and the third biological sample and the third difference between the second biological sample and the fourth biological sample, predict the progression-free survival and overall survival of the target patient using the trained prediction model.

[0049] Specifically, biological sample data before treatment were collected, such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2, and the changes in the measured values of biological samples before and after treatment were calculated. Figure 2As shown in the figure, decreased levels of GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 serve as input features for the model. The prediction model is trained using historical data, including biospecimen data from a patient population with known progression-free and overall survival (OS). Finally, the difference between the measured values in the target patient's biospecimen is input into the trained prediction model, which outputs the predicted PFS and OS for the target patient. Based on the predicted PFS and OS, physicians can adjust treatment plans to optimize long-term survival and improve patient survival and quality of life.

[0050] Through the coordination of the above steps, the present application can accurately predict the response of lung cancer patients to immunotherapy as well as the progression-free survival and overall survival of lung cancer patients.

[0051] Furthermore, in the above step S1, the effective data elements that affect the prediction effect of lung cancer immune phenotype are determined, see Figure 3 As shown, it includes: step S11: performing single-element linear regression analysis on each data element in the multimodal medical data, and screening out data elements related to the lung cancer immune phenotype prediction effect based on a preset first numerical value as a first screening result; step S12: adding the data elements in the first screening result to the multi-element linear regression model, and screening out data elements related to the lung cancer immune phenotype prediction effect again based on the first numerical value as a second screening result; step S13: using the data elements in the second screening result as data elements that affect the lung cancer immune phenotype prediction effect.

[0052] Specifically, by gradually extracting effective data elements that affect the prediction effect of lung cancer immune phenotype from multimodal medical data, the data processing dimension can be reduced and the prediction efficiency can be improved. First, in step S11, a single-element linear regression analysis is performed on each data element in the collected historical multimodal medical data of lung cancer patients. For example, assuming that the data elements include the patient's gender, age, blood levels of biological samples such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2, as well as tumor size and morphological characteristics in imaging data. Through single-element linear regression analysis, data elements that are significantly correlated with the prediction effect of lung cancer immune phenotype can be screened out, such as patient age, GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2. In this case, patient age, GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 will be used as the first screening results. Next, in step S12, data elements from the first screening results, such as patient age, GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2, are added to a multivariate linear regression model to further assess their correlation in a multivariate setting. For example, the multivariate linear regression model may show that GM-CSF levels are positively correlated with treatment efficacy, while CXCL2 levels are negatively correlated with treatment efficacy, and patient age has little correlation with treatment efficacy. In this way, data elements that are significantly correlated with predicted efficacy, such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2, can be further screened as the second screening results. Finally, in step S13, data elements from the second screening results, such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2, are identified as valid data elements that influence the prediction of lung cancer immunophenotypes. This approach improves model training efficiency by reducing interference from invalid data.

[0053] Furthermore, the above-mentioned step S11 also includes: performing univariate linear regression analysis on each data element in the multimodal medical data, calculating the first change amount, the first estimated range and the first probability value of each data element, and determining the data element whose first probability value is less than the first numerical value as a data element related to the prediction effect of lung cancer immune phenotype, and using it as the first screening result.

[0054] Specifically, the multimodal patient medical data collected includes blood biopsies such as GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2; imaging data such as tumor size and morphological characteristics; and clinical data such as age and medical history. For each data element, a linear relationship is calculated between it and the immunotherapy effect, such as progression-free survival or overall survival. The first change value represents the effect of a change in any data element on the predicted effect. For example, with the other data elements remaining unchanged, the average change in progression-free survival or overall survival is calculated for each unit change in one data element. For example, the expected change in progression-free survival is calculated for each 1 pg / mL increase in MMP2. The first estimate range is typically expressed as a 95% confidence interval. This means that the same MMP2 sampling is repeated multiple times, and the confidence interval is calculated each time to estimate a range for progression-free survival or overall survival, explaining the reliability of the estimate. The first probability value is used to determine whether the correlation between MMP2 and progression-free survival or overall survival is statistically regular. The significance level is typically set at 0.05. If the first value is less than 0.05, MMP2 is considered to be significantly associated with progression-free survival or overall survival. Suppose a univariate linear regression analysis of MMP2 yields the following results: first change: 0.45 ng / ml, first estimated range: 0.35-0.55 ng / ml, first probability value: 0.003. Because the first probability value of 0.003 is less than the significance level of 0.05, MMP2 is determined to be a data element associated with the predictive effect of lung cancer immunophenotypes. Step S11 allows for the preliminary screening of data elements that influence predictive effects from a large amount of multimodal medical data, reducing the complexity of subsequent analysis and improving model training efficiency.

[0055] Furthermore, the above-mentioned step S12 also includes: adding the data elements in the first screening result to the multi-element linear regression model, calculating the second change amount, the second estimation range and the second probability value of the first screening result, and determining the data elements whose second probability value is less than the first value as data elements related to the prediction effect of lung cancer immune phenotype, and using them as the second screening results.

[0056] Specifically, the data elements in the first screening result obtained in step S11 are further added to a multi-element linear regression model for analysis, evaluating the impact of multiple variables on the prediction effect when they coexist, thereby screening out data elements that are still significantly correlated under a multivariate environment. For example, assuming that the patient's age, GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 data elements are analyzed in a multi-element linear regression model, the second probability values of GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 are all less than 0.05, while the second probability value of patient age is 0.15, which is greater than 0.05. Therefore, GM-CSF, CHI3L1, CXCL2, VEGF, IFNα2, and MMP2 are determined to be valid data elements related to the prediction effect of lung cancer immunophenotype and are used as the second screening result. Through the above method, data elements that have a significant impact on the prediction effect under a multivariate environment can be further verified and screened.

[0057] Furthermore, the valid data elements are divided into first data and second data, including: taking samples in which the difference between the measurement values of patients before and after treatment with immune checkpoint inhibitors in the valid data elements is less than or equal to a first threshold as the first data, and taking samples in which the difference between the measurement values is greater than the first threshold as the second data.

[0058] Specifically, the difference in the change between the measured values of the biological samples before and after the patient receives treatment with the immune checkpoint inhibitor is compared with a preset first threshold. The first threshold is a reference value based on historical medical data or clinical trial statistics, which is used to determine which biological samples are suitable for predicting the immune response of the lung cancer patient based on the patient's baseline condition. For example, for patient A, the GM-CSF level before treatment is 60pg / ml and after treatment is 63pg / ml, with a difference of 3pg / ml, which is less than the first threshold of 5pg / ml, so the biological sample is classified as the first data. For patient A, the MMP2 level before treatment is 100ng / ml and after treatment is 120ng / ml, with a difference of 20ng / ml, which is greater than the first threshold of 5ng / ml, so the sample is classified as the second data. Through this method, step S2 can effectively distinguish samples that reflect the reference state and dynamic changes in treatment, providing a reference basis for subsequent treatment decisions.

[0059] Furthermore, a reference threshold value of the first data is set, see Figure 4As shown, the method includes: collecting first data of M lung cancer patients and first data of M healthy persons as test data, where M represents a positive integer greater than or equal to 50 and less than or equal to 10,000; in the test data, starting from the minimum value, selecting the value of each first data in sequence as the second threshold; taking the ratio of the number of first data with a value greater than or equal to the second threshold to M as the first ratio of lung cancer patients; and taking the ratio of the number of first data with a value less than the second threshold to M as the second ratio of healthy persons.

[0060] Based on the first ratio and the second ratio, a y value at each second threshold is calculated, and the second threshold that maximizes the y value is selected as a reference threshold for the first data. Calculating the y value at each second threshold includes: based on the first ratio and the second ratio, calculating the y value in the formula y=x1+x2-1 at each second threshold, where y is an indicator for evaluating the performance of the classification model for lung cancer patients and healthy subjects, x1 represents the first ratio, and x2 represents the second ratio.

[0061] Specifically, to ensure sufficient statistical samples for analysis while avoiding increased computational complexity due to excessive data volume, the first data of 200 lung cancer patients and the first data of 200 healthy individuals can be collected as test data. In the test data, starting with the minimum value, each first data value is sequentially selected as a temporary second threshold. For each selected second threshold, two ratios are calculated: the first ratio is the proportion of data in the first data of lung cancer patients that is greater than or equal to the second threshold. This ratio reflects the proportion of samples in the lung cancer patient population that is less than or equal to this threshold. The second ratio is the proportion of data in the first data of healthy individuals that is less than or equal to the second threshold. This ratio reflects the proportion of samples in the healthy population that is less than or equal to this threshold. Based on these two ratios, the y value at each second threshold is calculated. The y value is a comprehensive indicator used to measure the ability of the threshold to distinguish between patients and healthy individuals. The calculation formula for the y value involves the difference between the first and second ratios to maximize the discriminatory power. Finally, all possible second thresholds are traversed to find the second threshold that maximizes the y value and determine it as the reference threshold for the first data. By comparing with the reference threshold, the first data can provide information after immunotherapy before lung cancer patients receive immune checkpoint inhibitor treatment, quickly determine the patient's reference status, and thus provide a basis for subsequent diagnosis and treatment decisions.

[0062] Furthermore, training the prediction model includes: obtaining biological samples collected from historical lung cancer patients at a first time point and a second time point, and generating training data based on the difference between the biological samples collected at the first time point and the second time point and the annotated progression-free survival and overall survival of each historical lung cancer patient, creating a prediction model, and training, validating and testing the prediction model based on the training data.

[0063] Specifically, by integrating data from multiple time points before and after treatment, this solution can accurately predict the target patient's progression-free survival and overall survival, providing doctors with more reliable treatment decision support. Furthermore, this application can reduce the waste of medical resources and the financial burden on patients by reducing trial-and-error costs and unnecessary treatments.

[0064] Furthermore, the first data and the second data refer to one or more biological samples respectively, and the first data and the second data are different biological samples, and the biological samples refer to biomarkers.

[0065] Furthermore, the multimodal medical data also includes: clinical data including the age, gender, smoking history, tumor stage and previous treatment history of lung cancer patients; imaging data including high-resolution CT scans and PET-CT scans, and morphological characteristics of tumors are extracted from the imaging data, and morphological characteristics include tumor size, shape, texture and density; blood samples are obtained based on venous blood drawing from lung cancer patients, and blood data is extracted from the blood samples, and the blood samples contain multiple biological samples.

[0066] Specifically, clinical data covers lung cancer patients' medical history, symptoms, diagnosis, and previous treatment records, including age, gender, smoking history, family history, pathology (e.g., adenocarcinoma, squamous cell carcinoma), and whether they have received chemotherapy or radiotherapy. Imaging data includes lung images acquired using technologies like CT (computed tomography) and MRI (magnetic resonance imaging). These images can reveal the tumor's location, size, morphology, and relationship to surrounding tissues, providing a direct picture of the lesion. Blood data primarily involves the detection of various biomarkers from patient blood samples, such as CXCL2, VEGF, CHI3L1, and GM-CSF. Levels of these biomarkers are closely correlated with the lung cancer immune microenvironment and treatment response. For example, GM-CSF levels can reflect the degree of immune system activation, while VEGF levels are associated with tumor angiogenesis and metastatic potential. By integrating multimodal medical data from lung cancer patients, the problem of traditional lung cancer treatment, where a single data type fails to fully reflect the patient's condition and treatment efficacy, can be overcome.

[0067] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0068] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0069] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting and analyzing lung cancer immune phenotypes based on AI, characterized in that: The method comprises: Step S1: Collect and determine effective data elements that affect the prediction effect of lung cancer immune phenotype based on multimodal medical data of multiple historical lung cancer patients; Step S2: dividing the valid data elements into first data and second data, and setting a reference threshold for the first data; collecting a first biological sample and a second biological sample from a target patient at a first time point; and predicting an immune response of the target patient based on a first difference between a measured value of the first biological sample and the reference threshold for the first data; Step S3: Based on the immune response of the target patient, the target patient is treated with an immune checkpoint inhibitor, and a third biological sample and a fourth biological sample of the target patient are collected at a second time point after the first time point; Step S4: Based on the second difference between the first biological sample and the third biological sample and the third difference between the second biological sample and the fourth biological sample, predict the progression-free survival and overall survival of the target patient using the trained prediction model.

2. The method according to claim 1, characterized in that In step S1, determining valid data elements that affect the prediction effect of lung cancer immune phenotype includes: Step S11: performing single element linear regression analysis on each data element in the multimodal medical data, and screening out data elements related to the prediction effect of lung cancer immune phenotype based on a preset first value as a first screening result; Step S12: adding the data elements in the first screening result to a multi-element linear regression model, and again screening out data elements related to the lung cancer immune phenotype prediction effect based on the first value as a second screening result; Step S13: using the data elements in the second screening result as data elements that affect the prediction effect of the lung cancer immune phenotype.

3. The method according to claim 2, characterized in that The step S11 further includes: Perform univariate linear regression analysis on each data element in the multimodal medical data, calculate the first change, first estimated range and first probability value of each data element, and determine the data element whose first probability value is less than the first value as a data element related to the prediction effect of the lung cancer immune phenotype, and use it as the first screening result.

4. The method according to claim 2, characterized in that The step S12 further includes: The data elements in the first screening result are added to a multi-element linear regression model, and the second change amount, second estimation range and second probability value of the first screening result are calculated. The data elements whose second probability value is less than the first value are determined to be data elements related to the prediction effect of the lung cancer immune phenotype and are used as the second screening result.

5. The method according to claim 1, wherein Dividing the valid data element into first data and second data includes: The samples in which the difference between the measurement values of the patients before and after treatment with immune checkpoint inhibitors in the valid data elements is less than or equal to a first threshold are taken as the first data, and the samples in which the difference between the measurement values is greater than the first threshold are taken as the second data.

6. The method according to claim 1, wherein Setting a reference threshold of the first data includes: Collecting first data of M lung cancer patients and first data of M healthy people as test data, selecting the value of each first data in the test data, starting from the minimum value, as the second threshold, taking the ratio of the number of first data with a value greater than or equal to the second threshold to M as the first ratio of lung cancer patients, and taking the ratio of the number of first data with a value less than the second threshold to M as the second ratio of healthy people; Based on the first ratio and the second ratio, the y value under each second threshold is calculated, and the second threshold that maximizes the y value is selected as the reference threshold of the first data.

7. The method according to claim 6, characterized in that Calculate the y-value at each of the second thresholds: Based on the first ratio and the second ratio, the y value in the formula y=x1+x2-1 is calculated at each second threshold, where y refers to an indicator for evaluating the performance of the classification model for lung cancer patients and healthy people, x1 represents the first ratio, and x2 represents the second ratio.

8. The method according to claim 1, characterized in that Training a predictive model, including: Biological samples collected from historical lung cancer patients at the first time point and the second time point are obtained, and training data is generated based on the difference between the biological samples collected at the first time point and the second time point and the annotated progression-free survival and overall survival of each historical lung cancer patient, a prediction model is created, and the prediction model is trained, verified, and tested based on the training data.

9. The method according to claim 1, characterized in that The first data and the second data respectively refer to one or more biological samples, and the first data and the second data are different biological samples.

10. The method according to claim 1, characterized in that The multimodal medical data includes: clinical data, imaging data, and blood data; The clinical data includes the age, gender, smoking history, tumor stage and previous treatment history of the lung cancer patient; the imaging data includes high-resolution CT scans and PET-CT scans, and the morphological characteristics of the tumor, including the size, shape, texture and density of the tumor, are extracted from the imaging data; blood samples are obtained based on venous blood drawing from lung cancer patients, and the blood data is then extracted from the blood samples, and the blood samples include multiple biological samples.

Images