Method for predicting lung cancer EGFR genotype and immune molecule expression level

Abstract

The invention relates to the technical field of gene and immune molecule detection, in particular to a method for predicting lung cancer EGFR genotype and immune molecule expression level, and the method specifically comprises the following steps: S1, obtaining a CT image: obtaining the CT image of a non-small cell lung cancer patient, the CT image comprising a tumor area and a peritumor area; s2, image omics characteristics are extracted, wherein the image omics characteristics are extracted from the tumor area and the peritumor area respectively; s3, extracting deep network features: extracting the deep network features from the CT image by using a deep neural network based on an attention mechanism; and S4, constructing a prediction model: inputting the radiomics characteristics and the deep network characteristics into the prediction model, and outputting prediction results of EGFR genotypes and immune molecule expression levels. The method for predicting the lung cancer EGFR genotype and immune molecule expression level has the advantages of non-invasiveness, high efficiency, accurate prediction, clinical practicability, model interpretability and technical compatibility and flexibility.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of gene and immune molecule detection technology, and particularly relates to a method for predicting lung cancer EGFR genotype and immune molecule expression level by combining tumor and peripheral tissue information. BACKGROUND

[0002] Lung cancer is one of the highest incidence and mortality rates of malignant tumors worldwide, and non-small cell lung cancer (NSCLC) accounts for the vast majority of lung cancer cases. EGFR gene mutation and immune molecule (such as PD-1 / PD-L1) expression level are key factors in determining the efficacy of targeted therapy and immunotherapy for patients. At present, tissue biopsy and gene sequencing technology are mainly used to detect EGFR mutation status and immune molecule expression level in clinical practice, but these methods have the problems of strong invasiveness, long time consumption, high cost, and limited sample acquisition. In addition, traditional imaging diagnosis (such as CT) can only provide morphological information and cannot directly reflect the molecular characteristics of the tumor, resulting in limitations in early precise treatment decision-making. In recent years, imageomics and deep learning technology have shown great potential in medical image analysis, but existing technologies focus on tumor internal features and ignore important biological information of the peritumoral microenvironment. At the same time, existing prediction models often lack deep fusion of multi-modal features and have insufficient model interpretability, making it difficult to meet the clinical demand for precision and transparency. Therefore, there is an urgent need for a non-invasive, efficient and interpretable method to integrate imageomics features, deep learning features and clinical data of the tumor and peritumoral region to achieve precise prediction of EGFR genotype and immune molecule expression level. SUMMARY

[0003] The main purpose of the present application is to overcome the shortcomings of the prior art and provide a method for predicting lung cancer EGFR genotype and immune molecule expression level by combining tumor and peripheral tissue information. This method innovatively combines multi-dimensional data and interpretable analysis to provide reliable technical support for precise diagnosis and treatment of lung cancer, and has significant clinical application value and social benefits.

[0004] The technical scheme adopted by the present application to achieve the technical purpose is: a method for predicting lung cancer EGFR genotype and immune molecule expression level, specifically as follows: S1, obtaining a CT image: obtaining a CT image of a non-small cell lung cancer (NSCLC) patient, the CT image comprising a tumor region and a peritumoral region; providing original data input, the tumor region being used to detect internal features of the tumor, and the peritumoral region being used to capture microenvironment information.

[0005] S2, extracting imaging features: extracting imaging features from the tumor area and the peritumoral area respectively; quantifying the morphological, textural and other imaging phenotypes of the tumor and its surrounding tissues to provide structured features for subsequent models.

[0006] S3, extracting deep network features: using a deep neural network based on an attention mechanism to extract deep network features from the CT image; automatically learning high-dimensional abstract features of the tumor and peritumoral area, and the attention mechanism (CBAM) enhances the weight of key areas.

[0007] S4, constructing a prediction model: inputting the imaging features and deep network features into the prediction model to output the prediction results of EGFR genotype and immunomolecular expression level; fusing multi-dimensional features (imaging + deep learning) to achieve accurate prediction through classification algorithms.

[0008] Preferably, the peritumoral area is a range of 3mm outside the tumor boundary, ensuring the standardization of peritumoral feature extraction and covering the key areas of the tumor microenvironment.

[0009] Preferably, the imaging features include first-order statistical features, shape features, gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, neighborhood gray tone difference matrix (NGTDM) features and gray level dependence matrix (GLDM) features, which clearly define the mathematical method of feature extraction and cover tumor heterogeneity and spatial distribution information.

[0010] Preferably, the deep neural network based on the attention mechanism is a DenseNet121 network, and a convolution block attention module (CBAM) is added before its global average pooling layer. DenseNet 121 enhances feature reuse through dense connections, and the CBAM module focuses on key areas of the tumor and peritumoral area, improving feature discrimination.

[0011] Preferably, the prediction model also combines clinical features, including age, gender, smoking status, tumor location, lymphatic vessel invasion (LVI), pleural invasion (PI) and T stage. Clinical features cooperate with image features to improve the biological and clinical relevance of the model.

[0012] Preferably, the prediction model uses one of the classification algorithms of logistic regression, linear discriminant analysis, random forest, k-nearest neighbor, naive Bayes, support vector machine or multilayer perceptron to construct, providing multiple classifier options to adapt to different data distributions and prediction needs.

[0013] Preferably, the EGFR genotype includes wild type, mutant type, 19Del mutation, L858R mutation and T790M mutation, covering key mutation types related to targeted therapy.

[0014] Preferably, the immune molecule expression level is the expression state of PD-1 or PD-L1, which is directly related to the prediction of immunotherapy response.

[0015] Preferably, the prediction result is evaluated by the area under the receiver operating characteristic curve (ROC), accuracy, sensitivity, specificity, F1 score and precision to quantify the performance of the model and ensure clinical practicability.

[0016] Preferably, the method further comprises an explainability analysis of the prediction model, specifically comprising: Using the GradCAM algorithm to generate a heat map to show the contribution of different tissue regions to the prediction result, visualizing the deep learning feature attention area to assist doctors in understanding the model decision basis.

[0017] Using the SHAP method to analyze the importance of each feature to the model output, quantifying the contribution of radiomics, deep learning and clinical features to enhance the transparency of the model.

[0018] The working principle of the method for predicting the EGFR genotype and immune molecule expression level of lung cancer is as follows: Based on the multi-modal medical image analysis technology of artificial intelligence, the image features of the tumor and peritumoral region of the CT image are integrated, the deep learning features are combined with the clinical data to construct a prediction model. Specifically, the system first extracts quantitative radiomics features (such as texture, shape, etc.) and deep neural network features based on attention mechanism (DenseNet121+CBAM module) from the CT image, then fuses these features with patient clinical information, and establishes a prediction model through machine learning algorithm (such as LDA, SVM, etc.), finally outputs the prediction results of EGFR gene mutation status (such as 19Del, L858R, T790M, etc.) and PD-1 / PD-L1 immune molecule expression level, providing decision support for targeted therapy and immunotherapy of lung cancer patients.

[0019] Compared with the prior art, the present application has the following advantages: The method for predicting the EGFR genotype and immune molecule expression level of lung cancer is non-invasive and efficient, which avoids the trauma and time-consuming problem of traditional biopsy by using CT image combined with artificial intelligence technology, and provides a non-invasive and rapid molecular feature prediction method for patients.

[0020] The method for predicting lung cancer EGFR genotype and immune molecule expression level can perform multi-modal feature fusion, integrate imageomics features of tumor regions and peritumoral regions (expanded by 3 mm), deep learning features based on attention mechanism (DenseNet121+CBAM module), and clinical features, and significantly improve the prediction accuracy and biological relevance.

[0021] The method for predicting lung cancer EGFR genotype and immune molecule expression level has precise prediction and clinical practicability, covers EGFR genotype (such as 19Del, L858R, T790M) and immune molecule (PD-1 / PD-L1) expression states, is directly related to the efficacy prediction of targeted therapy and immunotherapy, and provides decision support for individualized treatment.

[0022] The method for predicting lung cancer EGFR genotype and immune molecule expression level has model interpretability, and through GradCAM heat map and SHAP value analysis, the key feature contribution area can be visualized and the importance of each feature can be quantified, thereby enhancing the transparency and clinical acceptability of the model.

[0023] The method for predicting lung cancer EGFR genotype and immune molecule expression level has technical compatibility and flexibility, supports various classification algorithms (such as logistic regression, random forest, etc.) and standardized evaluation indexes (such as AUC, sensitivity, etc.), and is suitable for different clinical scene requirements. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The structure flow chart of the method for predicting lung cancer EGFR genotype and immune molecule expression level.

[0025] Figure 2 The comparative result curve diagram of all models for predicting EGFR, 19Del and L858R mutations, wherein Figure 2 (a-b) are the ROC curve and DCA in the test set. Figure 2 (c) is the AUC distribution in the training set.

[0026] Figure 3 The comparative result curve diagram of all models for predicting drug resistance T790M mutation and immune molecule expression. Wherein Figure 3 (a-b) are the ROC curve and DCA in the test set. Figure 3 (c) is the AUC distribution in the training set.

[0027] Wherein: DETAILED DESCRIPTION

[0028] In order to make the objects, technical solutions and advantages of the present application clearer, the following further describes the present application with reference to the accompanying drawings and examples. However, it should be understood that the specific examples described herein are only used to explain the present application and are not used to limit the scope of the present application. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present application.

[0029] In the description of the present application, it should be noted that when an element is referred to as "fixed to" or "disposed on" another element, it can be directly on the other element or indirectly on the other element. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element.

[0030] In the description of the present application, it should be noted that the terms "center", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third", and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, "a plurality of" means two or more, unless otherwise explicitly specified and limited. The meaning of "several" is one or more, unless otherwise explicitly specified and limited.

[0031] In the description of the present application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "disposed", "mounted", "connected", "linked" should be broadly understood, for example, can be fixedly connected, can be detachably connected, or integrally connected; can be mechanically connected, can be electrically connected; can be directly connected, can be indirectly connected through an intermediate medium, can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances. Embodiment

[0032] Please refer to Figure 1 A method for predicting lung cancer EGFR genotype and immune molecule expression level, specifically as follows: S1, acquiring CT images: acquiring CT images of non-small cell lung cancer (NSCLC) patients, the CT images containing tumor regions and peritumoral regions; the CT images need to be preprocessed by standardization, including adjusting the image resolution to 160x160 pixels, and covering the maximum tumor slice region.

[0033] The peritumoral region is a range of 3mm outside the tumor boundary; the peritumoral region is automatically generated by an internal algorithm, which is used to capture the edge features of the tumor microenvironment and analyze them jointly with the tumor region features.

[0034] S2, extracting radiomics features: extracting radiomics features from the tumor region and peritumoral region respectively; the radiomics features are extracted by the PyRadiomics toolkit, and are subjected to triple screening of univariate analysis, mRMR algorithm and LASSO regression.

[0035] The radiomics features include first-order statistical features, shape features, gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, gray level size zone matrix (GLSZM) features, neighborhood gray tone difference matrix (NGTDM) features and gray level dependence matrix (GLDM) features; the features further include a derived feature group after wavelet transform, each group of features being processed by an experimentally determined 25 gray level bin width and a sitkBSpline interpolator.

[0036] S3, extracting deep network features: using a deep neural network based on an attention mechanism to extract deep network features from the CT images; the deep network features are input into the network after cropping the tumor region bounding box and enhancing the data, and are finally output by a fully connected layer.

[0037] The deep neural network based on the attention mechanism is a DenseNet121 network, and a convolution block attention module (CBAM) is added before the global average pooling layer of the DenseNet121 network; the CBAM module is composed of a channel attention submodule and a spatial attention submodule, which is used to enhance the focusing ability of the network on the tumor and peritumoral region; the DenseNet121 network is optimized by Xavier initialization weight, and the original fully connected layer is replaced by a 2-layer customized fully connected layer.

[0038] S4, constructing a prediction model: inputting the radiomics features and deep network features into a prediction model to output the prediction results of EGFR genotype and immune molecular expression level; the prediction model is optimized by grid search and 5-fold cross-validation by fusing radiomics scores (RadScore), deep network features and clinical features.

[0039] The prediction model also incorporates clinical features, including age, sex, smoking status, tumor location, lymphovascular invasion (LVI), pleural invasion (PI), and T stage; these clinical features are screened using the Akaike Information Criterion (AIC) and cascade-fused with RadScore and deep network features to form the IFAI-C comprehensive model.

[0040] Furthermore, in this embodiment, the prediction model is constructed using a classification algorithm selected from logistic regression, linear discriminant analysis, random forest, k-nearest neighbor, naive Bayes, support vector machine, or multilayer perceptron; the classification algorithm is robustly validated through 30 5-fold cross-validations, and the algorithm with the best AUC performance is ultimately selected for independent test set prediction.

[0041] Furthermore, in this embodiment, the EGFR genotype includes wild type, mutant type, 19Del mutation, L858R mutation and T790M mutation; the genotype is determined by gene sequencing of biopsy or surgical specimens, among which 19Del and L858R mutations are used for EGFR-TKIs treatment response stratification, and T790M mutation is used for drug resistance assessment.

[0042] Furthermore, in this embodiment, the immune molecule expression level is the expression status of PD-1 or PD-L1; the expression status is detected by immunohistochemistry, with a tumor proportion score (TPS) ≥ 1% as the positive threshold, and is aligned with the tumor infiltration level estimated by the CD8B gene.

[0043] Furthermore, in this embodiment, the prediction results are evaluated by the area under the receiver operating characteristic curve (ROC), accuracy, sensitivity, specificity, F1 score and precision; the evaluation also includes calibration curve analysis and decision curve analysis (DCA), and the 95% confidence interval is calculated by the bootstrap resampling method.

[0044] S6. The method further includes analyzing the interpretability of the prediction model, specifically including: The GradCAM algorithm is used to generate a heat map showing the contribution of different tissue regions to the prediction results; the heat map visualizes the attention distribution of deep network features in the tumor and peritumoral regions.

[0045] The SHAP method was used to analyze the importance of each feature to the model output; the SHAP value was used to quantify the incremental contribution and positive and negative correlation of clinical features, RadScore and deep network features to EGFR genotype and immune molecule expression.

[0046] The specific usage process of the method for predicting lung cancer EGFR genotype and immune molecule expression levels: First, CT images of non-small cell lung cancer (NSCLC) patients are acquired, and after standardized preprocessing, the tumor region is manually or semi-automatically segmented and expanded 3mm to generate the peritumoral region; Second, quantitative features (such as first-order statistics, texture features, etc.) are extracted from the tumor and peritumoral regions by imageomics tools, and deep features are extracted by attention mechanism-based DenseNet121 network. The two types of features are screened and fused in multiple rounds respectively; Subsequently, the screened imageomics score (RadScore), deep network features and clinical features (such as age, smoking status, etc.) are input into the cross-validated and optimized classification model (such as linear discriminant analysis), and the prediction results of EGFR genotype (such as 19Del, T790M) and PD-1 / PD-L1 expression state are output; Finally, the model performance is evaluated and the feature contribution is explained by ROC curve, SHAP value and heat map analysis, which assists the clinical decision of targeted therapy and immunotherapy.

[0047] It should be noted that although the above embodiments have been described in this paper, the patent protection scope of the present application is not limited thereby. Therefore, based on the innovative idea of the present application, the changes and modifications of the embodiments described in this paper, or the equivalent structures, equivalent processes or equivalent function transformations made by using the contents of the present application specification and drawings, directly or indirectly apply the above technical solutions to other related technical fields, are all included in the protection scope of the present application patent.

Claims

1. A method for predicting EGFR genotype and immune molecule expression level in lung cancer, characterized in that: The specific steps are as follows: S1. Obtaining a CT image: Obtaining a CT image of a patient with non-small cell lung cancer, wherein the CT image includes a tumor area and a peritumoral area; S2. Extract radiomics features: Extracting radiomic features from the tumor region and the peritumoral region respectively; S3. Extracting deep network features: extracting deep network features from the CT image using a deep neural network based on an attention mechanism; S4. Construct a prediction model: input the imaging genomics features and deep network features into the prediction model, and output the prediction results of EGFR genotype and immune molecule expression level.

2. The method for predicting lung cancer EGFR genotype and immune molecule expression level according to claim 1, characterized in that: The peritumoral area is the range extending 3 mm outside the tumor boundary.

3. The method for predicting EGFR genotype and immune molecule expression level in lung cancer according to claim 1, characterized in that: The imaging omics features include first-order statistical features, shape features, grayscale co-occurrence matrix features, grayscale run length matrix features, grayscale size area matrix features, neighborhood grayscale difference matrix features and grayscale dependence matrix features.

4. The method for predicting EGFR genotype and immune molecule expression level in lung cancer according to claim 1, characterized in that: The deep neural network based on the attention mechanism is a DenseNet121 network, and a convolution block attention module is added before its global average pooling layer.

5. The method for predicting lung cancer EGFR genotype and immune molecule expression level according to claim 1, characterized in that: The prediction model also incorporates clinical characteristics including age, sex, smoking status, tumor location, lymphovascular invasion, pleural invasion, and T stage.

6. The method for predicting EGFR genotype and immune molecule expression level in lung cancer according to claim 1, characterized in that: The prediction model is constructed using a classification algorithm selected from the group consisting of logistic regression, linear discriminant analysis, random forest, k-nearest neighbor, naive Bayes, support vector machine, and multilayer perceptron.

7. The method for predicting lung cancer EGFR genotype and immune molecule expression level according to claim 1, characterized in that: The EGFR genotypes include wild type, mutant type, 19Del mutation, L858R mutation and T790M mutation.

8. The method for predicting lung cancer EGFR genotype and immune molecule expression level according to claim 1, characterized in that: The immune molecule expression level is the expression status of PD-1 or PD-L1.

9. The method for predicting EGFR genotype and immune molecule expression level in lung cancer according to claim 1, characterized in that: The prediction results were evaluated by area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, F1 score and precision.

10. The method for predicting lung cancer EGFR genotype and immune molecule expression level according to claim 1, characterized in that: The method also includes interpretability analysis of the prediction model, specifically including: The GradCAM algorithm was used to generate a heat map showing the contribution of different tissue regions to the prediction results. The SHAP method is used to analyze the importance of each feature to the model output.

Images