Establishment method of pulmonary nodule malignant risk prediction model integrating iconography indexes and expired gas VOCs (Volatile Organic Compounds) data

Abstract

The invention discloses a method for establishing a pulmonary nodule malignant risk prediction model by integrating iconography indexes and expired gas VOCs (Volatile Organic Compounds) data, and belongs to the field of precancerous lesions of lung cancer and early prediction of lung cancer. In order to improve the accuracy of precancerous lesions of lung cancer and early prediction of lung cancer, the method comprises the following steps: acquiring a CT image of a patient with pulmonary nodules detected in lung CT, then processing the CT image of the patient with pulmonary nodules, and extracting image data of iconography indexes; collecting exhaled air VOCs of the pulmonary nodule patient, performing exhaled air VOCs index calibration and abundance analysis to obtain exhaled air data, and then performing multi-omics data arrangement in combination with image data to obtain a multi-omics data set; formulating a lung nodule malignant risk degree standard; and constructing a multi-omics deep learning model, and training the multi-omics deep learning model by using the data set of the pulmonary nodule malignant risk prediction model to obtain a pulmonary nodule malignant risk prediction model integrating iconography indexes and expired gas VOCs data.

Description

Technical Field

[0001] The present invention belongs to the field of lung cancer precursor lesions and early prediction of lung cancer, and specifically relates to a method for establishing a malignant risk prediction model for small lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data. Background Art

[0002] According to the statistics of WHO in 2023, lung cancer is the second most common malignant tumor in terms of new cases and the most deaths worldwide. The number of new cases and deaths in my country are both the highest among malignant tumors. Pulmonary nodules are closely related to lung cancer. According to the pathological properties of pulmonary nodules, they are mainly divided into benign nodules, precancerous nodules, and early lung cancer nodules. At present, there are many types of auxiliary diagnosis and risk prediction models for malignant pulmonary nodules. Basically, they are developed based on the dynamic follow-up artificial intelligence model system of deep learning. The main monitoring conditions are the sphericity, average CT value, and volume doubling time diameter change of pulmonary nodules under CT image characteristics. However, the accuracy, sensitivity, and specificity of the model's detection and follow-up results are still reported differently. The evaluation methods and results of different models vary greatly. At present, there is no standard overall follow-up evaluation system. Research and evaluation of precancerous nodules are rarely reported. Exhaled breath is a process of metabolism of human cells that produces some volatile organic compounds (VOCs). Currently, there have been studies on the early diagnosis of malignant tumors: due to the non-invasiveness of VOCs sampling, the high sensitivity and effectiveness of the detection results, it has gradually shown the application prospects of VOCs. In the case of malignant tumors, due to the unlimited proliferation and active metabolism of tumor cells, they will produce higher concentrations and release more persistent specific VOCs than normal tissues. Its diagnostic value is particularly prominent. In terms of lung tumors, since the VOCs produced by lung lesions are not directly detected by exhalation through blood exchange, the detection of exhaled breath VOCs is more suitable for the diagnosis and early warning of lung diseases. And in terms of assisting the diagnosis of lung cancer, since the metabolites after the formation of the initial cancer cells can more directly indicate the presence of tumor cells than ctDNA (circulating tumor DNA in peripheral blood) and CAC (abnormal chromosome fragments in peripheral blood), it has greater research and application value in the early diagnosis of lung cancer. However, it has not yet been applied to the dynamic monitoring of precancerous lesions and malignant risks of small lung nodules.

[0003] In summary, the existing technologies have the following defects in predicting the malignant risk of small pulmonary nodules:

[0004] Disadvantage 1: The existing model mainly relies on the imaging features of a single modality, which leads to a bottleneck in prediction accuracy and makes it difficult to apply the model in clinical practice.

[0005] Disadvantage 2: Different models use different imaging features, and the evaluation results of different models are difficult to compare and standardize.

[0006] Disadvantage 3: As a potential biomarker, exhaled breath VOCs have not yet been effectively integrated into the dynamic monitoring of the malignant risk of small lung nodules, and their diagnostic value has not been fully utilized.

[0007] Disadvantage 4: There is a lack of predictive models that effectively combine VOCs data with imaging indicators, and the potential value of these data is not fully utilized.

[0008] Disadvantage 5: The prediction results of the malignancy risk of small lung nodules are completely dependent on the model and fail to effectively combine clinical experience, which not only wastes knowledge resources but also increases the uncertainty of the prediction results. Summary of the invention

[0009] The problem to be solved by the present invention is to improve the accuracy of early prediction of lung cancer prelesions and lung cancer, and to propose a method for establishing a malignant risk prediction model for small lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data.

[0010] To achieve the above object, the present invention is implemented through the following technical solutions:

[0011] A method for establishing a risk prediction model for malignant lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data includes the following steps:

[0012] S1. collecting CT images of patients with small lung nodules detected by lung CT, and then processing the CT images of patients with small lung nodules, extracting image data of imaging indicators, and obtaining an image data set for standby use;

[0013] S2. Collect exhaled VOCs from patients with small pulmonary nodules, perform calibration and abundance analysis of exhaled VOCs indicators, obtain exhaled breath data, and then combine with imaging data to perform multi-omics data collation to obtain a multi-omics data set;

[0014] S3. Formulate a risk standard for malignancy of small pulmonary nodules, extract clinical decision data based on the risk standard for malignancy of small pulmonary nodules from the multi-omics data set obtained in step S2, and establish a data set for a malignancy risk prediction model for small pulmonary nodules by combining the image data obtained in step S1 and the exhaled breath data obtained in step S2;

[0015] S4. Construct a multi-omics deep learning model, and use the data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data;

[0016] S5. Visual application of the model: With the help of visualization tools, the malignancy risk prediction model of small lung nodules based on the comprehensive imaging indicators and exhaled breath VOCs data obtained in step S4 is visualized.

[0017] Furthermore, the specific implementation method of step S1 includes the following steps:

[0018] S1.1. Collect CT images of patients with small lung nodules detected by lung CT;

[0019] S1.2. Use the nodule segmentation software 3D-slicer to perform two-dimensional segmentation on the nodules in the CT image collected in step S1.1 to obtain a two-dimensional segmented CT image;

[0020] S1.3. Perform three-dimensional reconstruction on the CT images after two-dimensional segmentation, and use image analysis software Fusion 360 and Netfabb to perform stereo measurement on the extracted three-dimensional data to measure the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the pulmonary nodules;

[0021] S1.4. Perform normalized registration on lung nodules in CT images;

[0022] The calculation formula of sphericity S is as follows:

[0023]

[0024] Among them, V P is the nodule volume, S P is the surface area of ​​the nodule, 0<S<1, and S value of 1 corresponds to a perfect sphere;

[0025] Volume doubling time (VDT) refers to the time it takes for a tumor to double in volume. The calculation formula is as follows:

[0026]

[0027] Among them, V2 is the volume of the nodule at the second examination, V1 is the volume of the nodule at the baseline examination, and Δt is the interval between the two examinations;

[0028] The average CT value is the attenuation value of CT rays after passing through the nodule tissue;

[0029] The solid component ratio CTR refers to the ratio of solid components in pulmonary nodules. The visible components of pulmonary nodules under the mediastinal window are defined as solid components. Since the window position of the mediastinal window is generally 250-350HU and the window width is 30-50HU, the CT value > -400HU is regarded as the solid component. The image analysis software is used to calculate the CT value of all pixels of the segmented nodules. The ratio of the number of pixels N1 with CT value > -400HU to the total number of pixels N of the measured nodules is taken as CTR. The solid nodules are 1 and the ground glass nodules are 0. The calculation formula is as follows:

[0030] CTR=N1 / N.

[0031] Furthermore, the specific implementation method of step S2 includes the following steps:

[0032] S2.1. Set butanone, n-propanol, isoprene, ethylbenzene, styrene and hexanal as the calibration types, and use the exhaled breath concentration of patients clinically diagnosed with lung cancer as the calibration concentration;

[0033] S2.2. Based on the informed consent of the patient, collect exhaled VOCs from patients with small pulmonary nodules. Use a VOCs fluorine film sampling bag to collect >500ml of mixed gas exhaled by the patient calmly, and then determine the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the patient's pulmonary nodules;

[0034] S2.3. Analyze the gas composition of the exhaled air collected in step S2.2 using GC-MS gas chromatography-mass spectrometry;

[0035] S2.4. Perform cross-sectional grouping of lung nodules and calculate the average exhaled VOCs value of patients with small lung nodules in each group;

[0036] S2.5. Then, combined with the imaging features, multi-omics data are sorted to obtain a multi-omics data set.

[0037] Furthermore, the specific implementation method of step S3 includes the following steps:

[0038] S3.1. Establish the malignancy risk criteria for small pulmonary nodules based on the Fleischner Society's pulmonary nodule management guidelines (2017) and the Chinese expert consensus on the diagnosis of early lung cancer (2023);

[0039]

[0040] S3.2. Based on clinical experience, establish auxiliary malignancy risk advancement criteria;

[0041]

[0042] S3.3. Extract clinical decision data from the multi-omics dataset obtained in step S2 based on the criteria of steps S3.1 and S3.2;

[0043] S3.4. Use the image data obtained in step S1, the exhaled air data obtained in step S2, and the clinical decision data to form a data set for the malignancy risk prediction model of small lung nodules.

[0044] Furthermore, the specific implementation method of step S4 includes the following steps:

[0045] S4.1. Construct a multi-omics deep learning model using two sets of multi-layer perceptrons in parallel, followed by a fully connected layer and then a softmax layer;

[0046] S4.2. Set a first group of multi-layer perceptrons to encode 6 types of CT image data, set the 6 types of image data to fixed dimensions, input them into the prediction module to output the CT image data feature vector A, and the dimension of A is 200 dimensions;

[0047] A second multi-layer perceptron is set up to encode the six types of exhaled breath data, and the six types of exhaled breath features are set to fixed dimensions, which are input into the prediction module to output the exhaled breath data feature vector B, where the dimension of B is 200;

[0048] S4.3. Construct the comprehensive feature vector X, and the calculation expression is:

[0049] X=αA+βB+δC

[0050] Among them, α, β, and δ add up to 1, and the specific values ​​are obtained by model training; when α = 0.42, β = 0.32, and δ = 0.26, the results are promoted;

[0051] S4.4. Input the comprehensive feature vector X into the fully connected layer for linear transformation and activation function processing to obtain the output H of the fully connected layer. The calculation expression is:

[0052] H = RELU(W fc X+b fc )

[0053] Among them, W fc and b fc are the weights and biases of the fully connected layer;

[0054] Then the output H of the fully connected layer is input into the softmax layer for classification prediction to obtain the final classification prediction result Y. The calculation expression is:

[0055] Y = softmax(W softmax H+b softmax )

[0056] Among them, W softmax and b softmax is the weight and bias of the softmax layer;

[0057] S4.5. Set the calculation expression of the loss function L to:

[0058]

[0059] in, is the result of the i-th prediction, Y i is the ith actual result;

[0060] The data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 is used to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data.

[0061] Beneficial effects of the present invention:

[0062] The method for establishing a malignant risk prediction model for small lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data of the present invention uses a deep learning method to establish a multimodal AI model system based on the dynamic changes of three medical imaging parameters of lung nodules, namely, sphericity, average CT value, and volume doubling time, and quantitative values ​​of butanone, n-propanol, isoprene, ethylbenzene, styrene, and hexanal concentrations in exhaled breath VOCs. This system can more accurately and dynamically monitor the malignant risk of small lung nodules, and detect precancerous lesions and early lung cancer earlier and more accurately. The research on the onset patent not only fills the gap in the multi-omics follow-up research of small lung nodules, but also has wide application and promotion value.

[0063] The method for establishing a malignant risk prediction model for pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data of the present invention aims at the defects of the existing technology in predicting the malignant risk of pulmonary nodules. The present invention proposes a self-developed multimodal malignant risk prediction model for pulmonary nodules. The model integrates medical imaging parameters, volatile organic compounds (VOCs) data in exhaled breath and rich clinical experience through deep learning methods, and can more accurately realize dynamic monitoring of malignant risk of pulmonary nodules, and detect precancerous lesions and early lung cancer earlier and more accurately. This study not only fills the gap in multi-omics follow-up research on pulmonary nodules, but also has wide application and promotion value.

[0064] The method for establishing a malignant risk prediction model for pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data of the present invention can complete the malignant risk prediction of pulmonary nodules with only 6 imaging features and 6 exhaled breath VOCs data, and the prediction accuracy is improved by 5% compared with the existing single modality model, the accuracy can reach 96%, the misdiagnosis rate is 0.26%, and the missed diagnosis rate is 0.13%. While saving medical resources, it also reduces the economic burden on patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flowchart of a method for establishing a risk prediction model for malignant lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data according to the present invention;

[0066] Figure 2 The data collation process for VOSs modeling of the present invention;

[0067] Figure 3 The present invention is a flow chart of CT image modeling data organization. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0069] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0070] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 -Attached Figure 3 The detailed instructions are as follows: Specific implementation method one:

[0072] A method for establishing a risk prediction model for malignant lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data includes the following steps:

[0073] S1. collecting CT images of patients with small lung nodules detected by lung CT, and then processing the CT images of patients with small lung nodules, extracting image data of imaging indicators, and obtaining an image data set for standby use;

[0074] Furthermore, the specific implementation method of step S1 includes the following steps:

[0075] S1.1. Collect CT images of patients with small lung nodules detected by lung CT;

[0076] S1.2. Use the nodule segmentation software 3D-slicer to perform two-dimensional segmentation on the nodules in the CT image collected in step S1.1 to obtain a two-dimensional segmented CT image;

[0077] S1.3. Perform three-dimensional reconstruction on the CT images after two-dimensional segmentation, and use image analysis software Fusion 360 and Netfabb to perform stereo measurement on the extracted three-dimensional data to measure the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the pulmonary nodules;

[0078] S1.4. Perform normalized registration on lung nodules in CT images;

[0079] The calculation formula of sphericity S is as follows:

[0080]

[0081] Among them, V P is the nodule volume, S P is the surface area of ​​the nodule, 0<S<1, and S value 1 corresponds to a perfect sphere; Sphericity: Since tumor nodules have the particularity of spherical growth during their growth process, the change of sphericity (S) can be closer to the growth pattern of tumors, so this change trend is suitable for predicting the malignant risk of nodules;

[0082] Volume doubling time (VDT) refers to the time it takes for a tumor to double in volume. The calculation formula is as follows:

[0083]

[0084] Among them, V2 is the volume of the nodule at the second examination, V1 is the volume of the nodule at the baseline examination, and Δt is the interval between the two examinations; VDT refers to the time (days) it takes for the tumor volume to double, which represents the activity and invasiveness of tumor cells, and is determined by the cell proliferation cycle and the ratio of cell proliferation to loss. The doubling time of tumor nodules is significantly different from that of inflammatory nodules: the doubling time of acute inflammatory nodules is extremely short, generally less than 15 days, and the doubling time of chronic inflammatory nodules is slow, generally without significant changes for several years. The doubling time of tumor nodules is an average of 100-400 days (excluding indolent cancer nodules), so the volume doubling time can accurately reflect the specificity of tumor nodules;

[0085] The average CT value is the attenuation value of CT rays after passing through the nodule tissue; the unit of the average CT value is Hounsfield Unit (Hu), which can reflect the degree of tissue absorption of rays. It is a unified unit for expressing tissue density in CT imaging. Different tissue structures will present different CT values. As the CT value changes, it can also indicate the changes in the internal components of the nodule. Malignant nodules often originate from alveolar epithelial cells. As the number of malignant cells increases, the CT value of the nodule changes dynamically. Therefore, the average CT value is used to dynamically monitor the growth changes of nodules, which plays an important role in determining the development trend of nodules.

[0086] The solid component ratio CTR refers to the ratio of solid components in pulmonary nodules. The visible components of pulmonary nodules under the mediastinal window are defined as solid components. Since the window position of the mediastinal window is generally 250-350HU and the window width is 30-50HU, the CT value > -400HU is regarded as the solid component. The image analysis software is used to calculate the CT value of all pixels of the segmented nodules. The ratio of the number of pixels N1 with CT value > -400HU to the total number of pixels N of the measured nodules is taken as CTR. The solid nodules are 1 and the ground glass nodules are 0. The calculation formula is as follows:

[0087] CTR=N1 / N.

[0088] Because tumor cell clusters are mostly solid in CT images, as tumor cells proliferate, the proportion of solid components in the nodules continues to increase. Changes in CTR can very intuitively show the evolution of the invasiveness and malignancy of lung nodules. Current studies often use the ratio of the maximum diameter of the solid part to the maximum diameter of the nodule as CTR, but because nodules are mostly irregular in shape and the growth of solid components is also irregular, this measurement method has a large error and cannot reflect the actual situation of CTR. The visible component of the lung nodule under the mediastinal window can be defined as a solid component;

[0089] S2. Collect exhaled VOCs from patients with small pulmonary nodules, perform calibration and abundance analysis of exhaled VOCs indicators, obtain exhaled breath data, and then combine with imaging data to perform multi-omics data collation to obtain a multi-omics data set;

[0090] Furthermore, the specific implementation method of step S2 includes the following steps:

[0091] S2.1. Set butanone, n-propanol, isoprene, ethylbenzene, styrene and hexanal as the calibration types, and use the exhaled breath concentration of patients clinically diagnosed with lung cancer as the calibration concentration;

[0092] S2.2. Based on the informed consent of the patient, collect exhaled VOCs from patients with small pulmonary nodules. Use a VOCs fluorine film sampling bag to collect >500ml of mixed gas exhaled by the patient calmly, and then determine the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the patient's pulmonary nodules;

[0093] S2.3. Analyze the gas composition of the exhaled air collected in step S2.2 using GC-MS gas chromatography-mass spectrometry;

[0094] S2.4. Perform cross-sectional grouping of lung nodules and calculate the average exhaled VOCs value of patients with small lung nodules in each group;

[0095] S2.5. Then, combined with the imaging features, multi-omics data are sorted to obtain a multi-omics data set.

[0096] S3. Formulate a risk standard for malignancy of small pulmonary nodules, extract clinical decision data based on the risk standard for malignancy of small pulmonary nodules from the multi-omics data set obtained in step S2, and establish a data set for a malignancy risk prediction model for small pulmonary nodules by combining the image data obtained in step S1 and the exhaled breath data obtained in step S2;

[0097] Furthermore, the specific implementation method of step S3 includes the following steps:

[0098] S3.1. Establish the malignancy risk criteria for small pulmonary nodules based on the Fleischner Society's pulmonary nodule management guidelines (2017) and the Chinese expert consensus on the diagnosis of early lung cancer (2023);

[0099]

[0100] S3.2. Based on clinical experience, establish auxiliary malignancy risk advancement criteria;

[0101]

[0102] S3.3. Extract clinical decision data from the multi-omics dataset obtained in step S2 based on the criteria of steps S3.1 and S3.2;

[0103] S3.4. The image data obtained in step S1, the exhaled air data obtained in step S2 and the clinical decision data are used to form a data set for the malignant risk prediction model of small lung nodules, as shown in the following table:

[0104]

[0105]

[0106] S4. Construct a multi-omics deep learning model, and use the data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data;

[0107] Furthermore, the specific implementation method of step S4 includes the following steps:

[0108] S4.1. Construct a multi-omics deep learning model using two sets of multi-layer perceptrons in parallel, followed by a fully connected layer and then a softmax layer;

[0109] Two sets of multi-layer perceptrons (MLP): The first set of MLP is used to process CT image data, and the second set of MLP is used to process VOCs data. Each set of MLP contains two hidden layers, the first hidden layer contains 512 neurons, and the second hidden layer contains 256 neurons. Both use the ReLU activation function and output a 200-dimensional feature vector. Fully connected layer: contains 128 neurons and uses the ReLU activation function for nonlinear transformation.

[0110] S4.2. Set a first group of multi-layer perceptrons to encode 6 types of CT image data, set the 6 types of image data to fixed dimensions, input them into the prediction module to output the CT image data feature vector A, and the dimension of A is 200 dimensions;

[0111] A second multi-layer perceptron is set up to encode the six types of exhaled breath data, and the six types of exhaled breath features are set to fixed dimensions, which are input into the prediction module to output the exhaled breath data feature vector B, where the dimension of B is 200;

[0112] S4.3. Construct the comprehensive feature vector X, and the calculation expression is:

[0113] X=αA+βB+δC

[0114] Among them, α, β, and δ add up to 1, and the specific values ​​are obtained by model training; when α = 0.42, β = 0.32, and δ = 0.26, the results are promoted;

[0115] The fully connected layer is one of the most basic layers in a neural network, also known as a densely connected layer or a linear layer. Its main function is to multiply the input data with the weight matrix, add a bias term, and then perform a nonlinear transformation through an activation function. The fully connected layer can map the input data to a new feature space, thereby extracting and learning higher-level features. The Softmax layer is a commonly used activation function, mainly used for multi-category classification tasks. It converts each element of the input vector into a probability value between 0 and 1, and the sum of the probabilities of all elements is 1. In classification tasks, the Softmax function can interpret the output of the network as the probability distribution of each category.

[0116] S4.4. Input the comprehensive feature vector X into the fully connected layer for linear transformation and activation function processing to obtain the output H of the fully connected layer. The calculation expression is:

[0117] H = RELU(W fc X+b fc )

[0118] Among them, W fc and b fc are the weights and biases of the fully connected layer;

[0119] Then the output H of the fully connected layer is input into the softmax layer for classification prediction to obtain the final classification prediction result Y. The calculation expression is:

[0120] Y = softmax(W softmax H+b softmax )

[0121] Among them, W softmax and b softmax is the weight and bias of the softmax layer;

[0122] S4.5. Set the calculation expression of the loss function L to:

[0123]

[0124] in, is the result of the i-th prediction, Y i is the ith actual result;

[0125] The data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 is used to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data.

[0126] S5. Visual application of the model: With the help of visualization tools, the malignancy risk prediction model of small lung nodules based on the comprehensive imaging indicators and exhaled breath VOCs data obtained in step S4 is visualized.

[0127] The multimodal risk prediction model for malignant lung nodules proposed in this embodiment can detect the development and evolution of nodules earlier from the cell growth and metabolism level based on the imaging changes of lung nodules and the results of cell metabolism changes. It can be used to accurately monitor pre-lung cancer lesions, which is significantly earlier than the detection window of existing models and has greater benefits for patients. In addition, the present invention adopts dynamic monitoring of VOCs indicators, which is non-invasive, highly repeatable and acceptable to patients. The details are as follows:

[0128] Through research and a large number of experiments, only 6 imaging features and 6 exhaled VOCs data can be used to predict the risk of malignancy of small lung nodules, and the prediction accuracy is 5% higher than that of the existing single-modality model, reaching 96%, with a misdiagnosis rate of 0.26% and a missed diagnosis rate of 0.13%. This saves medical resources and reduces the economic burden on patients.

[0129] Predictions were made at the cell growth and metabolism levels and it was found that the cycle was significantly advanced.

[0130] For the first time, the exhaled breath and imaging features were jointly modeled, and clinical experience was introduced to participate in decision-making. This provides new research methods and tools for the multi-omics follow-up research of small lung nodules, filling the gap in existing research.

[0131] Twelve minimal features for predicting the risk of malignancy of small pulmonary nodules were provided, which facilitated the comparison and communication of results between different medical institutions and researchers.

[0132] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0133] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for establishing a risk prediction model for malignant lung nodules based on comprehensive imaging indicators and exhaled breath VOCs data, characterized in that: The steps include: S1. collecting CT images of patients with small lung nodules detected by lung CT, and then processing the CT images of patients with small lung nodules, extracting image data of imaging indicators, and obtaining an image data set for standby use; S2. Collect exhaled VOCs from patients with small pulmonary nodules, perform calibration and abundance analysis of exhaled VOCs indicators, obtain exhaled breath data, and then combine with imaging data to perform multi-omics data collation to obtain a multi-omics data set; S3. Formulate a risk standard for malignancy of small pulmonary nodules, extract clinical decision data based on the risk standard for malignancy of small pulmonary nodules from the multi-omics data set obtained in step S2, and establish a data set for a malignancy risk prediction model for small pulmonary nodules by combining the image data obtained in step S1 and the exhaled breath data obtained in step S2; S4. Construct a multi-omics deep learning model, and use the data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data; S5. Visual application of the model: With the help of visualization tools, the malignancy risk prediction model of small lung nodules based on the comprehensive imaging indicators and exhaled breath VOCs data obtained in step S4 is visualized.

2. The method for establishing a malignant risk prediction model for pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data according to claim 1, characterized in that: The specific implementation method of step S1 includes the following steps: S1.

1. Collect CT images of patients with small lung nodules detected by lung CT; S1.

2. Use the nodule segmentation software 3D-slicer to perform two-dimensional segmentation on the nodules in the CT image collected in step S1.1 to obtain a two-dimensional segmented CT image; S1.

3. Perform three-dimensional reconstruction on the CT images after two-dimensional segmentation, and use image analysis software Fusion 360 and Netfabb to perform stereo measurement on the extracted three-dimensional data to measure the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the pulmonary nodules; S1.

4. Perform normalized registration on lung nodules in CT images; The calculation formula of sphericity S is as follows: Among them, V P is the nodule volume, S P is the surface area of ​​the nodule, 0<S<1, and S value of 1 corresponds to a perfect sphere; Volume doubling time (VDT) refers to the time it takes for a tumor to double in volume. The calculation formula is as follows: Among them, V2 is the volume of the nodule at the second examination, V1 is the volume of the nodule at the baseline examination, and Δt is the interval between the two examinations; The average CT value is the attenuation value of CT rays after passing through the nodule tissue; The solid component ratio CTR refers to the ratio of solid components in pulmonary nodules. The visible components of pulmonary nodules under the mediastinal window are defined as solid components. Since the window position of the mediastinal window is generally 250-350HU and the window width is 30-50HU, the CT value > -400HU is regarded as the solid component. The image analysis software is used to calculate the CT value of all pixels of the segmented nodules. The ratio of the number of pixels N1 with CT value > -400HU to the total number of pixels N of the measured nodules is taken as CTR. The solid nodules are 1 and the ground glass nodules are 0. The calculation formula is as follows: CTR=N1 / N.

3. The method for establishing a malignant risk prediction model for pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data according to claim 1 or 2, characterized in that: The specific implementation method of step S2 includes the following steps: S2.

1. Set butanone, n-propanol, isoprene, ethylbenzene, styrene and hexanal as the calibration types, and use the exhaled breath concentration of patients clinically diagnosed with lung cancer as the calibration concentration; S2.

2. Based on the informed consent of the patient, collect exhaled VOCs from patients with small pulmonary nodules. Use a VOCs fluorine film sampling bag to collect >500ml of mixed gas exhaled by the patient calmly, and then determine the size, location, sphericity, average CT value, proportion of solid components, and volume doubling time of the patient's pulmonary nodules; S2.

3. Analyze the gas composition of the exhaled air collected in step S2.2 using GC-MS gas chromatography-mass spectrometry; S2.

4. Perform cross-sectional grouping of lung nodules and calculate the average exhaled VOCs value of patients with small lung nodules in each group; S2.

5. Then, combined with the imaging features, multi-omics data are sorted to obtain a multi-omics data set.

4. The method for establishing a risk prediction model for malignant pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data according to claim 3, characterized in that: The specific implementation method of step S3 includes the following steps: S3.

1. Establish the malignancy risk criteria for small pulmonary nodules based on the Fleischner Society's pulmonary nodule management guidelines (2017) and the Chinese expert consensus on the diagnosis of early lung cancer (2023); S3.

2. Based on clinical experience, establish auxiliary malignancy risk advancement criteria; S3.

3. Extract clinical decision data from the multi-omics dataset obtained in step S2 based on the criteria of steps S3.1 and S3.2; S3.

4. Use the image data obtained in step S1, the exhaled air data obtained in step S2, and the clinical decision data to form a data set for the malignancy risk prediction model of small lung nodules.

5. The method for establishing a risk prediction model for malignant pulmonary nodules based on comprehensive imaging indicators and exhaled breath VOCs data according to claim 4, characterized in that: The specific implementation method of step S4 includes the following steps: S4.

1. Construct a multi-omics deep learning model using two sets of multi-layer perceptrons in parallel, followed by a fully connected layer and then a softmax layer; S4.

2. Set a first group of multi-layer perceptrons to encode 6 types of CT image data, set the 6 types of image data to fixed dimensions, input them into the prediction module to output the CT image data feature vector A, and the dimension of A is 200 dimensions; A second multi-layer perceptron is set up to encode the six types of exhaled breath data, and the six types of exhaled breath features are set to fixed dimensions, which are input into the prediction module to output the exhaled breath data feature vector B, where the dimension of B is 200; S4.

3. Construct the comprehensive feature vector X, and the calculation expression is: X=αA+βB+δC Among them, α, β, and δ add up to 1, and the specific values ​​are obtained by model training; when α = 0.42, β = 0.32, and δ = 0.26, the results are promoted; S4.

4. Input the comprehensive feature vector X into the fully connected layer for linear transformation and activation function processing to obtain the output H of the fully connected layer. The calculation expression is: H=RELU(W fc X+b fc ) Among them, W fc and b fc are the weights and biases of the fully connected layer; Then the output H of the fully connected layer is input into the softmax layer for classification prediction to obtain the final classification prediction result Y. The calculation expression is: Y =softmax(W softmax H+b softmax ) Among them, W softmax and b softmax is the weight and bias of the softmax layer; S4.

5. Set the calculation expression of the loss function L to: in, is the result of the i-th prediction, Y i is the actual result of the ith case; The data set of the pulmonary nodule malignancy risk prediction model obtained in step S3 is used to train the multi-omics deep learning model to obtain a pulmonary nodule malignancy risk prediction model that integrates imaging indicators and exhaled breath VOCs data.

Images