A prediction model and method for the growth trend of subsolid pulmonary nodules

Abstract

The present invention relates to the technical field of prediction of growth trend of pulmonary subsolid nodules, and discloses a prediction model and prediction method for growth trend of pulmonary subsolid nodules, including a data acquisition module and a prediction analysis module. The prediction model and prediction method for growth trend of pulmonary subsolid nodules retain clinical data and lung CT image data that meet the screening criteria, substitute them into the VAE-Deepsurv algorithm model for training, obtain multiple versions of algorithm models through training, calculate the area under the ROC curve of multiple versions of algorithm models, retain the algorithm model with the largest area under the ROC curve as the final model, strictly control the feature quantity of the prediction model, classify the clinical data set and the image data set into the data collection module, calculate and generate the nodule growth data group, the individual risk index and the predicted growth data group by the prediction analysis module, judge the growth probability of pulmonary subsolid nodules, and generate corresponding reports, focusing on individual differences and high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of pulmonary subsolid nodule growth trend prediction, and in particular to a pulmonary subsolid nodule growth trend prediction model and prediction method. Background Art

[0002] Subsolid pulmonary nodules are focal, rounded, and increased-density subsolid pulmonary shadows with a diameter of ≤3 cm. The growth characteristics of subsolid pulmonary nodules may indicate different pathological conditions from benign to malignant. According to their nature, subsolid pulmonary nodules can be divided into two categories: benign and malignant. Benign nodules include granulomas and inflammatory lesions, while malignant nodules usually refer to the early manifestations of lung cancer. For the prediction of the growth trend of subsolid pulmonary nodules, imaging measurement methods are currently mainly used. The imaging-based measurement method is mainly CT scanning. CT scanning can provide high-resolution lung images. Through continuous scanning, the size changes of nodules can be observed, thereby predicting their growth rate. Artificial intelligence machine learning methods can identify and learn a large amount of imaging data of subsolid pulmonary nodules by training algorithm models. AI can help doctors identify nodules with potential risks more quickly and accurately.

[0003] At present, traditional subsolid pulmonary nodule growth trend prediction models and prediction methods mainly evaluate the growth probability of subsolid pulmonary nodules based on the growth rate of subsolid pulmonary nodules per unit time. However, in the actual prediction process, factors such as the size, morphology, density, underlying diseases, and family genetic history of the patient's lung nodules will lead to certain deviations in the prediction results. Traditional subsolid pulmonary nodule growth trend prediction methods do not take individual differences into account and have low prediction accuracy. Summary of the invention

[0004] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a growth trend prediction model and method for subsolid lung nodules, which have the advantages of strictly controlling the characteristic quantity of the prediction model, focusing on individual differences and high prediction accuracy. It solves the problem that the traditional growth trend prediction model for subsolid lung nodules has large prediction deviation and the traditional growth trend prediction method for subsolid lung nodules does not consider individual differences and has low accuracy.

[0005] (II) Technical solution

[0006] To achieve the above object, the present invention provides the following technical solution: a pulmonary subsolid nodule growth trend prediction model, wherein the pulmonary subsolid nodule growth trend prediction model is obtained by training the following steps: S1. Acquire clinical data and lung CT image data at multiple time points, clean all acquired data, and retain clinical data and lung CT image data that meet the screening criteria; S2. Extract the cleaned clinical data and lung CT image data in chronological order and substitute them into the VAE-Deepsurv algorithm model for training. Multiple versions of the VAE-Deepsurv algorithm model are trained and arranged from short to long according to the training time, and then marked with numbers. Among them, the one with the shortest training time is VAE-Deepsurv1, and the one with the longest training time is VAE-Deepsurv z , Indicates that the VAE-Deepsurv algorithm model that has completed training has versions; S3, control multiple versions of VAE-Deepsurv algorithm models to simultaneously predict the growth trend of patients' pulmonary subsolid nodules, compare the predicted results with the actual growth trend of patients' pulmonary subsolid nodules, and calculate the area under the ROC curve of multiple versions of VAE-Deepsurv algorithm models; S4. The VAE-Deepsurv algorithm model with the largest area under the ROC curve is retained as the final prediction model for the growth trend of subsolid pulmonary nodules.

[0007] A method for predicting the growth trend of pulmonary nodules comprises the following steps: Step 1: The data acquisition module obtains the patient's clinical data and lung CT image data at multiple time points through database import, cleans all the acquired data, retains the data that meets the screening criteria, and classifies them into clinical data sets and image data sets; Step 2: The prediction and analysis module substitutes the clinical data set and the imaging data set into the VAE-Deepsurv algorithm model, calculates and generates a nodule growth data set, and analyzes the growth characteristics of the subsolid pulmonary nodules at the current time point; Step 3: The VAE-Deepsurv algorithm model calculates and generates an individual risk index based on the clinical data set and the imaging data set, and analyzes the risk probability of the growth of subsolid pulmonary nodules in individual patients; Step 4: The VAE-Deepsurv algorithm model calculates and generates a predicted growth data set based on the calculated nodule growth data set, clinical data set, and imaging data set to predict the subsequent growth characteristics of pulmonary subsolid nodules; Step 5. The predictive analysis module is set with a characteristic threshold and a risk threshold, and compares the nodule growth data group, the individual risk index, and the predicted growth data group with the characteristic threshold and the risk threshold to determine the growth probability of the subsolid pulmonary nodules and generate a corresponding growth trend report or nodule prevention report.

[0008] Preferably, in step one, the data acquisition module cleans all acquired data according to age, classifies patients aged 50 and below into the youth group, and classifies patients aged 50 and above into the middle-aged and elderly group. The screening criteria require that both the youth group patients and the middle-aged and elderly group patients have 60 consecutive months of clinical data and imaging data.

[0009] Preferably, in step one, the clinical data set includes clinical data of young patients and middle-aged and elderly patients for 60 consecutive months, and the imaging data set includes lung CT image data of young patients and middle-aged and elderly patients for 60 consecutive months.

[0010] Preferably, in step 5, if the nodule growth data set and predicted growth data set When any value exceeds the characteristic threshold, the prediction and analysis module determines that the growth probability of the pulmonary subsolid nodule is high and generates a corresponding growth trend report.

[0011] Preferably, in step 5, if the nodule growth data set and predicted growth data set All values ​​in the feature threshold do not exceed the individual risk index When the risk threshold is exceeded, the predictive analysis module determines that the growth probability of subsolid pulmonary nodules is medium and generates a nodule prevention report accordingly.

[0012] Compared with the prior art, the present invention provides a lung nodule growth trend prediction model and prediction method, which has the following beneficial effects: 1. The present invention retains clinical data and lung CT image data that meet the screening criteria, extracts the cleaned clinical data and lung CT image data in chronological order, substitutes them into the VAE-Deepsurv algorithm model for training, and obtains multiple versions of the VAE-Deepsurv algorithm model through training. The multiple versions of the VAE-Deepsurv algorithm model are controlled to simultaneously predict the growth trend of the patient's lung subsolid nodules, and the prediction results are compared with the patient's actual lung subsolid nodule growth trend, and the areas under the ROC curves of the multiple versions of the algorithm model are calculated. The predictive ability of the multiple versions of the VAE-Deepsurv algorithm model is evaluated according to a unified standard, and the algorithm model with the largest area under the ROC curve is retained as the final model. The feature quantity of the prediction model is strictly controlled, and the subsequent prediction of the growth trend of the lung subsolid nodules is more accurate.

[0013] 2. The present invention uses a data acquisition module to obtain patient clinical data and lung CT image data at multiple time points through database import, cleans all acquired data, retains data that meets the screening criteria, and classifies them into clinical data sets and image data sets. The prediction analysis module substitutes the clinical data sets and image data sets into the VAE-Deepsurv algorithm model to calculate and generate a nodule growth data set. , the growth characteristics of subsolid lung nodules at the current time point are analyzed, and the VAE-Deepsurv algorithm model calculates and generates an individual risk index based on the clinical data set and the imaging data set The VAE-Deepsurv algorithm model calculates the risk probability of subsolid lung nodule growth in individual patients according to the nodule growth data set. , clinical data sets and imaging data sets, and computationally generate a predicted growth data set , predict the subsequent growth characteristics of pulmonary subsolid nodules, the prediction analysis module is set with feature thresholds and risk thresholds, and the nodule growth data group , Individual Risk Index , predicted growth data set By comparing the feature threshold and risk threshold, the growth probability of subsolid pulmonary nodules is determined, and a corresponding growth trend report or nodule prevention report is generated, focusing on individual differences and achieving high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a diagram of the training steps of the prediction model of the present invention; Figure 2 It is a step diagram of the prediction method of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Since the traditional prediction model and prediction method of the growth trend of pulmonary subsolid nodules mainly evaluates the growth probability of pulmonary nodules based on the growth rate of pulmonary subsolid nodules per unit time, in the actual prediction process, factors such as the size, shape, density, underlying diseases, and family genetic history of the patient's pulmonary nodules will lead to certain deviations in the prediction results. The traditional prediction method of the growth trend of pulmonary subsolid nodules does not take individual differences into account and has low prediction accuracy. Therefore, a prediction model and prediction method for the growth trend of pulmonary subsolid nodules are proposed. Please refer to Figure 1-Figure 2 , the prediction model is trained through the following steps: S1. Obtain clinical data and lung CT image data of patients of different ages at multiple time points, clean all acquired data, and retain clinical data and lung CT image data that meet the screening criteria; S2. Extract the cleaned clinical data and lung CT image data in chronological order, and substitute them into the VAE-Deepsurv algorithm model for training. The VAE-Deepsurv algorithm model performs convolution operations in three spatial dimensions based on the lung nodule volume data provided by the CT scan images, which can more accurately extract volume features. Multiple versions of the VAE-Deepsurv algorithm model obtained through training need to be arranged from short to long according to the training time, and then marked with numbers, the specific numbers are 1, 2, 3, ..., Among them, VAE-Deepsurv has the shortest training time. 1， The longest training time is VAE-Deepsurv z， Indicates that the VAE-Deepsurv algorithm model that has completed training has versions; S3. Control multiple versions of the VAE-Deepsurv algorithm model to simultaneously predict the growth trend of the patient's pulmonary subsolid nodules, and compare the predicted results with the patient's actual pulmonary subsolid nodule growth trend, and calculate the area under the ROC curve (AUC) of multiple versions of the VAE-Deepsurv algorithm model. The calculation formula is as follows: ; In the formula, Indicates the ranking of each subsolid pulmonary nodule that actually shows growth in the prediction results of all VAE-Deepsurv algorithm models. represents the sum of the rankings of all growing subsolid pulmonary nodules, M represents the number of subsolid pulmonary nodules that actually grew during follow-up, represents the average ranking of M growing subsolid pulmonary nodules if they are randomly sorted. represents the difference between the ranking sum of all growing pulmonary subsolid nodule samples and their average ranking sum under completely random ranking. This difference reflects the "advantage" of the actual growing pulmonary subsolid nodules predicted by the model over the random ranking. N represents the number of stable pulmonary subsolid nodules in actual follow-up. MN Used for normalization to ensure that the AUC value is between 0 and 1.

[0017] S4. The VAE-Deepsurv algorithm model with the largest area under the ROC curve is retained as the final prediction model for the growth trend of subsolid pulmonary nodules. The feature quantity of the prediction model is strictly controlled, and the subsequent prediction of the growth trend of pulmonary nodules is more accurate.

[0018] The prediction method for the growth trend of pulmonary subsolid nodules is based on the above-mentioned prediction model for the growth trend of pulmonary subsolid nodules. Figure 2 The method for predicting the growth trend of pulmonary subsolid nodules comprises the following steps: Step 1: The data acquisition module obtains clinical data and lung CT image data through database import. The data acquisition module cleans all acquired data according to age, and classifies patients aged 50 years and below into the youth group, and patients aged 50 years and above into the middle-aged and elderly group. The probability of growth of subsolid lung nodules in patients of different age groups is different. The screening criteria require that both the youth group and the middle-aged and elderly group patients have 60 consecutive months of clinical data and lung CT image data. The data that meets the screening criteria are retained and classified into clinical data sets and image data sets; The clinical data set includes the living environment data and family medical history data of young patients and middle-aged and elderly patients for 60 consecutive months. The expression of the clinical data set is , to Corresponding to each patient's age group, living environment data and family medical history data, Indicates the group corresponding to the patient's age group, Indicates the patient's living environment data and family medical history data, with subscripts 1 to Patients who meet the screening criteria have name; The image dataset includes lung CT image data of young patients and middle-aged and elderly patients for 60 consecutive months. The expression of the image dataset is: , to Corresponding to each patient's lung CT image data, Represents the patient's lung CT image data, subscript 1 to Patients who met the screening criteria had Name, life data set and image datasets All correspond to the same patient; Step 2: The prediction analysis module substitutes the clinical data set and the imaging data set into the VAE-Deepsurv algorithm model to calculate and generate the nodule growth data set. , the growth characteristics of the subsolid lung nodules at the current time point are analyzed and the calculation formula is as follows:

[0019] In the formula, represents the nodule growth data set, Indicates the life data set Clinical data of 60 consecutive patients for 60 months, Indicates the image dataset The lung CT image data of 60 consecutive patients were collected. It means that the VAE-Deepsurv algorithm model analyzes the growth characteristics of lung nodules at the current time point based on the patient's clinical data and lung CT image data for 60 consecutive months. The growth characteristics of lung subsolid nodules at the current time point include volume, shape, size and density, which is the nodule growth data set; Step 3: The VAE-Deepsurv algorithm model calculates and generates an individual risk index based on the clinical data set and the imaging data set , the risk probability of subsolid lung nodule growth in individual patients was analyzed and the calculation formula was as follows: The prediction analysis module is based on the life data set The living environment data and family medical history data of the patients for 60 consecutive months were scored. The factor scores included exposure factor scores. , Individual factor scores and family factor scores If the patient's living environment data contains smoking habits, the exposure factor score =1, if there is no smoking habit in the patient's living environment data, the exposure factor score If the patient's living environment data frequently contacts asbestos, silica dust or radioactive substances, the individual factor score is 0. =1, if the patient's living environment data is almost not exposed to asbestos, silica dust or radioactive substances, the individual factor score If the patient's immediate family members have lung cancer, the family factor score is 0. =1, if the patient's immediate family members do not have lung cancer, the family factor score is is 0,

[0020] In the formula, represents the individual risk index, It represents the fixed parameter used to measure the score of the factor in the individual risk index calculation formula. Indicates The ratio of the factor scores of the patient's living environment data and family history data for 60 consecutive months to the fixed parameters is the individual risk index, which comprehensively evaluates the exposure factors, individual factors and family factors that may affect the deterioration of lung disease, focusing on individual differences and having high prediction accuracy; Step 4: VAE-Deepsurv algorithm model calculates nodule growth data set , clinical data sets and imaging data sets, and computationally generate a predicted growth data set , predict the subsequent growth characteristics of pulmonary subsolid nodules, and the calculation formula is as follows:

[0021] In the formula, represents the predicted growth data set, The calculation formula for the predicted growth data set is for the nodule growth data set. The weight of The formula for calculating the predicted growth data set is for the individual risk index The weight of Indicates that the VAE-Deepsurv algorithm model is based on Weight and Weights, calculate nodule growth data set and individual risk index The weighted average of the predicted growth characteristics of pulmonary subsolid nodules at subsequent time points is used to predict the growth characteristics of pulmonary subsolid nodules at subsequent time points. The growth characteristics of pulmonary subsolid nodules at subsequent time points include volume, shape, size and density, which is the predicted growth data set. The prediction is made based on the growth characteristics of pulmonary subsolid nodules at the current time point and the exposure factors, individual factors and family factors that affect the growth of pulmonary subsolid nodules in many aspects, which effectively improves the prediction accuracy. Step 5. The prediction analysis module is equipped with feature thresholds and risk thresholds. The feature thresholds can be used to set reference thresholds for the size, shape, and density of subsolid lung nodules based on actual medical research results. The risk thresholds can also be used to set individual patient risk indexes based on actual medical research results. The prediction analysis module combines the nodule growth data set with the critical threshold , Individual Risk Index , predicted growth data set Compare the feature threshold and risk threshold to determine the growth probability of pulmonary subsolid nodules and generate corresponding growth trend reports or nodule prevention reports; If the nodule growth data set and predicted growth data set When any value in the feature threshold exceeds the feature threshold, the prediction analysis module determines that the growth probability of the pulmonary subsolid nodule is high, and generates a corresponding growth trend report. The growth trend report compares the nodule growth data group in detail. and predicted growth data set The difference between the two groups can be detected, and the medical staff can be assisted to judge whether the subsolid pulmonary nodules have a growth trend, and urge the patients to actively cooperate with the treatment. If the nodule growth data set and predicted growth data set All values ​​in the feature threshold do not exceed the individual risk index When the risk threshold is exceeded, the prediction analysis module determines that the growth probability of pulmonary subsolid nodules is medium, indicating that the patient has a higher probability of pulmonary subsolid nodule growth in the future than a normal healthy person, and generates a nodule follow-up strategy report accordingly. The nodule follow-up strategy report includes a nodule growth data set. , Individual Risk Index and predicted growth data set , providing professional advice to patients on maintaining lung health.

[0022] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A model for predicting the growth trend of pulmonary subsolid nodules, characterized by: The pulmonary subsolid nodule growth trend prediction model is obtained by training the following steps: S1. Obtain clinical data and lung CT image data of patients at multiple time points, clean all acquired data, and retain clinical data and lung CT image data that meet the screening criteria; S2, extracting the cleaned clinical data and lung CT image data in chronological order, substituting them into the VAE-Deepsurv algorithm model for training, and training to obtain multiple versions of the VAE-Deepsurv algorithm model; S3, control multiple versions of VAE-Deepsurv algorithm models to simultaneously predict the growth trend of patients' pulmonary subsolid nodules, compare the predicted results with the actual growth trend of patients' pulmonary subsolid nodules, and calculate the area under the ROC curve of multiple versions of VAE-Deepsurv algorithm models; S4. The VAE-Deepsurv algorithm model with the largest area under the ROC curve is retained as the final prediction model for the growth trend of subsolid pulmonary nodules.

2. A pulmonary subsolid nodule growth trend prediction model according to claim 1, characterized in that: In S2, multiple versions of the VAE-Deepsurv algorithm models are arranged from short to long according to the training time and marked with numbers, among which VAE-Deepsurv1 has the shortest training time and VAE-Deepsurv has the longest training time. z , Indicates that the VAE-Deepsurv algorithm model that has completed training has versions.

3. A method for predicting the growth trend of pulmonary subsolid nodules, applied to a model for predicting the growth trend of pulmonary subsolid nodules as claimed in any one of claims 1 to 2, characterized in that: The following steps are involved: Step 1: The data acquisition module obtains the patient's clinical data and lung CT image data at multiple time points through database import, cleans all the acquired data, retains the data that meets the screening criteria, and classifies them into clinical data sets and image data sets; Step 2: The prediction analysis module substitutes the clinical data set and the imaging data set into the VAE-Deepsurv algorithm model to calculate and generate the nodule growth data set. , analyze and obtain the growth characteristics of pulmonary subsolid nodules at the current time point; Step 3: The VAE-Deepsurv algorithm model calculates and generates an individual risk index based on the clinical data set and the imaging data set , the risk probability of subsolid lung nodule growth in individual patients was analyzed; Step 4: VAE-Deepsurv algorithm model calculates nodule growth data set , clinical data sets and imaging data sets, and computationally generate a predicted growth data set , predict the subsequent growth characteristics of subsolid pulmonary nodules; Step 5: The prediction analysis module is set with feature thresholds and risk thresholds, and the nodule growth data group , Individual Risk Index , predicted growth data set The characteristic threshold and risk threshold are compared to determine the growth probability of subsolid pulmonary nodules, and a corresponding growth trend report or nodule prevention report is generated.

4. A method for predicting the growth trend of pulmonary subsolid nodules according to claim 3, characterized in that: In the step one, the data acquisition module cleans all acquired data according to age, classifies patients aged 50 and below into the youth group, and classifies patients aged 50 and above into the middle-aged and elderly group. The screening criteria require that both the youth group patients and the middle-aged and elderly group patients have 60 consecutive months of clinical data and imaging data.

5. A method for predicting the growth trend of pulmonary subsolid nodules according to claim 4, characterized in that: In the step 1, the clinical data set includes clinical data of young patients and middle-aged and elderly patients for 60 consecutive months, and the imaging data set includes lung CT image data of young patients and middle-aged and elderly patients for 60 consecutive months.

6. A method for predicting the growth trend of pulmonary subsolid nodules according to claim 5, characterized in that: In the step three, The prediction analysis module is based on the clinical data set The clinical data of 60 consecutive patients were scored for factors, including exposure factor scores , Individual factor scores and family factor scores If the patient has a smoking habit in the clinical data, the exposure factor score =1, if the patient's clinical data frequently contacts asbestos, silica dust or radioactive substances, the individual factor score =1, if the patient's clinical data shows that a direct family member has lung cancer, the family factor score is 1.

7. A method for predicting the growth trend of pulmonary subsolid nodules according to claim 6, characterized in that: In step 5, if the nodule growth data set and predicted growth data set When any value exceeds the characteristic threshold, it indicates that the probability of lung nodule growth is high, and a corresponding growth trend report is generated.

8. A method for predicting the growth trend of pulmonary subsolid nodules according to claim 7, characterized in that: In step 5, if the nodule growth data set and predicted growth data set All values ​​in the feature threshold do not exceed the individual risk index When the risk threshold is exceeded, the predictive analysis module determines that the growth probability of lung nodules is medium and generates a nodule prevention report accordingly.

Images