Artificial intelligence-based pulmonary nodule growth prediction system and follow-up management system

Abstract

The invention belongs to the technical field of medical data processing, and particularly relates to a pulmonary nodule growth prediction system and follow-up management system based on artificial intelligence. The pulmonary nodule growth prediction system comprises an input module used for inputting CT image data and clinical feature data of a patient; the feature extraction module is used for carrying out feature extraction on the CT image data at each time point by adopting a feature extraction network; and the prediction module is used for fusing the features obtained by the feature extraction module by adopting a time sequence model, predicting parameters of the nodule growth model and obtaining a prediction curve of the nodule growth model. The invention further constructs a system capable of automatically generating the follow-up visit plan. The method has the advantages of being high in automation degree, accurate in prediction result and follow-up visit plan and capable of being dynamically updated at any time, nodule changes can be found in time, it is ensured that effective intervention means are adopted in the early stage, the survival rate and the treatment effect of patients are improved, and the method has good application prospects.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing technology, specifically relating to an artificial intelligence-based lung nodule growth prediction system and follow-up management system. Background Technology

[0002] Lung nodules are a common manifestation of early-stage lung cancer, and the growth rate of nodules (doubling time, DT) is an important indicator for assessing their malignant potential. Generally, the shorter the doubling time of a lung nodule, the higher its malignancy.

[0003] Once pulmonary nodules develop, patients require long-term follow-up. This is primarily because the nature of pulmonary nodules can change over time, including size, density, and morphology, and these changes are crucial for determining whether a nodule is benign or malignant. Current follow-up procedures and recommendations often rely on physician experience, are subjective, and lack standardized protocols. Furthermore, traditional follow-up management methods typically depend on fixed imaging follow-up cycles, lacking prediction of future nodule growth and failing to adjust the follow-up plan according to the nodule's actual growth rate. This can lead to missed diagnoses of early malignant nodules or inappropriate follow-up frequency.

[0004] Therefore, how to provide new diagnostic methods and systems for pulmonary nodules to accurately predict the future growth status of pulmonary nodules (such as the doubling time of nodules), and thus dynamically adjust follow-up strategies based on the prediction results, is an urgent problem to be solved in this field. Summary of the Invention

[0005] To address the problems of existing technologies, this invention provides an artificial intelligence-based lung nodule growth prediction system and follow-up management system.

[0006] An artificial intelligence-based lung nodule growth prediction system includes: The input module is configured to input the patient's CT image data and clinical feature data, wherein the CT image data includes at least two sets of data collected at different time points; The feature extraction module is configured to use a feature extraction network to extract features from the CT image data at each time point. The prediction module is configured to use a time series model to fuse the features obtained by the feature extraction module, predict the parameters of the nodule growth model, and obtain the prediction curve of the nodule growth model.

[0007] Preferably, the patient is a patient who has undergone or has not undergone surgery, the surgery being pulmonary nodule ablation or pulmonary nodule resection; When the patient is a patient who has undergone surgery, the CT imaging data includes at least one set of preoperative CT imaging data and at least one set of CT imaging data during follow-up.

[0008] Preferably, the prediction module also calculates the predicted value of the nodule doubling time based on the prediction curve of the nodule growth model.

[0009] Preferably, the nodule growth model is selected from the exponential growth model, the Gompertz growth model, the Logistic growth model, or a hybrid growth model composed of them.

[0010] Preferably, the expression for the exponential growth model is: in, t For time, t 0 represents the initial time. V ( t (time) t Nodule volume at time V 0 represents the initial volume. k1 The growth rate constant is e The base of the natural exponential function; The expression for the Gompertz growth model is: in, k2 The growth rate constant is b This is a parameter related to the change of growth rate over time; The expression for the Logistic growth model is: in, K For environmental carrying capacity, A For parameters related to the initial conditions, k2 is the growth rate constant.

[0011] Preferably, the expression for the hybrid growth model is: in, t For time, t 0 represents the initial time. t 1. To change the time threshold for characterizing nodule growth using different models, V ( t (time) t Nodule volume at time V 0 represents the initial volume. k1, k2 The growth rate constant is e The base of the natural exponential function; fort Nodule volume at time 1 b This is a parameter related to the change of growth rate over time; K For environmental carrying capacity, A These are parameters related to the initial conditions; The parameters predicted by the prediction model include t 1. k1, k2, b, K, A, V 0、 At least one of them.

[0012] Preferably, the feature extraction network is selected from ResNet, DenseNet, or MLP; The time series model is selected from LSTM, RNN or GRU.

[0013] Preferably, the clinical characteristic data are selected from at least one of age, gender, smoking history, and family medical history.

[0014] This invention also provides an artificial intelligence-based lung nodule follow-up management system, comprising: The doubling time prediction module integrates the aforementioned AI-based lung nodule growth prediction system to obtain prediction results for lung nodule growth. An image encoder is configured to process the patient's CT image data to obtain image features; A text encoder is configured to process structured images and / or structured clinicopathological information of patients to obtain text features; The image-text alignment module is configured to perform an image-text alignment operation on the image features and the image features to obtain aligned features. The follow-up plan formulation module is configured to input the aligned features and the predicted results of lung nodule growth into the large language generation model and output a follow-up plan.

[0015] Preferably, the large language generation model is selected from the BERT series model and the GPT series model; The follow-up plan includes at least one of the following: follow-up opinions, intervention opinions, and precautions.

[0016] This invention provides, for the first time, a method and system for predicting pulmonary nodule growth. It can obtain predicted values ​​of parameters related to pulmonary nodule growth based on multiple CT image data, generate predicted growth curves, and predict nodule doubling time. Furthermore, this invention also provides a follow-up management system that can automatically generate follow-up plans based on this predictive information, combined with CT image data, image reports, and clinicopathological information. The technical solution of this invention achieves the following beneficial technical effects: 1. Automation and intelligence: The system of this invention uses artificial intelligence algorithms to predict nodule growth models and nodule doubling time, and alerts high-risk nodules, which can reduce the subjectivity of human judgment.

[0017] 2. Accurate prediction of nodule doubling time: The system of the present invention incorporates CT image data from multiple time points into the model, taking into account the development trend of lung nodules. At the same time, in the preferred scheme, a hybrid growth model is constructed that uses different models to simulate nodule growth according to different nodule growth stages. This can more accurately predict the growth curve of nodules, thereby obtaining more accurate nodule doubling time.

[0018] 3. Multimodal Information Integration: The follow-up system of this invention fully utilizes image data and structured text information, extracting image features and text features through image encoders and text encoders respectively, and performing image-text alignment. The large language generation model can comprehensively analyze multimodal features, fully assess the malignancy risk of pulmonary nodules, and improve the accuracy and reliability of prediction.

[0019] 4. Personalized Intervention: Based on the doubling time prediction results and comprehensive analysis, personalized intervention recommendations and precautions are generated. Different follow-up plans are developed for patients with different risk levels, optimizing the allocation of medical resources and improving patient satisfaction.

[0020] 5. Dynamic Adjustment During Follow-up: New information is continuously collected during the follow-up process, the analysis process is constantly updated, and the follow-up plan and intervention measures are dynamically adjusted. This allows for timely detection of changes in nodules, ensuring effective intervention at an early stage, and improving patient survival rates and treatment outcomes.

[0021] In summary, this invention provides for the first time a system capable of predicting the growth curve of pulmonary nodules, accurately predicting the doubling time of nodules, and generating a follow-up plan, which can effectively improve the survival rate and treatment effect of patients and has great application prospects.

[0022] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.

[0023] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the workflow of the AI-based lung nodule growth prediction system in Example 1; Figure 2 This is a schematic diagram of the workflow of the AI-based lung nodule follow-up management system in Example 2. Detailed Implementation

[0025] It should be noted that the algorithms for data acquisition, transmission, storage and processing steps not specifically described in the embodiments, as well as the hardware structures and circuit connections not specifically described, can all be implemented using content already disclosed in the prior art.

[0026] Example 1: Artificial Intelligence-Based Lung Nodule Growth Prediction System The system in this embodiment includes: The input module is configured to input a patient's CT image data and clinical characteristic data. The CT image data includes at least two sets of data collected at different time points. The patient may or may not have undergone surgery, in which the surgery is pulmonary nodule ablation or pulmonary nodule resection. For patients who have undergone surgery, the CT image data includes at least one set of preoperative CT image data and at least one set of follow-up CT image data. The clinical characteristic data is selected from at least one of age, sex, smoking history, and family medical history.

[0027] The feature extraction module is configured to use a feature extraction network to extract features from the CT image data at each time point. The prediction module is configured to use a time series model to fuse the features obtained by the feature extraction module, predict the parameters of the nodule growth model, and obtain the prediction curve of the nodule growth model. As a preferred embodiment, the prediction module also calculates the predicted value of the nodule doubling time based on the prediction curve of the nodule growth model.

[0028] The choice of feature extraction network includes, but is not limited to, ResNet, DenseNet, or MLP. Feature extraction networks can capture subtle features in images, including shape features of lung nodules (such as roundness, irregularity, etc.), size features (such as diameter, volume, etc.), texture features (such as grayscale texture variations inside the nodule), and edge features (such as edge sharpness, smoothness, etc.).

[0029] The choice of time series models includes, but is not limited to, LSTM, RNN, or GRU. LSTM and other time series models can process sequential data and effectively capture the growth information and trends of lung nodules at different time points, providing a rich data foundation for growth model prediction and doubling time prediction. For example, lung nodule features (including imaging features and clinical information features) extracted at different time points can be arranged into sequential data and input into an LSTM model. LSTM can remember information from previous time points through its internal memory units, thereby better predicting the subsequent growth of lung nodules, especially demonstrating good control over the connection and transition between different stages in mixed growth models.

[0030] Nodule growth models can be selected based on the patient's condition, including but not limited to exponential growth models, Gompertz growth models, logistic growth models, or hybrid growth models composed of these.

[0031] The expression for the exponential growth model is: in, t For time, t 0 represents the initial time. V ( t (time) t Nodule volume at time V 0 represents the initial volume. k1 The growth rate constant is e The base of the natural exponential function; In the Gompertz growth model, the growth rate of lung nodules is not simply proportional to their current volume. Instead, the growth rate is gradually suppressed as the nodule volume increases. The model is expressed as follows: in, k2 This is a parameter describing the maximum growth rate of nodules (i.e., the growth rate constant). b This is a parameter related to the change of growth rate over time; Unlike the exponential growth model, the growth curve of lung nodules in the Gompertz growth model shows relatively rapid growth in the early stage, but then the growth rate gradually slows down due to factors such as its own volume, exhibiting an "S"-shaped curve characteristic of rapid growth followed by slow growth.

[0032] The Logistic growth model considers the influence of limiting factors such as carrying capacity on lung nodule growth. It assumes that the lung microenvironment containing the nodule has a maximum carrying capacity (i.e., environmental carrying capacity) for nodule growth; when the nodule volume approaches this carrying capacity, the growth rate tends to zero. The model is expressed as: in,K For environmental carrying capacity, A For parameters related to the initial conditions, k2 is the growth rate constant.

[0033] In real-world cases, the growth of pulmonary nodules may not simply follow one of the aforementioned classic growth models. For example, in some cases, the early stages of nodule growth may more closely resemble the exponential growth model, while in the middle and later stages, due to constraints from various factors such as the lung microenvironment, it may exhibit characteristics similar to the Gompertz or Logistic growth model. Therefore, as a preferred approach, a hybrid growth model can be constructed based on the growth characteristics of pulmonary nodules, assuming that in the early stages of nodule growth, its growth more closely follows the exponential growth model, and setting a time threshold. t 1. When the time threshold of the exponential growth phase has passed t After step 1, nodule growth enters either the Gompertz growth or Logistic growth phase. The expression for the hybrid growth model is: in, t For time, t 0 represents the initial time. t 1. To change the time threshold for characterizing nodule growth using different models, V ( t (time) t Nodule volume at time V 0 represents the initial volume. k1, k2 The growth rate constant is e The base of the natural exponential function; for t Nodule volume at time 1 b This is a parameter related to the change of growth rate over time; K For environmental carrying capacity, A These are parameters related to the initial conditions; The model parameters include t 1. k1, k2, b, K, A, V 0、 All of these can be predicted using predictive models.

[0034] In this embodiment, the method for predicting nodule growth trends based on the hybrid growth model described above can more accurately describe the actual growth of lung nodules throughout the entire growth cycle because it takes into account different growth mechanisms and limiting factors at different stages.

[0035] As a preferred approach, in this embodiment, the follow-up period can be one month, predicting the changes in nodules over 1 to 24 months to obtain a continuous change curve of nodule volume.

[0036] A flowchart illustrating the nodule doubling time prediction using the system of this embodiment is shown below. Figure 1 As shown in the figure, the doubling time of a nodule is an important indicator for assessing its growth rate. The shorter the doubling time, the faster the nodule grows, which may indicate an increased likelihood of malignancy. Validated on real-world datasets, this embodiment can accurately calculate the doubling time of a nodule based on the predicted nodule growth curve, determine whether there are any abnormalities in nodule growth, provide a reference for clinicians to determine follow-up frequency or intervention measures, and assist in diagnosis to assess the benign or malignant risk of the nodule.

[0037] Example 2: Artificial Intelligence-Based Lung Nodule Follow-up Management System Based on Example 1, this example further provides a follow-up management system, including: The doubling time prediction module integrates the AI-based lung nodule growth prediction system described in Example 1, used to obtain prediction results for lung nodule growth. These prediction results include the nodule growth curve and the nodule's doubling time.

[0038] The image encoder is configured to process the patient's CT image data to obtain image features, such as capturing key features of lung nodules in the images, such as their shape, size, and density.

[0039] A text encoder is configured to process a patient’s structured images and / or structured clinicopathological information to obtain text features; for example, extracting information such as nodule location and description from image reports, and obtaining key data such as the patient’s medical history and symptoms from clinical records.

[0040] The image-text alignment module is configured to perform image-text alignment operations on the image features and the image features to obtain aligned features; image-text alignment can better integrate multimodal information.

[0041] The follow-up plan formulation module is configured to input the aligned features and the predicted results of lung nodule growth into the large language generation model and output a follow-up plan.

[0042] The flowchart illustrating the system's process for generating follow-up plans in this embodiment is as follows: Figure 2 As shown.

[0043] Among them, the selection of large language generation models includes, but is not limited to, BERT series models and GPT series models.

[0044] The follow-up plan includes at least one of the following: follow-up recommendations, intervention recommendations, and precautions. For example, if the prediction results show that the nodule has a high risk of doubling in the short term, the system will automatically issue an alert to remind the doctor to intervene in a timely manner. Intervention recommendations may include immediate further imaging examinations, biopsies, or surgical evaluations; meanwhile, for low- to intermediate-risk nodules, the system will develop a personalized follow-up plan. Based on the nodule's growth rate and risk level, a reasonable follow-up interval and examination items are determined. For example, for slow-growing, low-risk nodules, the follow-up interval can be appropriately extended to reduce unnecessary imaging examinations and lower the patient's risk of radiation exposure. Precautions may include close monitoring of symptom changes and avoiding overexertion.

[0045] As can be seen from the above embodiments, this invention provides for the first time a system capable of accurately predicting the growth curve and doubling time of pulmonary nodules, and further, based on this system, a system capable of automatically generating follow-up plans is constructed. It has the advantages of high automation, accurate prediction results and follow-up plans, and the ability to be dynamically updated in real time. It can promptly detect changes in nodules, ensuring effective intervention at an early stage, thereby improving patient survival rates and treatment outcomes.

Claims

1. A pulmonary nodule growth prediction system based on artificial intelligence, characterized in that: include: An input module is configured to input CT image data and clinical characteristic data of a patient, wherein the CT image data includes at least two sets of data collected at different time points; The feature extraction module is configured to extract features using a feature extraction network for the CT image data at each time point; The prediction module is configured to use a time series model to fuse the features obtained by the feature extraction module, predict the parameters of the nodule growth model, and obtain a prediction curve of the nodule growth model.

2. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 1, characterized in that: The patient is a patient who has undergone or has not undergone surgery, and the surgery is pulmonary nodule ablation or pulmonary nodule resection; When the patient has undergone surgery, the CT image data includes at least one set of preoperative CT image data and at least one set of CT image data during a follow-up period.

3. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 1, characterized in that: The prediction module also calculates the predicted value of nodule doubling time based on the prediction curve of the nodule growth model.

4. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 1, characterized in that: The nodule growth model is selected from an exponential growth model, a Gompertz growth model, a Logistic growth model or a mixed growth model thereof.

5. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 4, characterized in that: The expression of the exponential growth model is: in, t For time, t 0 is the initial time, V ( t ) is the time t The nodule volume at V 0 is the initial volume, k1 is the growth rate constant, e The base of the natural exponential function; The expression of the Gompertz growth model is: in, k2 is the growth rate constant, b is a parameter related to the change of growth rate over time; The expression of the Logistic growth model is: in, K is the environmental carrying capacity, A are parameters related to the initial conditions, k2 is the growth rate constant.

6. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 4, characterized in that: The expression of the mixed growth model is: in, t For time, t 0 is the initial time, t 1 is the time threshold for changing different models to characterize nodule growth, V ( t ) is the time t The nodule volume at V 0 is the initial volume, k1, k2 is the growth rate constant, e The base of the natural exponential function; for t Nodule volume at 1 o'clock, b is a parameter related to the change of growth rate over time; K is the environmental carrying capacity, A are parameters related to the initial conditions; The parameters predicted by the prediction model include t 1. k1, k2, b, K, A, V 0. At least one of .

7. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 1, characterized in that: The feature extraction network is selected from ResNet, DenseNet or MLP; The time series model is selected from LSTM, RNN or GRU.

8. The artificial intelligence-based pulmonary nodule growth prediction system according to claim 1, characterized in that: The clinical characteristic data is selected from at least one of age, gender, smoking history, and family medical history.

9. A pulmonary nodule follow-up management system based on artificial intelligence, characterized in that: include: A doubling time prediction module, integrated with the artificial intelligence-based pulmonary nodule growth prediction system according to any one of claims 1 to 8, for obtaining a prediction result of pulmonary nodule growth; The image encoder is configured to process the CT image data of the patient to obtain image features; A text encoder is configured to process the patient's structured images and / or structured clinical pathology information to obtain text features; An image-text alignment module is configured to perform an image-text alignment operation on the image features and the image features to obtain aligned features; The follow-up plan formulation module is configured to input the aligned features and the prediction results of the lung nodule growth into a large language generation model and output a follow-up plan.

10. The pulmonary nodule follow-up management system according to claim 9, characterized in that: The large language generation model is selected from the BERT series model and the GPT series model; The follow-up plan includes at least one of the following information: follow-up opinions, intervention opinions, and precautions.

Images