Method and system for constructing pulmonary infection risk prediction model based on machine learning

Abstract

The invention is suitable for the technical field of model prediction, and provides a method and system for constructing a pulmonary infection risk prediction model based on machine learning, and the method comprises the following steps: collecting data information, and carrying out the standardization processing of the data information, the data information comprising structured data, image data, time series data and environment data; performing multi-modal feature extraction based on the data information to obtain an image feature F1, a physiological time sequence feature F2 and an environment feature F3; performing multi-modal feature fusion, determining weights of the image features, the physiological time sequence features and the environment features, and adaptively adjusting the weights of the environment features based on seasonal factors; processing class imbalance through focus loss, determining a time decay weight according to infection latency characteristics, and determining a total loss function; and carrying out model training by adopting a multi-layer perceptron to obtain a pulmonary infection risk prediction model. According to the invention, by introducing the seasonal factor and the seasonal sensitivity coefficient, adaptive adjustment of the environmental feature weight is realized.

Description

Technical Field

[0001] The present invention relates to the field of model prediction technology, and in particular to a method and system for constructing a lung infection risk prediction model based on machine learning. Background Art

[0002] Pulmonary infections, as a common respiratory disease, are associated with high morbidity and mortality, particularly among the elderly, children, and those with compromised immune systems. Despite advances in diagnosis and treatment, the continuous advancement of medical technology, early prediction and precise intervention for pulmonary infections remain significant clinical challenges. Traditional pulmonary infection risk assessment relies primarily on physician experience and limited biomarker testing. These methods often suffer from subjectivity, inaccuracy, and a failure to fully consider multiple influencing factors. In recent years, the rapid development of big data, artificial intelligence, and machine learning technologies has made it possible to leverage multimodal data for disease risk prediction. These data, drawn from a wide range of sources and containing a wealth of physiological, pathological, and environmental information, provide strong support for the construction of more accurate and comprehensive pulmonary infection risk prediction models. However, how to effectively integrate this multimodal data, extract valuable features, and address issues such as class imbalance and temporal dynamics in the data remain hot topics and challenges. Therefore, a method and system for constructing a pulmonary infection risk prediction model based on machine learning is needed to address these challenges. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a method and system for constructing a lung infection risk prediction model based on machine learning to solve the problems existing in the above-mentioned background technology.

[0004] The present invention is implemented as follows: a method for constructing a lung infection risk prediction model based on machine learning, the method comprising the following steps:

[0005] Collecting data information and performing standardization processing on the data information, wherein the data information includes structured data, image data, time series data and environmental data;

[0006] Multimodal feature extraction is performed based on data information to obtain image features F1, physiological time series features F2, and environmental features F3;

[0007] Perform multimodal feature fusion to determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. Adaptively adjust the weight of environmental features based on seasonal factors: w3' = w3 × (1 + α × S(t)), where α is the seasonal sensitivity coefficient and S(t) is the seasonal factor.

[0008] The class imbalance is handled by focal loss, the time decay weight is determined according to the characteristics of infection incubation period, and the total loss function is determined;

[0009] A multi-layer perceptron or Transformer architecture is used for model training to obtain a lung infection risk prediction model.

[0010] As a further solution of the present invention: the step of performing multimodal feature extraction based on data information to obtain image features F1, physiological time series features F2, and environmental features F3 specifically includes:

[0011] The texture features and morphological features are determined based on the image data. The texture features are obtained by calculating the contrast and entropy of the gray-level co-occurrence matrix, and the morphological features are obtained by the volume and sphericity of the lesion.

[0012] Determine time domain features and frequency domain features based on time series data. Time domain features include mean, standard deviation and trend, and frequency domain features are frequency band energy ratios.

[0013] An environmental risk score is determined based on various environmental parameters in the environmental data.

[0014] As a further solution of the present invention, the steps of performing multimodal feature fusion, determining the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively, and adaptively adjusting the weights of environmental features based on seasonal factors, specifically include:

[0015] Determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3;

[0016] Determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle;

[0017] The weights of environmental features are adaptively adjusted based on seasonal factors, and the weights are normalized after adjustment so that the sum of the adjusted weights is 1.

[0018] As a further solution of the present invention, the steps of processing category imbalance by focal loss, determining time decay weights according to infection incubation period characteristics, and determining the total loss function specifically include:

[0019] Determine the focal loss function and handle class imbalance through focal loss;

[0020] Determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time;

[0021] The total loss function is obtained based on the focal loss function and the time decay weight.

[0022] As a further solution of the present invention: the step of performing model training to obtain a lung infection risk prediction model specifically includes:

[0023] The input data of the model is determined to be the fusion feature FZ, and the output is the infection risk probability y;

[0024] Minimize the total loss and update the model parameters through the gradient descent method;

[0025] AUC-ROC, F1-score, precision and recall were used to evaluate the model performance.

[0026] As a further solution of the present invention: when standardizing data information, the image data is normalized to the [0,1] interval, and the feature matrix is extracted using 3D convolution; the time series data is generated using the sliding window method; and the environmental data is standardized using the Z-score.

[0027] Another object of the present invention is to provide a system for constructing a lung infection risk prediction model based on machine learning, the system comprising:

[0028] A data information acquisition module is used to collect data information and perform standardization processing on the data information. The data information includes structured data, image data, time series data and environmental data;

[0029] A multimodal feature extraction module is used to extract multimodal features based on data information to obtain image features F1, physiological time series features F2 and environmental features F3;

[0030] The multimodal feature fusion module is used to perform multimodal feature fusion and determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. The weight of environmental features is adaptively adjusted based on seasonal factors: w3'=w3×(1+α×S(t)), where α is the seasonal sensitivity coefficient and S(t) is the seasonal factor.

[0031] The loss function determination module is used to handle class imbalance through focal loss, determine the time decay weight according to the characteristics of infection incubation period, and determine the total loss function;

[0032] The prediction model building module is used to use a multi-layer perceptron or Transformer architecture to perform model training to obtain a lung infection risk prediction model.

[0033] As a further solution of the present invention: the multimodal feature extraction module includes:

[0034] An image feature extraction unit is used to determine texture features and morphological features based on image data. Texture features are obtained by calculating contrast and entropy values using a gray-level co-occurrence matrix, and morphological features are obtained by calculating the volume and sphericity of the lesion.

[0035] A time series feature extraction unit is used to determine time domain features and frequency domain features based on time series data. The time domain features include mean, standard deviation and trend, and the frequency domain features are frequency band energy ratios.

[0036] The environmental feature extraction unit is used to determine the environmental risk score based on various environmental parameters in the environmental data.

[0037] As a further solution of the present invention: the multimodal feature fusion module includes:

[0038] A fusion feature construction unit is used to determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3;

[0039] Seasonal factor determination unit, used to determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle;

[0040] The weight normalization unit is used to adaptively adjust the weights of environmental features based on seasonal factors, and perform weight normalization after adjustment so that the sum of the adjusted weights is 1.

[0041] As a further solution of the present invention: the loss function determination module includes:

[0042] A focal loss function unit is used to determine the focal loss function and handle category imbalance through focal loss;

[0043] The time decay weight unit is used to determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time;

[0044] The total loss function unit is used to obtain the total loss function based on the focal loss function and the time decay weight.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] By introducing seasonal factors and seasonal sensitivity coefficients, the present invention achieves adaptive adjustment of environmental feature weights, enabling the model to better adapt to seasonal changes in lung infection risk. The present invention employs a focal loss function, which effectively mitigates the class imbalance problem by increasing the penalty for difficult-to-classify samples and improves the model's ability to identify positive samples. Furthermore, the present invention also determines time-decay weights based on the characteristics of the infection incubation period. By introducing time-decay weights, the model can dynamically reflect the changing trends in infection risk over time, improving the timeliness and accuracy of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Flowchart of the method for building a lung infection risk prediction model based on machine learning.

[0048] Figure 2 Flowchart for extracting multimodal features in a method for building a lung infection risk prediction model based on machine learning.

[0049] Figure 3 Flowchart of multimodal feature fusion in the method of building a lung infection risk prediction model based on machine learning.

[0050] Figure 4 Flowchart for determining the total loss function in the method for building a lung infection risk prediction model based on machine learning.

[0051] Figure 5 Flowchart for model training in the method for building a lung infection risk prediction model based on machine learning.

[0052] Figure 6 Schematic diagram of the system for building a lung infection risk prediction model based on machine learning. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below with reference to 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 intended to limit the present invention.

[0054] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a lung infection risk prediction model based on machine learning, the method comprising the following steps:

[0056] S100, collecting data information and performing standardization processing on the data information, wherein the data information includes structured data, image data, time series data and environmental data;

[0057] S200, performing multimodal feature extraction based on the data information to obtain image features F1, physiological time series features F2, and environmental features F3;

[0058] S300, multimodal feature fusion is performed to determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. The weight of the environmental feature is adaptively adjusted based on seasonal factors: w3' = w3 × (1 + α × S(t)), where α is the seasonal sensitivity coefficient, α is a constant, and S(t) is the seasonal factor.

[0059] S400, handles class imbalance through focal loss, determines time decay weight according to infection incubation period characteristics, and determines the total loss function;

[0060] S500 uses a multi-layer perceptron or Transformer architecture to train models to obtain a lung infection risk prediction model.

[0061] It should be noted that existing lung infection risk assessments rely primarily on physicians' clinical experience and limited biomarker testing. These methods often suffer from subjectivity, inaccuracy, and difficulty in fully considering multiple influencing factors. The present invention aims to address these issues.

[0062] In an embodiment of the present invention, a large amount of data information is first collected and standardized. The data information includes structured data, image data, time series data and environmental data, wherein the structured data includes age, gender, underlying diseases (COPD, diabetes), immunosuppression status, recent surgical history, etc. The image data includes lung CT scans (DICOM format), lung segmentation is performed through pre-trained UNet, time series data includes continuously monitored respiratory rate, heart rate, body temperature (sampling frequency ≥ 1 time / hour), and environmental data includes ward air microbial concentration, temperature and humidity sensor data. Through multimodal feature extraction and fusion technology, the effective integration of different types of data is achieved, and the prediction accuracy of the model is improved. Then, multimodal feature extraction is performed based on the data information to obtain image features F1, physiological time series features F2 and environmental features F3. Next, multimodal feature fusion can be performed. First, the weights of image features, physiological time series features, and environmental features need to be determined as w1, w2, and w3, respectively. The weights of environmental features are adaptively adjusted based on seasonal factors, w3'=w3×(1+α×S(t)), where w3' is the adjusted weight of environmental features. Environmental factors have an important impact on the occurrence and development of lung infection, but traditional models often ignore this and use fixed feature weights. This solution introduces seasonal factors and seasonal sensitivity coefficients to achieve adaptive adjustment of environmental feature weights, enabling the model to better adapt to changes in lung infection risk in different seasons. Class imbalance is then handled through focal loss. In lung infection risk prediction, positive samples (i.e., patients who are actually infected) are often far less than negative samples (i.e., patients who are not infected), causing the model to easily favor negative samples during training, reducing prediction accuracy. The present invention uses a focal loss function to effectively alleviate the class imbalance problem by increasing the penalty for difficult-to-classify samples and improves the model's ability to recognize positive samples. In addition, the present invention will also determine the time decay weight according to the characteristics of the infection incubation period. The incubation period characteristics of lung infection mean that the risk of infection will change over time. Traditional models often ignore this point and adopt static prediction methods. The present invention introduces time decay weights to enable the model to dynamically reflect the changing trend of infection risk over time, thereby improving the timeliness and accuracy of the prediction. Finally, a multi-layer perceptron or Transformer architecture is used to train the model to obtain a lung infection risk prediction model. In view of the complexity and high dimensionality of multimodal data, advanced neural network architectures such as multi-layer perceptrons or Transformers are used for model training. These architectures have powerful feature learning and representation capabilities, and can more effectively capture complex patterns and relationships in the data, thereby improving the prediction performance of the model.

[0063] like Figure 2As shown, as a preferred embodiment of the present invention, the step of performing multimodal feature extraction based on data information to obtain image features F1, physiological time series features F2 and environmental features F3 specifically includes:

[0064] S201, determining texture features and morphological features based on the image data, wherein the texture features are obtained by calculating the contrast and entropy of the gray-level co-occurrence matrix, and the morphological features are obtained by the volume and sphericity of the lesion;

[0065] S202, determining time domain features and frequency domain features based on the time series data, where the time domain features include mean, standard deviation, and trend, and the frequency domain features include frequency band energy ratio;

[0066] S203: Determine an environmental risk score based on each environmental parameter in the environmental data.

[0067] In an embodiment of the present invention, when extracting features, the first step is to determine texture features and morphological features based on the image data. Texture features and morphological features constitute image features. Texture features are obtained by calculating contrast and entropy using the gray-level co-occurrence matrix (GLCM). Morphological features are obtained by the volume and sphericity of the lesion. The second step is to determine time domain features and frequency domain features based on the time series data. Time domain features include mean, standard deviation, and trend. Frequency domain features are the main frequency band energy ratios after STFT transformation. Time domain features and frequency domain features constitute time series features. Finally, the environmental risk score is determined based on the various environmental parameters in the environmental data.

[0068] like Figure 3 As shown, as a preferred embodiment of the present invention, multimodal feature fusion is performed to determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively, and the steps of adaptively adjusting the weights of environmental features based on seasonal factors specifically include:

[0069] S301, determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3;

[0070] S302, determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle;

[0071] S303 , adaptively adjusting the environmental feature weights based on seasonal factors, and normalizing the weights after adjustment so that the sum of the adjusted weights is 1.

[0072] In the embodiment of the present invention, the fusion feature FZ = w1×F1+w2×F2+w3×F3, w1+w2+w3=1, and the seasonal factor S(t) simulates seasonal fluctuations based on a sine function. t is the current time (in days), t0 is the seasonal reference time, and T is the seasonal period (e.g., 365 days). The environmental feature weights are then adaptively adjusted based on seasonal factors. After the adjustment, the weights are normalized, and w1 and w2 are adaptively adjusted so that the sum of the adjusted weights is 1.

[0073] like Figure 4 As shown, as a preferred embodiment of the present invention, the steps of processing category imbalance by focal loss, determining time decay weight according to infection incubation period characteristics, and determining the total loss function specifically include:

[0074] S401, determining a focal loss function, and processing category imbalance through focal loss;

[0075] S402, determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time;

[0076] S403: Obtain a total loss function based on the focal loss function and the time decay weight.

[0077] In this embodiment of the present invention, a focal loss function is introduced to address class imbalance. A time decay weight wt is also determined based on the characteristics of the infection incubation period: wt = e-λ(tc-ts), where λ is the decay coefficient, a constant value, tc is the current time, and ts is the sample collection time. Finally, the focal loss function and the time decay weight are combined to form the total loss function.

[0078] like Figure 5 As shown, as a preferred embodiment of the present invention, the step of performing model training to obtain a lung infection risk prediction model specifically includes:

[0079] S501, determining that the input data of the model is the fusion feature FZ, and the output is the infection risk probability y;

[0080] S502, minimizing the total loss by gradient descent method and updating the model parameters;

[0081] S503, AUC-ROC, F1-score, precision and recall are used to evaluate model performance.

[0082] In this embodiment of the present invention, the model's input data is the fused features FZ, and its output is the infection risk probability y. The infection risk probability y for each data set must be pre-labeled. Gradient descent is then used to minimize the total loss and update the model parameters. Finally, model performance is evaluated using AUC-ROC, F1-score, precision, and recall. This approach prioritizes the ability to identify minority classes (high-risk patients) to improve prediction accuracy and clinical practicality.

[0083] In an embodiment of the present invention, when standardizing data information, the image data is normalized to the [0, 1] interval, and a feature matrix is extracted using 3D convolution; a sliding window method (window length = 24h) is used to generate a time series for the time series data; and the environmental data is standardized using the Z-score.

[0084] like Figure 6 As shown, an embodiment of the present invention further provides a system for constructing a lung infection risk prediction model based on machine learning, the system comprising:

[0085] The data information acquisition module 100 is used to collect data information and perform standardization processing on the data information. The data information includes structured data, image data, time series data and environmental data;

[0086] A multimodal feature extraction module 200 is configured to extract multimodal features based on the data information to obtain image features F1, physiological time series features F2, and environmental features F3;

[0087] The multimodal feature fusion module 300 is used to perform multimodal feature fusion, determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively, and adaptively adjust the weight of environmental features based on seasonal factors, w3'=w3×(1+α×S(t)), where α is the seasonal sensitivity coefficient and S(t) is the seasonal factor;

[0088] A loss function determination module 400 is used to process class imbalance using focal loss, determine time decay weights based on infection latency characteristics, and determine a total loss function;

[0089] The prediction model building module 500 is used to use a multi-layer perceptron or Transformer architecture to perform model training to obtain a lung infection risk prediction model.

[0090] In the embodiment of the present invention, the multimodal feature extraction module 200 includes:

[0091] An image feature extraction unit is used to determine texture features and morphological features based on image data. Texture features are obtained by calculating contrast and entropy values using a gray-level co-occurrence matrix, and morphological features are obtained by calculating the volume and sphericity of the lesion.

[0092] A time series feature extraction unit is used to determine time domain features and frequency domain features based on time series data. The time domain features include mean, standard deviation and trend, and the frequency domain features are frequency band energy ratios.

[0093] The environmental feature extraction unit is used to determine the environmental risk score based on various environmental parameters in the environmental data.

[0094] In the embodiment of the present invention, the multimodal feature fusion module 300 includes:

[0095] A fusion feature construction unit is used to determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3;

[0096] Seasonal factor determination unit, used to determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle;

[0097] The weight normalization unit is used to adaptively adjust the weights of environmental features based on seasonal factors, and perform weight normalization after adjustment so that the sum of the adjusted weights is 1.

[0098] In the embodiment of the present invention, the loss function determination module 400 includes:

[0099] A focal loss function unit is used to determine the focal loss function and handle category imbalance through focal loss;

[0100] The time decay weight unit is used to determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time;

[0101] The total loss function unit is used to obtain the total loss function based on the focal loss function and the time decay weight.

[0102] The above is only a detailed description of the preferred embodiments of the present invention, which is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0103] It should be understood that, although the various steps in the flow chart of each embodiment of the present invention are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0104] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0105] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered merely as exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A method for constructing a lung infection risk prediction model based on machine learning, characterized in that: The method comprises the following steps: Collecting data information and performing standardization processing on the data information, wherein the data information includes structured data, image data, time series data and environmental data; Multimodal feature extraction is performed based on data information to obtain image features F1, physiological time series features F2, and environmental features F3; Perform multimodal feature fusion to determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. Adaptively adjust the weight of environmental features based on seasonal factors: w3' = w3 × (1 + α × S(t)), where α is the seasonal sensitivity coefficient, S(t) is the seasonal factor, and w3' is the adjusted environmental feature weight. The class imbalance is handled by focal loss, the time decay weight is determined according to the characteristics of infection incubation period, and the total loss function is determined; A multi-layer perceptron or Transformer architecture is used for model training to obtain a lung infection risk prediction model.

2. The method for constructing a lung infection risk prediction model based on machine learning according to claim 1, characterized in that: The step of performing multimodal feature extraction based on data information to obtain image features F1, physiological time series features F2, and environmental features F3 specifically includes: The texture features and morphological features are determined based on the image data. The texture features are obtained by calculating the contrast and entropy of the gray-level co-occurrence matrix, and the morphological features are obtained by the volume and sphericity of the lesion. Determine time domain features and frequency domain features based on time series data. Time domain features include mean, standard deviation and trend, and frequency domain features are frequency band energy ratios. An environmental risk score is determined based on various environmental parameters in the environmental data.

3. The method for constructing a lung infection risk prediction model based on machine learning according to claim 1, characterized in that: Perform multimodal feature fusion to determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. Adaptively adjust the weights of environmental features based on seasonal factors. Specifically, the steps include: Determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3; Determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle; The weights of environmental features are adaptively adjusted based on seasonal factors, and the weights are normalized after adjustment so that the sum of the adjusted weights is 1.

4. The method for constructing a lung infection risk prediction model based on machine learning according to claim 1, characterized in that: The steps of processing category imbalance by using focal loss, determining time decay weights according to infection incubation period characteristics, and determining the total loss function specifically include: Determine the focal loss function and handle class imbalance through focal loss; Determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time; The total loss function is obtained based on the focal loss function and the time decay weight.

5. The method for constructing a lung infection risk prediction model based on machine learning according to claim 1, characterized in that: The step of performing model training to obtain a lung infection risk prediction model specifically includes: The input data of the model is determined to be the fusion feature FZ, and the output is the infection risk probability y; Minimize the total loss and update the model parameters through the gradient descent method; AUC-ROC, F1-score, precision and recall were used to evaluate the model performance.

6. The method for constructing a lung infection risk prediction model based on machine learning according to claim 1, characterized in that: When standardizing data information, the image data is normalized to the [0,1] interval, and the feature matrix is extracted using 3D convolution. The sliding window method is used to generate time series for time series data. The environmental data is standardized using the Z-score.

7. A system for constructing a lung infection risk prediction model based on machine learning, characterized in that: The system comprises: A data information acquisition module is used to collect data information and perform standardization processing on the data information. The data information includes structured data, image data, time series data and environmental data; A multimodal feature extraction module is used to extract multimodal features based on data information to obtain image features F1, physiological time series features F2 and environmental features F3; The multimodal feature fusion module is used to perform multimodal feature fusion and determine the weights of image features, physiological time series features, and environmental features as w1, w2, and w3, respectively. The weight of environmental features is adaptively adjusted based on seasonal factors: w3'=w3×(1+α×S(t)), where α is the seasonal sensitivity coefficient, S(t) is the seasonal factor, and w3' is the adjusted environmental feature weight. The loss function determination module is used to handle class imbalance through focal loss, determine the time decay weight according to the characteristics of infection incubation period, and determine the total loss function; The prediction model building module is used to use a multi-layer perceptron or Transformer architecture to perform model training to obtain a lung infection risk prediction model.

8. The system for constructing a lung infection risk prediction model based on machine learning according to claim 7, characterized in that: The multimodal feature extraction module includes: An image feature extraction unit is used to determine texture features and morphological features based on image data. Texture features are obtained by calculating contrast and entropy values using a gray-level co-occurrence matrix, and morphological features are obtained by calculating the volume and sphericity of the lesion. A time series feature extraction unit is used to determine time domain features and frequency domain features based on time series data. The time domain features include mean, standard deviation and trend, and the frequency domain features are frequency band energy ratios. The environmental feature extraction unit is used to determine the environmental risk score based on various environmental parameters in the environmental data.

9. The system for constructing a lung infection risk prediction model based on machine learning according to claim 7, characterized in that: The multimodal feature fusion module includes: A fusion feature construction unit is used to determine the fusion feature FZ, FZ = w1×F1+w2×F2+w3×F3; Seasonal factor determination unit, used to determine the seasonal factor S(t), t is the current time, t0 is the seasonal reference time, and T is the seasonal cycle; The weight normalization unit is used to adaptively adjust the weights of environmental features based on seasonal factors, and perform weight normalization after adjustment so that the sum of the adjusted weights is 1.

10. The system for constructing a lung infection risk prediction model based on machine learning according to claim 7, characterized in that: The loss function determination module includes: A focal loss function unit is used to determine the focal loss function and handle category imbalance through focal loss; The time decay weight unit is used to determine the time decay weight wt according to the characteristics of the infection incubation period, wt = e- λ (tc-ts), where λ is the attenuation coefficient, tc is the current time, and ts is the sample collection time; The total loss function unit is used to obtain the total loss function based on the focal loss function and the time decay weight.

Images