Deep learning auxiliary prevention and treatment management platform for latent tuberculosis infection

Abstract

The invention discloses a deep learning auxiliary prevention and treatment management platform for latent tuberculosis infection, and relates to the technical field of latent tuberculosis infection auxiliary prevention. Comprising a data acquisition and dynamic monitoring module, a data preprocessing and organization module, a conflict feature extraction and analysis module, a deep learning conflict evaluation module and a dynamic weight adjustment and optimization module, and it is ensured that each input can be dynamically monitored in the data fusion process, so that basic data is provided for subsequent conflict detection, feature extraction and intelligent evaluation. According to the method, multi-modal data fusion parameters are monitored in real time, conflict features are extracted, data source conflicts are intelligently evaluated by means of a deep learning model, the weight is dynamically adjusted, and the data fusion process is optimized. According to the scheme, the risk assessment accuracy of latent tuberculosis infection is improved, misdiagnosis and missed diagnosis are avoided, and precise medical treatment and personalized treatment implementation are promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of auxiliary prevention of latent tuberculosis infection, and specifically to a deep learning-assisted prevention and treatment management platform for latent tuberculosis infection. Background Art

[0002] Latent tuberculosis infection (LTBI) occurs when the human immune system controls and suppresses the activity of Mycobacterium tuberculosis (MTB). Infected individuals experience no obvious clinical symptoms, but may develop active TB when immunity declines. Deep learning-assisted prevention and treatment management of LTBI involves using deep learning technology to analyze and model large amounts of patient data (such as imaging data, genomic data, and clinical records) to assist in early screening, risk assessment, treatment decision-making, and management optimization for LTBI. Deep learning algorithms can automatically identify risk factors for LTBI, predict the likelihood of LTBI converting to active TB, and recommend precise prevention and treatment strategies based on individualized health status. This approach can improve the accuracy and efficiency of LTBI management, reduce the risk of conversion to active TB, and ultimately help control the spread of TB.

[0003] Existing technologies have the following shortcomings: When analyzing large amounts of patient data such as imaging data, genomic data, and clinical records, existing technologies usually fuse these data into comprehensive features to comprehensively assess the risk of latent tuberculosis infection transforming into active tuberculosis. However, when there are conflicts between the features of different data sources, the model may find it difficult to handle them accurately. For example, imaging data shows that the lung lesions are mild, while genomic data or clinical records indicate that the patient has a higher susceptibility. This conflict may cause the model to deviate from the weights assigned to these features, thereby affecting the final prediction results. If this conflict is not effectively handled, it may cause the model to overestimate or underestimate the risk of conversion, further leading to unnecessary long-term treatment, drug side effects, or low-risk patients being neglected, ultimately missing the opportunity for early intervention, leading to missed diagnosis of active tuberculosis, and thus affecting public health safety.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a deep learning-assisted prevention and treatment management platform for latent tuberculosis infection. This platform uses real-time monitoring of multimodal data fusion parameters, extracts conflicting features using feature engineering, and intelligently assesses data source conflicts using deep learning models. It dynamically adjusts the weights of each data source to optimize the data fusion process. This solution effectively improves the accuracy and robustness of latent tuberculosis infection conversion risk assessment, avoids misdiagnosis and missed diagnosis, and promotes the implementation of precision medicine and personalized treatment, thereby addressing the above-mentioned background technology issues.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: a deep learning-assisted prevention and treatment management platform for latent tuberculosis infection, comprising a data acquisition and dynamic monitoring module, a data preprocessing and organization module, a conflict feature extraction and analysis module, a deep learning conflict assessment module, and a dynamic weight adjustment and optimization module:

[0007] The data acquisition and dynamic monitoring module acquires the fusion parameter information of multimodal data in real time, ensuring dynamic monitoring of each input during the data fusion process, thereby providing basic data for subsequent conflict detection, feature extraction, and intelligent evaluation;

[0008] The data preprocessing and organization module organizes the acquired multimodal data fusion parameter information into a unified data set and preprocesses the multimodal data fusion parameter information in the data set to facilitate subsequent analysis;

[0009] The conflict feature extraction and analysis module uses feature engineering technology to extract key features that reflect conflicts in data fusion from preprocessed data, and analyzes the extracted key features to provide strong support for subsequent conflict detection;

[0010] The deep learning conflict assessment module inputs the analyzed key features into a pre-learned deep learning model and uses the learning model to identify whether there are conflicts during the data fusion process;

[0011] The dynamic weight adjustment and optimization module, after detecting conflicts in the data fusion process and evaluating their severity, dynamically adjusts the weights according to the reliability of each data source and the severity of the conflict, reducing the negative impact of conflicting data sources on the final decision, thereby improving the accuracy and robustness of the fusion results.

[0012] Preferably, the specific steps of organizing the acquired multimodal data fusion parameter information into a unified data set are as follows: first, the fusion parameters of each data source are labeled and normalized to conform to a unified data structure;

[0013] Next, the data is denoised, missing values processed, and anomaly detected to eliminate potential quality issues;

[0014] Subsequently, the data were scaled uniformly using standardization techniques to ensure that parameters from different data sources were comparable and consistent in subsequent analyses;

[0015] Finally, the preprocessed data set is used for further feature extraction, conflict detection, and input into deep learning models to ensure the accuracy and reliability of the analysis results.

[0016] Preferably, key features reflecting conflicts in data fusion are extracted from the preprocessed data through feature engineering technology. The extracted features include the change in contribution of each data source to the final decision during the multimodal fusion process and the change in error generated during the multimodal data fusion process. The change in contribution of each data source to the final decision during the multimodal fusion process and the change in error generated during the multimodal data fusion process are analyzed under the detection window to generate a data source contribution change reference value and a data fusion error reference value respectively. The data source contribution change reference value is used to quantify the change in the influence and contribution of each data source in the multimodal data fusion process. The data fusion error reference value is used to measure the change in inconsistency over time in the multimodal data fusion process.

[0017] Preferably, the specific steps of analyzing the contribution change of each data source to the final decision in the multimodal fusion process under the detection window to generate a reference value of the data source contribution change are as follows:

[0018] In the multimodal data fusion process, the contribution of each data source to the final decision will be continuously adjusted as the data input changes. To quantify the contribution of the data source at each time point, a dynamic weight function is introduced. The calculation is calculated by considering the feature importance of the data source and the response of the fusion model. The calculation expression is as follows:

[0019]

[0020] , where W i (t) is the contribution weight of the i-th data source to the decision at time t, f i (t) is the characteristic strength of the i-th data source at time t, α i (t) is the dynamic adjustment factor related to the i-th data source at time t, f j (t) is the characteristic strength of the j-th data source at time t, α j (t) is the dynamic adjustment factor associated with the jth data source at time t, and N is the total number of all data sources used in the multimodal data fusion process;

[0021] Next, we perform a time series analysis on the dynamic contribution of each data source, calculate the change in the contribution of each data source, and introduce a contribution change index to quantify the change in the contribution of the data source between the current time and the previous time. The calculation expression is as follows:

[0022]

[0023] , where C i (t) is the contribution change index, which represents the contribution change index of the i-th data source at time t, W i (t-1) is the contribution weight of the i-th data source to the decision at time t-1, β i is the conflict adjustment factor of the i-th data source, δ i (t) is a time-dependent factor, which indicates the impact of external environmental changes on the i-th data source at the current moment;

[0024] Finally, the conflict of all data sources in multimodal data fusion is evaluated as a whole, and the contribution change index C of each data source is calculated. i (t) is integrated to generate a reference value of data source contribution change. The calculation expression is as follows:

[0025]

[0026] , where Data Source It is the reference value of the change in data source contribution.

[0027] Preferably, the specific steps of analyzing the error changes generated during the multimodal data fusion process in the detection window to generate a data fusion error reference value are as follows:

[0028] In the first step of generating the data fusion error index, the error of each data source is calculated. The error of each data source is expressed as the deviation of the data source from the final prediction result. The error calculation expression is as follows:

[0029]

[0030] , where E i is the error of the ith data source, f(x p ) is the prediction result of the i-th data source on the p-th sample, is the actual value of the pth sample, τ i is the error amplification index of the i-th data source, n is the total number of samples in the dataset, that is, the number of samples processed by the model when making predictions;

[0031] After calculating the errors of each data source, the obtained errors are weighted and fused to obtain a comprehensive error value. The error of each data source is weighted according to its importance to the fusion result. The calculation expression is as follows:

[0032]

[0033] , where E total is the total error after fusion of all data sources, w i is the weight of the i-th data source, β i is the error amplification index of the i-th data source, and N is the total number of all data sources used in the multimodal data fusion process;

[0034] Finally, the fused error value is converted into a data fusion error reference value, which is used to quantify the degree of conflict in the multimodal data fusion process. The fused error is mapped using a nonlinear function. The calculation expression is as follows:

[0035]

[0036] , where Data Fusion is the reference value of data fusion error, γ i is the weight coefficient of the i-th data source, is a weighted nonlinear transformation of the error of the i-th data source.

[0037] Preferably, the analyzed data source contribution change reference value and data fusion error reference value are input into a pre-learned deep learning model, and a fusion conflict risk coefficient is generated by the deep learning model. The multimodal data fusion process is intelligently evaluated by the fusion conflict risk coefficient to identify whether there is a conflict in the data fusion process.

[0038] Preferably, the multimodal data fusion process is intelligently evaluated by a pre-learned deep learning model, and the fusion conflict risk coefficient generated when identifying whether there is a conflict in the data fusion process is compared with the pre-set fusion conflict risk coefficient to divide the multimodal data fusion process. The division steps are as follows:

[0039] If the fusion conflict risk coefficient is greater than the preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having data fusion conflict; if the fusion conflict risk coefficient is less than or equal to the preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having no data fusion conflict.

[0040] Preferably, after detecting conflicts in the data fusion process and evaluating their severity, the specific steps of dynamically adjusting weights according to the reliability of each data source and the severity of the conflict are as follows:

[0041] After detecting a conflict in the data fusion process, we first need to assess the severity of the conflict and quantitatively evaluate the reliability of each data source. The evaluation formula is as follows:

[0042]

[0043] , where S i is the conflict severity index of data source i, which indicates the severity of the conflict generated by data source i in the multimodal data fusion process. i is the conflict sensitivity parameter of data source i, which adjusts its response to conflicts. Conflict is the fusion conflict risk coefficient, Fusion ref is the reference threshold of the fusion conflict risk coefficient, is the exponential growth factor of data source i;

[0044] When evaluating the reliability of data sources, we combine multiple weighted normalization methods and add weight adjustments for historical performance. We also consider the weighted impact of data quality and historical performance, and balance their importance through a composite function. The calculation expression is as follows:

[0045]

[0046] , where R i is the reliability score of data source i, Q i is the quality score of data source i, H i is the historical performance score of data source i, θ i is the stability coefficient of each data source, reflecting the weighted impact of historical performance on reliability assessment, V i is the historical variability of data source i, indicating the stability of the data source in history. is the mean of the historical variability of all data sources, w1 and w2 are the quality scores Q of data source i i and the historical performance score H of data source i i The weight coefficient of

[0047] In the final dynamic weight adjustment, in addition to adjusting the conflict severity and reliability, a composite weighted adjustment function is introduced. In the adjustment process, exponential weighting is added, the interaction between conflict severity and data reliability is considered, and enhanced contrast adjustment is introduced.

[0048] The weight of each data source is dynamically adjusted based on the severity of the conflict and the reliability evaluation results of the data source. The calculation formula is as follows:

[0049]

[0050] , where w i ' is the adjusted weight of the i-th data source, E i is the original weight, σ i is the adjustment coefficient, which controls the adjustment sensitivity of data source i, H i is the nonlinear weighting coefficient of the reliability score, μ i is the conflict severity adjustment coefficient, which controls the impact of conflict on weight adjustment, ρ i is a weighted index of conflict severity, adjusting the nonlinear effect of conflict severity on the weight.

[0051] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0052] The present invention effectively solves the data conflict problem existing in the multimodal data fusion process by acquiring multimodal data fusion parameters in real time and performing dynamic monitoring, thereby improving the accuracy and robustness of the risk assessment of converting latent tuberculosis infection to active tuberculosis. Specifically, the key features reflecting data conflicts are extracted through feature engineering technology, and intelligent evaluation is performed with the help of a pre-learned deep learning model to promptly identify conflicts between data sources. After the conflict is detected, the weight of each data source is dynamically adjusted based on the reliability of each data source and the severity of the conflict, reducing the negative impact of the conflicting data source on the final decision. This solution not only optimizes the accuracy of the multimodal data fusion process and avoids the risk of misdiagnosis and missed diagnosis, but also improves the adaptability of the model in a complex clinical data environment, ultimately promoting the implementation of precision medicine and personalized treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0054] Figure 1 This is a module diagram of the deep learning-assisted prevention and treatment management platform for latent tuberculosis infection of the present invention. DETAILED DESCRIPTION

[0055] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0056] The present invention provides Figure 1 The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection shown in the figure includes a data acquisition and dynamic monitoring module, a data preprocessing and organization module, a conflict feature extraction and analysis module, a deep learning conflict assessment module, and a dynamic weight adjustment and optimization module:

[0057] The data acquisition and dynamic monitoring module acquires the fusion parameter information of multimodal data in real time, ensuring dynamic monitoring of each input during the data fusion process, thereby providing basic data for subsequent conflict detection, feature extraction, and intelligent evaluation;

[0058] In multimodal data fusion, the quality and characteristics of various data sources, such as imaging data, genomic data, and clinical records, may change over time or at different stages of data collection. Real-time acquisition of fusion parameter information involves capturing multi-dimensional information such as the quality score, completeness, timeliness, and impact of each data source on the model.

[0059] The data preprocessing and organization module organizes the acquired multimodal data fusion parameter information into a unified data set and preprocesses the multimodal data fusion parameter information in the data set to facilitate subsequent analysis;

[0060] When organizing the acquired multimodal data fusion parameter information into a unified data set, the parameters from different data sources need to be formatted and standardized first to ensure that they can be compared and analyzed under the same framework. The specific steps include: First, the fusion parameters of each data source (such as the resolution of imaging data, the accuracy of genomic data, the timeliness of clinical data, etc.) are labeled and normalized to conform to a unified data structure, such as a table, time series, or data stream. Next, the data is denoised, missing values are processed, and anomalies are detected to eliminate potential quality issues such as missing fields, incorrect measurements, or inconsistent timestamps. Subsequently, the data is uniformly scaled using standardization techniques (such as Z-score standardization or Min-Max scaling) to ensure that the parameters of different data sources are comparable and consistent in subsequent analysis. The preprocessed data set is used for further feature extraction, conflict detection, and input into deep learning models to ensure the accuracy and reliability of the analysis results.

[0061] The creation of a data set involves integrating parameter information from different data sources (such as imaging, genomics, and clinical records) into a standardized, structured format for easy subsequent processing. This information may include the characteristic types of each data source (such as the resolution of imaging data, the accuracy of genomic data, and the timeliness of clinical data), as well as the relationships or interactions between them. The key to this step is to organize and uniformly format the data so that these parameters can be easily accessed in subsequent steps, providing sufficient data support for conflict detection and feature extraction.

[0062] Data preprocessing is a critical step in machine learning and deep learning. It involves removing redundant information, handling missing values, standardizing data, and removing noise. For example, clinical data may contain missing values, imaging data may be noisy, and genomic data may contain sequencing errors. The goal of preprocessing is to unify the data format and optimize its learning properties. Techniques such as missing value interpolation, normalization, data smoothing, and denoising may be applied at this stage to ensure that the data quality entering the feature extraction and analysis phase does not affect the final fusion results.

[0063] The conflict feature extraction and analysis module uses feature engineering technology to extract key features that reflect conflicts in data fusion from preprocessed data, and analyzes the extracted key features to provide strong support for subsequent conflict detection;

[0064] Through feature engineering technology, key features reflecting conflicts in data fusion are extracted from the preprocessed data. The extracted features include the changes in the contribution of each data source to the final decision during the multimodal fusion process and the changes in the errors generated during the multimodal data fusion process. The changes in the contribution of each data source to the final decision during the multimodal fusion process and the changes in the errors generated during the multimodal data fusion process are analyzed under the detection window to generate a data source contribution change reference value and a data fusion error reference value respectively. The data source contribution change reference value is used to quantify the changes in the influence and contribution of each data source in the multimodal data fusion process. The data fusion error reference value is used to measure the changes in the inconsistency in the multimodal data fusion process over time.

[0065] When the contribution of each data source to the final decision changes exponentially during multimodal fusion, it typically indicates potential data conflict within the current multimodal data fusion process. This is because, during a normal fusion process, the contributions of data sources should remain relatively stable, unless the performance of a particular data source undergoes a sudden change or its relationship with other data sources undergoes a significant shift. If the contribution of a particular data source increases exponentially, it may indicate significant inconsistencies or anomalies between that data source and the others. For example, imaging data may suddenly become overly important due to technical glitches or quality issues, leading the model to over-reliance on that data source in its decision-making. Alternatively, the contribution of genomic data may increase dramatically at a certain point, indicating a break in its expected relationship with imaging or clinical data, thereby affecting the fusion results. Conversely, a sharp drop in the contribution of certain data sources may also indicate a breakdown in synergy between data sources or an underestimation of the importance of that data source in the fusion process. Therefore, exponential changes often reveal potential conflicts between data sources, which may lead to incorrect predictions or decisions by the model, thus compromising the accuracy and stability of the final model.

[0066] The specific steps for analyzing the contribution changes of each data source to the final decision in the multimodal fusion process under the detection window to generate the reference value of the data source contribution change are as follows:

[0067] In the multimodal data fusion process, the contribution of each data source to the final decision will be continuously adjusted as the data input changes. To quantify the contribution of the data source at each time point, a dynamic weight function is introduced. The calculation is calculated by considering the feature importance of the data source and the response of the fusion model. The calculation expression is as follows:

[0068]

[0069] , where W i (t) is the contribution weight of the i-th data source to the decision at time t, f i (t) is the characteristic strength of the i-th data source at time t (e.g., data quality, signal-to-noise ratio, etc.), α i (t) is the dynamic adjustment factor related to the i-th data source at time t, reflecting the influence of the data source on the final decision, f j (t) is the characteristic strength of the j-th data source at time t, α j (t) is the dynamic adjustment factor associated with the jth data source at time t, and N is the total number of all data sources used in the multimodal data fusion process;

[0070] The purpose of this step is to provide basic data for the subsequent calculation of contribution changes.

[0071] Next, we perform a time series analysis on the dynamic contribution of each data source, calculate the change in the contribution of each data source, and introduce a contribution change index to quantify the change in the contribution of the data source between the current time and the previous time. The calculation expression is as follows:

[0072]

[0073] , where C i (t) is the contribution change index, which represents the contribution change index of the i-th data source at time t, W i (t-1) is the contribution weight of the i-th data source to the decision at time t-1, β i is the conflict adjustment factor of the i-th data source, δ i (t) is a time-dependent factor, which indicates the impact of external environmental changes on the i-th data source at the current moment;

[0074] The purpose of the above steps is to increase the contributions of those with large changes, highlight the abnormal situations, and thus help to detect potential conflicts.

[0075] Finally, the conflict of all data sources in multimodal data fusion is evaluated as a whole, and the contribution change index C of each data source is calculated. i (t) is integrated to generate a reference value of data source contribution change. The calculation expression is as follows:

[0076]

[0077] , where Data Source It is the reference value of the change in data source contribution.

[0078] By weighted summation of the contribution change index of each data source, a comprehensive conflict reference value β can be obtained. If the reference value is large, it indicates that there is significant conflict in the current multimodal data fusion process; otherwise, it indicates that the contributions between data sources are relatively consistent and the fusion process is normal.

[0079] The data source contribution change reference value, generated by analyzing the change in each data source's contribution to the final decision within the detection window during the multimodal fusion process, generally indicates data conflict in the current multimodal data fusion process. When the contribution of a particular data source changes significantly, especially when it fluctuates exponentially, it often indicates an inconsistency or anomaly in the relationship between the characteristics of that data source and other data sources. A large data source contribution change reference value reflects a dramatic shift in the influence of that data source on the final decision, which may be caused by data quality issues, inherent inconsistencies between data sources, or incorrect weight assignments. Such dramatic changes often indicate conflicting data sources in the multimodal fusion process, potentially leading to instability and increased error in model decisions. Conversely, a small and stable data source contribution change reference value indicates consistent interactions between data sources, the model is able to effectively integrate different data sources, and the decision-making process is relatively smooth, without significant data conflict.

[0080] An exponential increase in error during multimodal data fusion is often a key indicator of data conflict. When fusing data from different sources, the model integrates information from each source to make predictions or decisions. If the error for a particular data source increases exponentially during the fusion process, this may indicate significant inconsistency or deviation between that source and the others. When the error increases exponentially, it indicates that the inconsistencies between the different data sources have not been effectively addressed, leading to a significant decrease in the stability and accuracy of the fusion results. In this situation, conflicts between certain features or parameters become more prominent, possibly due to incorrect modeling of the relationships between the data sources or improper weighting. For example, if imaging and genomic data exhibit inconsistent characteristics for the same patient (e.g., no obvious lesions on imaging, but high susceptibility from genomic data), the model cannot find a reasonable balance between the two types of information, resulting in a rapid accumulation of error. This exponentially increasing error indicates that a conflict has occurred during the data fusion process, affecting the accuracy of the overall decision and even leading to inaccurate risk assessment and diagnosis. Therefore, when the error index increases exponentially, it not only reflects the conflict between data sources but may also indicate that the model is failing to properly handle this conflict, severely impacting the fusion results.

[0081] The specific steps for analyzing the error changes generated during the multimodal data fusion process under the detection window to generate the data fusion error reference value are as follows:

[0082] In the first step of generating the data fusion error index, the error of each data source is calculated. The calculated error will serve as the basis for subsequent calculations. The error of each data source is expressed as the deviation caused by the data source to the final prediction result. The error calculation expression is as follows:

[0083]

[0084] , where E i is the error of the ith data source, f(x p ) is the prediction result of the i-th data source on the p-th sample, is the actual value of the pth sample, representing the true situation of the sample, τ i is the error amplification index of the i-th data source, n is the total number of samples in the dataset, that is, the number of samples processed by the model when making predictions;

[0085] The purpose of this step is to calculate the error of each data source and perform nonlinear amplification on it, thereby providing basic data for subsequent conflict detection.

[0086] After calculating the errors of each data source, the obtained errors are weighted and fused to obtain a comprehensive error value. The error value will directly affect the generation of the final error reference value. The error of each data source is weighted according to its importance to the fusion result. The weight can be obtained through a pre-set reliability assessment or through dynamic learning. The calculation expression is as follows:

[0087]

[0088] , where E total is the total error after fusion of all data sources, w i is the weight of the i-th data source, indicating the proportion of the data source in the final decision, β i is the error amplification index of the i-th data source, which indicates the degree of nonlinear adjustment of the error of the data source, and N is the total number of all data sources used in the multimodal data fusion process;

[0089] This weighted fusion formula allows us to precisely control the contribution of each data source to the total error, based on its reliability and importance. This step ensures that the errors are weighted and adjusted during the fusion process, allowing the system to flexibly respond to differences and conflicts between data sources.

[0090] Finally, the fused error value is converted into a data fusion error reference value, which is used to quantify the degree of conflict in the multimodal data fusion process. The fused error is mapped using a nonlinear function. The calculation expression is as follows:

[0091]

[0092] , where Data Fusion is the reference value of data fusion error, γ i is the weight coefficient of the i-th data source, It is a weighted nonlinear transformation of the error of the i-th data source. Its purpose is to make the influence of the error on the reference value of the final data fusion error grow more nonlinearly, so as to avoid the data source with too large error dominating the result.

[0093] This formula effectively identifies and quantifies conflicts between different data sources by applying an exponential transformation and weighting to the errors. Specifically, as the error of a particular data source increases, the exponential growth accelerates, reflecting a higher level of conflict in the error index. This approach not only identifies the existence of conflicts but also quantifies their severity, helping to subsequently adjust and optimize the model's fusion strategy.

[0094] A larger data fusion error reference value, generated by analyzing the error changes during multimodal data fusion within a detection window, generally indicates data conflict in the current multimodal data fusion process. When multiple data sources (such as imaging data, genomic data, and clinical records) are being fused, the model makes decisions based on the weights and characteristics of each data source. If the error value of a data source gradually increases during the fusion process, and its growth trend is exponential, it indicates that the conflicts or inconsistencies between different data sources are gradually amplified. This is usually because the characteristics of the data sources cannot work together effectively, resulting in the accumulation of errors in the model's prediction results. For example, when imaging data and genomic data produce significantly different results for the risk assessment of the same patient, the model cannot correctly reconcile the differences between the two data sources, resulting in amplified errors. Conversely, a smaller and more stable data fusion error reference value indicates good synergy between the different data sources, and the model is able to integrate information from each data source more consistently, indicating that there are no obvious data conflicts in the current fusion process.

[0095] The deep learning conflict assessment module inputs the analyzed key features into a pre-learned deep learning model and uses the learning model to identify whether there are conflicts during the data fusion process;

[0096] The analyzed data source contribution change reference value and data fusion error reference value are input into the pre-learned deep learning model, and the fusion conflict risk coefficient is generated by the deep learning model. The multimodal data fusion process is intelligently evaluated by the fusion conflict risk coefficient to identify whether there is a conflict in the data fusion process.

[0097] Pre-learned deep learning models are models trained using deep learning algorithms (such as neural networks, convolutional neural networks, or recurrent neural networks) based on specific task requirements using large amounts of labeled data and multiple rounds of training. During the training phase, these models are fed with large amounts of historical data to learn complex nonlinear relationships between data and extract key features that influence target predictions (such as the risk of latent tuberculosis infection converting to active tuberculosis). Pre-learned models have learned through training how to identify correlations and potential conflicting patterns between different data sources (such as imaging data, genomic data, and clinical records) based on the input multimodal data. Through continuous training, the model can adaptively adjust its internal parameters, enabling it to make more accurate decisions when faced with new input data. In the context of multimodal data fusion, pre-learned deep learning models already possess the ability to effectively evaluate and judge based on features such as changes in data source contributions and error reference values. By inputting these features, the model can assess potential conflict risks in the current data fusion process based on previously learned patterns.

[0098] Specifically, the deep learning model uses key features such as changes in data source contributions and error reference values in historical data, combined with the knowledge accumulated by the model during training, to determine whether there are conflicts in the current data fusion process. For example, if certain feature combinations in historical data (such as the combination of imaging features and genomic features) often lead to high errors or risks, the model can identify these patterns and find similar situations in new data. Through this intelligent evaluation, the deep learning model can dynamically adjust and predict risks in the multimodal data fusion process, improving the accuracy and robustness of data fusion. The advantage of a pre-learned model is that it can not only handle large amounts of data and complex features, but also adapt to the heterogeneity and possible conflicts between data sources, and promptly identify and adjust inconsistent data sources, thereby effectively reducing errors and improving the final prediction ability.

[0099] The deep learning model is not limited here and can contribute to the change reference value Data Source And data fusion error reference value Data Fusion Perform comprehensive analysis to generate fusion conflict risk coefficient Conflict A specific implementation method is provided to implement the technical solution of the present invention.

[0100] Fusion Conflict Risk Factor Conflict The generation formula is as follows:

[0101]

[0102] , where k1 and k2 are the reference values of data source contribution change Data Source And data fusion error reference value Data Fusion The preset proportional coefficient, and k1 and k2 are both greater than 0.

[0103] The preset proportional coefficients (k1 and k2) are parameters used to measure the contribution of different data sources to the data fusion error. These two coefficients correspond to the reference value of the data source contribution change Data Source And data fusion error reference value Data Fusion Fusion Conflict Specifically, k1 controls the Source The error influence, k2 controls Data Fusion By setting k1 and k2, the model's sensitivity to different data sources and fusion errors can be adjusted in practice to balance their impact on the final conflict assessment. In practice, the values of k1 and k2 are typically determined through experimentation or tuning to ensure that the model accurately reflects the true contribution of data sources and fusion errors.

[0104] It can be seen from the fusion conflict risk coefficient that the larger the data source contribution change reference value generated after analyzing the contribution change of each data source to the final decision in the multimodal fusion process under the detection window, the larger the data fusion error reference value generated after analyzing the error change generated in the multimodal data fusion process under the detection window. This shows that the larger the fusion conflict risk coefficient generated when intelligently evaluating the multimodal data fusion process through the pre-selected deep learning model and identifying whether there is a conflict in the data fusion process, the greater the probability of data conflict in the multimodal data fusion process. Conversely, the smaller the probability of data conflict in the multimodal data fusion process.

[0105] The multimodal data fusion process is intelligently evaluated using a pre-learned deep learning model. The fusion conflict risk coefficient generated when identifying whether there is a conflict in the data fusion process is compared with the pre-set fusion conflict risk coefficient. The multimodal data fusion process is divided into the following steps:

[0106] If the fusion conflict risk coefficient is greater than the preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having data fusion conflict; if the fusion conflict risk coefficient is less than or equal to the preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having no data fusion conflict.

[0107] The dynamic weight adjustment and optimization module detects conflicts in the data fusion process and assesses their severity. It then dynamically adjusts the weights based on the reliability of each data source (such as data quality, historical performance, etc.) and the severity of the conflict, thereby reducing the negative impact of conflicting data sources on the final decision, thereby improving the accuracy and robustness of the fusion results.

[0108] The role of the dynamic weight adjustment and optimization module in multimodal data fusion is crucial, especially when there are data conflicts. It can effectively reduce the negative impact of conflicting data sources on the final decision and improve the accuracy and robustness of the overall decision. Specifically, when a conflict is detected in the multimodal data fusion process and its severity is assessed, the module will dynamically adjust the weight of each data source based on the reliability of each data source (such as data quality, historical performance, etc.) and the severity of the conflict. The core of this process lies in "dynamic" and "intelligent". It can flexibly adjust the importance of the data source in the fusion process according to its performance in specific situations, rather than invariably giving each data source the same weight.

[0109] For example, when the quality of imaging data is poor or there is a significant conflict between it and other data sources (such as genomic data), the dynamic weight adjustment module will automatically reduce the weight of the imaging data to reduce its impact on the final decision. Similarly, if the reliability of genomic data is high and it has a good historical performance in multiple scenarios, the module will increase the weight of genomic data, allowing it to play a greater role in the decision-making process. In this way, the system can not only reduce errors caused by data conflicts, but also ensure that the most reliable data sources are used in different scenarios, ultimately improving the accuracy of the fusion results.

[0110] Furthermore, the optimization module with dynamic weight adjustment enhances the robustness of the system. When data quality fluctuates or new challenges arise, adaptive adjustments are made to maintain the stability of the fusion process, preventing defects or deviations in individual data sources from impacting the reliability of the entire decision-making system. This self-optimizing and adaptable design enables the multimodal data fusion system to maintain efficient and accurate decision-making capabilities in complex and ever-changing real-world applications.

[0111] After detecting conflicts in the data fusion process and evaluating their severity, the specific steps for dynamic weight adjustment based on the reliability and severity of each data source are as follows:

[0112] After detecting a conflict in the data fusion process, we first need to assess the severity of the conflict and quantitatively evaluate the reliability of each data source. The evaluation formula is as follows:

[0113]

[0114] , where S i is the conflict severity index of data source i, which indicates the severity of the conflict generated by data source i in the multimodal data fusion process. i is the conflict sensitivity parameter of data source i, which adjusts its response to conflicts. Conflict is the fusion conflict risk coefficient, Fusion ref is the reference threshold of the fusion conflict risk coefficient, is the exponential growth factor of data source i, which controls the growth rate of the conflict severity index;

[0115] The above steps employ an exponential amplification approach, resulting in a nonlinear increase in conflict severity as it approaches a reference threshold. This reflects the dynamic amplification effect of conflict, particularly for conflict-sensitive data sources. In high-risk situations, changes in the conflict severity index become more sensitive and dramatic.

[0116] When evaluating the reliability of data sources, we combine multiple weighted normalization methods and add weight adjustments for historical performance. We also consider the weighted impact of data quality and historical performance, and balance their importance through a composite function. The calculation expression is as follows:

[0117]

[0118] , where R i is the reliability score of data source i, Q i is the quality score of data source i, which measures the accuracy and completeness of the data, etc. i is the historical performance score of data source i, considering the performance of the data source in past predictions, θ i is the stability coefficient of each data source, reflecting the weighted impact of historical performance on reliability assessment, V i is the historical variability of data source i, indicating the stability of the data source in history. is the mean of the historical variability of all data sources, w1 and w2 are the quality scores Q of data source i i and the historical performance score H of data source i i The weight coefficient determines the relative importance of the quality score and historical performance score in the overall reliability score;

[0119] Through this step, the reliability score relies not only on the quality and historical performance of the data source but also adjusts for data stability through historical variability. Higher historical variability indicates lower stability and a lower weighting of the impact. This formula ensures a comprehensive and multi-dimensional reliability assessment.

[0120] In the final dynamic weight adjustment, in addition to adjusting the conflict severity and reliability, a composite weighted adjustment function is introduced. In the adjustment process, exponential weighting is added, the interaction between conflict severity and data reliability is considered, and enhanced contrast adjustment is introduced.

[0121] The weight of each data source is dynamically adjusted based on the severity of the conflict and the reliability evaluation results of the data source. The calculation formula is as follows:

[0122]

[0123] , where w i ' is the adjusted weight of the i-th data source, E i is the original weight, which represents the initial contribution of the data source when there is no conflict, σ i is the adjustment coefficient, which controls the adjustment sensitivity of data source i, H i is the nonlinear weighting coefficient of the reliability score, which indicates the influence of reliability on the weight adjustment, μ i is the conflict severity adjustment coefficient, which controls the impact of conflict on weight adjustment, ρ i is a weighted index of conflict severity, adjusting the nonlinear effect of conflict severity on the weight.

[0124] This step comprehensively considers the interactive impact of data source reliability and conflict severity, dynamically optimizing weights through nonlinear adjustments. When conflicts are severe, weights decrease exponentially, while when data source reliability is low, weights also decrease accordingly. This complex and dynamic adjustment mechanism ensures a more accurate and robust final decision, reducing the negative impact of data conflicts and improving the accuracy of the overall fusion results.

[0125] The present invention effectively solves the data conflict problem existing in the multimodal data fusion process by acquiring multimodal data fusion parameters in real time and performing dynamic monitoring, thereby improving the accuracy and robustness of the risk assessment of converting latent tuberculosis infection to active tuberculosis. Specifically, the key features reflecting data conflicts are extracted through feature engineering technology, and intelligent evaluation is performed with the help of a pre-learned deep learning model to promptly identify conflicts between data sources. After the conflict is detected, the weight of each data source is dynamically adjusted based on the reliability of each data source and the severity of the conflict, reducing the negative impact of the conflicting data source on the final decision. This solution not only optimizes the accuracy of the multimodal data fusion process and avoids the risk of misdiagnosis and missed diagnosis, but also improves the adaptability of the model in a complex clinical data environment, ultimately promoting the implementation of precision medicine and personalized treatment.

[0126] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0127] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0128] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0129] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0130] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0132] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0133] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0134] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0135] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A deep learning-assisted prevention and treatment management platform for latent tuberculosis infection, characterized by: It includes data acquisition and dynamic monitoring module, data preprocessing and organization module, conflict feature extraction and analysis module, deep learning conflict assessment module and dynamic weight adjustment and optimization module: The data acquisition and dynamic monitoring module acquires the fusion parameter information of multimodal data in real time, ensuring dynamic monitoring of each input during the data fusion process, thereby providing basic data for subsequent conflict detection, feature extraction, and intelligent evaluation; The data preprocessing and organization module organizes the acquired multimodal data fusion parameter information into a unified data set and preprocesses the multimodal data fusion parameter information in the data set to facilitate subsequent analysis; The conflict feature extraction and analysis module uses feature engineering technology to extract key features that reflect conflicts in data fusion from preprocessed data, and analyzes the extracted key features to provide strong support for subsequent conflict detection; The deep learning conflict assessment module inputs the analyzed key features into a pre-learned deep learning model and uses the learning model to identify whether there are conflicts during the data fusion process; The dynamic weight adjustment and optimization module, after detecting conflicts in the data fusion process and evaluating their severity, dynamically adjusts the weights according to the reliability of each data source and the severity of the conflict, reducing the negative impact of conflicting data sources on the final decision, thereby improving the accuracy and robustness of the fusion results.

2. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 1 is characterized in that: The specific steps for organizing the acquired multimodal data fusion parameter information into a unified data set are as follows: first, the fusion parameters of each data source are labeled and normalized to conform to the unified data structure; Next, the data is denoised, missing values processed, and anomaly detected to eliminate potential quality issues; Subsequently, the data were scaled uniformly using standardization techniques to ensure that parameters from different data sources were comparable and consistent in subsequent analyses; Finally, the preprocessed data set is used for further feature extraction, conflict detection, and input into deep learning models to ensure the accuracy and reliability of the analysis results.

3. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 1 is characterized in that: Through feature engineering technology, key features reflecting conflicts in data fusion are extracted from the preprocessed data. The extracted features include the changes in the contribution of each data source to the final decision during the multimodal fusion process and the changes in the errors generated during the multimodal data fusion process. The changes in the contribution of each data source to the final decision during the multimodal fusion process and the changes in the errors generated during the multimodal data fusion process are analyzed under the detection window to generate a data source contribution change reference value and a data fusion error reference value respectively. The data source contribution change reference value is used to quantify the changes in the influence and contribution of each data source in the multimodal data fusion process. The data fusion error reference value is used to measure the changes in the inconsistency in the multimodal data fusion process over time.

4. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 3 is characterized in that: The specific steps for analyzing the contribution changes of each data source to the final decision in the multimodal fusion process under the detection window to generate the reference value of the data source contribution change are as follows: In the multimodal data fusion process, the contribution of each data source to the final decision will be continuously adjusted as the data input changes. To quantify the contribution of the data source at each time point, a dynamic weight function is introduced. The calculation is calculated by considering the feature importance of the data source and the response of the fusion model. The calculation expression is as follows: Where W i (t) is the contribution weight of the i-th data source to the decision at time t, f i (t) is the characteristic strength of the i-th data source at time t, α i (t) is the dynamic adjustment factor related to the i-th data source at time t, f j (t) is the characteristic strength of the j-th data source at time t, α j (t) is the dynamic adjustment factor associated with the jth data source at time t, and N is the total number of all data sources used in the multimodal data fusion process; Next, we perform a time series analysis on the dynamic contribution of each data source, calculate the change in the contribution of each data source, and introduce a contribution change index to quantify the change in the contribution of the data source between the current time and the previous time. The calculation expression is as follows: Where C i (t) is the contribution change index, which represents the contribution change index of the i-th data source at time t, W i (t-1) is the contribution weight of the i-th data source to the decision at time t-1, β i is the conflict adjustment factor of the i-th data source, δ i (t) is a time-dependent factor, which indicates the impact of external environmental changes on the i-th data source at the current moment; Finally, the conflict of all data sources in multimodal data fusion is evaluated as a whole, and the contribution change index C of each data source is calculated. i (t) is integrated to generate a reference value of data source contribution change. The calculation expression is as follows: Where, Data Source It is the reference value of the change in data source contribution.

5. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 3 is characterized in that: The specific steps for analyzing the error changes generated during the multimodal data fusion process under the detection window to generate the data fusion error reference value are as follows: In the first step of generating the data fusion error index, the error of each data source is calculated. The error of each data source is expressed as the deviation of the data source from the final prediction result. The error calculation expression is as follows: In, E i is the error of the ith data source, f(x p ) is the prediction result of the i-th data source on the p-th sample, is the actual value of the pth sample, τ i is the error amplification index of the i-th data source, n is the total number of samples in the dataset, that is, the number of samples processed by the model when making predictions; After calculating the errors of each data source, the obtained errors are weighted and fused to obtain a comprehensive error value. The error of each data source is weighted according to its importance to the fusion result. The calculation expression is as follows: Where, E total is the total error after fusion of all data sources, w i is the weight of the i-th data source, β i is the error amplification index of the i-th data source, and N is the total number of all data sources used in the multimodal data fusion process; Finally, the fused error value is converted into a data fusion error reference value, which is used to quantify the degree of conflict in the multimodal data fusion process. The fused error is mapped using a nonlinear function. The calculation expression is as follows: Where, Data Fusion is the reference value of data fusion error, γ i is the weight coefficient of the i-th data source, is a weighted nonlinear transformation of the error of the i-th data source.

6. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 3 is characterized in that: The analyzed data source contribution change reference value and data fusion error reference value are input into the pre-learned deep learning model, and the fusion conflict risk coefficient is generated by the deep learning model. The multimodal data fusion process is intelligently evaluated by the fusion conflict risk coefficient to identify whether there is a conflict in the data fusion process.

7. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 6 is characterized in that: The multimodal data fusion process is intelligently evaluated using a pre-learned deep learning model. The fusion conflict risk coefficient generated when identifying whether there is a conflict in the data fusion process is compared with the pre-set fusion conflict risk coefficient. The multimodal data fusion process is divided into the following steps: If the fusion conflict risk coefficient is greater than a preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having data fusion conflict; If the fusion conflict risk coefficient is less than or equal to a preset fusion conflict risk coefficient reference threshold, the current multimodal data fusion process is classified as having no data fusion conflict.

8. The deep learning-assisted prevention and treatment management platform for latent tuberculosis infection according to claim 7 is characterized in that: After detecting conflicts in the data fusion process and evaluating their severity, the specific steps for dynamic weight adjustment based on the reliability and severity of each data source are as follows: After detecting a conflict in the data fusion process, we first need to assess the severity of the conflict and quantitatively evaluate the reliability of each data source. The evaluation formula is as follows: Where S i is the conflict severity index of data source i, which indicates the severity of the conflict generated by data source i in the multimodal data fusion process. i is the conflict sensitivity parameter of data source i, which adjusts its response to conflicts. Conflict is the fusion conflict risk coefficient, Fusion ref is the reference threshold of the fusion conflict risk coefficient, is the exponential growth factor of data source i; When evaluating the reliability of data sources, we combine multiple weighted normalization methods and add weight adjustments for historical performance. We also consider the weighted impact of data quality and historical performance, and balance their importance through a composite function. The calculation expression is as follows: Where R i is the reliability score of data source i, Q i is the quality score of data source i, H i is the historical performance score of data source i, θ i is the stability coefficient of each data source, reflecting the weighted impact of historical performance on reliability assessment, V i is the historical variability of data source i, indicating the stability of the data source in history. is the mean of the historical variability of all data sources, w1 and w2 are the quality scores Q of data source i i and the historical performance score H of data source i i The weight coefficient of In the final dynamic weight adjustment, in addition to adjusting the conflict severity and reliability, a composite weighted adjustment function is introduced. In the adjustment process, exponential weighting is added, the interaction between conflict severity and data reliability is considered, and enhanced contrast adjustment is introduced. The weight of each data source is dynamically adjusted based on the severity of the conflict and the reliability evaluation results of the data source. The calculation formula is as follows: Where w i ' is the adjusted weight of the i-th data source, E i is the original weight, σ i is the adjustment coefficient, which controls the adjustment sensitivity of data source i, H i is the nonlinear weighting coefficient of the reliability score, μ i is the conflict severity adjustment coefficient, which controls the impact of conflict on weight adjustment, ρ i is a weighted index of conflict severity, adjusting the nonlinear effect of conflict severity on the weight.

Images