Computer AI auxiliary analysis follow-up visit data system and method

Abstract

The invention discloses a computer AI auxiliary analysis follow-up visit data system and method, and relates to the technical field of data processing, and the system comprises a data acquisition and preprocessing module, a dynamic fusion model construction module, a deep survival analysis module, an expected effect simulation module and a visual report generation and model iteration module. According to the method, multi-modal follow-up visit data and original data are integrated, a dynamic fusion model and a deep survival model are constructed, accurate long-term and short-term prognosis prediction and survival influence factor analysis are achieved, meanwhile, real-time risk monitoring and intervention effect simulation are conducted in combination with streaming computation, a visual visualization report is generated, the model is continuously iterated and optimized, and the real-time risk monitoring and intervention effect simulation is achieved. According to the method, the efficiency, accuracy and individuation level of medical services are remarkably improved, the workload of medical staff is relieved, and meanwhile, the satisfaction degree and credibility of patients can be improved by providing visual and accurate visual reports and individualized treatment schemes.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing. Specifically, it relates to a computer AI-assisted follow-up data analysis system and method. Background Art

[0002] Follow-up data refers to the relevant information on the health status and treatment effects of patients regularly collected by medical staff after the patients receive treatment or intervention. These data usually include the patients' physiological indicators, laboratory test results, imaging examinations, questionnaire survey results, and health data obtained through wearable devices, and are used to evaluate the patients' recovery, monitor the disease progression, and adjust the treatment plan.

[0003] Existing follow-up data processing methods have many limitations. They often rely on a single data source, and the data preprocessing process is complex and error-prone, thus affecting the accuracy and consistency of the data. In addition, most prediction models are constructed based on static features and it is difficult to effectively capture the temporal associations and causal relationships between features, which directly limits the accuracy of prediction. In terms of report generation, traditional methods usually present statically, lacking intuitiveness. Coupled with the low model update frequency, it is difficult to incorporate the latest data and user feedback in a timely manner.

[0004] More critically, the current prognosis prediction means lack real-time risk monitoring and intervention effect simulation functions, making it difficult to adjust and optimize the treatment plan in a timely manner. Similarly, the outdated report generation method and the lagging model update also hinder the efficient progress of the data analysis and report generation processes. These limitations have led medical staff to have to undertake heavy manual processing work with low efficiency. In addition, existing medical data analysis methods often cannot provide intuitive and personalized treatment plans, which greatly reduces the satisfaction and trust of patients.

[0005] Regarding the problems in the related art, no effective solution has been proposed yet. Summary of the Invention

[0006] Regarding the problems in the related art, the present invention proposes a computer AI-assisted follow-up data analysis system and method to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solutions adopted by the present invention are as follows: According to one aspect of the present invention, a computer AI-assisted follow-up data analysis system is provided. The system includes a data acquisition and preprocessing module, a dynamic fusion model construction module, a deep survival analysis module, an expected effect simulation module, and a visualization report generation and model iteration module; Among them, the data acquisition and preprocessing module is used to obtain and integrate follow-up data and original data, perform data alignment, complete multi-modal data acquisition, and perform preprocessing; A dynamic fusion model construction module, which is used to construct a dynamic fusion model of multi-dimensional clinical features based on the preprocessed follow-up data and original data, in combination with a causal convolutional network; A deep survival analysis module, which is used to construct a deep survival model based on the dynamic fusion model, in combination with a survival analysis algorithm, generate long-term and short-term prognosis predictions for patients, and use AI to obtain survival influencing factors; An expected effect simulation module, which is used to monitor risks in real time using stream computing based on the survival influencing factors obtained by AI, and simulate the expected effects of different intervention measures in combination with the deep survival model; A visualization report generation and model iteration module, which is used to generate a manipulable spatio-temporal visualization report according to the simulated expected effects, and continuously iterate and update the AI model.

[0008] Optionally, the data acquisition and preprocessing module includes a follow-up data acquisition sub-module, an original data acquisition sub-module, a data matching and integration sub-module, and a data preprocessing sub-module; Among them, the follow-up data acquisition sub-module is used to obtain follow-up data through medical institutions, and the follow-up data includes patients' regular reexamination records, questionnaires, and wearable device data; The original data acquisition sub-module is used to obtain original data through an API interface, and the original data includes electronic medical records, pathological sections, and gene sequencing; The data matching and integration sub-module is used to match and integrate the obtained follow-up data and original data based on the patient identity information and the visit time as the unique identifier to form a complete patient health profile; The data preprocessing sub-module is used to unify and preprocess the follow-up data and original data based on the matching and integration.

[0009] Optionally, unifying and preprocessing the follow-up data and original data based on the matching and integration includes: Unify different types of information in the follow-up data and original data to obtain multi-modal data; Based on the obtained multi-modal data, perform cleaning, denoising, and normalization processing.

[0010] Optionally, the dynamic fusion model construction module includes a clinical feature extraction sub-module, a feature dynamic fusion sub-module, and a fusion model construction sub-module; Among them, the clinical feature extraction sub-module is used to extract multi-dimensional clinical features based on the preprocessed follow-up data and original data, and the multi-dimensional clinical features include physiological indicators, laboratory test results, and imaging features; The feature dynamic fusion sub-module is used to dynamically fuse the extracted clinical features using a causal convolutional network to obtain the temporal relationship and causal relationship between the features; A fusion model construction sub-module, which is used to construct a dynamic fusion model of multi-dimensional clinical features based on the obtained temporal relationship and causal relationship.

[0011] Optionally, the deep survival analysis module includes a deep survival model construction sub-module, a prognosis prediction sub-module, and a survival influencing factor identification sub-module; Among them, the deep survival model construction sub-module is used to extract multi-modal features of patients through a dynamic fusion model, and construct a deep survival model using the partial likelihood loss function in the survival analysis algorithm; The prognosis prediction sub-module is used to predict the features of patients based on the generated deep survival model, and obtain the long-term and short-term prognosis predictions of patients; The survival influencing factor identification sub-module is used to identify the key factors affecting the survival of patients by using AI technology according to the obtained long-term and short-term prognosis predictions of patients.

[0012] Optionally, the deep survival model construction sub-module, which is used to extract multi-modal features of patients through a dynamic fusion model and construct a deep survival model using the partial likelihood loss function in the survival analysis algorithm, includes: The input layer receives the multi-modal features of patients extracted by the dynamic fusion model, and the hidden layer uses a fully connected network with residual connections to capture non-linear relationships; The output layer generates a short-term risk score and a long-term survival probability prediction in parallel, and introduces a time-aware attention mechanism to automatically weight the feature contributions of different time intervals.

[0013] Optionally, identifying the key factors affecting the survival of patients by using AI technology according to the obtained long-term and short-term prognosis predictions of patients includes: Based on the prediction results of the deep survival model, using the gradient-weighted class activation mapping technology, by calculating the weights of the prediction layer gradients back to the input feature layer, locating the key regions in the imaging data and the mutation time points in the laboratory indicators, generating a feature heat map to reveal the locally sensitive regions, and performing backpropagation analysis on the multi-modal features; Mask each dynamic fusion feature and then re-predict, calculate the difference in the contribution degrees of the features to the short-term death risk score and the long-term survival probability, and generate a feature importance ranking and an interaction effect matrix; Combining the contribution degree quantification results with the clinical data, and using the hierarchical clustering algorithm to identify the key factors affecting the survival of patients.

[0014] Optionally, the expected effect simulation module includes a real-time risk monitoring sub-module, an intervention effect simulation sub-module, and a risk assessment report generation sub-module; Among them, the real-time risk monitoring sub-module is used to monitor the health risks of patients in real time by using the flow computing technology based on the survival influencing factors obtained by AI; An intervention effect simulation sub-module, which is used to simulate the expected effects of different intervention measures in combination with a deep survival model; A risk assessment report generation sub-module, which is used to generate a real-time risk assessment report to provide a basis for clinical decision-making support.

[0015] Optionally, the visualization report generation and model iteration module includes a spatio-temporal visualization report generation sub-module and a model iteration and update sub-module; Among them, the spatio-temporal visualization report generation sub-module is used to generate a manipulable spatio-temporal visualization report according to the simulated expected effects, and display the patient's health status and the effects of intervention measures; The model iteration and update sub-module is used to continuously iterate and update the deep survival model by using AI technology, and optimize the deep survival model in combination with user feedback and actual effect evaluation.

[0016] According to another aspect of the present invention, a computer AI-assisted method for analyzing follow-up data is also provided, and the method includes the following steps: S1. Obtain and integrate follow-up data and original data, perform data alignment, complete multi-modal data collection and perform preprocessing; S2. Based on the preprocessed follow-up data and original data, combine a causal convolutional network to construct a dynamic fusion model of multi-dimensional clinical features; S3. Based on the dynamic fusion model, combine a survival analysis algorithm to construct a deep survival model, generate long-term and short-term prognosis predictions of patients, and use AI to obtain survival influencing factors; S4. Based on the survival influencing factors obtained by AI, use stream computing to monitor risks in real time, and combine the deep survival model to simulate the expected effects of different intervention measures; S5. Generate a manipulable spatio-temporal visualization report according to the simulated expected effects, and continuously iterate and update the AI model.

[0017] The beneficial effects of the present invention are as follows: 1. By integrating multi-modal follow-up data and original data, constructing a dynamic fusion model and a deep survival model, the present invention realizes accurate long-term and short-term prognosis prediction and survival influencing factor analysis. At the same time, combining stream computing to monitor risks and intervention effect simulation in real time, generating an intuitive visualization report, and continuously iterating and optimizing the model, significantly improving the efficiency, accuracy and personalization level of medical services, reducing the work burden of medical staff. At the same time, by providing an intuitive and accurate visualization report and a personalized treatment plan, it can also improve the satisfaction and trust of patients.

[0018] 2. Through the data acquisition and preprocessing module, the present invention can efficiently acquire and integrate follow-up data and original data, align the data, and complete the acquisition and preprocessing of multi-modal data, ensuring the accuracy and consistency of the data and providing a reliable basis for subsequent analysis. By using the dynamic fusion model construction module, based on the preprocessed data and combined with the causal convolutional network, a dynamic fusion model of multi-dimensional clinical features can be constructed, which can capture the temporal and causal relationships between features and improve the prediction accuracy of the model.

[0019] 3. Through the deep survival analysis module, the present invention can construct a deep survival model based on the dynamic fusion model and combined with the survival analysis algorithm, generate long-term and short-term prognosis predictions for patients, and use AI to obtain survival influencing factors, which helps doctors more accurately understand the patient's condition and prognosis, and thus formulate more reasonable treatment plans. The expected effect simulation module can, based on the survival influencing factors obtained by AI, use stream computing to monitor risks in real time and simulate the expected effects of different intervention measures in combination with the deep survival model, which helps doctors evaluate the effects of different treatment plans and thus select the optimal treatment strategy.

[0020] 4. Through the visualization report generation and model iteration module, the present invention can generate a manipulable spatio-temporal visualization report according to the simulated expected effects, display the patient's health status and the effects of intervention measures, which can help doctors more intuitively understand the patient's condition and treatment progress, and improve the efficiency and accuracy of clinical decision-making. It also has the function of model iteration and update, can continuously iterate and update the deep survival model using AI technology, and optimize the model performance in combination with user feedback and actual effect evaluation, ensuring that it always maintains the best state and improving the accuracy and reliability of prediction and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1 is a schematic block diagram of a computer AI-assisted follow-up data analysis system according to an embodiment of the present invention; Figure 2 is a flowchart of a computer AI-assisted follow-up data analysis method according to an embodiment of the present invention.

[0023] In the figure: 1. Data acquisition and preprocessing module; 2. Dynamic fusion model construction module; 3. Deep survival analysis module; 4. Expected effect simulation module; 5. Visualization report generation and model iteration module. Detailed implementation manners

[0024] To further illustrate the embodiments, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be combined with the relevant descriptions in the specification to explain the operation principle of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are usually used to represent similar components.

[0025] According to an embodiment of the present invention, a computer AI-assisted follow-up data analysis system and method are provided.

[0026] Now, the present invention will be further described in combination with the drawings and specific implementation manners. As Figure 1 shown, according to an embodiment of the present invention, a computer AI-assisted follow-up data analysis system is provided. The system includes a data acquisition and preprocessing module 1, a dynamic fusion model construction module 2, a deep survival analysis module 3, an expected effect simulation module 4, and a visualization report generation and model iteration module 5; Among them, the data acquisition and preprocessing module 1 is used to acquire and integrate follow-up data and original data, perform data alignment, complete multi-modal data collection, and perform preprocessing.

[0027] Preferably, the data acquisition and preprocessing module 1 includes a follow-up data acquisition sub-module, an original data acquisition sub-module, a data matching and integration sub-module, and a data preprocessing sub-module; Among them, the follow-up data acquisition sub-module is used to acquire follow-up data through medical institutions. The follow-up data includes patients' regular reexamination records, questionnaires, and wearable device data; The original data acquisition sub-module is used to acquire original data through an API interface. The original data includes electronic medical records, pathological sections, and gene sequencing; The data matching and integration sub-module is used to match and integrate the acquired follow-up data and original data based on the patient identity information and the visit time as the unique identifier to form a complete patient health profile; The data preprocessing sub-module is used to unify and preprocess the follow-up data and original data based on the matched and integrated data.

[0028] Preferably, unifying and preprocessing the follow-up data and original data based on the matched and integrated data includes: Unifying different types of information in the follow-up data and original data to obtain multi-modal data; Based on the acquired multimodal data, perform cleaning, denoising, and normalization processing.

[0029] It should be noted that in the data preprocessing stage, uniformly processing different types of information in the follow-up data and the original data means converting various types of data (such as text records, numerical data, image data, etc.) into a standard format that can be processed by the AI model. This step is crucial because it directly affects the accuracy and efficiency of subsequent analysis. In addition, cleaning, denoising, and normalization processing are key steps for improving the quality of the data. They are used to remove incorrect or redundant information, reduce noise interference, and convert data with different dimensions into the same scale, respectively, so as to ensure that the model can be trained and analyzed based on high-quality data. This series of preprocessing processes is an important foundation for ensuring that the system can accurately analyze the follow-up data, provide reliable predictions, and personalized treatment plans.

[0030] The dynamic fusion model construction module 2 is used to construct a dynamic fusion model of multi-dimensional clinical features based on the preprocessed follow-up data and the original data, in combination with the causal convolutional network.

[0031] Preferably, the dynamic fusion model construction module 2 includes a clinical feature extraction sub-module, a feature dynamic fusion sub-module, and a fusion model construction sub-module; Among them, the clinical feature extraction sub-module is used to extract multi-dimensional clinical features based on the preprocessed follow-up data and the original data. The multi-dimensional clinical features include physiological indicators, laboratory test results, and imaging features; The feature dynamic fusion sub-module is used to dynamically fuse the extracted clinical features using the causal convolutional network to obtain the temporal relationship and causal relationship between the features; The fusion model construction sub-module is used to construct a dynamic fusion model of multi-dimensional clinical features based on the obtained temporal relationship and causal relationship.

[0032] It should be noted that when the dynamic fusion model construction module 2 processes the follow-up data and the original data, the clinical feature extraction sub-module will carefully extract multi-dimensional clinical features including physiological indicators, laboratory test results, imaging features, etc. from these data. However, these features often change over time and there are complex temporal and causal relationships between them. The feature dynamic fusion sub-module can intelligently capture the dynamic changes and internal associations between these features by introducing the causal convolutional network, so as to more deeply understand the development process of the disease. Finally, the fusion model construction sub-module constructs a fusion model that can reflect the dynamic changes of clinical features based on these fine feature relationships. This process not only improves the model's ability to grasp the development law of the disease, but also provides a more accurate basis for subsequent analysis and prediction.

[0033] Among them, the feature dynamic fusion sub-module uses a dilated causal convolution network (Dilated Causal Convolution) to process unevenly sampled time series data: the receptive field range is controlled by the dilation coefficient d (such as d = 2 k increasing layer by layer), ensuring that the output only depends on historical information (causality constraint), and at the same time embedding a dynamic time warping (DTW) layer to align clinical indicators with different frequencies (such as daily blood pressure and monthly tumor markers, etc.), and its cost function is: ; In the formula, represents the minimum cumulative alignment cost between the first i points of sequence X and the first j points of sequence Y, x i represents the i-th data point in time series X, y j represents the j-th data point in time series Y, represents the Euclidean distance, represents the alignment of the i-th data point of sequence X with the j-th data point in sequence Y, represents the alignment of the j-th data point of sequence Y with the i-th data point of sequence X, represents the direct alignment of the i-th data point of sequence X with the j-th data point in sequence Y.

[0034] In addition, a cross-modal gating attention mechanism is introduced to calculate the interaction weights between radiomics features (such as MRI texture entropy) and laboratory indicators (such as white blood cell count): ; In the formula, represents the interaction weight, Q i represents the image feature vector, K j represents the laboratory indicator vector, d k represents the dimension scaling factor, and T represents the transpose symbol.

[0035] The fused features are passed to the downstream network through residual connections to avoid the problem of gradient disappearance. Finally, the dynamic fusion model eliminates the interference of confounding factors (such as the bias of age and gender on the efficacy evaluation) through adversarial training (Adversarial Training).

[0036] The deep survival analysis module 3 is used to construct a deep survival model based on the dynamic fusion model, combined with a survival analysis algorithm, generate long-term and short-term prognosis predictions for patients, and use AI to obtain survival influencing factors.

[0037] Preferably, the deep survival analysis module 3 includes a deep survival model construction sub-module, a prognosis prediction sub-module, and a survival influencing factor identification sub-module; Among them, the deep survival model construction sub-module is used to extract the multi-modal features of patients through a dynamic fusion model, and a deep survival model is constructed using the partial likelihood loss function in the survival analysis algorithm.

[0038] Preferably, the deep survival model construction sub-module is used to extract the multi-modal features of patients through a dynamic fusion model, and constructing a deep survival model using the partial likelihood loss function in the survival analysis algorithm includes: The input layer receives the multi-modal features of patients extracted by the dynamic fusion model, and the hidden layer uses a fully connected network with residual connections to capture non-linear relationships; The output layer generates short-term risk scores and long-term survival probability predictions in parallel, and introduces a time-aware attention mechanism to automatically weight the feature contributions in different time intervals.

[0039] The prognosis prediction sub-module is used to predict the features of patients based on the generated deep survival model, and obtain the long-term and short-term prognosis predictions of patients.

[0040] The survival influencing factor identification sub-module is used to identify the key factors affecting the survival of patients by using AI technology according to the long-term and short-term prognosis predictions of patients.

[0041] Preferably, identifying the key factors affecting the survival of patients by using AI technology according to the long-term and short-term prognosis predictions of patients includes: Based on the prediction results of the deep survival model, using the gradient-weighted class activation mapping technology, by calculating the weights of the prediction layer gradients backpropagated to the input feature layer, locate the key regions in the imaging data and the mutation time points in the laboratory indicators, generate a feature heat map to reveal the local sensitive regions, and perform backpropagation analysis on the multi-modal features; Mask each dynamic fusion feature and then re-predict, calculate the difference in the contribution degrees of the features to the short-term death risk score and the long-term survival probability, and generate a feature importance ranking and interaction effect matrix; Combine the contribution degree quantification results with the clinical data, and use the hierarchical clustering algorithm to identify the key factors affecting the survival of patients.

[0042] It should be explained that the survival analysis algorithm (DeepSurv algorithm) combines a deep neural network and a survival analysis model, and is mainly used to predict the survival time or survival probability of patients. It is a Cox proportional hazards deep neural network, aiming to simulate the interaction between the prognostic variables (covariates) of patients and the treatment effects, and is particularly suitable for evaluating the importance of prognostic covariates in events such as death or cancer recurrence, and can provide personalized treatment recommendations for patients.

[0043] In the deep survival analysis module, the AI technologies adopted mainly include Gradient-weighted Class Activation Mapping (Grad-CAM), feature masking and re-prediction methods, and hierarchical clustering algorithms. These technologies are applied in the process of identifying survival influencing factors: The Grad-CAM technology generates a feature heatmap by backpropagating the gradient of the prediction layer to the input feature layer, thereby revealing the key regions of the image data and the mutation time points of laboratory indicators; The feature masking method is to mask each dynamic fusion feature one by one and then re-predict, so as to calculate the contribution degree of each feature to the short-term death risk score and long-term survival probability, and generate a feature importance ranking and interaction effect matrix; Finally, the hierarchical clustering algorithm combines the quantization results and clinical data to further screen out the core factors affecting the patient's survival.

[0044] The deep survival analysis module combines a dynamic fusion model, a survival analysis algorithm, and AI technologies to predict the short-term and long-term prognosis of patients and identify the key factors affecting the patient's survival. Among them, the deep survival model construction sub-module not only receives and processes multi-modal patient features from the dynamic fusion model, but also introduces a partial likelihood loss function in model construction. This is a commonly used loss function in survival analysis, which can help the model better handle the truncation and censoring problems in survival data. In addition, this sub-module designs a fully connected network with residual connections to capture the non-linear relationships between features, and generates short-term risk scores and long-term survival probability predictions in parallel at the output layer. At the same time, a time-aware attention mechanism is used to automatically weight the feature contributions in different time intervals to improve the prediction accuracy of the model.

[0045] In the survival influencing factor identification sub-module, by performing backpropagation analysis on multi-modal features through the Gradient-weighted Class Activation Mapping (Grad-CAM) technology, the key regions in the image data and the mutation time points in laboratory indicators can be revealed. These key information is of great significance for understanding the changes in the patient's survival status. At the same time, by masking each dynamic fusion feature one by one and then re-predicting, the contribution degree difference of each feature to the short-term death risk score and long-term survival probability can be calculated, and then a feature importance ranking and interaction effect matrix can be generated. This helps to identify the key factors that have an important impact on the patient's survival. Finally, combining these quantization results and clinical data, the hierarchical clustering algorithm can be used to further screen out the core factors affecting the patient's survival.

[0046] The expected effect simulation module 4 is used to monitor risks in real time based on the survival influencing factors obtained by AI, and simulate the expected effects of different intervention measures in combination with the deep survival model.

[0047] Preferably, the expected effect simulation module 4 includes a real-time risk monitoring sub-module, an intervention effect simulation sub-module, and a risk assessment report generation sub-module; Among them, the real-time risk monitoring sub-module is used to monitor the health risks of patients in real time based on the survival influencing factors obtained by AI, using stream computing technology; The intervention effect simulation sub-module is used to simulate the expected effects of different intervention measures in combination with the deep survival model; The risk assessment report generation sub-module is used to generate a real-time risk assessment report to provide a basis for clinical decision-making support.

[0048] It should be noted that the expected effect simulation module 4 is an integrated module that integrates the functions of real-time risk monitoring, intervention effect simulation, and risk assessment report generation.

[0049] Among them, the real-time risk monitoring sub-module obtains the survival influencing factors using AI technology and realizes the real-time monitoring of patients' health risks with the help of stream computing technology, which can ensure the immediate perception and response to patients' health conditions. The intervention effect simulation sub-module combines the deep survival model to simulate the expected effects of different intervention measures, which helps the clinical team evaluate the potential effects of different treatment plans before making decisions. Finally, the risk assessment report generation sub-module integrates the risks monitored in real time and the simulated intervention effects into a real-time risk assessment report to provide a scientific and accurate basis for clinical decision-making.

[0050] The visualization report generation and model iteration module 5 is used to generate a manipulable spatio-temporal visualization report based on the simulated expected effects and continuously iterate and update the AI model.

[0051] Preferably, the visualization report generation and model iteration module 5 includes a spatio-temporal visualization report generation sub-module and a model iteration update sub-module; Among them, the spatio-temporal visualization report generation sub-module is used to generate a manipulable spatio-temporal visualization report based on the simulated expected effects to display the health conditions of patients and the effects of intervention measures; The model iteration update sub-module is used to continuously iterate and update the deep survival model using AI technology and optimize the deep survival model in combination with user feedback and actual effect evaluation.

[0052] It should be noted that the spatio-temporal visualization report generation sub-module in the visualization report generation and model iteration module 5 can not only generate an intuitive and manipulable spatio-temporal visualization report based on the simulated expected effects, clearly showing the changes in patients' health conditions and the effects of different intervention measures, enabling the clinical team to more intuitively understand the patients' situations and the potential impacts of treatment plans.

[0053] Meanwhile, the model iteration and update sub-module continuously iterates and updates the deep survival model through AI technology. This process is not only based on the performance evaluation of the model itself, but also fully combines user feedback and the effect evaluation of actual clinical applications to ensure that the deep survival model can continuously adapt to new clinical data and requirements, and continuously improve the prediction accuracy and clinical practicability.

[0054] According to another embodiment of the present invention, as Figure 2 shown, a computer AI-assisted method for analyzing follow-up data is also provided. The method includes the following steps: S1. Obtain and integrate follow-up data and original data, perform data alignment, complete multi-modal data collection and preprocessing; S2. Based on the preprocessed follow-up data and original data, combined with a causal convolutional network, construct a dynamic fusion model of multi-dimensional clinical features; S3. Based on the dynamic fusion model, combined with a survival analysis algorithm, construct a deep survival model, generate long-term and short-term prognosis predictions for patients, and use AI to obtain survival influencing factors; S4. Based on the survival influencing factors obtained by AI, use streaming computing to monitor risks in real time, and combined with the deep survival model, simulate the expected effects of different intervention measures; S5. According to the simulated expected effects, generate a manipulable spatio-temporal visualization report and continuously iterate and update the AI model.

[0055] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A computer AI-assisted analysis follow-up data system, characterized in that, Including: A data acquisition and preprocessing module, which is used to acquire and integrate follow-up data and raw data, perform data alignment, complete multi-modal data acquisition and perform preprocessing; A dynamic fusion model construction module, which is used to construct a dynamic fusion model of multi-dimensional clinical features based on the preprocessed follow-up data and raw data, in combination with a causal convolutional network; A deep survival analysis module, which is used to construct a deep survival model based on the dynamic fusion model, in combination with a survival analysis algorithm, generate long-term and short-term prognosis predictions for patients, and use AI to obtain survival influencing factors; An expected effect simulation module, which is used to monitor risks in real time by means of streaming computing based on the survival influencing factors obtained by AI, and simulate the expected effects of different intervention measures in combination with the deep survival model; A visualization report generation and model iteration module, which is used to generate a manipulable spatio-temporal visualization report according to the simulated expected effects, and continuously iterate and update the AI model.

2. The computer AI-assisted follow-up data analysis system according to claim 1, wherein The data acquisition and preprocessing module includes a follow-up data acquisition sub-module, a raw data acquisition sub-module, a data matching and integration sub-module, and a data preprocessing sub-module; Among them, the follow-up data acquisition sub-module is used to acquire follow-up data through medical institutions, and the follow-up data includes patients' regular reexamination records, questionnaires, and wearable device data; The raw data acquisition sub-module is used to acquire raw data through an API interface, and the raw data includes electronic medical records, pathological sections, and gene sequencing; The data matching and integration sub-module is used to match and integrate the acquired follow-up data and raw data based on the patient identity information and the visit time as the unique identifier to form a complete patient health profile; The data preprocessing sub-module is used to unify and preprocess the follow-up data and raw data based on the matched and integrated ones.

3. The computer AI-assisted follow-up data analysis system according to claim 2, wherein, The unification and preprocessing based on the matched and integrated follow-up data and raw data includes: Unifying different types of information in the follow-up data and raw data to obtain multi-modal data; Based on the acquired multi-modal data, performing cleaning, denoising, and normalization processing.

4. A computer AI-assisted follow-up data analysis system according to claim 1, characterized in that, The dynamic fusion model construction module includes a clinical feature extraction sub-module, a feature dynamic fusion sub-module, and a fusion model construction sub-module; Among them, the clinical feature extraction sub-module is used to extract multi-dimensional clinical features based on the preprocessed follow-up data and raw data, and the multi-dimensional clinical features include physiological indicators, laboratory test results, and imaging features; The feature dynamic fusion sub-module is used to dynamically fuse the extracted clinical features by using a causal convolutional network to obtain the temporal relationship and causal relationship between the features; The fusion model construction sub-module is used to construct a dynamic fusion model of multi-dimensional clinical features based on the obtained temporal relationship and causal relationship.

5. The computer AI-assisted follow-up data analysis system according to claim 1, wherein, The deep survival analysis module includes a deep survival model construction sub-module, a prognosis prediction sub-module, and a survival influencing factor identification sub-module; Among them, the deep survival model construction sub-module is used to extract multi-modal features of patients through a dynamic fusion model, and construct a deep survival model by using the partial likelihood loss function in the survival analysis algorithm; The prognostic prediction sub-module is used to predict the characteristics of patients based on the generated deep survival model, and obtain the long-term and short-term prognostic predictions of patients. The survival influencing factor identification sub-module is used to identify the key factors affecting the survival of patients by using AI technology according to the obtained long-term and short-term prognostic predictions of patients.

6. The computer AI-assisted follow-up data analysis system according to claim 5, wherein, The deep survival model construction sub-module is used to extract the multi-modal features of patients through a dynamic fusion model, and construct a deep survival model by using the partial likelihood loss function in the survival analysis algorithm, including: The input layer receives the multi-modal features of patients extracted by the dynamic fusion model, and the hidden layer uses a fully connected network with residual connections to capture non-linear relationships. The output layer generates short-term risk scores and long-term survival probability predictions in parallel, and introduces a time-aware attention mechanism to automatically weight the feature contributions in different time intervals.

7. The computer AI-assisted follow-up data analysis system according to claim 5, wherein The identifying the key factors affecting the survival of patients by using AI technology according to the obtained long-term and short-term prognostic predictions of patients includes: Based on the prediction results of the deep survival model, using the gradient-weighted class activation mapping technology, by calculating the weights of the prediction layer gradients back to the input feature layer, locating the key regions in the imaging data and the mutation time points in the laboratory indicators, generating a feature heat map to reveal the locally sensitive regions, and performing backpropagation analysis on the multi-modal features. Mask each dynamic fusion feature one by one and then re-predict, calculate the difference in the contribution degrees of the features to the short-term death risk score and the long-term survival probability, and generate a feature importance ranking and an interaction effect matrix. Combining the contribution degree quantification results with the clinical data, using the hierarchical clustering algorithm to identify the key factors affecting the survival of patients.

8. The computer AI-assisted follow-up data analysis system according to claim 1, characterized in that The expected effect simulation module includes a real-time risk monitoring sub-module, an intervention effect simulation sub-module, and a risk assessment report generation sub-module. Among them, the real-time risk monitoring sub-module is used to monitor the health risks of patients in real time by using the streaming computing technology based on the survival influencing factors obtained by AI. The intervention effect simulation sub-module is used to simulate the expected effects of different intervention measures in combination with the deep survival model. The risk assessment report generation sub-module is used to generate a real-time risk assessment report to provide a basis for clinical decision-making support.

9. The computer AI-assisted follow-up data analysis system according to claim 1, wherein, The visualization report generation and model iteration module includes a spatio-temporal visualization report generation sub-module and a model iteration update sub-module. Among them, the spatio-temporal visualization report generation sub-module is used to generate a manipulable spatio-temporal visualization report according to the simulated expected effects, and display the health status of patients and the effects of intervention measures. The model iteration update sub-module is used to continuously iterate and update the deep survival model by using AI technology, and optimize the deep survival model in combination with user feedback and actual effect evaluation.

10. A method for computer AI-assisted analysis of follow-up data, using the computer AI-assisted analysis of follow-up data system described in any one of claims 1-9, characterized in that, This method includes the following steps: S1. Obtain and integrate the follow-up data and the original data, perform data alignment, complete the multi-modal data collection and perform preprocessing. S2. Based on the preprocessed follow-up data and the original data, combined with the causal convolutional network, construct a dynamic fusion model of multi-dimensional clinical features. S3. Based on the dynamic fusion model, combined with the survival analysis algorithm, construct a deep survival model, generate the long-term and short-term prognostic predictions of patients, and use AI to obtain the survival influencing factors. S4. Based on the survival influencing factors obtained by AI, real-time risk monitoring is carried out using stream computing, and the expected effects of different intervention measures are simulated by combining with a deep survival model; S5. According to the simulated expected effects, a controllable spatio-temporal visualization report is generated, and the AI model is continuously iteratively updated.

Images