Digital intelligent tumor prevention and treatment management platform and management method

Abstract

The invention discloses a digital intelligent tumor prevention and treatment management platform and management method, and relates to the technical field of intelligent medical treatment, and the method comprises the steps: collecting the multi-modal data of a patient in real time, carrying out the standardized cleaning and space-time alignment processing, and generating a space-time tagging data set; inputting the multi-modal joint feature representation into an anti-fact inference engine, simulating potential effects under intervention of different treatment schemes, calculating individual processing effects, and generating an anti-fact rehabilitation suggestion set; according to the anti-fact rehabilitation suggestion set, a dynamic causal enhancement decision model is constructed, real-time physiological data feedback of the patient is continuously received through a near-end strategy optimization algorithm, and a treatment scheme is dynamically adjusted. According to the method, through multi-modal data space-time alignment and feature fusion, in combination with a tumor biological mechanism and a machine learning algorithm, the accuracy and clinical credibility of individualized tumor treatment scheme recommendation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent healthcare, and in particular to a digital and intelligent tumor prevention and control management platform and a management method. Background Art

[0002] Currently, based on the joint analysis of medical imaging omics, genomic sequencing, and wearable device monitoring data, the paradigm shift from single biomarker analysis to multi-dimensional dynamic assessment has been gradually realized. Existing technologies construct a correlation model between imaging features and gene expression through deep neural networks, and generate treatment recommendations by combining clinical decision tree algorithms, showing certain application value in tumor staging prediction and drug sensitivity analysis.

[0003] However, existing methods still face technical bottlenecks in cross-modal data integration and causal mechanism mining: on the one hand, there are significant differences between the high spatial resolution of medical images and the time series characteristics of gene sequencing and wearable devices. Traditional registration algorithms are difficult to eliminate the spatio-temporal scale deviation between multi-source data, resulting in a high error rate in the dynamic association modeling of imaging features and gene mutations; on the other hand, existing causal inference models lack a constraint mechanism for the topological relationship of cross-modal feature spaces. When analyzing complex causal chains such as gene mutations driving lesion evolution and patient behavior affecting treatment response, they are easily interfered by confounding factors and misjudged, directly affecting the clinical adaptability of individualized treatment plans and the control of safety thresholds. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a digital and intelligent tumor prevention and control management method to solve the problems of insufficient spatio-temporal alignment accuracy of multi-modal tumor data and high causal association misjudgment rate.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a digital and intelligent tumor prevention and control management method, which includes: collecting multi-modal data of patients in real time, performing standardized cleaning and spatio-temporal alignment processing to generate a spatio-temporal tagged data set; inputting the spatio-temporal tagged data set into a spatio-temporal causal alignment network, using a temporal convolutional network to extract temporal features, and constructing a multi-modal causal association graph through a causal discovery algorithm to identify the causal chain between imaging features, gene mutations, and patient behaviors, and outputting a causal intensity quantization matrix; based on the causal intensity quantization matrix, screening associated nodes with a causal intensity exceeding a preset causal intensity threshold, and forcing the alignment of the imaging slice feature vector and the gene mutation embedding vector through a cross-modal contrast loss function to generate a multi-modal joint feature representation fused with causal constraints; inputting the multi-modal joint feature representation into a counterfactual reasoning engine, simulating the potential effects under different treatment plan interventions, calculating the individual treatment effect, and generating a counterfactual rehabilitation advice set; according to the counterfactual rehabilitation advice set, constructing a dynamic causal reinforcement decision model, and continuously receiving real-time physiological data feedback of the patient through a proximal policy optimization algorithm to dynamically adjust the treatment plan.

[0007] As a preferred embodiment of the digital and intelligent tumor prevention and control management method of the present invention, wherein: the standardized cleaning and spatio-temporal alignment processing includes performing standardized processing on the imaging features of the lesion area, and using a dynamic time warping algorithm to align the vital sign monitoring data and the behavior log time axis.

[0008] As a preferred embodiment of the digital and intelligent tumor prevention and control management method of the present invention, wherein: the causal discovery algorithm refers to establishing a causal pointing rule based on the timestamp priority principle, and constructing a multi-modal causal association graph through conditional independence testing.

[0009] As a preferred embodiment of the digital and intelligent tumor prevention and control management method of the present invention, wherein: the cross-modal contrast loss function includes the following steps, Based on the causal intensity quantization matrix, screening causal edges, and extracting the imaging slice feature vector and the gene mutation embedding vector as positive sample pairs; Uniformly sampling from non-causally associated cross-modal nodes to generate a negative sample pair set; Based on the constructed positive and negative sample pair sets, adopting a contrast learning framework weighted by causal intensity, constructing a contrast loss function, minimizing the contrast loss through an Adam optimizer, and monitoring the decline rate of the positive sample pair distance.

[0010] As a preferred embodiment of the digital and intelligent tumor prevention and control management method of the present invention, wherein: the counterfactual reasoning engine adopts the following steps, Based on the causal edges retained in the causal intensity quantization matrix, selecting gene mutation binary features and lesion volume as intervenable variables; Input the intervention variable into the double-robust estimation framework, combine the gradient boosting regression tree to predict the potential outcome, calibrate the counterfactual prediction bias through the real data residuals, and verify the mean absolute error.

[0011] As a preferred solution of the digital intelligent tumor prevention and control management method described in the present invention, wherein: the multi-modal data includes medical images, gene sequencing results, continuous vital sign monitoring data, and behavioral logs reported by patients themselves.

[0012] As a preferred solution of the digital intelligent tumor prevention and control management method described in the present invention, wherein: the counterfactual rehabilitation recommendation set includes drug dose adjustment plans, follow-up cycle optimization strategies, and behavioral intervention plans.

[0013] In a second aspect, the present invention provides a digital intelligent tumor prevention and control management platform, including a data preprocessing module, a causal modeling module, a feature fusion module, a counterfactual reasoning module, and a dynamic decision-making module. The data preprocessing module is used to collect the multi-modal data of patients in real time, and perform standardized cleaning and spatio-temporal alignment processing to generate a spatio-temporal labeled data set; the causal modeling module is used to input the spatio-temporal labeled data set into a spatio-temporal causal alignment network, extract temporal features using a temporal convolutional network, and construct a multi-modal causal association graph through a causal discovery algorithm, identify the causal chain between image features, gene mutations, and patient behaviors, and output a causal strength quantification matrix; the feature fusion module is used to screen the associated nodes with causal strength exceeding a preset causal strength threshold based on the causal strength quantification matrix, and force-align the image slice feature vector and the gene mutation embedding vector through a cross-modal contrast loss function to generate a multi-modal joint feature representation with fused causal constraints; the counterfactual reasoning module is used to input the multi-modal joint feature representation into a counterfactual reasoning engine, simulate the potential effects under different treatment plan interventions, calculate the individual treatment effect, and generate a counterfactual rehabilitation recommendation set; the dynamic decision-making module is used to construct a dynamic causal reinforcement decision model based on the counterfactual rehabilitation recommendation set, continuously receive real-time physiological data feedback of patients through a proximal policy optimization algorithm, and dynamically adjust the treatment plan.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, and the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the digital intelligent tumor prevention and control management method described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, any step of the digital intelligent tumor prevention and control management method described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: By performing spatio-temporal alignment and feature fusion on multi-modal medical images, genomic sequencing data, and clinical monitoring data, the dynamic relationships of biomarkers with significant clinical relevance during tumor evolution are accurately captured, enhancing the reliability of causal inference. By calculating the covariance matrix of gene mutation frequencies, lesion volume change rates, and physiological index fluctuations, spatio-temporal correlation features of multi-modal data are extracted, and multi-dimensional features with clear pathological manifestations are refined from them, enhancing the adaptability and generalization ability of the dynamic decision-making model to individualized treatment plans. By combining regularization constraints based on tumor biology with a causal inference model, integrating pathological mechanisms and machine learning algorithms, the biological rationality and interpretability of the causal inference model are enhanced, enabling the treatment recommendation results output by the dynamic decision-making model to have verifiable clinical credibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0018] Figure 1 It is a flowchart of the digital intelligent tumor prevention and control management method.

[0019] Figure 2 It is a flowchart of multi-modal data processing.

[0020] Figure 3 It is a flowchart of causal feature fusion.

[0021] Figure 4 It is a flowchart of dynamic decision optimization. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0023] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention, but the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0024] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a digital tumor prevention and treatment management method, including: S1: Collect multimodal data of patients in real time, perform standardized cleaning and spatiotemporal alignment processing to generate a spatiotemporal labeled dataset.

[0026] Specifically, the following steps are included: S1.1: Deploy medical imaging equipment, gene sequencers, wearable devices, and mobile applications to synchronize clocks through the Network Time Protocol (NTP protocol) and automatically generate standard timestamps for date and time representation (ISO8601) when collecting multimodal data.

[0027] The multimodal data include medical images, gene sequencing results, continuous vital signs monitoring data and patient self-reported behavior logs.

[0028] S1.2: Use the medical image registration (Elastix) toolkit to perform multi-resolution B-spline registration on medical images, align different modality images to the patient's anatomical coordinate system and mark the lesion area, and extract the image features of the lesion area (such as CT value, volume); Based on the timestamp difference between medical images and gene sequencing results, gene mutations are associated with lesion areas.

[0029] For example, when the timestamp difference is ≤ 24 hours, it is determined to be valid.

[0030] S1.3: Dynamic time warping was used to align continuous vital sign monitoring data (sampled at 10 Hz) with the patient’s self-reported behavioral log timeline, vital sign data were resampled to 1 Hz, and missing values ​​were filled using cubic spline interpolation.

[0031] S1.4: Perform Z-score standardization on the imaging features of the lesion area, convert the gene mutation data into binary features, remove abnormal vital signs (such as heart rate > 220 bpm) and perform interpolation processing.

[0032] S1.5: Store the standard timestamp (ISO 8601) for date and time representation, annotated lesion areas, processed image features and gene mutation data, aligned continuous vital sign monitoring data, and behavioral log data into HDF5 files grouped by patient ID.

[0033] Among them, the HDF5 file is the specific storage form of spatiotemporal labeled data sets.

[0034] S2: Input the spatio-temporal labeled dataset into the spatio-temporal causal alignment network, use the temporal convolutional network to extract temporal features, and construct a multi-modal causal association graph through a causal discovery algorithm to identify the causal chain among image features, gene mutations, and patient behaviors, and output a causal intensity quantization matrix.

[0035] Specifically, it includes the following steps: S2.1: Read the time series of continuous vital sign monitoring data and the aligned behavior log data from the spatio-temporal labeled dataset (HDF5 file).

[0036] For the time series of continuous vital sign monitoring data (such as heart rate, blood oxygen saturation), use the dilated causal convolution layer of the temporal convolutional network (TCN) and set the dilation factor .

[0037] Among them, is the number of layers, ; a causal convolution layer with a kernel size of 3 outputs a multi-scale temporal feature vector.

[0038] Perform one-hot encoding on the aligned behavior log data (such as taking medicine, exercising), align it to a 1Hz sampling rate along the time axis, and generate behavior event sequence features.

[0039] Example: Input a heart rate sequence of length 7200 (2-hour data), and output a 256-dimensional temporal feature vector after 3 layers of TCN. The behavior event "taking medicine" is encoded as [0,1,0,...,0], and a 7200-dimensional sparse sequence is generated after alignment.

[0040] S2.2: Use the PC algorithm (Peter-Clark algorithm) to perform conditional independence tests on image features, gene mutations, and behavior event sequence features.

[0041] Establish a causal direction rule based on the timestamp priority principle: Based on the representation method of the recorded date and time (ISO8601) standard timestamp, if a number of pre-variables have timestamps earlier than the result variable (such as the gene sample collection time is earlier than the behavior event time), then add a directed edge → .

[0042] Specifically, use the G-test statistic (significance level set to 0.05) to test the independence between variables.

[0043] For example, gene mutation "EGFR L858R" (timestamp 14:25:30Z) → behavior event "worsening fatigue" (timestamp 14:30:00Z), G-test p = 0.03, add a causal edge.

[0044] Based on all the causal edges that pass the test → ), construct a multi-modal causal association graph in the form of a directed acyclic graph, where the nodes represent variables such as imaging features, gene mutations, and behavioral events, and the edges represent statistically verified causal relationships.

[0045] S2.3: Based on the constructed multi-modal causal association graph, force-align the extracted imaging features with the generated gene mutation binary features.

[0046] Define the triplet loss (anchor, positive sample, and negative sample).

[0047] Specifically, the anchor: a randomly selected imaging slice feature vector; the positive sample: the gene mutation embedding causally associated with the anchor; the negative sample: a randomly sampled non-associated gene mutation embedding.

[0048] The loss function , is expressed as: ; In the formula, represents the anchor feature vector, represents the positive sample feature vector, represents the negative sample feature vector, represents the L2 norm distance, is a preset constant threshold (e.g., γ = 0.5) used to control the minimum difference in the distances between positive and negative samples, represents taking the maximum value of the input value and 0; For each edge → in the multi-modal causal association graph, calculate the causal strength based on conditional mutual information: ; Among them, is 's set of parent nodes, represents conditional mutual information; It should be noted that conditional mutual information is an index in information theory that measures the statistical dependence degree between two variables (such as and ) given other variables . Its core logic is: when the set of parent nodes of is known, 's additional information contribution to .

[0049] Set the causal strength threshold , retain 's edges, filter low-significance associations (such as 's edges are removed), and generate a causal strength quantification matrix.

[0050] It should be noted that to set the causal strength threshold , first, based on the constructed multimodal causal association graph, calculate the causal strength → of each edge , and obtain the complete causal strength distribution; by analyzing the distribution characteristics of the causal strength values, draw a histogram to observe its central tendency and dispersion; select the significantly high-value interval (such as the top 30% high-value interval) based on the inflection point characteristics of the distribution curve as the retention range; calculate the corresponding numerical threshold 0.7 from this distribution; use the retained causal edges to construct a simplified causal graph and verify its medical rationality; finally, determine that the causal strength threshold is set to .

[0051] Store the causal strength quantification matrix as a Numpy file (.npz).

[0052] S3: Based on the causal strength quantification matrix, screen the associated nodes whose causal strength exceeds the preset causal strength threshold, and force-align the image slice feature vectors and gene mutation embedding vectors through a cross-modal contrast loss function to generate a multimodal joint feature representation with fused causal constraints.

[0053] Specifically, it includes the following steps: S3.1: Read the generated causal strength quantification matrix (.npz file).

[0054] Call the image features (CT value, volume, etc.) and the converted binary features of gene mutations as the original features to be aligned.

[0055] Based on the preset causal strength threshold , retain all causal edges that satisfy .

[0056] For example: Retain the edge with causal strength (such as gene mutation "EGFR L858R" → image feature "abnormal CT value area" extracted by S1.2); remove the low-significance edge with M = 0.62 (such as gene mutation "ALK fusion" → behavior event "aggravated cough").

[0057] Only contains the retained causal edges, where the node set covers image slice feature vectors, gene mutations, and behavior event sequence features.

[0058] S3.2: Uniformly sample from non-causally associated cross-modal nodes to generate a set of negative sample pairs.

[0059] Among them, non-causally associated cross-modal nodes refer to cross-modal variable combinations (such as gene mutation features and imaging features) that satisfy the causal strength conditions in the multi-modal causal association graph constructed in the present invention, and there is no significant causal relationship between these variable pairs verified by the PC algorithm.

[0060] Specifically, for positive sample pair generation: If there is an edge → (such as gene mutation → imaging slice feature vector) in the simplified causal graph, then extract the and feature vectors from the gene mutation feature and the imaging slice feature vector respectively to form a set of positive sample pairs.

[0061] For negative sample pair generation: Traverse all cross-modal node combinations, screen the gene mutation-imaging slice feature vector pairs that satisfy , and select a set of negative sample pairs that match the number in the set of positive sample pairs from the screening results by uniform sampling to ensure data balance.

[0062] Based on the constructed sets of positive and negative sample pairs, adopt a contrastive learning framework weighted by causal strength to construct a contrastive loss function , expressed as: ; In the formula, represents the set of positive sample pairs, represents the set of negative sample pairs, represents the gene mutation embedding vector (gene mutation binary feature), represents the imaging slice feature vector (extraction of lesion area imaging features), represents the negative sample distance threshold, represents the cross-modal feature vector in the negative sample pair, represents the anchor (Anchor) feature index in the loss function, represents the contrast feature index in the negative sample pair, represents the target feature index in the positive sample pair; It should be noted that the extraction of lesion area imaging features refers to the process of locating the lesion area from medical images (such as CT / MRI) and extracting its quantitative features; the imaging slice feature vector value refers to the final output form of "extraction of lesion area imaging features", that is, the high-dimensional vector encoded by the neural network.

[0063] S3.3: Input the lesion region image features (CT value, volume) extracted by the medical image registration (Elastix) toolkit into a pre-trained 3D image classification neural network (3D ResNet-50) model (ImageNet pre-trained weights), and output a 512-dimensional feature vector; Input positive sample pairs (gene embeddings, image features) and negative sample pairs for alignment training; Optimization objective: Minimize the defined contrastive loss L to reduce the distance between positive sample pairs and increase the distance between negative sample pairs; Training parameters: Adam optimizer (learning rate ), batch size 32, iterate 1000 times.

[0064] Monitor the decline rate of the distance between positive sample pairs (calculated based on the loss function in S3.2). If the decline rate is < 1% within 10 adjacent epochs, it is determined to converge.

[0065] S3.4: Concatenate the aligned image slice feature vectors and gene embedding vectors, input them into a fully connected layer for dimensionality reduction, and output a 128-dimensional joint feature , and the expression is: ; In the formula, represents the weight matrix (initialized according to the He normal distribution), represents the bias term; Randomly select samples, visualize the joint feature space through t-distributed stochastic neighbor embedding (t-SNE), and check whether strongly causally associated nodes (such as gene mutations and image features retained in S2.3) are clustered; For example, the distance between the gene mutation data "EGFR L858R" (S1.4) and the image feature "abnormal CT value region" in the joint feature space should be significantly less than that of unassociated nodes.

[0066] Represent the joint feature and store it in the HDF5 file generated in S1.5 to expand the original dataset.

[0067] S4: Input the multi-modal joint feature representation into a counterfactual inference engine, simulate the potential effects under different treatment plan interventions, calculate the individual treatment effect, and generate a counterfactual rehabilitation advice set.

[0068] Specifically, it includes the following steps: S4.1: Read the multi-modal joint feature representation (128-dimensional in the HDF5 file ) and the causal strength quantization matrix (.npz file).

[0069] Based on the causal edges retained in the causal strength quantization matrix , screen the set of intervenable variables.

[0070] Define the intervention variable and the outcome variable.

[0071] Specifically, for the definition of the intervention variable: select the root nodes in the causal graph that directly affect vital signs or behavior logs (such as binary features of gene mutations and lesion volume); for the definition of the outcome variable: vital sign monitoring data (such as blood glucose value, blood oxygen saturation) and key events in behavior logs (such as exercise duration, medication compliance).

[0072] Establish a mapping function between the intervention variable and the outcome variable.

[0073] Specifically, for each intervention variable (such as the binary feature of gene mutation "EGFR L858R"), define its adjustable range (such as the presence / absence of mutation); for continuous intervention variables (such as lesion volume in imaging features), define the adjustment range (such as ±10% of the current value).

[0074] Example: Intervention variable = gene mutation "EGFR L858R" (taking 0 or 1), outcome variable = mean continuous blood glucose monitoring (mg / dL) S4.2: Extract historical intervention-outcome pairs from the spatio-temporal tagged dataset.

[0075] Specifically, input the multi-modal joint feature representation (128 dimensions) and the current state of the intervention variable ; output vital sign monitoring data (such as blood glucose value time series) and behavior log events (such as number of exercises).

[0076] Train a potential outcome prediction model using the Doubly Robust Estimation framework.

[0077] Specifically, use the Gradient Boosting Regressor to predict the potential outcome after adjusting the intervention variable ; calibrate the counterfactual prediction bias based on the residuals between the real observed data and the prediction results.

[0078] Furthermore, verify the potential outcome prediction model, divide the training set and the validation set (8:2), calculate the mean absolute error (MAE) to evaluate the prediction accuracy. If the prediction error exceeds the clinically allowable range (such as blood glucose prediction error > 5 mg / dL), increase the sample size or adjust the depth of the regression tree.

[0079] It should be noted that the "clinically acceptable range" refers to the error margin acceptable in clinical decision-making for a certain indicator (such as blood glucose, tumor volume, etc.) specified in medical practice or treatment guidelines, ensuring that the prediction results have practical application value.

[0080] Based on the validated potential outcome prediction model, a set of candidate treatment plans is generated.

[0081] Among them, the set of candidate treatment plans includes drug intervention plans, behavior intervention plan bases, and follow-up strategy plans. Specifically: Drug intervention plan: If there is a gene mutation → a strong causal edge of vital signs , extract the adjustable dose range of the corresponding drug from the clinical knowledge base; Behavior intervention plan: For behavior variables with significant causal associations (such as exercise duration → blood glucose), calculate the safe adjustment interval according to ±2 standard deviations; Follow-up strategy plan: According to the predicted change rate of the lesion volume, divide the reexamination interval levels by gradient. Output a structured set of candidate plans for counterfactual simulation.

[0082] S4.3: For each candidate treatment plan, modify the value of the intervention variable .

[0083] For example, if the intervention is related to gene mutation (such as the use of EGFR inhibitors), change the gene mutation feature from 1 → 0 (assuming the drug inhibits mutation expression); if the intervention is related to imaging features (such as radiotherapy reducing the lesion), reduce the lesion volume feature by 10%.

[0084] Calculate the difference between the potential outcome after intervention and the current state outcome.

[0085] For multiple outcome variables (such as blood glucose, blood oxygen, exercise duration), perform a weighted sum (such as blood glucose weight 0.6, blood oxygen weight 0.3). Screen the intervention plans with the top 10% in the ranking of the individual treatment effect (ITE) (such as "increasing the metformin dose by 5mg can reduce blood glucose by 15mg / dL"); Among them, "ITE" is the abbreviation of "Individual Treatment Effect", which refers to a quantitative index of the personalized effect difference generated by different treatment plans for a specific patient compared with the baseline state.

[0086] Combined with the constraints of medical guidelines, exclude suggestions beyond the reasonable range (such as dose adjustment exceeding the upper limit of the pharmacopoeia); Among them, the constraints of medical guidelines refer to the treatment plan selection boundaries clearly defined in the treatment guidelines by authoritative medical institutions (such as the Chinese Medical Association, NCCN, etc.), including mandatory rules such as drug dose ranges, lists of contraindications, and treatment indications.

[0087] Classify and store them as a JSON file according to the intervention type as a set of counterfactual rehabilitation suggestions.

[0088] Among them, the counterfactual rehabilitation advice set includes the following specific intervention plans.

[0089] Specifically, the drug dosage adjustment plan: According to the patient's individual conditions, combined with the drug metabolism characteristics and changes in clinical indicators, dynamically adjust the drug dosage. For example, for patients with hypercholesterolemia, the dosage of statins can be appropriately increased to optimize the blood lipid regulation effect, while monitoring liver function and creatine kinase levels to ensure drug safety.

[0090] Follow-up cycle optimization strategy: Based on medical imaging data and the dynamic evolution trend of lesions, customize the reexamination frequency individually. If it is detected that the lesion progression speed accelerates, shorten the imaging follow-up interval to more closely track the disease changes and adjust the treatment plan in a timely manner.

[0091] Behavioral intervention plan: Integrate the patient's daily behavior data and physiological indicators to design an adaptive health management plan. For example, when it is detected that the blood oxygen saturation drops, it is recommended to increase the duration of low-intensity aerobic exercise and combine it with breathing training to gradually improve cardiopulmonary function.

[0092] S5: According to the counterfactual rehabilitation advice set, construct a dynamic causal reinforcement decision-making model, continuously receive real-time physiological data feedback of the patient through the proximal policy optimization algorithm, and dynamically adjust the treatment plan.

[0093] Specifically, it includes the following steps: S5.1: Construct a state space based on real-time physiological data and the counterfactual rehabilitation advice set, and define an executable action space.

[0094] Specifically, extract the patient's real-time physiological data (such as blood glucose value, blood oxygen saturation), current treatment plan parameters (such as drug dosage, follow-up cycle), key intervention variables in the counterfactual rehabilitation advice set (such as gene mutation status, lesion volume adjustment advice), and the retained edges in the causal strength quantification matrix from the HDF5 file .

[0095] Align the patient's real-time physiological data, current treatment plan parameters, key intervention variables, and the retained edges in the causal strength quantification matrix according to the time stamp to form a state vector.

[0096] Among them, the state vector includes dimensions: vital sign values, drug dosage, gene mutation binary status, lesion volume, and causal edge strength values.

[0097] Furthermore, according to the JSON file of the counterfactual rehabilitation advice set, map the executable actions to a discrete action space.

[0098] Among them, the action types include: drug dosage adjustment, follow-up cycle adjustment, and behavioral intervention parameters.

[0099] Example: The state vector includes a blood glucose value of 126 mg / dL, a blood oxygen saturation of 98%, a metformin dose of 500 mg / d, a status 1 of the gene mutation "EGFR L858R", a lesion volume of 3.2 cm³, and a causal edge "EGFR L858R → abnormal CT value area" with an intensity of 0.85. The action types can be selected as "metformin dose + 5 mg", "follow-up period shortened by 3 days", and "exercise duration increased by 15 minutes".

[0100] S5.2: Initialize the policy and value network parameters based on the state space and action space.

[0101] Specifically, when constructing: The policy network adopts a two-layer fully connected structure. The first fully connected layer (FC1-256, ReLU activation) maps the state vector to a 256-dimensional feature space, and the second fully connected layer (FC2-128, Softmax output) generates an action probability distribution.

[0102] Furthermore, the value network adopts the first fully connected layer (FC1-256, ReLU activation) and the second fully connected layer (FC2-1, linear output) with the same hierarchical configuration but independent parameters for state value estimation.

[0103] Among them, the network structure is consistent with the policy network, and the parameters are independently initialized.

[0104] Configure the Adam optimizer: The learning rate of the policy network is 1e-4, the learning rate of the value network is 3e-4, the batch size is 32, and the discount factor γ = 0.99.

[0105] S5.3: Execute multi-objective reward-driven policy training based on network parameters and real-time data.

[0106] Real-time collect the latest physiological data of the patient (through wearable devices and HDF5 files), update the state vector, input it into the policy network to generate an action probability distribution, and select the action type according to the softmax probability sampling.

[0107] Monitor the changes in vital signs after executing the action type, and calculate the mean square error (MSE) with the target value.

[0108] Based on the causal edge between gene mutation and risk in the causal intensity quantization matrix , calculate the probability increment of the cumulative risk of gene mutation, and after weighting, superimpose it on the total reward.

[0109] Store the state transition records in an independent dataset of the HDF5 file, and store them partitioned by timestamp.

[0110] S5.4: Implement policy optimization and dynamic safety verification based on experience replay data.

[0111] Randomly sample a batch of data (batch size 32) from the experience replay data and calculate the advantage function.

[0112] Calculate the policy gradient loss, restricting the probability ratio of the new and old policies to be in the interval [0.8, 1.2] to prevent policy mutations.

[0113] Minimize the prediction error of the value function and calculate the mean squared error (MSE) loss.

[0114] If the action type causes a physiological index to exceed the limit (such as heart rate > 180 bpm), immediately terminate and roll back to the previous effective policy.

[0115] S5.5: Based on the optimization results, update the policy parameters and trigger periodic retraining.

[0116] Save the weights of the trained policy network and the value network to independent groups in the HDF5 file respectively.

[0117] Record the action execution log (timestamp, action type, actual effect, reward value) to the / logs group of the HDF5 file, partitioned by patient ID and date.

[0118] When the amount of experience replay data reaches 1000 or the patient's physiological state changes significantly (such as a new gene mutation is detected), trigger a full-data retraining.

[0119] This embodiment also provides a digital intelligent tumor prevention and control management platform, including: a data preprocessing module, a causal modeling module, a feature fusion module, a counterfactual reasoning module, and a dynamic decision-making module. The data preprocessing module is used to collect the patient's multi-modal data in real time, and perform standardized cleaning and spatio-temporal alignment processing to generate a spatio-temporal labeled data set; the causal modeling module is used to input the spatio-temporal labeled data set into a spatio-temporal causal alignment network, extract temporal features using a temporal convolutional network, and construct a multi-modal causal association graph through a causal discovery algorithm, identify the causal chain between imaging features, gene mutations, and patient behavior, and output a causal strength quantization matrix; the feature fusion module is used to filter the associated nodes with a causal strength exceeding a preset causal strength threshold based on the causal strength quantization matrix, and force-align the imaging slice feature vectors and gene mutation embedding vectors through a cross-modal contrast loss function to generate a multi-modal joint feature representation with fused causal constraints; the counterfactual reasoning module is used to input the multi-modal joint feature representation into a counterfactual reasoning engine, simulate the potential effects under different treatment plan interventions, calculate the individual treatment effect, and generate a counterfactual rehabilitation advice set; the dynamic decision-making module is used to construct a dynamic causal reinforcement decision model based on the counterfactual rehabilitation advice set, continuously receive real-time physiological data feedback of the patient through the proximal policy optimization algorithm, and dynamically adjust the treatment plan.

[0120] This embodiment also provides a computer device, which is applicable to the situation of the digital and intelligent tumor prevention and control management method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the digital and intelligent tumor prevention and control management method proposed in the above embodiment.

[0121] The computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0122] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the digital and intelligent tumor prevention and control management method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0123] In summary, the present invention processes multi-modal medical images, genomic sequencing data, and clinical monitoring data through spatio-temporal alignment and feature fusion, accurately captures the dynamic relationships of biomarkers with significant clinical relevance during tumor progression, and improves the reliability of causal inference. By calculating the covariance matrix of gene mutation frequencies, lesion volume change rates, and physiological index fluctuations, spatio-temporal correlation features of multi-modal data are extracted, and multi-dimensional features with clear pathological manifestations are refined therefrom, enhancing the adaptability and generalization ability of the dynamic decision-making model to individualized treatment plans. By combining the regularization constraint based on tumor biology with the causal inference model, integrating pathological mechanisms and machine learning algorithms, the biological rationality and interpretability of the causal inference model are enhanced, enabling the treatment recommendation results output by the dynamic decision-making model to have verifiable clinical credibility.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A digitalized tumor prevention and control management method, characterized in that: including, real-time collection of multimodal data of patients, and performing standardized cleaning and spatio-temporal alignment processing to generate a spatio-temporal labeled dataset; inputting the spatio-temporal labeled dataset into a spatio-temporal causal alignment network, using a temporal convolutional network to extract temporal features, and constructing a multimodal causal association graph through a causal discovery algorithm, identifying the causal chain between imaging features, gene mutations and patient behaviors, and outputting a causal intensity quantification matrix; based on the causal intensity quantification matrix, screening associated nodes with causal intensity exceeding a preset causal intensity threshold, and forcing the alignment of the imaging slice feature vector and the gene mutation embedding vector through a cross-modal contrast loss function to generate a multimodal joint feature representation incorporating causal constraints; inputting the multimodal joint feature representation into a counterfactual reasoning engine, simulating the potential effects under different treatment plan interventions, calculating the individual treatment effect, and generating a counterfactual rehabilitation advice set; according to the counterfactual rehabilitation advice set, constructing a dynamic causal reinforcement decision-making model, and continuously receiving real-time physiological data feedback of the patient through a proximal policy optimization algorithm to dynamically adjust the treatment plan.

2. The digitalized tumor prevention and control management method according to claim 1, wherein: The standardized cleaning and spatio-temporal alignment processing includes performing standardized processing on the imaging features of the lesion area, and using a dynamic time warping algorithm to align the vital sign monitoring data and the behavior log time axis.

3. The digital intelligent tumor prevention and control management method according to claim 1, characterized in that: The causal discovery algorithm refers to establishing a causal pointing rule based on the timestamp priority principle, and constructing a multimodal causal association graph through conditional independence testing.

4. The digital intelligent tumor prevention and control management method according to claim 3, characterized in that: The cross-modal contrast loss function includes the following steps. Based on the causal intensity quantification matrix, screening causal edges, and extracting the imaging slice feature vector and the gene mutation embedding vector as positive sample pairs; uniformly sampling from non-causally associated cross-modal nodes to generate a negative sample pair set; based on the constructed positive and negative sample pair set, adopting a contrast learning framework weighted by causal intensity, constructing a contrast loss function, minimizing the contrast loss through an Adam optimizer, and monitoring the decline rate of the positive sample pair distance.

5. The digital intelligent tumor prevention and control management method according to claim 4, wherein: The counterfactual reasoning engine adopts the following steps. Based on the causal edges retained in the causal intensity quantification matrix, selecting the gene mutation binary feature and the lesion volume as intervenable variables; inputting the intervention variables into a double-robust estimation framework, combining a gradient boosting regression tree to predict potential results, calibrating the counterfactual prediction bias through real data residuals, and verifying the mean absolute error.

6. The digital intelligent tumor prevention and control management method according to claim 1, characterized in that: The multimodal data includes medical images, gene sequencing results, continuous vital sign monitoring data, and behavior logs self-reported by patients.

7. The digital and intelligent tumor prevention and control management method according to claim 1, wherein: The counterfactual rehabilitation advice set includes a drug dose adjustment plan, a follow-up cycle optimization strategy, and a behavior intervention plan.

8. A digital intelligent tumor prevention and control management platform, based on the digital intelligent tumor prevention and control management method according to any one of claims 1 to 7, characterized in that: including a data preprocessing module, a causal modeling module, a feature fusion module, a counterfactual deduction module, and a dynamic decision-making module. The data preprocessing module is used for real-time collection of multimodal data of patients, and performing standardized cleaning and spatio-temporal alignment processing to generate a spatio-temporal labeled dataset; The causal modeling module is used for inputting the spatio-temporal labeled dataset into a spatio-temporal causal alignment network, using a temporal convolutional network to extract temporal features, and constructing a multimodal causal association graph through a causal discovery algorithm, identifying the causal chain between imaging features, gene mutations and patient behaviors, and outputting a causal intensity quantification matrix; The feature fusion module is used to screen associated nodes with causal strength exceeding a preset causal strength threshold based on the causal strength quantization matrix, and force the alignment of the image slice feature vectors and the gene mutation embedding vectors through a cross-modal contrast loss function to generate a multi-modal joint feature representation incorporating causal constraints; The counterfactual inference module is used to input the multi-modal joint feature representation into a counterfactual inference engine, simulate potential effects under different treatment plan interventions, calculate the individual treatment effect, and generate a counterfactual rehabilitation advice set; The dynamic decision-making module is used to construct a dynamic causal reinforcement decision model based on the counterfactual rehabilitation advice set, and continuously receive real-time physiological data feedback of the patient through a proximal policy optimization algorithm to dynamically adjust the treatment plan.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the digital intelligent tumor prevention and control management method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the digital intelligent tumor prevention and control management method according to any one of claims 1 to 7.

Images