Alzheimer's disease course dynamic prediction method, device, equipment and medium

Abstract

This application relates to a method, device, equipment, and medium for dynamic prediction of Alzheimer's disease course. The method includes: acquiring multidimensional detection data of a patient; determining the patient's disease subtype and disease stage based on the multidimensional detection data and clinical diagnostic criteria; performing time-series analysis on the multidimensional detection data according to the disease subtype and the disease stage to obtain the patient's disease progression pattern; predicting the patient's future disease trajectory using a preset disease trajectory prediction model based on the disease progression pattern to obtain the patient's Alzheimer's disease course prediction result; and generating a recommended intervention plan for the patient based on the Alzheimer's disease course prediction result and preset clinical intervention rules. This method enables dynamic prediction of the Alzheimer's disease course and provides personalized intervention suggestions for patients, improving clinical treatment outcomes.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical diagnosis and predictive analysis, and in particular relates to methods, devices, equipment and media for dynamic prediction of the course of Alzheimer's disease. Background Technology

[0002] With the continuous development of the medical and health field, especially the advancement of diagnostic technologies for neurodegenerative diseases, clinical assessment-based diagnostic techniques for Alzheimer's disease have emerged. These techniques typically rely on single or limited-dimensional test data such as neuropsychological scale scores and brain imaging examinations, combined with standard clinical guidelines, to perform static diagnosis and determine the patient's current stage of the disease. Their characteristics are real-time and standardized assessment procedures, but they lack the ability to capture the dynamic evolution of the disease course.

[0003] Traditional techniques for assessing the course of Alzheimer's disease primarily rely on regular clinical follow-ups and discrete time-point scale assessments, such as using tools like the Mini-Mental State Examination (MMSE) or ADAS-Cog (Alzheimer's Disease Assessment Scale - Cognitive Section) to score cognitive function, combined with MRI or PET imaging data, and using predefined thresholds to divide the disease stage. The process is often manual, rule-based, and lacks comprehensive data integration, failing to systematically incorporate multidimensional information.

[0004] Current traditional methods suffer from the following problems: They are largely based on static and isolated data points, failing to achieve temporal analysis and dynamic prediction of disease progression, and making it difficult to anticipate disease deterioration trends. Their diagnostic dimensions are narrow, failing to fully integrate multi-source data from neuropsychology, imaging, body fluid biomarkers, and physiological signals, leading to biased assessment results and susceptibility to subjective factors. Traditional methods lack personalized modeling capabilities, unable to generate dynamic intervention plans based on patient-specific disease progression patterns, limiting the accuracy and foresight of clinical diagnosis and treatment. These problems collectively restrict the effectiveness of Alzheimer's disease management, necessitating the introduction of intelligent dynamic prediction technologies to improve treatment outcomes. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, equipment, and medium for dynamic prediction of the course of Alzheimer's disease that can solve the above problems.

[0006] Firstly, this application provides a method for dynamically predicting the course of Alzheimer's disease, including:

[0007] Acquire multidimensional test data from patients, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker test data, and physiological signal data;

[0008] Based on multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage are determined;

[0009] Based on the disease subtype and disease stage, time-series analysis of multidimensional detection data is performed to obtain the patient's disease progression pattern;

[0010] Based on the disease progression pattern, the patient's future disease trajectory is predicted through a pre-set disease trajectory prediction model, thus obtaining the patient's Alzheimer's disease course prediction results.

[0011] Based on the Alzheimer's disease course prediction results, and according to the preset clinical intervention rules, a recommended intervention plan for the patient's disease course is generated.

[0012] In one embodiment, the clinical diagnostic criteria include disease subtype judgment features for each disease subtype, data analysis dimensions corresponding to each disease subtype, and disease course stage classification feature thresholds corresponding to each disease subtype, wherein the disease subtype judgment features are a set of feature vectors predefined based on clinical diagnostic guidelines.

[0013] Based on multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage were determined, including:

[0014] Neuropsychological scale scores are mapped to quantitative feature values ​​to obtain neuropsychological features;

[0015] Brain region morphological features are extracted from brain imaging measurement data to obtain brain imaging features;

[0016] Concentration feature codes were extracted from body fluid biomarker detection data to obtain body fluid biomarker features;

[0017] Extracting time-frequency domain features from physiological signal data yields physiological signal characteristics;

[0018] By integrating neuropsychological features, brain imaging features, humoral biomarker features, and physiological signal features, the basic disease characteristics are obtained;

[0019] The cosine similarity algorithm is used to calculate the feature similarity between basic symptom features and disease subtype judgment features;

[0020] Based on feature similarity and combined with a preset feature similarity threshold, the disease subtype is determined;

[0021] Data analysis dimensions and disease stage segmentation thresholds were retrieved from clinical diagnostic criteria;

[0022] For each target dimension in the data analysis dimension, according to the feature selection rules corresponding to that target dimension, matching feature data are selected from the basic symptom features to obtain disease stage feature data.

[0023] Based on the characteristic threshold for disease stage division, the characteristic data of disease stage are used to determine the threshold and thus determine the disease stage.

[0024] In one embodiment, time-series analysis is performed on multidimensional detection data based on disease subtype and disease stage to obtain the patient's disease progression pattern, including:

[0025] Based on the disease subtype and disease stage, and combined with the preset time window rules, the time window length and sliding step size for time series analysis are determined.

[0026] For each target dimension in the data analysis dimension, matching feature data is selected from the multidimensional detection data according to the feature selection rules corresponding to that target dimension, and the time series data to be analyzed is obtained.

[0027] Based on the time window length and sliding step size, the time series data to be analyzed is divided into sliding window segments to generate multiple data subsequences;

[0028] A dynamic time warping algorithm is used to perform temporal alignment and feature compensation on data subsequences to construct disease evolution sequences;

[0029] The disease progression sequence is input into a preset time-series pattern recognition model for pattern recognition to obtain the disease development pattern.

[0030] In one embodiment, the method further includes:

[0031] Obtain the clinical evolution characteristics and clinical course time series data corresponding to each disease subtype;

[0032] The feature dimensions and state transition constraints of the temporal pattern recognition model are determined based on clinical evolution characteristics.

[0033] Based on the feature dimension, the state space structure of the temporal pattern recognition model is constructed to obtain the initial model framework;

[0034] The state transition constraints are embedded into the initial model framework to obtain the initial temporal pattern recognition model;

[0035] Based on clinical disease time series data, the initial time series pattern recognition model is iteratively trained by adjusting the state space connection weights to obtain a preset time series pattern recognition model. The iterative training stops when the model output error is less than a preset error threshold.

[0036] In one embodiment, based on the disease progression pattern, the patient's future disease trajectory is predicted using a pre-set disease trajectory prediction model to obtain Alzheimer's disease course prediction results, including:

[0037] For each stage of the disease, features of the disease development pattern are extracted to generate a disease progression feature set;

[0038] The disease progression feature set is input into the disease trajectory prediction model to obtain the probability distribution of the patient's disease status at a preset time node in the future. The state space structure of the disease trajectory prediction model is pre-calibrated according to the clinical evolution law corresponding to the disease subtype.

[0039] Based on the probability distribution of disease course status, a disease course trajectory curve is generated by fitting.

[0040] By referring to the clinical course stage description system corresponding to the disease subtype, the disease course trajectory curve is mapped to the corresponding disease stage to obtain the Alzheimer's disease course prediction results.

[0041] In one embodiment, based on the Alzheimer's disease course prediction results, a course intervention recommendation plan is generated according to preset clinical intervention rules, including:

[0042] For each stage of Alzheimer's disease progression prediction, the severity scores of the starting point, ending point, and mutation point of that stage are extracted from the disease trajectory curve.

[0043] Based on the severity score and the disease course curve, the average, rate of change, and trend curvature of the severity score were calculated.

[0044] By integrating the mean, rate of change, and trend curvature, the disease status characteristics of each stage of the disease course are obtained;

[0045] For each stage of the disease, based on the characteristics of the disease state and according to the preset clinical intervention rules, disease intervention nodes and intervention plan parameters are generated;

[0046] By integrating Alzheimer's disease course prediction results, disease intervention nodes, and intervention program parameters, a recommended intervention program for the disease course is generated.

[0047] In one embodiment, Alzheimer's disease course prediction results, disease intervention nodes, and intervention plan parameters are integrated to generate a disease intervention recommendation plan, including:

[0048] The comprehensive quantitative representation matrix of the disease intervention recommendation plan is calculated using the following formula:

[0049]

[0050] Where S is the comprehensive quantitative representation matrix of the disease course intervention recommendation plan, m is the total number of disease stages, and k is the index of the disease stage. Let be the weighting coefficient for the k-th stage of the disease. This is a temporal fusion operator, used to perform tensor product operations on the disease progression prediction results of the k-th disease stage and the fusion features of intervention node-intervention protocol parameters. This is the quantization vector for the disease progression prediction result at the k-th stage of the disease. This is a composite operator for intervention nodes and intervention protocol parameters, used for nonlinear mapping and fusion of intervention nodes and intervention protocol parameters at the k-th stage of the disease. This represents the quantified value of the intervention node at the k-th stage of the disease progression. This is the set of intervention parameters for the k-th stage of the disease. This is the intervention suitability correction coefficient for the k-th stage of the disease. This is a disease stage-intervention node correlation function. The function outputs the prediction result for the k-th disease stage and the correlation scalar of the intervention node. This is an intervention protocol parameter-disease stage adaptation function, which outputs a quantitative value of the fit between the intervention protocol parameters and the clinical characteristics of the disease stage at the k-th disease stage. Let be the clinical feature threshold vector for the k-th stage of the disease. This is a global correction factor. This is a global association regularization function;

[0051] Based on the comprehensive quantitative characterization matrix and combined with the pre-set intervention protocol parameter library, a disease course intervention recommendation plan is generated.

[0052] Secondly, this application also provides a device for dynamically predicting the course of Alzheimer's disease, comprising:

[0053] The multidimensional data acquisition module is used to acquire multidimensional detection data of patients, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data.

[0054] The subtype stage determination module is used to determine the patient's disease subtype and disease stage based on multidimensional detection data and clinical diagnostic criteria;

[0055] The time-series pattern analysis module is used to perform time-series analysis on multidimensional detection data based on disease subtype and disease stage to obtain the patient's disease progression pattern.

[0056] The disease progression prediction module is used to predict the patient's future disease progression based on the disease development pattern and through a preset disease progression prediction model, so as to obtain the patient's Alzheimer's disease progression prediction results.

[0057] The intervention plan generation module is used to generate suggested intervention plans for patients based on the Alzheimer's disease course prediction results and according to preset clinical intervention rules.

[0058] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described dynamic prediction method for the course of Alzheimer's disease.

[0059] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for dynamically predicting the course of Alzheimer's disease.

[0060] The aforementioned method, device, computer equipment, and storage medium for dynamic prediction of Alzheimer's disease progression acquire multidimensional detection data, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data, achieving comprehensive coverage of the patient's condition. Based on the multidimensional detection data, and combined with clinical diagnostic criteria, the disease subtype and disease stage are dynamically determined. Through time-series analysis of the multidimensional detection data, the disease development pattern is obtained, and sliding window and dynamic time warping algorithms are introduced to capture the dynamic evolution of the disease progression. A pre-set disease trajectory prediction model is used to predict the patient's future disease trajectory, generating Alzheimer's disease progression prediction results and providing prospective disease warnings. Based on the prediction results and pre-set clinical intervention rules, personalized disease intervention suggestions are generated, achieving closed-loop management from prediction to intervention, improving the accuracy and proactivity of diagnosis and treatment. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of the Alzheimer's disease course dynamic prediction method of the present invention;

[0063] Figure 2 This is a structural diagram of the Alzheimer's disease progression dynamic prediction device of the present invention;

[0064] Figure 3 This is a structural diagram of an Alzheimer's disease progression dynamic prediction device according to an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0066] In one embodiment, such as Figure 1 As shown, a method for dynamically predicting the course of Alzheimer's disease is provided. This embodiment illustrates the application of this method to a terminal, but it is understood that the method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. The hardware architecture of the implementation environment includes medical testing equipment, an edge computing terminal, and a server. When it is necessary to dynamically predict the course of Alzheimer's disease and formulate intervention plans for patients in clinical practice, the medical testing equipment transmits the collected patient data to the edge computing terminal. After the terminal integrates the data, it uploads it to the server. The server executes the prediction method and generates prediction results and intervention plans, which are then sent back to the edge computing terminal. Doctors can view and adjust the plans through the terminal, realizing collaborative interaction between the terminal and the server.

[0067] In this embodiment, the method includes the following steps:

[0068] S01. Obtain multidimensional detection data from the patient, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data.

[0069] The multidimensional detection data includes neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data. The neuropsychological scale scores can quantify cognitive function status through assessment tools such as MMSE or ADAS-Cog. Brain imaging measurement data can capture the morphological features of brain regions using imaging technologies such as MRI or PET. Body fluid biomarker detection data can reflect pathological progress by analyzing the concentration of biomarkers such as Aβ or tau protein in blood or cerebrospinal fluid. Physiological signal data can extract time-frequency domain features from EEG or electrocardiogram signals to monitor physiological status. In practice, raw data can be collected through various medical testing devices, integrated by edge computing terminals, and transmitted to the server to provide a foundation for subsequent analysis.

[0070] S02, based on multidimensional detection data and combined with clinical diagnostic criteria, determines the patient's disease subtype and disease stage.

[0071] The clinical diagnostic criteria include a predefined set of disease subtype judgment feature vectors for each disease subtype, corresponding data analysis dimensions, and disease stage classification feature thresholds. In implementation, neuropsychological scale scores, brain imaging measurements, body fluid biomarker detection data, and physiological signal data from multidimensional data can be mapped or extracted into quantitative features (such as neuropsychological features, brain region morphological features, concentration feature encoding, and time-frequency domain features), and integrated into basic symptom features. A similarity algorithm (such as cosine similarity) is used to calculate the matching degree between these features and the predefined disease subtype judgment features to determine the disease subtype. Based on the data analysis dimensions and feature thresholds, basic symptom features are screened and thresholds are determined to classify disease stages. Through multidimensional data fusion and clinical rules, the determination of disease subtypes and disease stages is automated, improving diagnostic efficiency and accuracy.

[0072] S03, based on disease subtype and disease stage, performs time-series analysis on multidimensional detection data to obtain the patient's disease progression pattern.

[0073] Temporal analysis is a process of extracting dynamic evolution patterns from multi-source heterogeneous medical data based on time series mining technology. Disease development patterns characterize the evolutionary features and patterns of diseases in the time dimension. In implementation, the time window parameters (window length and sliding step size) of temporal analysis can be adaptively determined according to the disease subtype and disease stage. Feature data related to each analysis dimension are selected from multi-dimensional data to form the temporal sequence to be analyzed. Multiple data sub-sequences are generated by using a sliding window. The dynamic time warping algorithm is used for temporal alignment and feature compensation to construct a standardized disease evolution sequence. The sequence is input into a preset temporal pattern recognition model (such as a pattern classifier based on a state space model) for automatic pattern recognition, thereby extracting the disease development pattern that reflects the individualized disease progression trend of patients.

[0074] S04. Based on the disease progression pattern, the patient's future disease trajectory is predicted through a preset disease trajectory prediction model, thus obtaining the patient's Alzheimer's disease course prediction results.

[0075] Among them, the disease trajectory prediction model is a computational model based on the pre-calibrated state space structure of clinical evolution laws, used to output the probability distribution of the disease status at future time points. In implementation, features can be extracted from the disease development pattern to generate a disease evolution feature set covering different disease stages. This feature set is input into the disease trajectory prediction model, and the probability distribution of the patient's disease status at the preset future time point is obtained through the state transition calculation inside the model. Based on this probability distribution, a continuous probability density curve is generated as the disease trajectory curve. Referring to the clinical disease stage description system corresponding to the disease subtype, the key points on the trajectory curve are mapped to the specific disease stage to obtain a structured Alzheimer's disease course prediction result with time dimension.

[0076] S05, based on the Alzheimer's disease course prediction results, generates a recommended course intervention plan for the patient according to the preset clinical intervention rules.

[0077] The proposed intervention plan can be represented by a comprehensive quantitative representation matrix, which integrates prediction results, intervention nodes, and intervention plan parameters to provide personalized clinical guidance. The clinical intervention rules are based on predefined decision logic in medical guidelines, including parameters such as intervention suitability correction coefficients and clinical feature threshold vectors. In practice, the severity scores of each stage of the disease (such as scores for the start point, end point, and mutation point) can be extracted from the disease prediction results, and their statistical characteristics (such as average, rate of change, and trend curvature) can be calculated to obtain the disease status characteristics. Based on these characteristics and clinical intervention rules, quantitative values ​​of intervention nodes and a set of intervention plan parameters are generated. The prediction results, intervention nodes, and parameters can be integrated through a nonlinear mapping fusion function to generate a structured proposed intervention plan for the disease course, achieving precise intervention.

[0078] The aforementioned method for dynamic prediction of Alzheimer's disease progression overcomes the limitations of traditional methods, such as narrow diagnostic dimensions and isolated data, by acquiring multidimensional patient data, including neuropsychological scale scores, brain imaging measurements, body fluid biomarker data, and physiological signal data. It dynamically determines disease subtypes and disease stages by combining clinical diagnostic criteria, and improves the objectivity and accuracy of diagnosis by utilizing predefined disease subtype judgment characteristics and threshold determination mechanisms. Time-series analysis of multidimensional data based on disease subtypes and disease stages, incorporating sliding window partitioning and dynamic time warping algorithms, constructs a disease evolution sequence and identifies development patterns, effectively capturing the dynamic evolution of the disease progression. A pre-defined disease trajectory prediction model predicts future trajectories based on disease development patterns, generating structured disease prediction results and providing proactive early warning of disease deterioration trends. Personalized intervention recommendations are generated based on the prediction results and clinical intervention rules, forming a closed-loop management system from dynamic prediction to precise intervention, enhancing the initiative and personalization of diagnosis and treatment.

[0079] In one embodiment, the clinical diagnostic criteria include disease subtype judgment features for each disease subtype, data analysis dimensions corresponding to each disease subtype, and disease course stage classification feature thresholds corresponding to each disease subtype, wherein the disease subtype judgment features are a set of feature vectors predefined based on clinical diagnostic guidelines.

[0080] Based on multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage were determined, including:

[0081] S11.1 Map the neuropsychological scale scores to quantitative feature values ​​to obtain neuropsychological features;

[0082] S11.2 Extract brain region morphological features from brain imaging measurement data to obtain brain imaging features;

[0083] S11.3 Extract the concentration feature encoding of body fluid biomarker detection data to obtain body fluid biomarker features;

[0084] S11.4 Extract the time-frequency domain features of the physiological signal data to obtain the physiological signal features;

[0085] S12 integrates neuropsychological features, brain imaging features, body fluid biomarker features, and physiological signal features to obtain basic disease characteristics;

[0086] S13, using the cosine similarity algorithm, calculates the feature similarity between basic symptom features and disease subtype judgment features;

[0087] S14. Based on feature similarity and combined with a preset feature similarity threshold, determine the disease subtype;

[0088] S15, retrieve data analysis dimensions and disease stage segmentation thresholds from clinical diagnostic criteria;

[0089] S16, For each target dimension in the data analysis dimension, according to the feature selection rule corresponding to the target dimension, filter out matching feature data from the basic symptom features to obtain disease stage feature data;

[0090] S17, Based on the characteristic threshold for disease stage division, threshold determination is performed on the characteristic data of disease stage to determine the disease stage.

[0091] For example, neuropsychological scale scores can be converted into standardized quantitative feature values ​​according to preset quantitative mapping rules to obtain neuropsychological features that reflect the patient's cognitive function status. Brain imaging measurement data can be processed using brain region morphology extraction algorithms to identify and extract core features such as volume and morphological changes in key brain regions like the hippocampus and temporal lobe, thus obtaining brain imaging features. For body fluid biomarker detection data, concentration feature encoding is used to convert the concentration detection results of biomarkers such as Aβ and tau proteins in blood and cerebrospinal fluid into body fluid biomarker features according to preset encoding rules. Time-frequency domain analysis methods are used to analyze physiological signal data, extracting frequency and time-domain variation features from EEG, ECG, and other data to obtain physiological signal features. Finally, according to feature dimension alignment and fusion rules, neuropsychological features, brain imaging features, body fluid biomarker features, and... Physiological signal features are integrated into a basic symptom feature vector. A cosine similarity algorithm is used to calculate the feature similarity between this basic symptom feature vector and the feature vectors for each disease subtype in clinical diagnostic criteria. The calculated feature similarity is compared with a preset feature similarity threshold. Disease subtypes with feature similarity exceeding the threshold are identified as the patient's disease subtype. Data analysis dimensions and disease stage classification feature thresholds corresponding to this disease subtype are retrieved from the clinical diagnostic criteria. For each target dimension in the data analysis, feature data matching the target dimension is selected from the basic symptom features according to pre-defined feature selection rules that match the target dimension, and integrated into disease stage feature data. The disease stage feature data is then compared with the disease stage classification feature thresholds one by one. Based on the threshold range of the feature data, the patient's current disease stage is determined.

[0092] In one embodiment, time-series analysis is performed on multidimensional detection data based on disease subtype and disease stage to obtain the patient's disease progression pattern, including:

[0093] S21. Based on the disease subtype and disease stage, and combined with the preset time window rules, determine the time window length and sliding step size for time series analysis.

[0094] S22, For each target dimension in the data analysis dimension, according to the feature selection rule corresponding to the target dimension, filter out matching feature data from the multidimensional detection data to obtain the time series data to be analyzed;

[0095] S23, Based on the time window length and sliding step size, the time series data to be analyzed is divided into sliding window segments to generate multiple data subsequences;

[0096] S24 uses a dynamic time warping algorithm to perform temporal alignment and feature compensation on data subsequences to construct a disease evolution sequence;

[0097] S25, input the disease progression sequence into the preset time-series pattern recognition model, perform pattern recognition, and obtain the disease development pattern.

[0098] Specifically, the preset time window rules are pre-defined rules based on the differences in the clinical progression rate of different disease subtypes and the pathological changes at each stage of the disease. For example, for the amnesic Alzheimer's disease subtype (a disease subtype), the pathological changes in the early stages of the disease are relatively slow. In the time window rules, the corresponding time window length for this stage is 3 months, and the sliding step is 1 month. For the progressive stage of the non-amnesic subtype, because the pathological changes are faster, the time window rules are set to a time window length of 2 months and a sliding step of 15 days. Based on this, combined with the current patient's disease subtype and disease stage, the time window length and sliding step required for this time series analysis are matched and determined from the preset time window rules. Step size; For each target dimension in the aforementioned determined data analysis dimensions (such as neuropsychological dimension, brain imaging dimension, etc.), according to the feature selection rules corresponding to that target dimension (e.g., the feature selection rule for the neuropsychological dimension is to select continuous detection data of MMSE scale scores and ADAS-Cog scale scores, and the feature selection rule for the brain imaging dimension is to select continuous measurement data of hippocampal volume and temporal lobe cortex thickness), feature data that meet the rules are extracted and selected from the multidimensional detection data. These feature data are then arranged in chronological order to obtain the time series data to be analyzed; according to the determined time window length and sliding step size, a sliding window is applied to the time series data to be analyzed. The data segmentation process involves dividing the data into 10 subsequences. For example, if the time series data to be analyzed consists of 12 consecutive months of MMSE scores (tested monthly), with a time window length of 3 months and a sliding step of 1 month, then starting from month 1, data is extracted in 3-month windows (months 1-3, 2-4… 10-12), generating a total of 10 data subsequences. The Dynamic Time Warping (DTW) algorithm is then used to process these subsequences. This algorithm calculates the time warping distance between different subsequences to correct temporal misalignment caused by minor differences in detection time intervals, thus achieving temporal alignment. Simultaneously, for the few missing detection values ​​in the subsequences, a neighbor-based detection method is used. The linear interpolation method is used for feature compensation. All data subsequences after time alignment and feature compensation are integrated in chronological order to construct a disease evolution sequence that reflects the changes in the patient's condition over time. The preset time pattern recognition model is a model pre-trained based on the clinical evolution characteristics and clinical course time sequence data corresponding to each disease subtype (such as Hidden Markov Model, Long Short-Term Memory Network Model, etc.). The constructed disease evolution sequence is input into the model. The model identifies and analyzes the trend and rate of change of pathological features in the sequence and outputs the corresponding pattern classification result. This result is the patient's disease development pattern (such as slow progression, moderate progression, rapid progression, etc.).

[0099] In one embodiment, the method further includes:

[0100] S31, obtain the clinical evolution characteristics and clinical course time series data corresponding to each disease subtype;

[0101] S32, Determine the feature dimensions and state transition constraints of the temporal pattern recognition model based on clinical evolution characteristics;

[0102] S33. Based on the feature dimension, construct the state space structure of the temporal pattern recognition model to obtain the initial model framework;

[0103] S34, embed the state transition constraints into the initial model framework to obtain the initial temporal pattern recognition model;

[0104] S35, based on clinical course time series data, the initial time series pattern recognition model is iteratively trained by adjusting the state space connection weights to obtain a preset time series pattern recognition model. The iterative training stops when the model output error is less than a preset error threshold.

[0105] For example, by connecting to a multi-center clinical database (such as a database containing Alzheimer's cases from the neurology departments of 10 top-tier hospitals nationwide), clinically diagnosed case data for each disease subtype (such as amnesic Alzheimer's disease, non-amnesic Alzheimer's disease, etc.) can be screened. The clinical progression characteristics specifically refer to the pathological progression patterns unique to each subtype. For instance, the clinical progression characteristics of the amnesic subtype include early stage characterized by episodic memory decline (a monthly decrease of 0.2-0.3 points in the memory dimension of the MMSE scale), mid-stage accompanied by language function decline (a monthly decrease of 0.15-0.2 points in the language dimension), and late stage with comprehensive cognitive decline. The clinical course timeline data is the timeline from the initial diagnosis to the end of follow-up for each case. The dataset comprises continuous monitoring data (with a monitoring period of at least 2 years), including neuropsychological scale scores (MMSE, ADAS-Cog) every 3 months, brain imaging data (hippocampal volume, temporal lobe cortex thickness) every 6 months, and body fluid biomarker data (cerebrospinal fluid Aβ42 / Aβ40 ratio, plasma p-tau181 concentration) every 12 months. This data is then categorized by disease subtype to form a structured dataset. Based on the clinical evolution characteristics of each disease subtype, the feature dimensions and state transition constraints of the temporal pattern recognition model are determined. Taking the amnesic subtype as an example, considering its evolutionary characteristics of "memory decline → language decline → overall decline," the feature dimensions are determined to be three core dimensions: MMSE memory dimension... The study analyzed the rate of change in clinical scores, the rate of change in ADAS-Cog language dimension scores, and the rate of hippocampal atrophy. Based on the shortest duration of each stage in the clinical evolution of this subtype (at least 12 months for early stage and at least 18 months for mid-stage), state transition constraints were set as follows: "The transition from 'early stage' to 'mid-stage' in the model must meet the following time-space constraints: a duration ≥ 12 months requires 1 time-space constraint (1 step every 3 months, i.e., ≥ 4 steps); the transition from 'mid-stage' to 'late stage' requires 18 months requires 6 time-space constraints (≥ 6 steps)." Based on the determined feature dimensions, a state space structure for a temporal pattern recognition model (using a Hidden Markov Model as an example) was constructed, and the model's states were divided into those related to clinical progression. The disease progression is divided into three stages: "early stage," "intermediate stage," and "late stage." Each stage corresponds to an observation probability distribution range of three feature dimensions (e.g., the observation probability distribution range of the MMSE memory dimension score change rate in the early stage is [-0.3, -0.2] points / month). An initial model framework is formed by defining the number of stages, the observation distribution of feature dimensions for each stage, and the initial transition paths between stages. The determined state transition constraints are embedded into the initial model framework and implemented by adjusting the state transition probability matrix. For example, in the initial transition probability matrix, the initial value of the transition probability from "early stage → intermediate stage" is set to 0 (if ≥4 time steps are not met), and if the condition is met, an initial probability of 0 is assigned according to a preset rule.3. Set the initial transition probability of "mid-term state → late-term state" to 0 (if the requirement of ≥6 time steps is not met), and assign an initial probability of 0.25 when the requirement is met. Set the probability of other state transition paths that do not meet the constraints to 0 to obtain the initial time-series pattern recognition model. Iteratively train the initial time-series pattern recognition model based on clinical disease progression time-series data. During training, the gradient descent algorithm can be used to adjust the state space connection weights of the model (i.e., the transition probability weights between states and the observation probability weights of feature dimensions in each state). In each iteration, input the feature sequence from the clinical disease progression time-series data into the model, calculate the mean squared error (MSE) between the model's output state sequence and the labeled sequence of the actual clinical disease progression stage (as the model output error), and compare this error with a preset error threshold (e.g., 0.05). If the error is greater than the threshold, continue adjusting the weights and proceed to the next iteration; if the error is less than or equal to the threshold, stop the iteration. The resulting model is the preset time-series pattern recognition model.

[0106] In one embodiment, based on the disease progression pattern, the patient's future disease trajectory is predicted using a pre-set disease trajectory prediction model to obtain Alzheimer's disease course prediction results, including:

[0107] S41, for each stage of the disease, feature extraction of the disease development pattern is performed to generate a disease progression feature set;

[0108] S42, input the disease progression feature set into the disease trajectory prediction model to obtain the probability distribution of the patient's disease status at a future preset time node. The state space structure of the disease trajectory prediction model is pre-calibrated according to the clinical evolution law corresponding to the disease subtype.

[0109] S43, Based on the probability distribution of disease course status, fit and generate disease course trajectory curve;

[0110] S44, referring to the clinical course stage description system corresponding to the disease subtype, maps the course trajectory curve to the corresponding course stage to obtain the Alzheimer's disease course prediction result.

[0111] Specifically, for each stage of the patient's current and past disease course (e.g., early and middle stages), feature extraction algorithms (such as sliding window-based statistical feature extraction algorithms) are used to extract features of the disease progression pattern. The extracted features cover the changing patterns of core indicators within each disease stage, such as the monthly average change rate of MMSE scale scores and the annual rate of hippocampal volume atrophy in the early stage, and the quarterly change rate of ADAS-Cog scale scores and the annual decrease in the cerebrospinal fluid Aβ42 / Aβ40 ratio in the middle stage. These extracted features are then categorized and organized according to the disease stage to form a structured dataset containing both time and feature dimensions. The disease progression feature set; the preset disease trajectory prediction model is a time-series prediction model built on a deep learning framework (such as a long short-term memory recurrent neural network model). Its state space structure is pre-labeled according to the clinical progression pattern corresponding to the patient's disease subtype (such as amnesic Alzheimer's disease). For example, for the amnesic subtype, the clinical pattern of "early stage mainly characterized by memory decline → middle stage accompanied by language function decline → late stage with comprehensive cognitive impairment" is used. The model's state space is labeled with three core states: "early stage - middle stage - late stage". The transition logic between each state is related to the disease progression rate of the subtype (such as the average duration of the early stage is 1).Matching the disease progression feature set (5-2 years, with an average duration of 2-3 years for the middle stage), the generated disease progression feature set is input into the model. Through learning and extrapolating the temporal features in the feature set, the model outputs the probability data of the patient's disease progression status (early, middle, and late) at preset future time points (e.g., 3 months, 6 months, 1 year, and 2 years), i.e., the probability distribution of disease progression status (e.g., 92% probability of being in the early stage and 8% probability of being in the middle stage in the next 3 months, 35% probability of being in the early stage and 65% probability of being in the middle stage in the next year). A Gaussian process regression fitting algorithm can be used to curve-fit the obtained disease progression status probability distribution. During the fitting process, time is used as the horizontal axis (unit: month or year), and the probability value of each disease progression status is used as the vertical axis (value range: 0-1). By minimizing the fitting error (e.g., root mean square error), the curve parameters are adjusted to generate a disease progression trajectory curve that continuously reflects the changing trend of the probability of each disease progression status at different future time points. For example, in the next 1-2 years, the probability of the early stage gradually decreases from 35% to 5%, and the probability of the middle stage gradually decreases from 5% to 65%. The probability of a disease progression increases from 65% to 90%, and the probability of a late-stage state increases from 0 to 5%. The clinical course stage description system corresponding to the disease subtypes is a structured descriptive standard based on authoritative clinical guidelines (such as the NIA-AA Alzheimer's disease diagnostic guidelines). This system clearly defines the core indicator threshold ranges for each course stage (e.g., early stage MMSE score 21-26, hippocampal volume decreased by 5%-10% compared to the normal mean; mid-stage MMSE score 10-20, hippocampal volume decreased by 10%-20% compared to the normal mean). Referring to this system, the course state with the highest probability at each time point in the generated course trajectory curve is matched and mapped with the corresponding course stage in the system. For example, the "mid-stage state" with the highest probability in the next year is mapped to "moderate cognitive impairment stage (i.e., mid-stage)" in the description system, and the "late-stage state" with the highest probability in the next two years is mapped to "severe cognitive impairment stage (i.e., late-stage)", forming a clear conclusion containing the corresponding course stage at each future time point, i.e., the Alzheimer's disease course prediction result.

[0112] In one embodiment, based on the Alzheimer's disease course prediction results, a course intervention recommendation plan is generated according to preset clinical intervention rules, including:

[0113] S51, for each stage of Alzheimer's disease course prediction results, obtain the severity score of the starting point, ending point and mutation point of the stage from the course trajectory curve;

[0114] S52, based on the severity score and the disease course curve, calculate the average, rate of change and trend curvature of the severity score;

[0115] S53 integrates the mean, rate of change, and trend curvature to obtain the disease status characteristics of each stage of the disease course;

[0116] S54 generates disease intervention nodes and intervention plan parameters based on disease status characteristics and preset clinical intervention rules for each stage of the disease course.

[0117] S55 integrates Alzheimer's disease course prediction results, disease intervention nodes, and intervention program parameters to generate recommended disease intervention programs.

[0118] For example, for each disease stage clearly defined in the disease course prediction results, the severity scores of key nodes in each stage are extracted from the fitted disease course trajectory curve (taking a continuous curve with the MMSE scale score as the core indicator as an example). The starting point is the score of the first time node corresponding to that stage on the trajectory curve (e.g., the starting point of the mid-stage is an MMSE score of 18 points in the next 6 months), and the ending point is the score of the last time node corresponding to that stage (e.g., the ending point of the mid-stage is an MMSE score of 12 points in the next 1 year). Mutation points can be determined by identifying time nodes where the slope of the trajectory curve changes abruptly (e.g., the score changes from a point in the next 8 months). The score suddenly drops from 16 to 14, with the slope changing from -0.5 points / month to -1 point / month; this point is the mutation point, corresponding to a score of 14. The corresponding score data is directly obtained by reading the coordinate values ​​of the trajectory curve at the aforementioned node. Based on the extracted severity score and the complete time range covered by the disease trajectory curve for that stage, three key indicators are calculated. The average is the arithmetic mean of the scores at all time nodes within that stage (one node per month). For example, in the mid-stage, there are 6 nodes within 6 months, and the average is (18+17+16+14+13+12) / 6=15 points. The rate of change is the (terminal point score). - The starting score is divided by the duration of the stage (in months). The trend curvature is obtained by taking the second derivative of the fitting function of the trajectory curve. The average curvature within the stage is taken as the trend curvature (e.g., the average curvature of the mid-stage is -0.03, and the negative sign indicates that the score decline trend is accelerating). The feature vector integration method is used to integrate the average, rate of change, and trend curvature of each disease stage to form the disease status feature of that stage (e.g., the disease status feature of the mid-stage is [15 points, -1 point / month, -0.03]), ensuring that the feature data is structured and can be called by subsequent rules. The preset clinical intervention rules can be based on NIA-AA. The structured decision-making rules developed based on Alzheimer's disease clinical intervention guidelines and multicenter clinical data, such as the rule that "when the mid-stage disease status characteristics meet the criteria of 'MMSE average score of 12-20, change rate ≤ -0.8 points / month, trend curvature ≤ -0.02', the intervention node is set as clinical assessment + intervention every 2 weeks, and the intervention protocol parameters are donepezil 10mg / day (orally in the morning) + 60 minutes of cognitive training once a day (including memory association and language repetition modules)". Based on this, the disease status characteristics of each stage are matched with the conditions in the preset clinical intervention rules. If the mid-stage characteristics are [15 points, -1 point / month, -0.02], the intervention will be implemented.

[03] If the above rules are met, corresponding disease intervention nodes (every 2 weeks) and intervention parameters (donepezil dosage, cognitive training duration and modules) are generated. The Alzheimer's disease course prediction results (e.g., "in the middle stage in the next 6 months to 1 year, and into the late stage in 1 to 2 years"), intervention nodes for each stage (every 2 weeks for the middle stage, weekly for the late stage), and intervention parameters (drug dosage, training plan, and follow-up examination items for different stages) are integrated into a structured document. Intervention effect evaluation indicators are also added (e.g., monthly monitoring of MMSE score changes, and hippocampal volume re-examination every 3 months), forming a recommended disease intervention plan that includes "prediction stage - intervention time - intervention measures - parameter details - evaluation criteria".

[0119] In one embodiment, Alzheimer's disease course prediction results, disease intervention nodes, and intervention plan parameters are integrated to generate a disease intervention recommendation plan, including:

[0120] S61, calculate the comprehensive quantitative representation matrix of the disease intervention recommendation plan using the following formula:

[0121]

[0122] Where S is the comprehensive quantitative representation matrix of the disease course intervention recommendation plan, m is the total number of disease stages, and k is the index of the disease stage. Let be the weighting coefficient for the k-th stage of the disease. This is a temporal fusion operator, used to perform tensor product operations on the disease progression prediction results of the k-th disease stage and the fusion features of intervention node-intervention protocol parameters. This is the quantization vector for the disease progression prediction result at the k-th stage of the disease. This is a composite operator for intervention nodes and intervention protocol parameters, used for nonlinear mapping and fusion of intervention nodes and intervention protocol parameters at the k-th stage of the disease. This represents the quantified value of the intervention node at the k-th stage of the disease progression. This is the set of intervention parameters for the k-th stage of the disease. This is the intervention suitability correction coefficient for the k-th stage of the disease. This is a disease stage-intervention node correlation function. The function outputs the prediction result for the k-th disease stage and the correlation scalar of the intervention node. This is an intervention protocol parameter-disease stage adaptation function, which outputs a quantitative value of the fit between the intervention protocol parameters and the clinical characteristics of the disease stage at the k-th disease stage. Let be the clinical feature threshold vector for the k-th stage of the disease. This is a global correction factor. This is a global association regularization function;

[0123] S62, based on the comprehensive quantitative characterization matrix and combined with the preset intervention protocol parameter library, generates a disease course intervention suggestion plan.

[0124] Specifically, the total number of disease stages m to be included in the calculation can be determined based on the patient's disease course prediction results (if the prediction includes two stages, intermediate and late, then m=2), and a weight coefficient can be assigned to each stage (distinguished by index k). —The weighting is set based on the degree of impact of each stage on disease prognosis; for example, mid-stage intervention is more critical for delaying progression. The value is higher than that of the later stage. The disease progression prediction results for each stage (such as duration and core score range) are converted into a quantitative vector. Intervention nodes (such as intervention frequency) are converted into quantitative values. Intervention parameters (such as drug dosage and training duration) are organized into a parameter set. ; through the intervention node-intervention scheme parameter composite operator Will and Nonlinear fusion is performed to obtain a fusion value that reflects the overall characteristics of the intervention measures, and a temporal fusion operator is also used. Combine the fusion value with Combine, and according to Weighting allows for a more accurate capture of the correlation between prediction results and intervention measures. An intervention suitability correction coefficient is introduced. (Based on the clinical fit between the stage and the intervention), correlation function between disease stage and intervention node. (calculate and (degree of correlation), intervention protocol parameters - disease stage adaptation function (judge Does it meet the clinical characteristic threshold vector for this stage? These three factors are combined to correct for the fit between intervention measures and stages; through a global correction coefficient. and global association regularization function The integrated results of all stages are globally adjusted to ensure that the information of each stage is not conflicting and the overall logic is coherent. The resulting comprehensive quantitative representation matrix S can comprehensively and accurately cover the prediction results, intervention nodes, intervention parameters and the correlation between various elements. The preset intervention plan parameter library stores the correspondence between different combinations of quantitative features and actual clinical intervention measures (such as specific fusion values ​​and correlation ranges corresponding to specific drug dosages and training plans). By matching the quantitative indicators in the S matrix with the correspondence in the parameter library, the specific intervention operation content of each stage can be extracted. Combined with the disease course prediction results, a complete disease course intervention suggestion plan containing "prediction stage - intervention frequency - intervention measure details (drugs, training, etc.)" can be formed.

[0125] The aforementioned method for dynamic prediction of Alzheimer's disease progression, by acquiring multidimensional detection data encompassing neuropsychological scale scores, brain imaging measurements, body fluid biomarker detection data, and physiological signal data, overcomes the limitations of traditional methods that rely on narrow diagnostic dimensions and limited single-dimensional data. It achieves comprehensive multidimensional coverage of the patient's condition, addressing the issues of biased assessment results and susceptibility to subjective factors. Based on multidimensional detection data and clinical diagnostic criteria, the method determines the patient's disease subtype and stage. For each subtype and stage, time-series analysis of the multidimensional data is conducted to extract disease progression patterns, changing the traditional method's assessment based on static, isolated data points. By capturing the dynamic evolution of the disease progression, it achieves dynamic prediction of the disease, effectively warning of disease deterioration trends. Based on the disease progression patterns, a pre-set disease trajectory prediction model is used to obtain disease progression prediction results, and combined with clinical intervention rules, personalized disease progression intervention recommendations are generated. This compensates for the lack of personalized modeling capabilities in traditional methods, forming a closed loop from dynamic prediction to precise intervention, significantly improving the accuracy and foresight of Alzheimer's disease clinical diagnosis and treatment.

[0126] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0127] Based on the same inventive concept, this application also provides an Alzheimer's disease dynamic prediction device for implementing the aforementioned method for dynamic prediction of Alzheimer's disease progression. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more embodiments of the Alzheimer's disease dynamic prediction device provided below can be found in the limitations of the Alzheimer's disease dynamic prediction method described above, and will not be repeated here.

[0128] In one exemplary embodiment, such as Figure 2 As shown, a device for dynamic prediction of the course of Alzheimer's disease is provided, comprising:

[0129] The multidimensional data acquisition module 101 is used to acquire multidimensional detection data of patients, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data and physiological signal data.

[0130] The subtype stage determination module 102 is used to determine the patient's disease subtype and disease stage based on multidimensional detection data and clinical diagnostic criteria.

[0131] The time-series pattern analysis module 103 is used to perform time-series analysis on multidimensional detection data based on disease subtype and disease stage to obtain the patient's disease progression pattern.

[0132] The disease course trajectory prediction module 104 is used to predict the patient's future disease course trajectory based on the disease development pattern and through a preset disease course trajectory prediction model, so as to obtain the patient's Alzheimer's disease course prediction result.

[0133] The intervention plan generation module 105 is used to generate a suggested intervention plan for the patient's disease course based on the Alzheimer's disease course prediction results and according to the preset clinical intervention rules.

[0134] In one embodiment, in the subtype stage determination module 102, the clinical diagnostic criteria include disease subtype determination features for each disease subtype, data analysis dimensions corresponding to each disease subtype, and disease course stage division feature thresholds corresponding to each disease subtype, wherein the disease subtype determination features are a set of feature vectors predefined based on clinical diagnostic guidelines.

[0135] Based on multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage were determined, including:

[0136] Neuropsychological scale scores are mapped to quantitative feature values ​​to obtain neuropsychological features;

[0137] Brain region morphological features are extracted from brain imaging measurement data to obtain brain imaging features;

[0138] Concentration feature codes were extracted from body fluid biomarker detection data to obtain body fluid biomarker features;

[0139] Extracting time-frequency domain features from physiological signal data yields physiological signal characteristics;

[0140] By integrating neuropsychological features, brain imaging features, humoral biomarker features, and physiological signal features, the basic disease characteristics are obtained;

[0141] The cosine similarity algorithm is used to calculate the feature similarity between basic symptom features and disease subtype judgment features;

[0142] Based on feature similarity and combined with a preset feature similarity threshold, the disease subtype is determined;

[0143] Data analysis dimensions and disease stage segmentation thresholds were retrieved from clinical diagnostic criteria;

[0144] For each target dimension in the data analysis dimension, according to the feature selection rules corresponding to that target dimension, matching feature data are selected from the basic symptom features to obtain disease stage feature data.

[0145] Based on the characteristic threshold for disease stage division, the characteristic data of disease stage are used to determine the threshold and thus determine the disease stage.

[0146] In one embodiment, the root time-series pattern analysis module 103 is further configured to:

[0147] Based on the disease subtype and disease stage, and combined with the preset time window rules, the time window length and sliding step size for time series analysis are determined.

[0148] For each target dimension in the data analysis dimension, matching feature data is selected from the multidimensional detection data according to the feature selection rules corresponding to that target dimension, and the time series data to be analyzed is obtained.

[0149] Based on the time window length and sliding step size, the time series data to be analyzed is divided into sliding window segments to generate multiple data subsequences;

[0150] A dynamic time warping algorithm is used to perform temporal alignment and feature compensation on data subsequences to construct disease evolution sequences;

[0151] The disease progression sequence is input into a preset time-series pattern recognition model for pattern recognition to obtain the disease development pattern.

[0152] In one embodiment, such as Figure 3 As shown, the time series pattern analysis module 103 also includes a recognition model construction unit 106, used for:

[0153] Obtain the clinical evolution characteristics and clinical course time series data corresponding to each disease subtype;

[0154] The feature dimensions and state transition constraints of the temporal pattern recognition model are determined based on clinical evolution characteristics.

[0155] Based on the feature dimension, the state space structure of the temporal pattern recognition model is constructed to obtain the initial model framework;

[0156] The state transition constraints are embedded into the initial model framework to obtain the initial temporal pattern recognition model;

[0157] Based on clinical disease time series data, the initial time series pattern recognition model is iteratively trained by adjusting the state space connection weights to obtain a preset time series pattern recognition model. The iterative training stops when the model output error is less than a preset error threshold.

[0158] In one embodiment, the disease progression prediction module 104 is further configured to:

[0159] For each stage of the disease, features of the disease development pattern are extracted to generate a disease progression feature set;

[0160] The disease progression feature set is input into the disease trajectory prediction model to obtain the probability distribution of the patient's disease status at a preset time node in the future. The state space structure of the disease trajectory prediction model is pre-calibrated according to the clinical evolution law corresponding to the disease subtype.

[0161] Based on the probability distribution of disease course status, a disease course trajectory curve is generated by fitting.

[0162] By referring to the clinical course stage description system corresponding to the disease subtype, the disease course trajectory curve is mapped to the corresponding disease stage to obtain the Alzheimer's disease course prediction results.

[0163] In one embodiment, the intervention plan generation module 105 is further configured to:

[0164] For each stage of Alzheimer's disease progression prediction, the severity scores of the starting point, ending point, and mutation point of that stage are extracted from the disease trajectory curve.

[0165] Based on the severity score and the disease course curve, the average, rate of change, and trend curvature of the severity score were calculated.

[0166] By integrating the mean, rate of change, and trend curvature, the disease status characteristics of each stage of the disease course are obtained;

[0167] For each stage of the disease, based on the characteristics of the disease state and according to the preset clinical intervention rules, disease intervention nodes and intervention plan parameters are generated;

[0168] By integrating Alzheimer's disease course prediction results, disease intervention nodes, and intervention program parameters, a recommended intervention program for the disease course is generated.

[0169] In one embodiment, the intervention plan generation module 105 is further configured to:

[0170] The comprehensive quantitative representation matrix of the disease intervention recommendation plan is calculated using the following formula:

[0171]

[0172] Where S is the comprehensive quantitative representation matrix of the disease course intervention recommendation plan, m is the total number of disease stages, and k is the index of the disease stage. Let be the weighting coefficient for the k-th stage of the disease. This is a temporal fusion operator, used to perform tensor product operations on the disease progression prediction results of the k-th disease stage and the fusion features of intervention node-intervention protocol parameters. This is the quantization vector for the disease progression prediction result at the k-th stage of the disease. This is a composite operator for intervention nodes and intervention protocol parameters, used for nonlinear mapping and fusion of intervention nodes and intervention protocol parameters at the k-th stage of the disease. This represents the quantified value of the intervention node at the k-th stage of the disease progression. This is the set of intervention parameters for the k-th stage of the disease. This is the intervention suitability correction coefficient for the k-th stage of the disease. This is a disease stage-intervention node correlation function. The function outputs the prediction result for the k-th disease stage and the correlation scalar of the intervention node. This is an intervention protocol parameter-disease stage adaptation function, which outputs a quantitative value of the fit between the intervention protocol parameters and the clinical characteristics of the disease stage at the k-th disease stage. Let be the clinical feature threshold vector for the k-th stage of the disease. This is a global correction factor. This is a global association regularization function;

[0173] Based on the comprehensive quantitative characterization matrix and combined with the pre-set intervention protocol parameter library, a disease course intervention recommendation plan is generated.

[0174] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the previously described method for dynamically predicting the course of Alzheimer's disease.

[0175] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0176] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0177] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for dynamic prediction of the course of Alzheimer's disease, characterized in that, The method includes: Acquire multidimensional detection data of patients, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data; Based on the multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage were determined. Based on the disease subtype and the disease stage, the multidimensional detection data are analyzed over time to obtain the patient's disease progression pattern; Based on the disease progression pattern, the patient's future disease trajectory is predicted using a preset disease trajectory prediction model, thus obtaining the patient's Alzheimer's disease course prediction result. Based on the Alzheimer's disease course prediction results, and according to preset clinical intervention rules, a course intervention recommendation plan is generated for the patient.

2. The method according to claim 1, characterized in that, The clinical diagnostic criteria include disease subtype judgment features for each disease subtype, data analysis dimensions corresponding to each disease subtype, and disease course stage classification feature thresholds corresponding to each disease subtype, wherein the disease subtype judgment features are a set of feature vectors predefined based on clinical diagnostic guidelines; Based on the multidimensional detection data and combined with clinical diagnostic criteria, the patient's disease subtype and disease stage are determined, including: The neuropsychological scale scores are mapped to quantitative feature values ​​to obtain neuropsychological features; Brain region morphological features are extracted from the brain imaging measurement data to obtain brain imaging features; The concentration feature codes of the body fluid biomarker detection data are extracted to obtain the body fluid biomarker features; Extract the time-frequency domain features of the physiological signal data to obtain the physiological signal features; By integrating the aforementioned neuropsychological features, brain imaging features, body fluid biomarker features, and physiological signal features, the basic disease characteristics are obtained; The cosine similarity algorithm is used to calculate the feature similarity between the basic symptom features and the disease subtype judgment features; Based on the feature similarity and combined with a preset feature similarity threshold, the disease subtype is determined; Retrieve the data analysis dimensions and the disease stage segmentation characteristic thresholds from the clinical diagnostic criteria; For each target dimension in the data analysis dimensions, matching feature data is selected from the basic symptom features according to the feature selection rules corresponding to the target dimension to obtain disease stage feature data. Based on the disease stage segmentation feature threshold, the disease stage feature data is subjected to threshold determination to determine the disease stage.

3. The method according to claim 2, characterized in that, The step of performing time-series analysis on the multidimensional detection data based on the disease subtype and the disease stage to obtain the patient's disease progression pattern includes: Based on the disease subtype and the disease stage, and in conjunction with preset time window rules, the time window length and sliding step size of the time series analysis are determined. For each target dimension in the data analysis dimensions, matching feature data is selected from the multidimensional detection data according to the feature selection rules corresponding to the target dimension to obtain the time series data to be analyzed; Based on the time window length and the sliding step size, the time series data to be analyzed is divided into sliding window segments to generate multiple data subsequences; A dynamic time warping algorithm is used to perform temporal alignment and feature compensation on the data subsequences to construct a disease evolution sequence; The disease progression sequence is input into a preset time-series pattern recognition model for pattern recognition to obtain the disease development pattern.

4. The method according to claim 3, characterized in that, The method further includes: Obtain the clinical evolution characteristics and clinical course time series data corresponding to each of the disease subtypes; The feature dimensions and state transition constraints of the temporal pattern recognition model are determined based on the clinical evolution characteristics. Based on the aforementioned feature dimensions, the state space structure of the temporal pattern recognition model is constructed to obtain the initial model framework; The state transition constraints are embedded into the initial model framework to obtain the initial temporal pattern recognition model; Based on the aforementioned clinical course time series data, the initial time series pattern recognition model is iteratively trained by adjusting the state space connection weights to obtain the preset time series pattern recognition model. The iterative training stops when the model output error is less than a preset error threshold.

5. The method according to claim 1, characterized in that, The step of predicting the patient's future disease trajectory based on the disease progression pattern using a preset disease trajectory prediction model to obtain Alzheimer's disease course prediction results includes: For each of the aforementioned disease stages, feature extraction is performed on the disease development pattern to generate a disease progression feature set; The disease progression feature set is input into the disease trajectory prediction model to obtain the probability distribution of the patient's disease status at a preset time node in the future. The state space structure of the disease trajectory prediction model is pre-calibrated according to the clinical evolution law corresponding to the disease subtype. Based on the probability distribution of the disease course status, a disease course trajectory curve is fitted and generated; Referring to the clinical course stage description system corresponding to the disease subtype, the course trajectory curve is mapped to the corresponding course stage to obtain the Alzheimer's disease course prediction result.

6. The method according to claim 5, characterized in that, Based on the Alzheimer's disease course prediction results, and according to preset clinical intervention rules, a course intervention recommendation plan is generated, including: For each stage of the Alzheimer's disease course prediction results, the severity scores of the starting point, ending point, and mutation point of that stage are extracted from the course trajectory curve. Based on the severity score of the illness and the disease course curve, the average value, rate of change, and trend curvature of the severity score of the illness are calculated. By integrating the average value, the rate of change, and the trend curvature, the disease status characteristics of each stage of the disease are obtained; For each of the aforementioned disease stages, based on the characteristics of the disease state, and according to preset clinical intervention rules, disease intervention nodes and intervention plan parameters are generated; By integrating the Alzheimer's disease course prediction results, the disease course intervention nodes, and the intervention plan parameters, the recommended disease course intervention plan is generated.

7. The method according to claim 6, characterized in that, The process of integrating the Alzheimer's disease course prediction results, the disease course intervention nodes, and the intervention plan parameters to generate the disease course intervention suggestion plan includes: The comprehensive quantitative representation matrix of the disease intervention recommendation plan is calculated using the following formula: Where S is the comprehensive quantitative representation matrix of the disease course intervention recommendation plan, m is the total number of disease stages, and k is the index of the disease stage. Let be the weighting coefficient for the k-th stage of the disease. This is a temporal fusion operator, used to perform tensor product operations on the disease progression prediction results of the k-th disease stage and the fusion features of the intervention node-intervention plan parameters. This is the quantization vector for the disease progression prediction result at the k-th stage of the disease. This is a composite operator for intervention nodes and intervention protocol parameters, used for nonlinear mapping and fusion of intervention nodes and intervention protocol parameters at the k-th stage of the disease. This represents the quantified value of the intervention node at the k-th stage of the disease progression. This is the set of intervention parameters for the k-th stage of the disease. This is the intervention suitability correction coefficient for the k-th stage of the disease. This is a disease stage-intervention node correlation function, which outputs the prediction result of the k-th disease stage and the correlation scalar of the intervention node. This is an intervention protocol parameter-disease stage adaptation function, which outputs a quantitative value of the fit between the intervention protocol parameters and the clinical characteristics of the disease stage at the k-th disease stage. Let be the clinical feature threshold vector for the k-th stage of the disease. This is a global correction factor. This is a global association regularization function; Based on the comprehensive quantitative characterization matrix and combined with a pre-defined intervention protocol parameter library, the recommended intervention protocol for the disease course is generated.

8. A device for dynamically predicting the course of Alzheimer's disease, characterized in that, The device includes: The multidimensional data acquisition module is used to acquire multidimensional detection data of patients, including neuropsychological scale scores, brain imaging measurement data, body fluid biomarker detection data, and physiological signal data. The subtype stage determination module is used to determine the patient's disease subtype and disease stage based on the multidimensional detection data and in combination with clinical diagnostic criteria; The time-series pattern analysis module is used to perform time-series analysis on the multidimensional detection data based on the disease subtype and the disease stage to obtain the patient's disease development pattern. The disease course trajectory prediction module is used to predict the patient's future disease course trajectory based on the disease development pattern and through a preset disease course trajectory prediction model, so as to obtain the patient's Alzheimer's disease course prediction result. The intervention plan generation module is used to generate a suggested intervention plan for the patient's disease course based on the Alzheimer's disease course prediction results and according to preset clinical intervention rules.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the 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 a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images