Alzheimer's disease risk prediction method, device and system

Abstract

The invention belongs to the technical field of risk prediction, and particularly relates to an Alzheimer's disease risk prediction method, device and system. The method comprises the following steps: acquiring various detection data of a to-be-detected object; inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the to-be-detected object with the time accumulation operator; inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the to-be-detected object with the time accumulation operator; and inputting the first risk prediction result and the second risk prediction result into a full connection layer, and outputting a third risk prediction result of the to-be-detected object at present and within a period of time in the future. The method is high in prediction precision and high in practicability, is helpful to timely make related preparations and prevention measures through early diagnosis of the Alzheimer's disease patients so as to slow down disease deterioration and improve the survival probability of the to-be-detected object, and has a good application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk prediction, and specifically relates to a method, device and system for predicting the risk of Alzheimer's disease. Background Art

[0002] In recent years, more and more researchers have used artificial intelligence methods to conduct predictive research on Alzheimer's Disease (AD). Deep Learning (DL) has shown great potential in clinical decision support for AD. Compared with other shallow learning models, the main advantage of DL is that they can directly learn the most predictive features from the raw data of a given labeled example dataset. However, studies have shown that the weights of deep models affect the results through multi-layer combinations, so they do not produce clinically meaningful solutions. In order to make the research more interpretable and clinically meaningful, machine learning methods are more advantageous. In addition, in the field of artificial intelligence, when using some machine learning methods such as PCA dimensionality reduction and deep learning methods for feature processing, the data will lose its original structure, which is not clinically meaningful and interpretable in the medical industry. In the AD prediction scenario, researchers often use baseline data combined with traditional machine learning methods for research, but these methods cannot consider time series data and cannot give phased prediction results.

[0003] At present, the current AD classification prediction research focuses on mining the features that affect the prediction results. For example, Chinese patent CN116994100A discloses a model training method, device, electronic device and storage medium. The method includes: obtaining a sample fundus image of the user that meets the model training conditions, and the sample fundus image is annotated with the real disease category of Alzheimer's disease; inputting the sample fundus image into the disease prediction model to be trained; calling the feature screening and vascular segmentation layer to perform feature screening and vascular segmentation and annotation processing on the sample fundus image to obtain an attention feature map and a vascular annotation feature map; calling the feature fusion layer to perform feature fusion processing on the attention feature map and the vascular annotation feature map to obtain the predicted disease probability corresponding to the sample fundus image; based on the real disease category and the predicted disease probability, calculating the loss value of the disease prediction model to be trained; when the loss value is within a preset range, obtaining a disease prediction model for predicting Alzheimer's disease. This patent can improve the prediction accuracy of Alzheimer's disease, however, this method ignores the research on the prediction model itself. As for the prediction model itself, the existing technology has the following problems: At present, the AD risk prediction method studied using longitudinal multi-task learning only considers the temporal correlation of longitudinal multi-tasks, and does not consider the temporal accumulation effect unique to chronic diseases. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a method, device and system for predicting the risk of Alzheimer's disease. By designing a time accumulation operator, the time accumulation effect is integrated into the model calculation, thereby obtaining a more accurate early diagnosis result.

[0005] To achieve the above object, in a first aspect of the present invention, the present invention provides a method for predicting the risk of Alzheimer's disease, the method comprising:

[0006] Acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and no longer collected during subsequent follow-up processes;

[0007] Inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the object to be detected with the time accumulation operator;

[0008] Inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the subject to be detected with the time accumulation operator;

[0009] The first risk prediction result and the second risk prediction result are input into a fully connected layer, and a third risk prediction result of the object to be detected within the current period and a future period is output.

[0010] In a second aspect of the present invention, the present invention further provides a risk prediction device for Alzheimer's disease, wherein:

[0011] A data acquisition module is used to acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and no longer collected during subsequent follow-up processes;

[0012] A first classification module, used for inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the object to be detected with the time accumulation operator;

[0013] A second classification module, used for inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the object to be detected with the time accumulation operator;

[0014] The third classification module is used to input the first risk prediction result and the second risk prediction result into the fully connected layer, and output the third risk prediction result of the object to be detected in the current and future period of time.

[0015] In the third aspect of the present invention, the present invention also provides a risk prediction system for Alzheimer's disease, comprising a processor, a communication interface and a memory, wherein the processor, the communication interface and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: the present invention designs a time accumulation operator by analyzing the conceptual characteristics of the time accumulation effect, and the operator can integrate the medical knowledge of the time accumulation effect into model calculation and model prediction; the present invention has high prediction accuracy and high practicality, and through early diagnosis of Alzheimer's patients, it helps to make relevant preparations and preventive measures in time to slow down the deterioration of the disease and improve the survival probability of the subjects to be detected, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a scenario of a method for predicting the risk of Alzheimer's disease provided in an embodiment of the present application;

[0018] Figure 2 A flow chart of a method for predicting the risk of Alzheimer's disease provided by an embodiment of the present invention;

[0019] Figure 3 Schematic diagram of input and output of the second classification model in an embodiment of the present invention;

[0020] Figure 4 Schematic diagram of input and output of the risk prediction model in an embodiment of the present invention;

[0021] Figure 5 A structural diagram of a device for predicting the risk of Alzheimer's disease provided by an embodiment of the present invention;

[0022] Figure 6 A structural diagram of the Alzheimer's disease risk prediction system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The embodiment of the present invention obtains various examination data of the object to be predicted, divides the data according to the data source and data form, and inputs the data into the constructed first classification model and the second classification model; then splices them through the fully connected layer to obtain the disease diagnosis results of the patient to be predicted in the future.

[0025] See also Figure 1 , Figure 1 A scenario diagram of a risk prediction method for Alzheimer's disease provided in an embodiment of the present application. The server can receive various test data of the object to be tested; input the baseline data and the time accumulation operator into the first classification model, and output the first risk prediction result of the object to be tested with the time accumulation operator; input the follow-up data and the time accumulation operator into the second classification model, and output the second risk prediction result of the object to be tested with the time accumulation operator; input the first risk prediction result and the second risk prediction result into the fully connected layer, and output the third risk prediction result of the object to be tested in the current and future period. Figure 2 FIG. 4 is a flow chart of a method for predicting the risk of Alzheimer's disease in an embodiment of the present invention; Figure 1 As shown, the method includes:

[0026] 101. Acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and not collected again during subsequent follow-up processes;

[0027] In an embodiment of the present invention, various test data of multiple objects to be detected can be obtained. In order to ensure the reliability of various test data, the acquisition channels of various test data may include but are not limited to public websites, medical institution databases, etc. These various test data can be obtained through three categories of data: demographic baseline data, imaging data, and laboratory test results. The three categories of data, namely, demographic baseline data, imaging data, and laboratory test results, are obtained based on different fields and document contents (such as general information, chief complaint, surgical records, daily course of disease) containing a variety of different target (such as diagnosis, surgery, symptoms, etc.) models.

[0028] In some possible embodiments, a data set that meets the actual production environment can be selected to complete AD risk prediction. The embodiment of the present invention selects the ADNI database. The Alzheimer's Disease Neuroimaging Program (ADNI) is a longitudinal multicenter study that aims to develop clinical, imaging, genetic and biochemical biomarkers for early detection and tracking of Alzheimer's disease (AD). In some possible embodiments, the data used in the present invention include MRI, scale examination data, plasma data, physical examination data, vital signs data, and baseline symptom data. Among them, MRI data, scale examination data, plasma data and part of the physical examination data are follow-up data, and the rest are baseline data. The MRI data uses the ADNI1:Complete 1Yr3T data set in the database. The data set is sampled once every six months, and each patient has three sampling samples. The patient's initial examination time is specified as Time1, the review time after half a year is Time2, and the review time after one year is Time3. The remaining data are screened according to the patients in the MRI data set, and a total of 114 patients are selected.

[0029] In a preferred embodiment of the present invention, based on the acquired detection data, the detection data can also be preprocessed; it can be extracted by manual annotation, natural language processing (NLP), etc. to obtain structured data (or formatted data). Data cleaning can also be performed to obtain cleaned data. Among them, data cleaning includes any combination of secondary characterization, screening, denoising, interpolation / filling, outlier analysis, clustering, and dimensionality reduction.

[0030] In some embodiments, the present invention uses FreeSurfer image analysis software to perform MRI cortical reconstruction and volume segmentation, and obtains 40 ROI areas and 62 brain areas, wherein the embodiment of the present invention selects ROI total volume, slice mean, and slice standard deviation as the features of the ROI area, and selects the number of vertices, surface area, gray matter volume, average thickness, thickness standard deviation, mean curvature, Gaussian curvature, folding index, and intrinsic curvature as brain area structural features, and finally obtains a total of 678 MRI features. After integrating all numerical features, missing value processing and normalization are performed. There are three cases of data missing, including missing samples in follow-up data, missing continuous values, and missing discrete values. Among the 114 patients, 57 had missing samples, of which 20 were missing in the second sampling and 37 were missing in the third sampling. For missing samples, the embodiment of the present invention uses the grouped average value to fill in. The missing of continuous eigenvalues ​​is also filled with the grouped average value. For the missing of discrete eigenvalues, the majority value is used to fill in. All continuous feature values ​​were normalized using the mean-standard deviation method, including 703 follow-up data features and 72 baseline data features.

[0031] In an embodiment of the present invention, in order to better obtain the time correlation of multiple tasks and the time accumulation effect unique to chronic diseases, the embodiment of the present invention constructs a time accumulation operator based on the follow-up time of the follow-up data; the present invention believes that the manifestation of the time accumulation effect in the classification prediction model should be that the parameters of each follow-up time period consider the influence of the parameters of the historical follow-up time period; and the influence corresponding to the risk prediction results of each historical follow-up time period will decrease with the time distance from the current follow-up time, and the sum of the influences of all follow-up times is 1.

[0032] The embodiment of the present invention designs a time accumulation operator: according to the influence of the test result of each follow-up time on the test result of another follow-up time, the weight of each follow-up time relative to another follow-up time is calculated; according to the weight of any follow-up time relative to another follow-up time, the time accumulation operator under each follow-up time is constructed. Specifically, it can be expressed as:

[0033]

[0034] stω k. =1,ω∈[0,1]

[0035] Where T represents the time accumulation operator, ω 1,2 It represents the weight of the first follow-up time relative to the second follow-up time. This weight reflects the influence of the first follow-up time on the risk prediction results of the second follow-up time.

[0036] In some embodiments of the present invention, the follow-up period of the data used in the present invention is 36 months, with sampling once every 12 months, for a total of 3 times. The design of the time accumulation operator is mainly considered from three perspectives: current time weight, time window size and distance attenuation:

[0037] Design=(a,d,f)

[0038] Among them, a represents the size of the current time weight, that is, the impact of the time accumulation effect on the prediction result; d represents the size of the time window. The larger the time window, the greater the impact of the historical time considered by the time accumulation effect on the current time; f represents whether the weight decays, that is, whether the weight in the time window decays as the distance from the current time increases. The time accumulation operator used in the embodiment of the present invention is Design = (0.9, 3, false), and its formula is:

[0039]

[0040] Based on the time accumulation operator of the above embodiment, the present invention can fully consider the cumulative effect of follow-up data at different time points and improve the accuracy of longitudinal prediction. By weighted integration of the test results of historical time periods, the time accumulation operator can capture the dynamic change trend of the patient's disease course and provide more comprehensive information support for disease risk prediction in the current and future period. In addition, the operator ensures that the model can still effectively utilize early follow-up data in a longer follow-up period by flexibly setting the time window size and weight attenuation function, thereby achieving more accurate risk prediction.

[0041] 102. Input the baseline data and the time accumulation operator into a first classification model, and output a first risk prediction result of the object to be detected with the time accumulation operator;

[0042] In an embodiment of the present invention, the first classification model is a decision tree baseline data auxiliary model; the baseline data is input into the first classification model to obtain an initial risk detection result; the initial risk detection result is copied to obtain a zeroth risk detection result with the same dimension as the time accumulation operator; the zeroth risk detection result is combined with the time accumulation operator to obtain a first risk detection result with a time accumulation operator; the baseline data includes the number of samples and feature dimensions; the first risk detection result includes the number of samples, follow-up time and number of types.

[0043] Specifically, the present invention takes into account the output of baseline data Since the baseline data has no time dimension, the input and output forms of the baseline data are a two-dimensional tensor, that is, the number of samples and the feature dimension. The input data X and the model parameter matrix W are input into the first classification model to obtain the corresponding output data Y b ∈R n×c , where the input data X is the baseline data, and the model parameter matrix W is the parameter matrix of the first classification model; in order to fuse the output of the first classification model with the output of the second classification model, it is necessary to combine the output with the time accumulation operator to obtain the same as Y f The same form of baseline data assists the model output Y b ∈R n×t×c The decision tree model used in the present invention is implemented based on sklearn. The selected decision tree model uses mean square error as the feature selection criterion, randomly selects feature partitioning points, and has a maximum tree depth of 3.

[0044] like Figure 3 As shown, Figure 3This is a schematic diagram of the input and output structure of the first classification model of an embodiment of the present invention. It can be found that the initial risk detection result (c1, c2, c3) is copied, and the number of copies is the number of follow-up times n. In this way, a tensor matching the time dimension can be generated, in which the value of each feature dimension at different follow-up times t1, t2, t3 remains unchanged; that is, the zeroth risk detection result with the same dimension as the time accumulation operator can be obtained; the zeroth risk detection result is combined with the time accumulation operator T to obtain the first risk detection result with the time accumulation operator; the time accumulation operator captures the cumulative impact of different follow-up times on the time dimension. After matrix operation, the first risk detection result obtained includes the cumulative effect at each follow-up time, so that the output risk prediction result can comprehensively consider the cumulative effect of baseline data and follow-up data, and provide more accurate risk prediction results and risk prediction basis. 103. The follow-up data and the time accumulation operator are input into the second classification model, and the second risk prediction result of the object to be detected with the time accumulation operator is output;

[0045] In an embodiment of the present invention, the follow-up data are respectively cross-multiplied with each classification parameter matrix in the model parameter matrix to obtain a multi-classification matrix of each follow-up data; the multi-classification matrix of each follow-up data is dot-multiplied with a time accumulation operator and summed horizontally to obtain a multi-classification time accumulation matrix of each follow-up data; the multi-classification time accumulation matrix of each follow-up data is spliced ​​to obtain a classification prediction result of each follow-up data; the classification prediction result of each follow-up data is integrated to obtain a second risk prediction result of the object to be detected with a time accumulation operator; the follow-up data includes the number of samples, feature dimensions and follow-up time; the second risk prediction result includes the number of samples, follow-up time and number of types.

[0046] Specifically, the second classification model of the embodiment of the present invention is a longitudinal multi-task follow-up data model; the longitudinal multi-task learning used regards the prediction task of each time point of the sample as a risk task, requiring the input data to be in the form of a three-dimensional tensor, that is, in addition to the number of samples and feature dimensions, a one-dimensional follow-up number is added, that is, Similarly, the model parameter matrix It is also a three-dimensional tensor, namely the number of classifications, feature dimensions and number of follow-up visits. Input data X, model parameter matrix W, and time accumulator T are input into the second classification model to obtain the corresponding output data Y f , where the input data X is the follow-up data including the number of samples, feature dimensions and follow-up time, and the model parameter matrix W is the parameter matrix of the second classification model.

[0047] The specific calculation process includes: input data and each classification parameter matrix Perform matrix cross multiplication to obtain the matrix of c classifications of n samples matrix Perform matrix dot multiplication with the time accumulation operator T and sum horizontally to obtain the time accumulation output of c categories of n samples Output the accumulated time of c classifications for each sample Splicing to get the classification prediction matrix for each sample Finally, the output of the entire longitudinal multi-task follow-up data model is

[0048] It is understandable that the classification model is learned through the training set to establish a mapping from the input space X to the output space Y (discrete value). The principle is to learn a classification function or classification model based on the existing training data set, and the function or model can map new input data to predefined categories. The first classification model and the second classification model of the embodiment of the present invention can be existing classification models, such as logistic regression, decision tree, support vector machine (SVM), random forest, neural network and other models; the present invention mainly adjusts the multi-input and multi-output tensors of the second classification model, integrates the time accumulation operator, and learns the cumulative effect of various detection data in time, so as to obtain more accurate early diagnosis results.

[0049] 104. Input the first risk prediction result and the second risk prediction result into a fully connected layer, and output a third risk prediction result of the object to be detected in the current period and a future period.

[0050] In an embodiment of the present invention, the present invention uses a fully connected layer to perform decision fusion on the outputs of the longitudinal multi-task follow-up data model and the decision tree baseline data auxiliary model.

[0051] In an embodiment of the present invention, the risk prediction results are gradually fully connected according to the follow-up time points; the first risk prediction result and the second risk prediction result of the current follow-up time are weighted and integrated with the first risk prediction result and the second risk prediction result of the previous several follow-up times through a preset weight matrix. The preset weight matrix can be obtained through a trained perceptron; the perceptron can be obtained through two two-dimensional matrices, namely the first risk prediction result and the second risk prediction result, as training data and training.

[0052] The present invention uses a fully connected layer to combine the outputs of the longitudinal multi-task follow-up data model and the decision tree baseline data auxiliary model as the fully connected layer input X fc =(Y f ,Y b) process includes: considering that the weights of different models may be different at different times, the fully connected layer does not directly perform a fully connected weighted sum of the outputs of the two models, but performs a full connection at different times, that is, the fully connected layer parameters are The training of the fully connected layer uses the mean square error as the loss function and is optimized using the stochastic gradient descent algorithm. Finally, the output Y of the AD risk prediction model with time cumulative effect is obtained, which is the third risk prediction result.

[0053] In the preferred embodiment of the present invention, considering that the medical structure can generate AD test data of different types and standards, most AD risk predictions at this stage only use one of the data modes of follow-up data or baseline data for classification prediction; and each test data has different sources and different structures. In order to process these heterogeneous test data, Figure 4 Schematic diagram of input and output of the risk detection model of the embodiment of the present invention. Figure 4 As shown, each training module can correspond to a risk prediction model. The risk prediction model includes a first classification model, a second classification model and a fully connected layer; Figure 4 The overall structure of the risk detection model of the present invention is shown in . The model first inputs different types of test data (such as MRI data, scale data, and physiological data) into multiple independent training modules (Trainer 1, Trainer 2, Trainer n). Each training module performs independent training according to the input test data type and outputs the corresponding prediction results. In this process, the time accumulation operator T is also introduced into each training module to enhance the model's ability to process time dimension information. Then, these prediction results are integrated through a decision fusion strategy (Decision ConvergenceStrategy). The fusion strategy will weight them according to the weights w1, w2,...Wn of different prediction results, and finally output the comprehensive prediction result Y finaly , this process enables the model to obtain information from different data sources and more accurately predict the risk of Alzheimer's disease through time accumulation effects.

[0054] It can be understood that the preferred embodiment of the present invention, based on the combination of the time accumulation operator, considers the complexity of data in the actual production environment, and trains multi-source heterogeneous data according to data source and data form to provide decision support for doctors. This embodiment solves the problem of adaptability of a single model to various test data by dividing the actual inspection data of the object to be inspected according to data source and data form and learning them separately; by improving the longitudinal multi-task learning model, the prediction result has the effect of early diagnosis.

[0055] In some embodiments of the present invention, the first classification model, the second classification model and the fully connected layer constitute a risk prediction model; the present invention can also optimize and update the model parameters of the first classification model, the second classification model and the fully connected layer according to the loss function, so that the model can predict more accurate risk results. In an embodiment of the present invention, the first classification model is optimized by sklearn; the second classification model is optimized by mean square error loss and regularization; the fully connected layer is optimized by mean square error loss.

[0056] The present invention constructs the optimized loss function as:

[0057]

[0058] Among them, L represents the loss function of the risk prediction model, represents the mean square error loss of the risk prediction model, λ1, λ2 and λ3 are regularization parameters, γ is the risk prediction model, X is the input data matrix, W is the model parameter matrix, T is the time accumulation operator; Y is the output data matrix, ||W||1 is the Lasso coefficient regularization term, is the group sparse regularization term, w kij represents the weight of the i-th feature dimension in the k-th risk task at the j-th follow-up time, c represents the number of risk tasks, that is, the number of risk types, m represents the feature dimension, and t represents the number of follow-up times, allowing joint feature selection and task-specific feature selection for multiple tasks during model optimization. T ||1 is the temporal smoothing regularization term, taking into account the temporal correlation between tasks, and R is the temporal smoothing matrix.

[0059] The specific values ​​are as follows:

[0060] If the follow-up time i to be selected is consistent with the current follow-up time j, the time smoothing matrix Rij takes a value of 1, indicating that the influence of the follow-up time i to be selected on the current follow-up time j is autocorrelated;

[0061] If the follow-up time i to be selected is the time point before the current follow-up time j, the time smoothing matrix Rij takes a value of -1, indicating that the influence of the follow-up time i to be selected on the current follow-up time j is mutually correlated.

[0062] It can be expressed as:

[0063]

[0064] Among them, i represents the follow-up time to be selected. This is the row index of the matrix, representing the characteristics or output of the model at a certain point in time. j represents the current follow-up time j or the historical follow-up time. This is the column index of the matrix, representing the influence of other time points associated with a certain time point. By adjusting the difference between the current time point and the previous time point, smoothing in the time dimension is achieved. It ensures that the model has a smooth transition when changing over time, avoiding drastic fluctuations between time points, thereby improving the stability of the model and the accuracy of prediction.

[0065] It can be understood that the present invention controls the weight of the current time point during self-correlation, while the cross-correlation is used to adjust the influence of the previous time point on the current time point to ensure a smooth transition in the time dimension.

[0066] In some embodiments, the first classification model can be directly optimized by sklearn. The second classification model and the fully connected layer can also include other losses, such as mean absolute error, cross entropy loss, etc.

[0067] Based on the above process, the present invention divides and processes the various test data of the object to be detected, inputs them into the risk prediction model, and obtains the disease prediction output through the first classification model, the second classification model, and the fully connected layer. Where t represents the predictable future time, and c represents the disease progression. There are three types of Alzheimer's disease: normal cognitive level (CN), mild cognitive impairment (MCI) and Alzheimer's disease (AD).

[0068] In some embodiments of the present invention, the present invention also provides a risk prediction device for Alzheimer's disease, such as Figure 5 As shown, the risk prediction device 200 includes:

[0069] The data acquisition module 201 is used to acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and no longer collected during the subsequent follow-up process;

[0070] A first classification module 202, used for inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the object to be detected with the time accumulation operator;

[0071] A second classification module 203, used for inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the subject to be detected with the time accumulation operator;

[0072] The third classification module 204 is used to input the first risk prediction result and the second risk prediction result into the fully connected layer, and output the third risk prediction result of the object to be detected in the current period and a future period.

[0073] Among them, the risk prediction device for Alzheimer's disease can be integrated in a computer device, and the computer device can be a terminal or server device. The terminal can be a terminal device such as a smart phone, a tablet computer, a laptop computer, a personal computer (PC for short). The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN for short), and big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and the present application does not limit this. In some embodiments of the present invention, a risk prediction system 300 for Alzheimer's disease is also provided, such as Figure 6 As shown. The Alzheimer's disease risk prediction system 300 may include at least one processor 301 and at least one memory 302 and a corresponding communication interface 303. Those skilled in the art will understand that Figure 6 The Alzheimer's disease risk prediction system 300 shown in the figure does not constitute a limitation of the system, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0074] The processor 301 is the control center of the risk prediction system 300 for Alzheimer's disease. By running or executing the software programs and / or modules stored in the memory 302 and calling the data stored in the memory 302, the various functions of the risk prediction system 300 for Alzheimer's disease and processing data are executed, thereby monitoring the risk prediction system 300 for Alzheimer's disease as a whole. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 302 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 301.

[0075] The data of each patient in the follow-up data set is equivalent to a matrix of size 3*703, with 3 rows representing samples of 3 periods of time and 703 columns representing 703 preprocessed features; the optimization algorithm selected for the longitudinal multi-task follow-up data model is the Nesterov accelerated gradient descent method, with a learning rate of 0.01, a weight decay of 0.1, and a momentum of 0.1; the decision tree baseline data auxiliary model is implemented based on sklearn, and the selected decision tree model uses mean square error as the feature selection criterion, randomly selects feature partitioning points, and has a maximum tree depth of 3. The number of iterations of the entire training is 5000.

[0076] The following are specific experimental results, comparing the evaluation parameters of the present invention with other traditional machine learning algorithms and their algorithm combinations. As shown in Table 1:

[0077] Table 1 Comparison results of the accuracy of the six time accumulation operators designed by the present invention on the longitudinal multi-task model

[0078]

[0079]

[0080] For the ADNI dataset, the average accuracy of the present invention in each time group reached 76.83%, which performed best among all model combinations, followed by the MTL+KNN model combination. In real medical scenarios, the distribution of different data sets and the characteristics of the attributes are different, which leads to different optimal model combinations. And from the experimental results, it can be analyzed that the algorithm model using the time accumulation operator has increased the average accuracy of each time group by 22.78%, and the KNN algorithm and LR algorithm have achieved a large improvement, especially in the prediction of the Time3 group, the accuracy improvement is most obvious, which proves the effectiveness of the time accumulation operator and the superiority of the present invention in the early diagnosis and prediction of Alzheimer's disease. The present invention is implemented on the ADNI dataset, which proves that the present invention is superior to the traditional machine learning single model and other models that do not consider the time accumulation effect in terms of the accuracy of the prediction results of early diagnosis of AD and the generalization ability of the model.

[0081] It should be noted that, in response to the shortcomings of existing early diagnosis and prediction methods for Alzheimer's disease, the present invention proposes an Alzheimer's disease risk prediction method, device and system, applies the time accumulation effect to the AD risk prediction scheme, and designs a time accumulation operator from three perspectives by analyzing the conceptual characteristics of the time accumulation effect. The time accumulation operator is integrated into the risk prediction model calculation, and the experimental results prove the effectiveness and superiority of the time accumulation operator designed by the present invention.

[0082] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the risk of Alzheimer's disease, characterized in that: include: Acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and no longer collected during subsequent follow-up processes; Inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the object to be detected with the time accumulation operator; Inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the subject to be detected with the time accumulation operator; The first risk prediction result and the second risk prediction result are input into a fully connected layer, and a third risk prediction result of the object to be detected within the current period and a future period is output.

2. The method for predicting the risk of Alzheimer's disease according to claim 1, characterized in that: The calculation process of the time accumulation operator includes calculating the weight of each follow-up time relative to another follow-up time according to the influence of the detection result of each follow-up time on the detection result of another follow-up time; and constructing the time accumulation operator at each follow-up time according to the weight of any follow-up time relative to another follow-up time.

3. The method for predicting the risk of Alzheimer's disease according to claim 1 or 2, characterized in that: Inputting the baseline data and the time accumulation operator into the first classification model, and outputting the first risk prediction result of the object to be detected with the time accumulation operator includes inputting the baseline data into the first classification model to obtain an initial risk detection result; copying the initial risk detection result to obtain a zeroth risk detection result with the same dimension as the time accumulation operator; combining the zeroth risk detection result with the time accumulation operator to obtain a first risk detection result with the time accumulation operator; the baseline data includes the number of samples and feature dimensions; The first risk detection result includes sample number, follow-up time and type number.

4. The method for predicting the risk of Alzheimer's disease according to claim 1 or 2, characterized in that: Inputting the follow-up data and the time accumulation operator into the second classification model, and outputting the second risk prediction result of the subject to be detected with the time accumulation operator includes: Perform matrix cross multiplication on the follow-up data with each classification parameter matrix in the model parameter matrix to obtain a multi-classification matrix for each follow-up data; perform dot multiplication on the multi-classification matrix of each follow-up data with the time accumulation operator and sum them horizontally to obtain a multi-classification time accumulation matrix for each follow-up data; concatenate the multi-classification time accumulation matrices of each follow-up data to obtain a classification prediction result for each follow-up data; integrate the classification prediction results of each follow-up data to obtain a second risk prediction result of the object to be detected with a time accumulation operator; the follow-up data includes the number of samples, feature dimensions and follow-up time; the second risk prediction result includes the number of samples, follow-up time and number of types.

5. The method for predicting the risk of Alzheimer's disease according to claim 4, characterized in that: The first risk prediction result and the second risk prediction result are input into the fully connected layer, and the third risk prediction result of the object to be detected in the current and future period of time is output, including gradually fully connecting the risk prediction results according to the follow-up time; the first risk prediction result and the second risk prediction result of the current follow-up time are weighted and integrated with the first risk prediction results and the second risk prediction results of the previous several follow-up times through a preset weight matrix.

6. The method for predicting the risk of Alzheimer's disease according to claim 1, characterized in that: The first classification model, the second classification model and the fully connected layer constitute a risk prediction model; the risk prediction model is based on mean square error loss optimization and regularization optimization.

7. The method for predicting the risk of Alzheimer's disease according to claim 6, characterized in that: The risk prediction model is expressed as follows based on mean square error loss optimization and regularization optimization: Among them, L represents the loss function of the risk prediction model, represents the mean square error loss of the risk prediction model, λ1, λ2 and λ3 are regularization parameters, γ is the risk prediction model, X is the input data matrix, W is the model parameter matrix, T is the time accumulation operator; Y is the output data matrix, W1 is the Lasso coefficient regularization term, is the group sparse regularization term, w kij represents the weight of the i-th feature dimension in the k-th risk task at the j-th follow-up time, c represents the number of risk tasks, that is, the number of risk types, m represents the feature dimension, t represents the number of follow-up times, and RW T 1 is the time smoothing regularization term, and R is the time smoothing matrix.

8. The method for predicting the risk of Alzheimer's disease according to claim 7, characterized in that: The specific values ​​of the time smoothing matrix are as follows: if the follow-up time to be selected is consistent with the current follow-up time, the value of the time smoothing matrix is ​​1, indicating that the influence of the follow-up time to be selected on the current follow-up time is autocorrelated; If the follow-up time to be selected is a time point before the current follow-up time, the time smoothing matrix takes a value of -1, indicating that the influence of the follow-up time to be selected on the current follow-up time is mutually correlated.

9. A device for predicting the risk of Alzheimer's disease, characterized in that: The device: A data acquisition module is used to acquire various test data of the subject to be detected; the various test data include follow-up data and baseline data; the follow-up data include data collected from the subject to be detected at multiple follow-up times, and the baseline data include data collected once when the subject to be detected enters the queue and no longer collected during subsequent follow-up processes; A first classification module, used for inputting the baseline data and the time accumulation operator into a first classification model, and outputting a first risk prediction result of the object to be detected with the time accumulation operator; A second classification module, used for inputting the follow-up data and the time accumulation operator into a second classification model, and outputting a second risk prediction result of the object to be detected with the time accumulation operator; The third classification module is used to input the first risk prediction result and the second risk prediction result into the fully connected layer, and output the third risk prediction result of the object to be detected in the current and future period of time.

10. A risk prediction system for Alzheimer's disease, characterized in that: It comprises a processor, a communication interface and a memory, wherein the processor, the communication interface and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method as claimed in any one of claims 1 to 8.

Images