CN116894812A - Individualized disease prediction methods, devices, media and equipment based on sequence learning - Google Patents

Abstract

The application provides a method, a device and a medium for predicting an individualized disease based on sequence learning, and electronic equipment, wherein the method comprises the following steps: acquiring brain CT detection graphs of patients with Alzheimer's disease at different disease stages, and constructing a data set; marking the data set to identify ventricle areas and brain white matter areas around the brain; performing feature labeling on the ventricle area image, and counting ventricle expansion features to form first feature data; performing feature labeling on the brain white matter region image, and counting white matter atrophy features to form second feature data; calculating the percentage of the area of the first ventricle expansion characteristic and/or the white matter atrophy characteristic to the whole brain area, and constructing third characteristic data; the first characteristic data, the second characteristic data and the third characteristic data are input into a preset fusion depth convolution and LSTM model of the RNN circulating neural network for training so as to construct a prediction model of Alzheimer's disease, and the prediction accuracy is improved.

Description

Individualized disease prediction methods, devices, media and equipment based on sequence learning

Technical field

The invention belongs to the field of artificial intelligence medical application technology, and in particular relates to an individualized disease prediction method, device, medium and electronic equipment based on sequence learning.

Background technique

Alzheimer's disease is a serious Alzheimer's disease. In recent years, the high incidence rate has caused a rapid increase in the number of patients. The quality of life of patients decreased significantly. Despite high medical and social costs, treatment options are limited. Effective early diagnosis is lacking. Therefore, early screening and diagnosis can detect most patients before they have obvious symptoms and seize the best and effective treatment time. Early detection and early treatment of patients are particularly important.

This invention considers integrating multiple technologies for early screening and prediction of Alzheimer's disease.

Contents of the invention

In view of the shortcomings of the existing technology, the present invention proposes an individualized disease prediction method, device, medium, and electronic equipment based on sequence learning, which can predict diseases early and improve the accuracy of disease prediction.

In order to achieve the above objectives, on one hand, the present invention provides an individualized disease prediction method based on sequence learning for Alzheimer's disease prediction, including:

Obtain brain CT examination pictures of Alzheimer's disease patients at different disease stages and construct a data set;

Label the data set to identify ventricular regions and white matter regions surrounding the brain;

Mark the features of the ventricular region image, calculate ventricular enlargement features, and form first feature data;

Mark the image of the brain white matter region with features, calculate the white matter atrophy features, and form second feature data;

Calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area, and construct third characteristic data;

The first feature data, the second feature data, and the third feature data are input into a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the prediction model of Alzheimer's disease. .

Optionally, before labeling the data set, it also includes:

Preprocess the data set images,

The preprocessed data set is segmented into tissue types.

Optionally, mark the ventricular region images with features to calculate ventricular enlargement features,

A standard brain model is used to standardize the images of the ventricular region of different individuals, measure the ventricular volume, gyrus volume or lesion volume, and identify the ventricular enlargement characteristics.

Optionally, labeling the data set also includes:

Registration of brain CT scans obtained at different time points for the same patient; and/or

Register and fuse brain CT detection images with images obtained by other imaging technologies.

Optionally, use new biomarker methods to mark the data set, including:

Through bioinformatics analysis of the genome, transcriptome and proteome data of Alzheimer's disease patients and normal people, we can find key genes, messenger RNA and proteins related to Alzheimer's disease.

Optionally, the LSTM model includes: input layer, hidden layer and output layer;

The first feature data, the second feature data, and the third feature data serve as the input of the input layer, the output of the input layer serves as the input of the hidden layer, and the output of the hidden layer serves as the output layer As input, the output layer outputs Alzheimer's disease prediction results;

The hidden layer includes:

a memory unit,

An input gate module is connected to the input end of the memory unit and controls the flow of the first characteristic data, the second characteristic data, and the third characteristic data into the memory unit;

A forgetting module, connected to the memory unit, controls whether the information in the memory unit at the previous moment is accumulated into the memory unit at the current moment,

The output information of the input gate module is combined with the output information of the forgetting module to generate the memory unit;

An output gate module is connected to the output end of the memory unit and controls the information in the memory unit at the current moment to flow into the next hidden layer or the output layer.

Optionally, the loss function during the training process uses the mean square error loss function, and the gradient descent algorithm uses the Adam gradient descent algorithm.

On the other hand, the present invention also provides an individualized disease prediction device based on sequence learning, which adopts the above individualized disease prediction method based on sequence learning and at least includes:

The data set construction module is used to obtain brain CT examination pictures of Alzheimer's disease patients at different disease stages and construct a data set;

A feature labeling module for labeling the data set and identifying the ventricular region and the white matter region surrounding the brain; and

Mark the features of the ventricular region image, calculate ventricular enlargement features, and form first feature data;

Mark the image of the brain white matter region with features, calculate the white matter atrophy features, and form second feature data;

Calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area, and construct third characteristic data;

A model building module for inputting the first feature data, the second feature data, and the third feature data into a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the Alzheimer's disease Predictive models for Alzheimer's disease.

Another aspect of the present invention also provides a storage medium for storing a computer program for executing the above-mentioned personalized disease prediction method based on sequence learning.

Another aspect of the present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above is achieved. An individualized disease prediction method based on sequence learning.

It can be seen from the above solutions that the advantages of the present invention are:

The invention provides an individualized disease prediction method based on sequence learning, which constructs a data set by acquiring brain CT detection maps of Alzheimer's disease patients at different disease stages; and then marks the data set to identify the ventricular region, and the brain white matter area around the brain; then feature marking is performed on the image of the cerebral ventricle area, and the characteristics of ventricular enlargement are statistically calculated to form the first characteristic data; the image of the cerebral white matter area is characterized and marked, and the characteristics of white matter atrophy are statistically calculated to form the third characteristic data. two characteristic data; and simultaneously calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area to construct a third characteristic data; combine the first characteristic data and the second characteristic data The data, and the third feature data are input to a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the prediction model of Alzheimer's disease. This method uses an LSTM model that combines deep convolution and RNN recurrent neural network for training to build a prediction model for Alzheimer's disease to solve the problem of unclear early detection of Alzheimer's disease, achieve early prediction, and improve prediction Accuracy.

Description of the drawings

Figure 1 is a schematic flow chart of an individualized disease prediction method based on sequence learning provided by an embodiment of the present invention;

Figure 2 is the structure diagram of the LSTM model;

Figure 3 shows the error rate results of the LSTM model;

Figure 4 is the model prediction chart;

Figure 5 is a framework diagram of the sequence learning-based individualized disease prediction device of the present invention;

Figure 6 is a schematic structural diagram of the electronic device;

in:

11-Input layer

12-hidden layer

121-Memory unit

122-Input gate module

123-Forgetting module

124-Output gate module

13-Output layer

300-Personalized disease prediction device based on sequence learning

301-Dataset Building Blocks

302-Feature Marking Module

303-Model Building Blocks

400-Electronic equipment

401-processor

402-memory.

Detailed ways

In order to make the above-mentioned features and effects of the present invention more clear and understandable, examples are given below and are described in detail with reference to the accompanying drawings.

RNN (Recurrent Neural Network) is the predecessor of LSTM (Long Short-Term Memory Network), but compared with LSTM, it has shortcomings such as inability to learn long-term dependencies well, vanishing gradient problem, unstable training, weak performance, narrow application range, and difficulty in interpretation. Two major limitations of RNNs are vanishing gradients and difficulty in learning long-term dependencies. This makes it less expressive, difficult to process longer sequence data, and its application is also more limited. LSTM has overcome these two problems to a large extent, has stronger performance and wider application range, and has gradually replaced RNN as the mainstream model for sequence learning. LSTM is gradually surpassing RNN and becoming the mainstream tool in the field of sequence learning. The present invention is based on the LSTM model that combines deep convolution and RNN recurrent neural network, and provides an individualized disease prediction method based on sequence learning, which enables early prediction of Alzheimer's disease and aims at solving the problem of early detection of Alzheimer's disease. The obvious problem improves the accuracy of prediction, helps patients with early Alzheimer's disease to carry out appropriate medical intervention, delays the onset of disease, and reduces the investment cost of Alzheimer's disease.

Figure 1 shows a schematic flowchart of an individualized disease prediction method based on sequence learning provided by an embodiment of the present invention.

An individualized disease prediction method based on sequence learning for Alzheimer's disease prediction, including:

S1. Obtain brain CT examination pictures of Alzheimer's disease patients at different disease stages and construct a data set.

In practice, Alzheimer's disease can be divided into four stages based on the severity of symptoms: normal, mild, moderate and severe. Each disease stage has the following main characteristics: 1. Normal stage: There is no obvious decline in memory or cognitive ability, and self-care ability in daily life remains normal. 2. Mild Alzheimer's disease: A certain degree of memory decline begins to occur, especially recent memory, but the impact on life is not significant. The self-care ability in daily life is basically maintained, but the ability to handle complex tasks is slightly reduced. 3. Moderate Alzheimer's disease: The decline in memory, language expression ability and judgment is intensified, and certain help is needed to complete daily activities. The patient may start asking the same questions repeatedly or have trouble understanding complex concepts. Partial self-care ability is reduced. 4. Severe Alzheimer's disease: Cognitive function is severely damaged, and the ability to take care of oneself is almost lost. Unable to handle the most basic activities of daily living and requiring round-the-clock nursing care. Most people will lose the ability to express language and have difficulty recognizing relatives and friends. The main characteristics of the above different stages are as follows: (1) Memory decline: starting from mild to severe to almost complete loss. (2) Ability to take care of oneself: from mild to basically maintained, to severe to almost complete loss. (3) Language and expression: From mild to slight decline, to severe loss of most language abilities. (4) Judgment and thinking: From mild to severe, most of the judgment and understanding abilities are lost. (5) Behavior and emotions: changes from moderate to severe to complete dependence on others for care. There are large differences in the degree of cognitive function damage and decline in self-care ability between each stage, which provides reference for clinical diagnosis and treatment.

Therefore, in a specific implementation, in order to ensure the integrity of the data, this embodiment can collect brain CT detection images of patients at different stages of normal, mild, moderate and severe to construct a training data set.

S2. Mark the data set and identify the ventricular area and the white matter area around the brain.

One of the important characteristics of Alzheimer's disease is ventricular enlargement and white matter atrophy around the brain. Therefore, in specific implementation, the characteristics of the ventricular region and the white matter region of the brain can be specifically identified as model training features for disease prediction.

In addition, in this embodiment, a new biomarker method can be used to mark the data set, and through bioinformatics analysis of the genome, transcriptome and proteome data of Alzheimer's disease patients and normal people, it is found that Key genes, messenger RNAs, and proteins associated with Alzheimer's disease were identified to identify the ventricular regions and white matter regions surrounding the brain. The new biomarker method is used to cross-analyze large-scale multi-omics data to find potential biomarker molecules. It is especially suitable for integrating genome, transcriptome, proteome and metabolome data to discover organisms with more clinical significance. landmark. Combining the patient's multi-omics data and clinical information, bioinformatics methods can be used to establish a prediction model for ventricular-related diseases, which can be used to evaluate the probability of disease occurrence or progression, provide personalized prognosis, and facilitate accurate diagnosis and treatment of the disease. .

In addition, before labeling the data set, it is necessary to preprocess the data set images, image segmentation, pixel classification, image registration and fusion, etc. Preprocessing includes operations such as smoothing, filtering, and supplementing missing areas of CT images to eliminate noise, enhance image quality and continuity, and prepare data for subsequent analysis and processing; image segmentation requires organizing the preprocessed data set. Segmentation, such as gray matter, white matter and cerebrospinal fluid segmentation, to facilitate feature extraction, image segmentation provides an important basis for quantitative analysis and computer-aided diagnosis. Pixel classification is the process of assigning each pixel to a specific category based on image features. It uses methods such as machine learning to achieve the classification and identification of gray matter, white matter and pathological tissue. Image registration and fusion includes the process of registering and fusing brain CT examination maps obtained at different time points of the same patient; and/or registering and fusion of brain CT examination maps with images obtained by other imaging technologies. Image registration enhances the image quality The information between them is conducive to the detection and tracking of lesions.

S3. Mark the features of the ventricular region image, calculate the features of ventricular enlargement, and form the first feature data;

Features are marked on the brain white matter region images, and white matter atrophy features are statistically calculated to form second feature data.

In a specific implementation, the ventricular region images and brain white matter region images are identified through the above steps, and then the ventricular region images are characterized, and the characteristics of ventricular enlargement are statistically calculated. Specifically, a standard brain model can be used to classify the ventricles of different individuals. Standardize the regional images to measure the ventricular volume, gyral volume or lesion volume to identify the ventricular enlargement characteristics and construct a first feature data set; and perform feature marking on the brain white matter regional images to calculate the white matter atrophy characteristics, Specifically, a standard brain model can be used to standardize the brain white matter region images of different individuals, and the white matter volume can be measured to identify the white matter atrophy characteristics and construct a second feature data set.

S4. Calculate the percentage of the first ventricular enlargement feature and/or the white matter atrophy feature in the entire brain region, and construct third feature data.

In this embodiment, after constructing the first characteristic data about the characteristics of ventricular enlargement and the second characteristic data about the characteristics of white matter atrophy, in order to further improve the correlation of the characteristic data and improve the model recognition accuracy, the third characteristic data is further determined. The percentage of an area with ventricular enlargement characteristics and/or white matter atrophy characteristics in the entire brain area is used as the third characteristic data. The first feature data, the second feature data, and the third feature data are combined as model training features and input into the training model for disease prediction.

S5. Input the first feature data, the second feature data, and the third feature data to the preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the Alzheimer's disease model. Predictive model.

The long short-term memory network LSTM is a special RNN (recurrent neural network) that can learn long-term dependencies. LSTM learns selective forgetting and memory information through the "gate" structure, and can model the dependencies between distant elements in long sequence data. This is something that traditional RNN cannot achieve, making it excellent in tasks such as time series prediction. It can also solve the vanishing gradient problem. In long sequence data, the gradient of traditional RNN is easy to "disappear" during the backpropagation process, making it impossible to learn. LSTM avoids this problem by selecting important information through the gate structure, allowing it to learn longer time series. In addition, the training process of LSTM is more stable, easy to converge, and faster than other RNNs. This makes it easier to implement in practical applications. Better performance. LSTM solves the problems of traditional RNN in learning long sequence dependencies and gradient disappearance by introducing the "gate" mechanism, and has strong expressive power. It is widely used in sequence learning tasks and has achieved strong results. It is currently one of the more mature and commonly used sequence learning models. Its structure is also relatively clear and the process is easier to explain, which enhances its reliability and credibility.

Therefore, in this embodiment, in order to improve the prediction accuracy of the model, an LSTM model that combines deep convolution and RNN recurrent neural network is used for training, and a prediction model of Alzheimer's disease is constructed to solve the problem of unclear early detection of Alzheimer's disease. problem, achieving early predictions.

Specifically, the LSTM model structure is shown in Figure 2, which includes an input layer 11, a hidden layer 12, and an output layer 13, in which the first feature data, the second feature data, and the third feature data are used as inputs to the input layer. , the output of the input layer is used as the input of the hidden layer, the output of the hidden layer is used as the input of the output layer, and the output layer outputs the Alzheimer's disease prediction result. For hidden layer 12, it specifically includes:

a memory unit 121,

An input gate module 122 is connected to the input end of the memory unit 121 and controls the flow of the first characteristic data, the second characteristic data, and the third characteristic data into the memory unit;

A forgetting module 123, connected to the memory unit 121, controls whether the information in the memory unit at the previous moment is accumulated into the memory unit at the current moment,

The output information of the input gate module is combined with the output information of the forgetting module to generate the memory unit;

An output gate module 124 is connected to the output end of the memory unit 121 and controls the information in the memory unit at the current moment to flow into the next hidden layer 12 or the output layer 13 .

Specifically, the LSTM model formula is:

f _t =sigmoid (W _f ·[h _t-1 , x _t ]+b _f )

i _t =sigmoid (W _i ·[h _t-1 , x _t ]+b _i )

o _t =Sigmoid (W _o ·[h _t-1 , x _t ]+b _o )

h _t =o _t ×tanh(c _t )

Among them, f _t , i _t and o _t represent the forgetting gate (i.e. forgetting module), input gate (i.e. input module) and output gate (i.e. output module) respectively; x _t represents the input sequence, i.e. the first feature data, the second Feature data, and the third feature data; h _t represents the hidden state, c _t , represents the memory unit; W _f , _Wi , and W _o represent the recursive connection weights of the threshold, b _f , _bi , and _bo represent the bias vector; tanh represents the hyperbolic tangent activation function.

In addition, the loss function during the training process uses the mean square error loss function, and the gradient descent algorithm uses the Adam gradient descent algorithm.

Figure 3 shows the error rate results of the model. It can be seen that after continuous training, after about 20 cycles, the loss of data and the error rate have dropped to a lower level, and in subsequent training, Data loss and error rates also decrease more slowly. The final error is also kept at a low level, ensuring accuracy.

Figure 4 shows the prediction map obtained after model prediction. It is clear from the prediction graph that the probability of developing Alzheimer's disease in the next few years, especially in the second year after testing, will be very high. Therefore, with appropriate medical intervention before then, the incidence of Alzheimer's disease can be reduced by 30%, delayed by 5 years, and reduced by 50%.

In summary, the invention provides an individualized disease prediction method based on sequence learning, which constructs a data set by obtaining brain CT detection maps of Alzheimer's disease patients at different disease stages; and then marks the data set, Identify the ventricular area and the white matter area around the brain; then mark the image of the cerebral ventricular area with features, and count the characteristics of ventricular enlargement to form the first feature data; mark the image with features of the white matter area of the brain, and count the white matter atrophy characteristics to form second characteristic data; and simultaneously calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area to construct third characteristic data; and combine the first characteristic The data, the second feature data, and the third feature data are input to a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the prediction model of Alzheimer's disease. This method uses an LSTM model that combines deep convolution and RNN recurrent neural network for training to build a prediction model for Alzheimer's disease to solve the problem of unclear early detection of Alzheimer's disease, achieve early prediction, and improve prediction Accuracy.

In addition, the above embodiments of the present invention can be applied to terminal devices with the function of the personalized disease prediction method based on sequence learning. The terminal devices may include personal terminals, host computer terminals, etc., and the embodiments of the present invention are not limited thereto. The terminal can support Windows, Android (Android), IOS, Windows Phone and other operating systems.

Referring to Figure 5, Figure 5 shows an individualized disease prediction device 300 based on sequence learning. The personalized disease prediction method based on sequence learning can be applied to personal terminals and host computer terminal equipment, which can be implemented through As shown in Figure 1, the individualized disease prediction method based on sequence learning, the individualized disease prediction device based on sequence learning provided by the embodiment of the present application can realize each process of the above individualized disease prediction method based on sequence learning.

An individualized disease prediction device 300 based on sequence learning, using the above-mentioned personalized disease prediction method based on sequence learning, at least includes:

The data set construction module 301 is used to obtain brain CT detection maps of Alzheimer's disease patients at different disease stages and construct a data set;

The feature labeling module 302 is used to label the data set and identify the ventricular area and the white matter area around the brain; and

Mark the features of the ventricular region image, calculate ventricular enlargement features, and form first feature data;

Mark the image of the brain white matter region with features, calculate the white matter atrophy features, and form second feature data;

Calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area, and construct third characteristic data;

The model building module 303 is used to input the first feature data, the second feature data, and the third feature data to a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the Al Predictive models for Alzheimer's disease.

In addition, it should be understood that in the personalized disease prediction device 300 based on sequence learning according to the embodiment of the present application, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned function allocation can be divided into different groups according to needs. The functional modules are completed, that is, the personalized disease prediction device 300 based on sequence learning can be divided into functional modules different from the modules illustrated above to complete all or part of the functions described above.

FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

In addition, as shown in Figure 6, the embodiment of the present application also provides an electronic device 400, including a processor 401, a memory 402, and programs or instructions stored on the memory 402 and executable on the processor 401. When the program or instruction is executed by the processor 401, the steps of the above-mentioned personalized disease prediction method based on sequence learning are implemented, and the same technical effect can be achieved.

Wherein, the processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage media, such as computer read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk, etc.

In addition, embodiments of the present application also provide a readable storage medium on which a program or instructions are stored. When the program or instructions are executed by a processor, the steps of the above sequence learning-based personalized disease prediction method are implemented. , and can achieve the same technical effect.

It should be noted that, in this document, the terms "comprising", "comprises" or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article or apparatus that includes that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM, disk , optical disk), including several instructions to cause a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of this application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms can be made without departing from the purpose of this application and the scope protected by the claims, all of which fall within the protection of this application.

Claims (10)

1. An individualized disease prediction method based on sequence learning for Alzheimer's disease prediction, which is characterized by: Obtain brain CT examination pictures of Alzheimer's disease patients at different disease stages and construct a data set; Label the data set to identify ventricular regions and white matter regions surrounding the brain; Mark the features of the ventricular region image, calculate ventricular enlargement features, and form first feature data; Mark the image of the brain white matter region with features, calculate the white matter atrophy features, and form second feature data; Calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area, and construct third characteristic data; The first feature data, the second feature data, and the third feature data are input into a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the prediction model of Alzheimer's disease. . 2. The method according to claim 1, characterized in that before labeling the data set, it further includes: Preprocess the data set images, The preprocessed data set is segmented into tissue types. 3. The method according to claim 1, characterized in that the ventricular region image is marked with features to calculate ventricular enlargement features, A standard brain model is used to standardize the images of the ventricular region of different individuals, measure the ventricular volume, gyrus volume or lesion volume, and identify the ventricular enlargement characteristics. 4. The method of claim 1, wherein labeling the data set further includes: Registration of brain CT scans obtained at different time points for the same patient; and/or Register and fuse brain CT detection images with images obtained by other imaging technologies. 5. The method according to claim 1, characterized in that, The data set was labeled using novel biomarker methods, including: Through bioinformatics analysis of the genome, transcriptome and proteome data of Alzheimer's disease patients and normal people, we can find key genes, messenger RNA and proteins related to Alzheimer's disease. 6. The method according to claim 1, wherein the LSTM model includes: an input layer, a hidden layer and an output layer; The first feature data, the second feature data, and the third feature data serve as the input of the input layer, the output of the input layer serves as the input of the hidden layer, and the output of the hidden layer serves as the output layer As input, the output layer outputs Alzheimer's disease prediction results; The hidden layer includes: a memory unit, An input gate module is connected to the input end of the memory unit and controls the flow of the first characteristic data, the second characteristic data, and the third characteristic data into the memory unit; A forgetting module, connected to the memory unit, controls whether the information in the memory unit at the previous moment is accumulated into the memory unit at the current moment, The output information of the input gate module is combined with the output information of the forgetting module to generate the memory unit; An output gate module is connected to the output end of the memory unit and controls the information in the memory unit at the current moment to flow into the next hidden layer or the output layer. 7. The method according to claim 6, characterized in that, The loss function during the training process uses the mean square error loss function, and the gradient descent algorithm uses the Adam gradient descent algorithm. 8. An individualized disease prediction device based on sequence learning, characterized in that the method of individualized disease prediction based on sequence learning according to any one of claims 1 to 7 is adopted, at least including: The data set construction module is used to obtain brain CT examination pictures of Alzheimer's disease patients at different disease stages and construct a data set; A feature labeling module for labeling the data set and identifying the ventricular region and the white matter region surrounding the brain; and Mark the features of the ventricular region image, calculate ventricular enlargement features, and form first feature data; Mark the image of the brain white matter region with features, calculate the white matter atrophy features, and form second feature data; Calculate the percentage of the first ventricular enlargement characteristic and/or the white matter atrophy characteristic area in the entire brain area, and construct third characteristic data; A model building module for inputting the first feature data, the second feature data, and the third feature data into a preset LSTM model that combines deep convolution and RNN recurrent neural network for training to build the Alzheimer's disease Predictive models for Alzheimer's disease. 9. A storage medium, characterized in that it is used to store a computer program for executing the individualized disease prediction method based on sequence learning according to any one of claims 1 to 7. 10. An electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that claim 1 is implemented when the processor executes the computer program The individualized disease prediction method based on sequence learning according to any one of ~7.