Remote training program push method, device, computer equipment and storage medium

Abstract

The present application relates to a method, apparatus, computer equipment and storage medium for pushing a remote training program. The method includes: obtaining cognitive function assessment information and medical related information sent by a first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical related information includes medical image information and biological data information; determining abnormal stage information based on the cognitive function assessment information and medical related information; determining a training program that matches the abnormal stage information based on a variety of training materials in a database, wherein the training program includes training materials related to multiple cognitive domains; pushing the training program to the first terminal for the first user to perform training operations according to the training program. This method can automatically recommend appropriate assessment and training programs to users, realize remote cognitive assessment and training, and enable users to complete home training without the aid of special medical equipment.

Description

Technical Field

[0001] The present application relates to the field of intelligent interaction technology, and in particular to a method, apparatus, computer equipment, and storage medium for pushing a remote training program. Background Art

[0002] Currently, when users need to undergo specific training based on their physical condition, they must visit a professional institution, undergo a series of tedious examinations, and then receive on-site guidance and training based on the results. However, this method requires users to frequently travel to the on-site clinic, which is time-consuming and labor-intensive, and inconvenient. Summary of the Invention

[0003] Based on this, it is necessary to provide a method, device, computer equipment and storage medium for pushing a remote training plan that can remotely push the training plan to address the above technical problems.

[0004] A method for pushing a remote training program, the method comprising:

[0005] Obtaining cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information;

[0006] Determining abnormal stage information based on the cognitive function assessment information and medical related information;

[0007] Determining a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains;

[0008] The training plan is pushed to the first terminal so that the first user can perform training operations according to the training plan.

[0009] In one embodiment, determining abnormal stage information based on the cognitive function assessment information and medical related information includes:

[0010] Performing image preprocessing on medical image information to obtain a preprocessed image;

[0011] Inputting the preprocessed image into a pre-trained long short-term memory neural network model, and extracting features by the long short-term memory neural network model to obtain a hidden representation of image features; the image features include band change features, edge features, and grayscale features;

[0012] Based on the hidden representation of the biological information and image features, a fully connected neural network is used for classification to obtain prediction stage information;

[0013] Abnormal stage information of the first user is determined according to the cognitive function evaluation information and the predicted stage information.

[0014] In one embodiment, the training step of the long short-term memory neural network model includes:

[0015] Obtaining a current sample image, and using image features in the current sample image as input values ​​at a current moment;

[0016] According to the hidden representation of the image feature at the previous moment, a forget gate performs a weighted operation on the input value to obtain a first intermediate value, an input gate performs a weighted operation on the input value to obtain a second intermediate value and a third intermediate value, and an output gate performs a weighted operation on the input value to obtain a fourth intermediate value;

[0017] Determining the unit state of the image feature at the current moment according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value;

[0018] Obtaining an output value at the current moment according to the unit state of the image feature at the current moment and the fourth intermediate value; wherein the output value is a hidden representation of the image feature at the current moment;

[0019] The next sample image is obtained, the image features in the next sample image are used as the input values ​​at the next moment, and the hidden representation based on the image features at the previous moment is returned. The steps of performing a weighted operation on the input value by the forgetting gate to obtain a first intermediate value, performing a weighted operation on the input value by the input gate to obtain a second intermediate value and a third intermediate value, and performing a weighted operation on the input value by the output gate to obtain a fourth intermediate value are continued until the training termination condition is met and the iteration is stopped to obtain a trained long short-term memory neural network model.

[0020] In one embodiment, the step of training the fully connected neural network includes:

[0021] Inputting the image features of the sample image, the hidden representation of the image features, and the unit state into a fully connected neural network, and outputting an abnormality stage prediction result; the abnormality stage prediction result represents the probability value of being in different abnormal stages;

[0022] Determining image features to be deleted based on a difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, and determining adjustment parameters based on the image features to be deleted;

[0023] Determining a reward function based on the abnormal stage prediction result and the adjustment parameters;

[0024] With the goal of maximizing the reward function, the adjustment parameters of the fully connected neural network are adjusted until the training termination conditions are met, and the trained fully connected neural network is obtained.

[0025] In one embodiment, the method further comprises:

[0026] Sending evaluation standard information to the first terminal, so that the first terminal can visually display evaluation content to the first user; the evaluation standard information is used to guide the first user to perform an evaluation test on the first terminal;

[0027] The visual display includes at least one of picture display, text display, video display, and voice display.

[0028] In one embodiment, the method further comprises:

[0029] Based on an interaction request initiated by the first terminal or the second terminal, a communication connection is established between the first terminal and the second terminal, so that the first terminal receives an evaluation instruction sent by the second terminal and performs a remote evaluation according to the evaluation instruction; wherein the interaction request includes at least one of a chat request, a voice request, and a video request.

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

[0031] Pushing the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and

[0032] Receive the modified training plan sent by the second terminal, and push the modified training plan to the first terminal.

[0033] A device for pushing a remote training program, comprising:

[0034] an acquisition module, configured to acquire cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information;

[0035] a determination module, configured to determine abnormal stage information based on the cognitive function assessment information and medical related information;

[0036] The determination module is further configured to determine a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains;

[0037] The push module is used to push the training plan to the first terminal so that the user can perform training operations according to the training plan.

[0038] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0039] Obtaining cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information;

[0040] Determining abnormal stage information based on the cognitive function assessment information and medical related information;

[0041] Determining a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains;

[0042] The training plan is pushed to the first terminal so that the first user can perform training operations according to the training plan.

[0043] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0044] Obtaining cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information;

[0045] Determining abnormal stage information based on the cognitive function assessment information and medical related information;

[0046] Determining a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains;

[0047] The training plan is pushed to the first terminal so that the first user can perform training operations according to the training plan.

[0048] The above-mentioned remote training program push method, apparatus, computer equipment, and storage medium involve a first user performing a remote assessment on a first terminal, thereby obtaining cognitive function assessment information and medical-related information sent by the first terminal, and determining abnormal stage information based on the cognitive function assessment information and medical-related information. Based on the determined abnormal stage information, a matching training program is then determined and pushed to the first terminal, whereupon the first user can then perform training operations according to the training program. Thus, the system can automatically recommend appropriate assessment and training programs to users, enabling remote cognitive assessment and training. This allows users to conveniently complete home training without having to travel to a site or utilize special medical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a diagram of an application environment for a method for pushing a remote training solution in one embodiment;

[0050] Figure 2 1 is a flow chart of a method for pushing a remote training program in one embodiment;

[0051] Figure 3 A schematic diagram of a principle framework in which a server determines abnormal stage information based on cognitive function assessment information and medical-related information in one embodiment;

[0052] Figure 4 A flowchart illustrating steps for a server to determine abnormal stage information based on cognitive function assessment information and medical related information in one embodiment;

[0053] Figure 5 1 is a flow chart of the training steps of a long short-term memory neural network model in one embodiment;

[0054] Figure 6 A schematic diagram of the structure of a long short-term memory neural network model in one embodiment;

[0055] Figure 7 1 is a flow chart of the training steps of a fully connected neural network in one embodiment;

[0056] Figure 8 A structural block diagram of a device for pushing a remote training program in one embodiment;

[0057] Figure 9 A structural block diagram of a device for pushing a remote training program in another embodiment;

[0058] Figure 10 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0060] The present application provides a method, apparatus, computer equipment and storage medium for pushing remote training plans. By combining big data analysis with neural network models, the system can automatically push suitable assessment and training plans, which are more accurate and adaptable than manually selecting training plans based on experience, allowing users to achieve remote assessment and training using a terminal without relying on special medical equipment.

[0061] The terms "first" and "second" involved in the embodiments of the present application are used in this application to describe different terminals, but these terminals should not be limited by these terms. These terms are only used to distinguish one terminal from another terminal. For example, a first terminal can be referred to as a second terminal, and similarly, a second terminal can be referred to as a first terminal without departing from the scope of the various described embodiments, but unless the context clearly indicates otherwise, they are not the same terminal. A similar situation also exists with a first user and a second user.

[0062] In the embodiments of the present application, the first user refers to a person who needs training, such as a patient or a patient's family member, who can use the first terminal to receive a training plan and perform training operations accordingly. Conversely, the second user refers to a person who can provide training advice, such as a therapist or doctor, who can use the second terminal to view the training plan and / or modify or revise the training plan.

[0063] The first / second terminal may be, but is not limited to, various electronic devices such as personal computers, laptops, smartphones, tablets, smart TVs, and portable wearable devices. In some embodiments, the first / second terminal may be loaded with an APP application (such as a mini-program, etc.) or a client with webpage / website access capabilities, and provide a remote assessment portal, display training plans to the first user, and display training plans to the second user through the APP application or client. In some embodiments, the electronic device / processing device includes components such as a memory, a memory controller, one or more processing units (CPUs), a peripheral interface, an RF circuit, an audio circuit, a speaker, a microphone, an input / output (I / O) subsystem, a display, other output or control devices, and external ports, which communicate via one or more communication buses or signal lines.

[0064] The training plan refers to training materials that guide the first user on how to conduct training. The training materials include training materials related to cognitive domains such as memory, speech function, attention, executive function, visual-spatial function, emotion, sleep, mental behavior, and daily living abilities, and can be in the form of one or more of video tutorials, text tutorials, and voice guidance.

[0065] The push method of the remote training program provided in this application can be applied to Figure 1 In the application environment shown, the server 101 is connected to the first terminal 102 and the second terminal 103 via a network. The network may be the Internet, a mobile network, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), one or more intranets, or a suitable combination thereof. The embodiment of the present application does not limit the type or protocol of the communication network between the server and the electronic device on the first / second user side.

[0066] The server 101 is configured with a data processing system (or platform system, etc.). In some embodiments, the server can be arranged on one or more physical servers according to various factors such as function and load. Among them, when distributed on multiple physical servers, the server can be composed of servers based on cloud architecture. For example, servers based on cloud architecture include public cloud (Public Cloud) servers and private cloud (Private Cloud) servers, wherein public or private cloud servers include Software-as-a-Service (Software as a Service, SaaS), Platform-as-a-Service (Platform as a Service, PaaS) and Infrastructure-as-a-Service (Infrastructure as a Service, IaaS), etc. The server can also be composed of distributed or centralized server clusters. For example, a server cluster consists of at least one physical server. Multiple virtual servers are configured in each physical server, and each virtual server runs at least one functional module to implement the push method of the above-mentioned remote training solution, and each virtual server communicates through the network.

[0067] In one embodiment, Figure 2 As shown, a method for pushing a remote training program is provided, which is applied to Figure 1 The following steps are used as an example to illustrate the server in the example:

[0068] Step S202, obtaining cognitive function evaluation information and medical related information sent by the first terminal; the cognitive function evaluation information is obtained by the first user performing remote evaluation on the first terminal; the medical related information includes medical image information and biometric information.

[0069] The cognitive function assessment information includes, but is not limited to, assessment results of cognitive domain-related functions such as memory, speech function, attention, executive function, visual space, emotion, sleep, mental behavior, and daily living ability. The cognitive function assessment information is obtained by the first user performing a remote assessment on the first terminal. For example, the first terminal displays a cognitive function assessment scale to the first user on a display screen; after the first user completes filling in the cognitive function assessment scale, the first terminal obtains the cognitive function assessment information of the first user. The cognitive function assessment scale includes, but is not limited to, one or more of the Ascertain Dementia 8 (AD8), the Mini-Mental State Examination (MMSE), the Montreal Cognitive Assessment (MOCA), the Activity of Daily Living Scale (ADL), and the Memory and Executive Scale. For another example, the first terminal prompts the first user to perform a remote assessment through voice, obtains the voice information of the first user's answer according to the voice prompt, converts it into cognitive function assessment information, and sends it to the server. Of course, this is not limited to this. The above implementation is only an example and does not limit the scope of the embodiments of the present application. For example, after the first terminal and the second terminal establish a communication connection, voice communication between the first user and the second user can be realized. Thus, the first user completes the remote evaluation under the guidance of the second user, and the first terminal sends the cognitive function evaluation information of the first user to the server.

[0070] Medically relevant information refers to relevant medical data obtained after the diagnosis of the first user, including one or more of medical image information, biometric information, and life and dietary habit information. Medical image information includes one or more of imaging data and electroencephalogram data, such as electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI). Biometric information includes, but is not limited to, one or more of epidemiological data, baseline data, laboratory data, and Aβ42 and tau protein biomarker data.

[0071] Specifically, the first terminal sends the cognitive function assessment information to the server; the server receives and obtains the cognitive function assessment information and medical related information associated with the first user for subsequent processing.

[0072] Step S204: determining abnormal stage information based on the cognitive function assessment information and medical related information.

[0073] Among them, the abnormal stage information is used to represent the information of the stage of abnormality. Abnormal stages can be mild, moderate or severe. For example, in the case of Alzheimer's disease, the degree of deterioration of cognitive ability and physical function can be divided into three stages. The first stage is the mild stage, which is manifested by symptoms such as memory loss, decreased judgment ability, time disorientation, poor visual-spatial ability of complex structures, and a small vocabulary; the second stage is the moderate stage, which is manifested by severe damage to short and long-term memory, decreased visual-spatial ability of simple structures, time and place disorientation, calculation inability, various neurological symptoms, and visible symptoms such as aphasia, apraxia and agnosia; the third stage is the severe stage, which is manifested by symptoms such as severe memory loss, inability to take care of oneself in daily life, mutism, and limb rigidity.

[0074] Specifically, the server inputs the obtained cognitive function assessment information and medical-related information of the first user into a stage judgment model for processing, and the stage judgment model predicts the corresponding abnormal stage information. Thus, the server obtains abnormal stage information that matches the first user. The stage judgment model, for example, is a judgment model composed of big data. Based on the first user's cognitive function assessment information and medical-related information, combined with tens of thousands of cognitive function assessment information and medical-related information in the big data and corresponding abnormal stage information, it outputs a result indicating the abnormal stage that the first user may be in, thereby obtaining the abnormal stage information.

[0075] For another example, the stage judgment model is an artificial intelligence model, such as a long short-term memory neural network model (LSTM), or an improved model based on the LSTM neural network model, such as a neural network model formed by combining the LSTM neural network model with a fully connected neural network. For example, Figure 3 As shown, the server inputs the obtained cognitive function assessment information and medical-related information of the first user into a long short-term memory neural network model for processing, obtaining an intermediate result. This intermediate result is then input into a fully connected neural network, which then outputs an abnormality stage prediction result. The server then combines the abnormality stage prediction result output by the fully connected neural network with the cognitive function assessment information to obtain final abnormality stage information. In step S206, a training plan matching the abnormality stage information is determined based on the various training materials in the database. The training plan includes training materials related to multiple cognitive domains.

[0076] Specifically, after determining the abnormal stage information corresponding to the first user, the server determines a training program that matches the abnormal stage information, so that the first user can perform training accordingly. Specifically, the server can pre-establish a database and store a variety of training materials (including but not limited to one or more of video training materials, text training materials, and audio training materials, or a combination thereof) in the database; after determining the abnormal stage information, the server searches and matches the database to determine a matching training program.

[0077] In one embodiment, professional personnel can pre-build training plans that match different abnormal stage information and upload them to a server, which then stores them in a database. For example, for stage A, a professional can manually combine video, audio, and other files corresponding to various training materials and send them to the server as the training plan for stage A, which is then stored. After the server determines that the abnormal stage information corresponding to the first user is stage A, it searches the database and directly matches the training plan for stage A.

[0078] For example, assuming the abnormal stage information includes stages A to H, the training program corresponding to stage A includes video / audio / text training materials for sustained attention training, logical thinking training, reasoning ability training, long-term memory training, advanced daily life training, and applied mathematics training. The training program corresponding to stage B includes video / audio / text training materials for sustained attention training, shifting training, spatial memory training, character memory training, calculation practice training, maze training, and poetry comprehension training. The training program corresponding to stage C includes video / audio / text training materials for divided attention training, work training, verbal memory training, short-term memory training, and transportation training. The training program corresponding to stage D includes video / audio / text training materials for shifting attention training, sustained attention training, short-term memory training, drawing training, maze training, and poetry comprehension training. The training program corresponding to stage E includes video / audio / text training materials for shifting attention training, sustained attention training, control training, spatial memory training, character memory training, calculation practice training, puzzle training, and home life skills training. The training program for Phase F includes training materials in the form of video / audio / text for sustained attention training, task training, control training, short-term memory training, person memory training, place memory training, drawing training, spatial orientation training, shopping activity training, and home life training. The training program for Phase G includes training materials in the form of video / audio / text for sustained attention training, concept formation training, short-term memory training, spatial orientation training, shopping activity training, home life training, and verbal memory training. The training program for Phase H includes training materials in the form of video / audio / text for sustained attention training, shopping activity training, home life training, daily living skills training, short-term memory training, and spatial vision training.

[0079] Therefore, the server determines the training plan in the database that matches the abnormal stage information and contains training materials related to multiple cognitive domains, and realizes automatic and intelligent adjustment of the training plan according to the different characteristics of the abnormal stage information, facilitating targeted training of multiple cognitive domains.

[0080] Step S208: Push the training plan to the first terminal so that the first user can perform training operations according to the training plan.

[0081] Specifically, the server sends the determined training plan to the first terminal, and the first terminal displays the training plan to the first user in a visual manner, so that the first user can perform training operations according to the training plan. The training operation refers to the first user performing corresponding training according to the training plan.

[0082] The above-mentioned method for pushing remote training plans involves a first user performing a remote assessment on a first terminal, thereby obtaining cognitive function assessment information and medical-related information sent by the first terminal, and determining abnormal stage information based on the cognitive function assessment information and medical-related information. Based on the determined abnormal stage information, a matching training plan is then determined and pushed to the first terminal, whereupon the first user can then perform training operations according to the training plan. This system can automatically recommend appropriate assessment and training plans to the user, enabling remote cognitive assessment and training, allowing users to complete home training without the need for special medical equipment.

[0083] In a specific embodiment, the present application embodiment utilizes multi-linear multi-data interoperability technology for computers, apps, and mini-programs to build a MySQL (relational database management system) environment on an Alibaba Cloud server, and uses a MySQL database visualization management tool to build a complete software system consisting of the first user's information system, scale system, attachment system, video system, multimedia system, cognitive training system, remote data system, and data encryption transmission system. The first user can conduct a remote cognitive assessment on a tablet or mobile phone, and through the server's big data analysis and artificial intelligence technology, a suitable training plan is pushed, and the training plan is adaptively adjusted based on the first user's cognitive assessment characteristics.

[0084] When the primary or secondary user operates on the corresponding primary or secondary terminal, the client software installed on the terminal monitors the user's operation events and responds. When data is needed for analysis, the client software sends a request to the server interface. The server performs the corresponding add, delete, query, and modify operations on the MySQL database based on the request, and then packages the data into a JSON format and returns it to the client software. By routing the request through the Windows Communication Foundation (WCF) service, rather than requiring the client to connect directly to the database, this method minimizes the pressure on the MySQL connection pool and ensures database data security to a certain extent.

[0085] In order to further improve the accuracy and efficiency of identifying abnormal stage information, in some embodiments, such as Figure 4 As shown, the server determines abnormal stage information based on cognitive function assessment information and medical related information, including the following steps:

[0086] Step S402: performing image preprocessing on the medical image information to obtain a preprocessed image.

[0087] Medical image information includes EEG images and functional magnetic resonance imaging (fMRI). Specifically, the server first needs to perform image preprocessing on the EEG and fMRI images to facilitate neural network analysis and mining of the changing characteristics of the alpha band in the EEG images and the hidden edge features and grayscale features in the fMRI images. This also helps improve the accuracy of the results and eliminate the influence of noise.

[0088] For example, for EEG images, the server first uses a low-pass filter to filter the data, reducing high-frequency variations while retaining more useful low-frequency structure. Secondly, to ensure the temporal sparsity of the data, the image needs to be subsampled to reduce the correlation between the temporal dimension and the data. For functional magnetic resonance imaging (fMRI), the server first performs correction and registration to minimize mismatches between images. Next, the fMRI image is normalized and smoothed to eliminate the influence of interfering signals. This results in the server obtaining preprocessed medical image information, known as the preprocessed image.

[0089] In step S404, the pre-processed image is input into a pre-trained long short-term memory neural network model, and the long short-term memory neural network model performs feature extraction to obtain a hidden representation of the image features.

[0090] Image features include band variation, edge features, and grayscale features. Band variation is primarily the α band variation, which has a frequency distribution of 8-13 Hz and consists of two bands: μ1 (8-10 Hz) and μ2 (10-13 Hz), with an amplitude of approximately 20-100 μV.

[0091] Specifically, the server inputs the preprocessed image into a pre-trained long short-term memory neural network model, performs feature extraction through the long short-term memory neural network model, and processes it through the input gate, output gate, and forget gate in the long short-term memory neural network model to output a hidden representation of the image features.

[0092] Step S406 , based on the hidden representation of the biometric information and image features, a fully connected neural network is used for classification to obtain prediction stage information.

[0093] Specifically, the server uses the output obtained by processing the long short-term memory neural network model, that is, the hidden representation of the image features, as the input of the fully connected neural network, which is classified by the fully connected neural network to obtain prediction stage information.

[0094] Step S408: Determine abnormal stage information of the first user based on the cognitive function evaluation information and the predicted stage information.

[0095] Specifically, the server then combines the predicted stage information obtained by the fully connected neural network classification with the cognitive function assessment information of the first user to comprehensively determine the abnormal stage information that the first user is in. For example, if the server determines that the predicted stage information obtained by the fully connected neural network classification matches the abnormal stage information reflected by the cognitive function assessment information of the first user, the server will use the predicted stage information as the abnormal stage information of the first user. For another example, if the server determines that the predicted stage information obtained by the fully connected neural network classification is stage E in the aforementioned embodiment, and based on the cognitive function assessment information, it is determined that the stage information of the first user may be in stage C in the aforementioned embodiment, the server can determine that the abnormal stage information of the first user is stage E or stage D, etc.

[0096] In the above embodiment, the medical image information is analyzed by combining the long short-term memory neural network model with the fully connected neural network, which improves the recognition accuracy of abnormal stage information. Compared with the traditional manual on-site determination method, it avoids the influence of human subjective factors and is more efficient.

[0097] In some embodiments, as Figure 5 As shown in Figure 2, the training steps of the long short-term memory neural network model include:

[0098] Step S502: Acquire the current sample image, and use the image features in the current sample image as input values ​​at the current moment.

[0099] In step S504, based on the hidden representation of the image features at the previous moment, the forget gate performs a weighted operation on the input value to obtain a first intermediate value, the input gate performs a weighted operation on the input value to obtain a second intermediate value and a third intermediate value, and the output gate performs a weighted operation on the input value to obtain a fourth intermediate value.

[0100] Step S506 , determining the unit state of the image feature at the current moment according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value.

[0101] Step S508 , obtaining an output value at the current moment according to the unit state of the image feature at the current moment and the fourth intermediate value; wherein the output value is a hidden representation of the image feature at the current moment.

[0102] Step S510, obtain the next sample image, use the image features in the next sample image as the input value at the next moment, and return to the hidden representation based on the image features at the previous moment, and continue to perform the steps of performing weighted operation on the input value by the forgetting gate to obtain the first intermediate value, performing weighted operation on the input value by the input gate to obtain the second intermediate value and the third intermediate value, and performing weighted operation on the input value by the output gate to obtain the fourth intermediate value, until the iteration is stopped when the training termination condition is met, and a trained long short-term memory neural network model is obtained.

[0103] The training termination conditions include, but are not limited to, the number of training times reaching a preset number, the training duration reaching a preset duration, or the accuracy reaching a preset threshold.

[0104] Specifically, the server obtains multiple medical images as training samples and inputs them into a long short-term memory (LSTM) neural network model for training. During a particular training session, the server first obtains a sample image from that training session as the current sample image and performs feature extraction using the LSTM neural network model to obtain the image features of the current sample image. Because the LSTM neural network is a recurrent neural network, the current input must be combined with the previous output. Therefore, based on the hidden representation of the image features at the previous moment, the server performs a weighted operation on the input values ​​using a forget gate to obtain a first intermediate value, a weighted operation on the input values ​​using an input gate to obtain a second intermediate value and a third intermediate value, and a weighted operation on the input values ​​using an output gate to obtain a fourth intermediate value. The server then determines the unit state of the image features at the current moment based on the unit state of the image features at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value. The server then combines the unit state of the image features at the current moment with the fourth intermediate value to obtain the output value at the current moment, i.e., the hidden representation of the image features at the current moment. At the same time, the server stores the hidden representation of the image features at the current moment for use as input values ​​for the next training, thereby performing an iterative cycle.

[0105] During the next training, the server obtains the next sample image, uses the image features in the next sample image as the input value at the next moment, and returns to step S504 to continue execution until the training termination condition is met and the iteration is stopped to obtain a trained long short-term memory neural network model.

[0106] For example, Figure 6 As shown, the server inputs the sample image into the long short-term memory neural network model, and the long short-term memory neural network model performs feature extraction to obtain the image feature x of the current sample image. t (including all the image features of the sample image) as the input value at the current moment (i.e., moment t). Then, the hidden representation h of the image features at the previous moment (i.e., moment t-1) obtained through the previous training ist-1 , the input value x is given by the forget gate t Perform weighted operation to obtain the first intermediate value f t Among them, the forget gate input value x t Perform weighted operation to obtain the first intermediate value f t The calculation formula can be expressed as follows:

[0107] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0108] Among them, σ represents the sigmoid function, W f is the weight matrix of the forget gate, b f is the bias parameter.

[0109] In the LSTM neural network model, the forget gate performs this process. This process uses the sigmoid function to multiply the cell state by the output of the sigmoid function, taking the current input value and the previous output value. If the sigmoid function outputs 0, that information is forgotten; otherwise, it continues to be passed through the cell state. Simultaneously, the input gate processes this information, updating the old cell state. The forget gate previously determined which information to forget or add, which is then implemented by the input gate, and the output gate determines the final output value.

[0110] For example, the long short-term memory neural network model is composed of an input gate for the input value x t Perform weighted operation to obtain the second intermediate value i t and the third median Among them, the input gate is the input value x t Perform weighted operations to obtain the second intermediate value i t and the third median The calculation formula can be expressed as follows:

[0111] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0112]

[0113] Among them, W i and W C The weight matrix of the input gate, b i and b C is the bias parameter of the input gate, is a candidate value vector used to decide which information can be added to the cell state.

[0114] For example, the long short-term memory neural network model is composed of an output gate for the input value x t Perform weighted operation to obtain the fourth intermediate value o t . Among them, the output gate is for the input value x t Perform weighted operations to obtain the fourth intermediate value o t The calculation formula can be expressed as follows:

[0115] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0116] Among them, W o is the weight matrix of the output gate, b o is the bias parameter of the output gate.

[0117] After obtaining the first intermediate value, the second intermediate value, and the third intermediate value, the server calculates the unit state C of the image feature at the previous moment. t-1 , determine the unit state C of the image feature at the current moment t That is, first use the old state C t-1 Multiply the corresponding points by f t , that is, C t-1 ·f t , used to discard the information that has been decided to be forgotten, and then add Final unit state C t . It can be expressed by the formula:

[0118]

[0119] Then, the server combines the unit state C of the image features at the current moment t and the fourth intermediate value o t , get the output value at the current moment, that is, the hidden representation h of the image feature at the current moment t At the same time, the server’s hidden representation of the image features at the current moment h t The values ​​are stored for use as input for the next training, thus performing an iterative cycle.

[0120] In the above embodiment, analyzing medical image information through a long short-term memory neural network model can improve the accuracy of identifying abnormal stage information. Compared with traditional remote manual diagnosis, it avoids the influence of human subjective factors and has higher accuracy and efficiency.

[0121] In some embodiments, as Figure 7 As shown in Figure 2, the training steps of a fully connected neural network include:

[0122] In step S702, the image features of the sample image, the hidden representation of the image features, and the unit state are input into a fully connected neural network, and an abnormal stage prediction result is output; the abnormal stage prediction result represents the probability value of being in different abnormal stages.

[0123] Step S704 : determining image features to be deleted based on the difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, and determining adjustment parameters based on the image features to be deleted.

[0124] Step S706: Determine the reward function based on the abnormal stage prediction result and the adjustment parameters.

[0125] In step S708, the adjustment parameters of the fully connected neural network are adjusted with the goal of maximizing the reward function until the training termination condition is met, thereby obtaining a trained fully connected neural network.

[0126] Specifically, the server inputs the image features of the sample image extracted by the LSTM neural network model, the hidden representation of the image features processed by the LSTM neural network model, and the unit states in the LSTM neural network model into a fully connected neural network. The fully connected neural network performs classification and outputs an abnormality stage prediction result. The abnormality stage prediction result represents the probability of being in different abnormality stages. Typically, the abnormality stage prediction result output is a probability distribution. For example, for the three abnormality stages c1, c2, and c3, the abnormality stage prediction result P(y|X) output is the probability distribution of the disease stage prediction, where y∈{c1,c2,c3}. For each sample image, the server pre-labels the actual abnormality stage to which the sample image belongs. Therefore, the server can compare the abnormality stage prediction result output by the fully connected neural network with the actual abnormality stage to which the sample image belongs and, based on the difference between the two, determine whether to retain or delete the image features of the sample image. Based on the image features to be deleted, the server can determine the adjustment parameters and construct a reward function based on the abnormality stage prediction result.

[0127] Thus, the server can perform training based on the reward value returned by the constructed reward function, with the goal of maximizing the reward function. By adjusting the adjustment parameters of the fully connected neural network, the server can continuously iterate until the training termination conditions are met, thereby obtaining a trained fully connected neural network. The training termination conditions include, but are not limited to, reaching a preset number of training times, reaching a preset training duration, or reaching a preset accuracy threshold.

[0128] For example, the reward function can be calculated by the following formula:

[0129]

[0130] Among them, R L is the reward value, and P(c|X) is the anomaly prediction result obtained by inputting the sample image features. L′ refers to the number (number of times) of features deleted from a sample. It is used to delete a feature of a sample image when its impact on the final result is minimal. Accordingly, L is the number of complete features in the sample image. γ is a hyperparameter; a larger γ tends to delete features; conversely, a smaller γ tends to retain features. b is also a hyperparameter, which is manually set to make the reward value positive or negative.

[0131] In the above embodiment, the self-learning of the neural network is completed through deep reinforcement learning, the long short-term memory neural network is combined with the fully connected neural network, a stage judgment model is constructed, and then the model prediction results and the actual results are compared and analyzed, thereby improving the accuracy of identifying abnormal stages.

[0132] In some embodiments, the method further includes: sending assessment criteria information to the first terminal, so that the first terminal can visually display the assessment content to the first user; wherein the assessment criteria information is used to guide the first user to perform the assessment test on the first terminal, including but not limited to a cognitive function assessment scale, an assessment guidance video, or other materials that can guide the first user in performing the assessment. The visual display includes but is not limited to one or more of image display, text display, video display, and audio display, or a combination of these.

[0133] Specifically, the server can send evaluation standard information to the first terminal based on the remote evaluation request of the first terminal, so that the first terminal can visually display the evaluation content to the first user. For example, the first user performs touch and other operations on the first terminal, and the first terminal sends a remote evaluation request to the server accordingly. After the server receives the remote evaluation request sent by the first terminal, it searches the database for corresponding evaluation standard information (such as various scales) and sends the evaluation standard information to the first terminal. After the first terminal receives the evaluation standard information sent by the server, it visually displays it to the first user through a display screen, for example, displaying a scale to the first user through a page, or displaying a scale to the first user in combination with an audio prompt, or playing a video to the first user to guide the first user to perform a remote evaluation.

[0134] In the above embodiment, by sending the evaluation standard information to the first terminal, the user can complete the remote evaluation without going to the hospital, and the evaluation is convenient and flexible.

[0135] In some embodiments, the method further includes: establishing a communication connection between the first terminal and the second terminal based on an interaction request initiated by the first terminal or the second terminal, so that the first terminal receives an evaluation instruction sent by the second terminal and performs a remote evaluation based on the evaluation instruction. The interaction request includes, but is not limited to, one or more of a chat request, a voice request, and a video request. The evaluation instruction is used to guide the first user in performing the remote evaluation, such as a voice instruction, a text tutorial, or a video tutorial sent by the second user.

[0136] Specifically, the server may also receive an interaction request initiated by the first terminal and request a communication connection from the second terminal. After the second user agrees to the communication connection on the second terminal, the server may establish a communication connection between the first terminal and the second terminal. Thus, the first user may receive an evaluation instruction sent by the second user via the second terminal and complete the evaluation in accordance with the evaluation instruction. Alternatively, the server may receive an interaction request initiated by the second terminal and request a communication connection from the first terminal. After the first user agrees to the communication connection on the first terminal, the server may establish a communication connection between the first terminal and the second terminal. Thus, the second user may send an evaluation instruction to the first user, and the first user may complete the evaluation in accordance with the evaluation instruction.

[0137] In the above embodiment, a communication connection is established between the first or second user through an interactive request initiated by either the first or second user, enabling proactive and timely remote assessments. For example, the second user can proactively learn about the first user's situation when they schedule a return visit or discover an anomaly. Alternatively, the first user can proactively conduct a remote assessment when needed, creating a highly interactive experience and providing a positive user experience. Furthermore, the second user can remotely guide the cognitive function assessment based on professional advice, making the first user's remote assessment results more accurate, and subsequent training plans more effective and targeted.

[0138] In some embodiments, the above method further includes: pushing the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and receiving the modified training plan sent by the second terminal, and pushing the modified training plan to the first terminal.

[0139] Specifically, after determining the corresponding training plan, the server can also push the training plan to the second terminal so that the second user can view or modify the training plan on the second terminal. When the second user modifies the training plan, the second terminal sends the modified training plan to the server. The server receives the modified training plan from the second terminal and pushes the modified training plan to the first terminal so that the first user can perform training operations according to the modified training plan.

[0140] In the above embodiment, by sending the training plan obtained by automatic matching to the second user, the second user can conduct further summary analysis based on the cognitive function assessment information stored in the database of the server, and conduct professional analysis in combination with the imaging data, so as to make professional modifications or adjustments to the training plan, thereby making the training plan more accurate and targeted.

[0141] It should be understood that although Figure 2 、 Figure 4 、 Figure 6-Figure 7 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 2 、 Figure 4 、 Figure 6-Figure 7 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The order of execution of these steps or stages is not necessarily one by one, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0142] In one embodiment, Figure 8 As shown, a push device 800 for a remote training solution is provided, comprising: an acquisition module 801, a determination module 802, and a push module 803, wherein:

[0143] The acquisition module 801 is used to obtain cognitive function evaluation information and medical related information sent by the first terminal; the cognitive function evaluation information is obtained by the user performing remote evaluation on the first terminal; the medical related information includes medical image information and biometric information.

[0144] The determination module 802 is configured to determine abnormal stage information based on the cognitive function assessment information and medical related information.

[0145] The determination module 802 is further configured to determine a training program that matches the abnormal stage information based on the various training materials in the database, wherein the training program includes training materials related to multiple cognitive domains.

[0146] The push module 803 is used to push the training plan to the first terminal so that the user can perform training operations according to the training plan.

[0147] In one embodiment, the determination module is further used to perform image preprocessing on medical image information to obtain a preprocessed image; the preprocessed image is input into a pre-trained long short-term memory neural network model, and the long short-term memory neural network model performs feature extraction to obtain a hidden representation of the image features; the image features include band change features, edge features, and grayscale features; based on the hidden representation of the biological information and the image features, a fully connected neural network is used for classification to obtain predicted stage information; and based on the cognitive function assessment information and the predicted stage information, the abnormal stage information of the first user is determined.

[0148] In one embodiment, Figure 9 As shown, the above-mentioned device also includes a training module 804, which is used to obtain a current sample image and use the image features in the current sample image as the input value at the current moment; according to the hidden representation of the image feature at the previous moment, the forget gate performs a weighted operation on the input value to obtain a first intermediate value, the input gate performs a weighted operation on the input value to obtain a second intermediate value and a third intermediate value, and the output gate performs a weighted operation on the input value to obtain a fourth intermediate value; according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value, the unit state of the image feature at the current moment is determined; according to the image feature at the current moment, the unit state of the image feature is determined. The unit state of the feature and the fourth intermediate value are added to obtain the output value at the current moment; wherein the output value is the hidden representation of the image feature at the current moment; the next sample image is obtained, the image feature in the next sample image is used as the input value at the next moment, and the process returns to the hidden representation according to the image feature at the previous moment, the forget gate performs a weighted operation on the input value to obtain the first intermediate value, the input gate performs a weighted operation on the input value to obtain the second intermediate value and the third intermediate value, and the output gate performs a weighted operation on the input value to obtain the fourth intermediate value, and the iteration is stopped when the training termination condition is met to obtain a trained long short-term memory neural network model.

[0149] In one embodiment, the training module is further used to input the image features of the sample image, the hidden representation of the image features, and the unit state into the fully connected neural network, and output the abnormal stage prediction result; the abnormal stage prediction result represents the probability value of being in different abnormal stages; based on the difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, determine the image features to be deleted, and determine the adjustment parameters based on the image features to be deleted; determine the reward function based on the abnormal stage prediction result and the adjustment parameters; with the goal of maximizing the reward function, adjust the adjustment parameters of the fully connected neural network until the training is terminated when the training termination condition is met, and obtain a trained fully connected neural network.

[0150] In one embodiment, the above-mentioned device also includes a sending module for sending evaluation standard information to the first terminal so that the first terminal can visually display the evaluation content to the first user; the evaluation standard information is used to guide the first user to perform an evaluation test on the first terminal; wherein the visual display includes at least one of picture display, text display, video display, and voice display.

[0151] In one embodiment, the above-mentioned device also includes an interaction module, which is used to establish a communication connection between the first terminal and the second terminal based on an interaction request initiated by the first terminal or the second terminal, so that the first terminal can receive an evaluation instruction sent by the second terminal and perform a remote evaluation based on the evaluation instruction; wherein the interaction request includes at least one of a chat request, a voice request, and a video request.

[0152] In one embodiment, the push module is further used to push the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and receive the modified training plan sent by the second terminal, and push the modified training plan to the first terminal.

[0153] The specific definition of the remote training program push device can be found in the definition of the remote training program push method above, and will not be repeated here. The various modules in the above-mentioned remote training program push device can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0154] In one embodiment, a computer device is provided. The computer device may be the server in the aforementioned embodiment, and its internal structure diagram may be as shown in FIG. Figure 10 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as cognitive function assessment information, medical-related information, and training materials. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for pushing a remote training program is implemented.

[0155] Those skilled in the art will understand that Figure 10The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0156] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: obtaining cognitive function assessment information and medical related information sent by a first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical related information includes medical image information and biometric information; determining abnormal stage information based on the cognitive function assessment information and the medical related information; determining a training plan that matches the abnormal stage information based on a variety of training materials in a database, wherein the training plan includes training materials related to a plurality of cognitive domains; and pushing the training plan to the first terminal so that the first user can perform training operations according to the training plan.

[0157] In one embodiment, when the processor executes the computer program, it also implements the following steps: performing image preprocessing on the medical image information to obtain a preprocessed image; inputting the preprocessed image into a pre-trained long short-term memory neural network model, and performing feature extraction by the long short-term memory neural network model to obtain a hidden representation of the image features; the image features include band change features, edge features, and grayscale features; based on the hidden representation of the biometric information and the image features, a fully connected neural network is used for classification to obtain predicted stage information; and determining the abnormal stage information of the first user based on the cognitive function assessment information and the predicted stage information.

[0158] In one embodiment, when the processor executes the computer program, the following steps are further implemented: obtaining a current sample image and using the image features in the current sample image as the input value at the current moment; performing a weighted operation on the input value by a forget gate to obtain a first intermediate value, performing a weighted operation on the input value by an input gate to obtain a second intermediate value and a third intermediate value, and performing a weighted operation on the input value by an output gate to obtain a fourth intermediate value; determining the unit state of the image feature at the current moment based on the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value; determining the unit state of the image feature at the current moment based on the image feature at the current moment; performing a weighted operation on the input value by a forget gate to obtain a first intermediate value, a weighted operation on the input value by an input gate to obtain a second intermediate value and a third intermediate value, and a weighted operation on the input value by an output gate to obtain a fourth intermediate value. The unit state of the feature and the fourth intermediate value are used to obtain the output value at the current moment; wherein the output value is the hidden representation of the image feature at the current moment; the next sample image is obtained, the image feature in the next sample image is used as the input value at the next moment, and the process of returning to the hidden representation according to the image feature at the previous moment, performing a weighted operation on the input value by the forgetting gate to obtain the first intermediate value, performing a weighted operation on the input value by the input gate to obtain the second intermediate value and the third intermediate value, and performing a weighted operation on the input value by the output gate to obtain the fourth intermediate value is continued until the iteration is stopped when the training termination condition is met, and a trained long short-term memory neural network model is obtained.

[0159] In one embodiment, when the processor executes the computer program, the following steps are further implemented: inputting the image features, hidden representations of the image features, and unit states of the sample image into a fully connected neural network, and outputting an abnormal stage prediction result; the abnormal stage prediction result represents the probability value of being in different abnormal stages; based on the difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, determining the image features to be deleted, and determining the adjustment parameters based on the image features to be deleted; determining a reward function based on the abnormal stage prediction result and the adjustment parameters; adjusting the adjustment parameters of the fully connected neural network with the goal of maximizing the reward function until the training is terminated when the training termination condition is met, thereby obtaining a trained fully connected neural network.

[0160] In one embodiment, when the processor executes the computer program, it also implements the following steps: sending evaluation standard information to the first terminal so that the first terminal can visually display the evaluation content to the first user; the evaluation standard information is used to guide the first user to perform an evaluation test on the first terminal; wherein the visual display includes at least one of picture display, text display, video display, and voice display.

[0161] In one embodiment, when the processor executes the computer program, it further implements the following steps: establishing a communication connection between the first terminal and the second terminal based on an interaction request initiated by the first terminal or the second terminal, so that the first terminal receives an evaluation instruction sent by the second terminal and performs a remote evaluation based on the evaluation instruction; wherein the interaction request includes at least one of a chat request, a voice request, and a video request.

[0162] In one embodiment, when the processor executes the computer program, it further implements the following steps: pushing the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and receiving the modified training plan sent by the second terminal, and pushing the modified training plan to the first terminal.

[0163] The above-mentioned computer equipment enables remote cognitive assessment and training, allowing users to complete home training without the help of special medical equipment.

[0164] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: obtaining cognitive function assessment information and medical-related information sent by a first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biological information; determining abnormal stage information based on the cognitive function assessment information and the medical-related information; determining a training plan that matches the abnormal stage information based on a variety of training materials in a database, the training plan including training materials related to multiple cognitive domains; and pushing the training plan to the first terminal for the first user to perform training operations according to the training plan.

[0165] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: performing image preprocessing on the medical image information to obtain a preprocessed image; inputting the preprocessed image into a pre-trained long short-term memory neural network model, and extracting features by the long short-term memory neural network model to obtain a hidden representation of the image features; the image features include band change features, edge features, and grayscale features; based on the hidden representation of the biometric information and the image features, a fully connected neural network is used for classification to obtain predicted stage information; and determining the abnormal stage information of the first user based on the cognitive function assessment information and the predicted stage information.

[0166] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: obtaining a current sample image and using the image features in the current sample image as the input value at the current moment; performing a weighted operation on the input value by a forget gate to obtain a first intermediate value, performing a weighted operation on the input value by an input gate to obtain a second intermediate value and a third intermediate value, and performing a weighted operation on the input value by an output gate to obtain a fourth intermediate value; determining the unit state of the image feature at the current moment according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value; determining the unit state of the image feature at the current moment according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value; The unit state of the image feature and the fourth intermediate value are used to obtain the output value at the current moment; wherein the output value is the hidden representation of the image feature at the current moment; the next sample image is obtained, the image feature in the next sample image is used as the input value at the next moment, and the process of returning to the hidden representation according to the image feature at the previous moment, performing a weighted operation on the input value by the forgetting gate to obtain the first intermediate value, performing a weighted operation on the input value by the input gate to obtain the second intermediate value and the third intermediate value, and performing a weighted operation on the input value by the output gate to obtain the fourth intermediate value is continued until the training termination condition is met and the iteration is stopped to obtain a trained long short-term memory neural network model.

[0167] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: inputting the image features, hidden representations of the image features, and unit states of the sample image into a fully connected neural network, and outputting an abnormal stage prediction result; the abnormal stage prediction result represents the probability value of being in different abnormal stages; based on the difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, determining the image features to be deleted, and determining the adjustment parameters based on the image features to be deleted; determining a reward function based on the abnormal stage prediction result and the adjustment parameters; adjusting the adjustment parameters of the fully connected neural network with the goal of maximizing the reward function until the training is terminated when the training termination condition is met, thereby obtaining a trained fully connected neural network.

[0168] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: sending evaluation standard information to the first terminal so that the first terminal can visually display the evaluation content to the first user; the evaluation standard information is used to guide the first user to perform an evaluation test on the first terminal; wherein the visual display includes at least one of picture display, text display, video display, and voice display.

[0169] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: establishing a communication connection between the first terminal and the second terminal based on an interaction request initiated by the first terminal or the second terminal, so that the first terminal receives an evaluation instruction sent by the second terminal and performs a remote evaluation based on the evaluation instruction; wherein the interaction request includes at least one of a chat request, a voice request, and a video request.

[0170] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: pushing the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and receiving the modified training plan sent by the second terminal, and pushing the modified training plan to the first terminal.

[0171] The above-mentioned computer-readable storage medium realizes remote cognitive assessment and training, allowing users to complete home training without the help of special medical equipment.

[0172] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0173] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for pushing a remote training program, characterized in that: The method comprises: Obtaining cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a first user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information; Determining abnormal stage information based on the cognitive function assessment information and medical related information; Determining a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains; Pushing the training plan to the first terminal so that the first user can perform training operations according to the training plan; Determining abnormal stage information based on the cognitive function assessment information and medical related information includes: Performing image preprocessing on medical image information to obtain a preprocessed image; Inputting the preprocessed image into a pre-trained long short-term memory neural network model, and extracting features by the long short-term memory neural network model to obtain a hidden representation of image features; the image features include band change features, edge features, and grayscale features; Based on the hidden representation of the biological information and image features, a fully connected neural network is used for classification to obtain prediction stage information; determining abnormal stage information of the first user based on the cognitive function evaluation information and the predicted stage information; The training steps of the long short-term memory neural network model include: Obtaining a current sample image, and using image features in the current sample image as input values ​​at a current moment; According to the hidden representation of the image feature at the previous moment, a forget gate performs a weighted operation on the input value to obtain a first intermediate value, an input gate performs a weighted operation on the input value to obtain a second intermediate value and a third intermediate value, and an output gate performs a weighted operation on the input value to obtain a fourth intermediate value; Determining the unit state of the image feature at the current moment according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value; Obtaining an output value at the current moment according to the unit state of the image feature at the current moment and the fourth intermediate value; wherein the output value is a hidden representation of the image feature at the current moment; Obtaining a next sample image, using the image features in the next sample image as the input value at the next moment, and returning to the hidden representation based on the image features at the previous moment, performing a weighted operation on the input value by a forget gate to obtain a first intermediate value, performing a weighted operation on the input value by an input gate to obtain a second intermediate value and a third intermediate value, and performing a weighted operation on the input value by an output gate to obtain a fourth intermediate value, and continuing to perform the steps until a training termination condition is met, stopping the iteration, and obtaining a trained long short-term memory neural network model; The training steps of the fully connected neural network include: Inputting the image features of the sample image, the hidden representation of the image features, and the unit state into a fully connected neural network, and outputting an abnormality stage prediction result; the abnormality stage prediction result represents the probability value of being in different abnormal stages; Determining image features to be deleted based on a difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, and determining adjustment parameters based on the image features to be deleted; Determining a reward function based on the abnormal stage prediction result and the adjustment parameters; With the goal of maximizing the reward function, the adjustment parameters of the fully connected neural network are adjusted until the training termination conditions are met, and the trained fully connected neural network is obtained.

2. The method according to claim 1, characterized in that The method further comprises: Sending evaluation standard information to the first terminal, so that the first terminal can visually display evaluation content to the first user; the evaluation standard information is used to guide the first user to perform an evaluation test on the first terminal; The visual display includes at least one of picture display, text display, video display, and voice display.

3. The method according to claim 1, characterized in that The method further comprises: Based on an interaction request initiated by the first terminal or the second terminal, a communication connection is established between the first terminal and the second terminal, so that the first terminal receives an evaluation instruction sent by the second terminal and performs a remote evaluation according to the evaluation instruction; wherein the interaction request includes at least one of a chat request, a voice request, and a video request.

4. The method according to claim 1, wherein The method further comprises: Pushing the training plan to the second terminal so that the second user can modify the training plan on the second terminal; and Receive the modified training plan sent by the second terminal, and push the modified training plan to the first terminal.

5. A remote training program push device, characterized in that: The device comprises: an acquisition module, configured to acquire cognitive function assessment information and medical-related information sent by the first terminal; the cognitive function assessment information is obtained by a user performing a remote assessment on the first terminal; the medical-related information includes medical image information and biometric information; a determination module, configured to determine abnormal stage information based on the cognitive function assessment information and medical related information; The determination module is further configured to determine a training program that matches the abnormal stage information based on a plurality of training materials in a database, wherein the training program includes training materials related to a plurality of cognitive domains; A push module, configured to push the training plan to the first terminal, so that the user can perform training operations according to the training plan; The determination module is further configured to perform image preprocessing on the medical image information to obtain a preprocessed image; input the preprocessed image into a pre-trained long short-term memory neural network model, and extract features using the long short-term memory neural network model to obtain a hidden representation of image features; the image features include band change features, edge features, and grayscale features; perform classification using a fully connected neural network based on the biometric information and the hidden representation of the image features to obtain predicted stage information; and determine abnormal stage information of the first user based on the cognitive function assessment information and the predicted stage information; The training module is used to obtain a current sample image and use the image features in the current sample image as the input value at the current moment; according to the hidden representation of the image feature at the previous moment, the forget gate performs a weighted operation on the input value to obtain a first intermediate value, the input gate performs a weighted operation on the input value to obtain a second intermediate value and a third intermediate value, and the output gate performs a weighted operation on the input value to obtain a fourth intermediate value; according to the unit state of the image feature at the previous moment, the first intermediate value, the second intermediate value, and the third intermediate value, the unit state of the image feature at the current moment is determined; according to the unit state of the image feature at the current moment and the first intermediate value, the second intermediate value, and the third intermediate value, the unit state of the image feature at the current moment is determined; four intermediate values, obtaining an output value at the current moment; wherein the output value is a hidden representation of the image features at the current moment; obtaining a next sample image, using the image features in the next sample image as the input value at the next moment, and returning to the hidden representation based on the image features at the previous moment, performing a weighted operation on the input value by a forget gate to obtain a first intermediate value, performing a weighted operation on the input value by an input gate to obtain a second intermediate value and a third intermediate value, and performing a weighted operation on the input value by an output gate to obtain a fourth intermediate value, and continuing to perform the steps until the training termination condition is met, stopping the iteration, and obtaining a trained long short-term memory neural network model; The training module is further configured to input the image features of the sample image, the hidden representation of the image features, and the unit state into a fully connected neural network, and output an abnormal stage prediction result; the abnormal stage prediction result represents the probability value of being in different abnormal stages; based on the difference between the abnormal stage prediction result and the abnormal stage to which the sample image actually belongs, determine the image features to be deleted, and determine the adjustment parameters based on the image features to be deleted; determine a reward function based on the abnormal stage prediction result and the adjustment parameters; and adjust the adjustment parameters of the fully connected neural network with the goal of maximizing the reward function until the training is terminated when the training termination condition is met, thereby obtaining a trained fully connected neural network.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images