Cognitive disorder detection method, device, equipment, medium and product

Abstract

The invention relates to a cognitive impairment detection method, device, equipment, medium and product, and the method comprises the following steps: an obtaining step: obtaining information data about the cognitive state of a user through a plurality of voice interaction tasks; an extraction step of performing feature extraction on the obtained information data based on a machine learning mode to obtain key features, namely feature vectors, of nodes corresponding to each voice interaction task; a processing step: processing the nodes through the two graph attention networks, updating the feature representation of the key feature of each node, and obtaining a new feature vector of each node; and a detection step: through the new feature vector of each node, calculating to obtain a global feature representation reflecting the characteristics of the whole graph, and evaluating the cognitive ability according to the global feature representation so as to detect the cognitive impairment. According to the invention, the accuracy and adaptability of detection in various environments can be significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, equipment, medium, and product for detecting cognitive impairment based on voice interaction. Background Art

[0002] Speech ability generally refers to a person's ability to speak or communicate verbally. It is the process of producing sound through the body's various vocal structures, enabling interaction. The control of speech itself is the result of the coordinated efforts of various brain regions, with different brain regions performing their respective roles throughout the speech process, ultimately completing the entire conversation. Precisely because of the close connection between speech ability and various brain regions, the link between speech disorders and brain damage in neurological diseases has garnered considerable attention in neuroscience and clinical medicine.

[0003] Mild cognitive impairment (MCI) is a type of cognitive impairment. Patients experience problems with memory and cognition, with declines in one or more cognitive functions, but activities of daily living are not significantly affected. Approximately 50% of patients with MCI will develop Alzheimer's disease (AD) within five years. MCI patients may experience difficulties expressing and understanding language, such as a reduced vocabulary, disorganized grammatical structures, and shortened speech. These language changes may be early warning signs of cognitive decline and, if left untreated, can easily progress to severe cognitive impairment. In the early stages of AD, brain regions involved in language processing and memory are affected, leading to speech impairments such as decreased verbal fluency, loss of noun memory, and amnesia for common words. The presence of these language impairments not only provides clues for diagnosing AD but also opens the door to earlier therapeutic intervention. Parkinson's disease (PD) is currently believed to be linked to brain regions involved in motor control. However, a growing number of studies have shown that Parkinson's disease is also associated with a decline in language and cognitive function. Patients may experience speech disorders such as voice changes, slower speech speed, and decreased ability to perform complex language tasks. These symptoms can help detect Parkinson's disease early and provide guidance for the development of personalized treatment plans.

[0004] Therefore, for neurodegenerative diseases, especially MCI, AD and PD, early language disorders cannot be ignored. They are of great significance in the early warning of the disease and provide valuable clues for timely monitoring and intervention of the disease.

[0005] Voice interaction tasks involve the use of technologies such as speech recognition, natural language processing, and deep learning to achieve a more objective, accurate, and fine-grained assessment of cognitive abilities. This relatively intelligent approach allows us to capture subtle differences in a user's language expressions, speaking speed, intonation, and other aspects during voice interaction, providing a more comprehensive insight into their cognitive state. This allows us to assess their speech abilities and identify potential speech disorders early on. Effective speech recognition technology can accurately convert spoken content into text, providing a data foundation for subsequent cognitive analysis. Natural language processing technology can further parse and understand the meaning of user expressions, extracting key information about cognitive abilities. Deep learning technology plays a key role in both speech recognition and natural language processing. Through deep neural network learning and pattern recognition, it can capture more complex and abstract language features, thereby improving the accuracy of assessments.

[0006] Voice interaction tasks combine multiple effective methods to not only provide objective quantitative indicators of cognitive ability, but also enable a more granular assessment of user cognitive abilities, including but not limited to attention, memory, and language comprehension. By implementing real-time, dynamic, and process-based voice interaction tasks, new possibilities can be opened up for further development in cognitive science and medicine.

[0007] In summary, voice interaction tasks can provide more opportunities for early detection and intervention of speech disorders and neurodegenerative diseases, and are expected to promote progress in cognitive science and medicine, and provide new ideas and methods for disease warning and diagnosis.

[0008] Currently, methods that rely on speech recognition or voice interaction often require specialized expertise to develop relevant rules and are often subject to a degree of subjective influence, making their design and implementation relatively cumbersome. Furthermore, currently employed models are often overly simplistic, failing to fully reflect the fine-grained cognitive capabilities of different tasks. This simplicity can lead to unsatisfactory results when handling complex speech tasks or voice interaction processes, particularly due to their relatively poor robustness to changing contexts and speech characteristics.

[0009] However, existing cognitive impairment detection technologies have the following defects:

[0010] (1) Current speech recognition and interaction systems often rely on complex rules that must be developed by experts with specialized knowledge. Since it is difficult for non-professionals to adjust or optimize these rules, it not only increases the difficulty of system development but also limits the widespread application of these systems. Furthermore, this reliance on expert knowledge may result in the system being unable to flexibly adapt to diverse language expressions and dialect differences, thus affecting its effectiveness across different regions and populations.

[0011] (2) Existing voice interaction models are often too simple and may not be able to accurately process voice information, nor may they be able to understand complex language structures, ambiguous pronunciations, or subtle changes in context. Overly simple models limit their application in complex real-world scenarios, especially when dealing with changing contexts and voice features, where their performance and accuracy may be significantly reduced. In addition, low system complexity leads to insufficient robustness, which means that they may not be able to effectively cope with interference from factors such as background noise, changes in the speaker's emotions, or accents, thus affecting the assessment of cognitive ability.

[0012] (3) Current technologies are often unable to accurately distinguish and measure subtle differences in different cognitive tasks when assessing cognitive function, such as specific abilities in memory, attention, pronunciation, language, etc. This lack of fine-grained analysis makes it difficult for the system to provide accurate assessments of cognitive function, especially in distinguishing mild cognitive impairments such as MCI. As a result, these technologies may not accurately capture early signs of cognitive decline, thus affecting the early diagnosis and intervention of neurodegenerative diseases such as MCI, AD, and PD.

[0013] It can be seen that whether it is possible to provide an improved cognitive impairment detection technology based on voice interaction based on the deficiencies in the existing technology has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0014] Problems to be solved by the invention

[0015] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a cognitive impairment detection system based on process-based voice interaction that is more accurate in cognitive ability assessment, especially for early warning of speech disorders and neurodegenerative diseases. To this end, a method for fine-grained cognitive ability assessment and graph neural network (GNN) is proposed, which can carefully evaluate and analyze the user's cognitive abilities in memory, attention, pronunciation, language, and other aspects. In addition, by utilizing graph neural network technology, the present invention can not only provide a corresponding cognitive ability assessment for each process, but also effectively deal with complex situations in actual applications, such as user differences and data incompleteness, thereby significantly improving the accuracy and adaptability of the system in various environments.

[0016] Methods for solving problems

[0017] A first aspect of the present invention relates to a method for detecting cognitive impairment, comprising the following steps:

[0018] Acquisition step: obtaining information data about the user's cognitive status through multiple voice interaction tasks;

[0019] The extraction step uses machine learning to extract features from the acquired information data to obtain the key features of the node corresponding to each voice interaction task, namely the feature vector;

[0020] In the processing step, two graph attention networks are used to process the nodes, update the feature representation of the key features of each node, and obtain the new feature vector of each node;

[0021] In the detection step, the new feature vector of each node is used to calculate the global feature representation that reflects the characteristics of the entire graph. The cognitive ability is evaluated based on the global feature representation, thereby detecting cognitive impairment.

[0022] Preferably, the multiple voice interaction tasks include: vowel pronunciation, story reading, listening comprehension, picture description, process description, recent event recall, and free statement.

[0023] Preferably, in the extraction step, feature extraction is performed on the acquired information data based on a support vector machine approach.

[0024] Preferably, a kernel function is used for feature extraction in a support vector machine.

[0025] Preferably, a multi-head attention mechanism is adopted in the graph attention network.

[0026] Preferably, the global feature is represented as the arithmetic mean of the feature vectors of all nodes.

[0027] A second aspect of the present invention relates to a cognitive impairment detection device, comprising:

[0028] An acquisition module is used to obtain information data about the user's cognitive status through multiple voice interaction tasks;

[0029] The extraction module is used to extract features from the acquired information data based on machine learning to obtain the key features of the node corresponding to each voice interaction task, namely the feature vector;

[0030] The processing module is used to process the nodes through two graph attention networks, update the feature representation of the key features of each node, and obtain a new feature vector for each node;

[0031] The detection module is used to calculate a global feature representation reflecting the characteristics of the entire graph through the new feature vector of each node, evaluate cognitive ability based on the global feature representation, and thus detect cognitive impairment.

[0032] A third aspect of the present invention relates to a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the cognitive impairment detection method of the first aspect are implemented.

[0033] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cognitive impairment detection method of the first aspect.

[0034] A fifth aspect of the present invention relates to a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the cognitive impairment detection method of the first aspect.

[0035] Effects of the Invention

[0036] The improved cognitive impairment detection method provided by this invention reduces the need for human intervention, thereby reducing potential subjective errors and improving the objectivity and repeatability of the assessment. Furthermore, the use of a graph attention network to dynamically weight features not only makes the assessment results more accurate but also reflects more complex cognitive abilities, enabling more accurate assessments of a user's cognitive abilities and the early detection of symptoms of neurodegenerative diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flowchart of the cognitive impairment detection method according to the first embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the principle of the cognitive impairment detection method according to the first embodiment of the present invention.

[0039] Figure 3 This is a flowchart of the cognitive impairment detection method according to the first embodiment of the present invention.

[0040] Figure 4 This is a flowchart of the cognitive impairment detection method according to the first embodiment of the present invention.

[0041] Figure 5 This is a structural diagram of a computer device according to a third embodiment of the present invention. DETAILED DESCRIPTION

[0042] Embodiments of the present invention will now be described more fully with reference to the accompanying drawings, in which embodiments of the present invention are shown. However, the present invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein.

[0043] The terms used herein are for the purpose of describing specific embodiments only and are not intended to limit the present disclosure. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that when used herein, the term "comprising" specifies the presence of stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0044] Unless otherwise defined, the terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs. The terms used herein should be interpreted as having the same meaning as that in the context of this specification and the relevant art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as such herein.

[0045] Hereinafter, the cognitive impairment detection method according to the present invention will be described in detail.

[0046] Figure 1 FIG. 1 is a flow chart of a method for detecting cognitive impairment according to a first embodiment of the present invention. Figure 1 As shown, the specific process of the cognitive impairment detection method is as follows: first, an acquisition step (step S100) is performed to obtain information data about the user's cognitive status through multiple voice interaction tasks. Then an extraction step (step S101) is performed to perform feature extraction on the acquired information data based on a machine learning method to obtain the key features of the node corresponding to each voice interaction task, i.e., the feature vector. Then a processing step (step S102) is performed to process the nodes through two graph attention networks, update the feature representation of the key features of each node, and obtain a new feature vector for each node. Finally, a detection step (step S103) is performed to calculate a global feature representation reflecting the characteristics of the entire graph through the new feature vector of each node, and perform an assessment of cognitive ability based on the global feature representation, thereby detecting cognitive impairment.

[0047] Step S100 is described. Specifically, Figure 2 As shown, the multiple voice interaction tasks preferably include: vowel pronunciation, story reading, listening comprehension, picture description, process description, recent event recall, free statement, and may also include other voice interaction tasks. Initially, the user completes a series of voice and language tasks, each of which is designed to assess a specific aspect of cognitive ability. By executing the corresponding voice interaction tasks in sequence, the system can obtain key information about the user's cognitive state, and then perform subsequent related evaluation tasks based on the key information obtained. Examples of these voice interaction tasks are introduced below.

[0048] Vowel pronunciation: The user continuously pronounces the six vowels in Chinese Pinyin for as long as possible, hoping to obtain the duration and stability of the user's pronunciation.

[0049] Reading a story: Users first read a text aloud, and then repeat the text as completely and accurately as possible without the help of the text, so as to achieve the purpose of assessing users' memory, language comprehension and reproduction ability.

[0050] Listening comprehension: First, a descriptive audio clip is played, and then the participant repeats the main content heard and their own understanding, thereby assessing the user's listening comprehension and information processing ability.

[0051] Picture Description: This task involves describing the content of a single image. During this task, users are presented with a single image, such as a picture of a person stealing cookies, and asked to describe its contents. The other part involves presenting multiple potentially related images and having them describe their content. This assesses the user's visual information processing and language expression abilities, which are evaluated through the level of detail, accuracy, and coherence of the description.

[0052] Process description: Users describe an orderly process or activity, such as the entire process of cooking a dish or sowing a certain type of crop, to evaluate the logic and structured language expression ability of the user's description.

[0053] Recent Recall: Users recall and describe a recent event, thereby assessing their short-term memory and narrative ability.

[0054] Free expression: There are no requirements for users, allowing them to freely talk about any topic. It is used to evaluate the user's ability to conceive and express their inner thoughts freely.

[0055] These tasks comprehensively assess the user's fine-grained cognitive abilities in multiple aspects, including memory, attention, pronunciation, and language. Each task can provide key information in the corresponding cognitive field and plays an important role in subsequent cognitive ability assessment.

[0056] Step S101 is explained. Specifically, after step S100, a large amount of user voice-related data is collected, and the purpose of feature extraction is to identify and extract key information reflecting the user's cognitive state from various voice and language data. By carefully analyzing the user's performance in various tasks, such as the persistence of vowel pronunciation, memory and retelling ability in narrative tasks, and the level of detail in image descriptions, many features can be obtained for each task. However, not all of these features are worth further analysis, and it is necessary to select more relevant features and remove redundant or irrelevant features through feature selection.

[0057] Manually screening for effective features is obviously unreasonable, as manual feature extraction is often time-consuming and impractical for a system. Therefore, we use automated feature extraction, applying machine learning methods to analyze the raw features obtained, and identify which features, from a large number of potentially effective features, are most likely to predict or explain the user's cognitive state.

[0058] The machine learning method is preferably support vector machine (SVM), or it can also be Lasso regression or decision tree. In the case of support vector machine, Figure 3 As shown, it is preferred to use a kernel function for feature extraction. The kernel function converts the samples of the input space into a high-dimensional space and converts the operations of the high-dimensional vector into the operations of the low-dimensional vector, thereby saving a huge amount of calculation. Figure 3 The final features extracted by the kernel function are the key features.

[0059] Automatic feature extraction methods standardize the contributions of different features, evaluate each feature's contribution to the model, and select those with the greatest impact on the target variable. By reducing the number of features in the dataset, redundant or irrelevant features that could negatively impact model performance are eliminated, and the computational burden of subsequent model training is reduced. Furthermore, reducing feature dimensionality effectively prevents model overfitting and accelerates subsequent model training and prediction.

[0060] Through automatic feature extraction and feature selection methods, the assessment system can effectively extract, select, integrate and format key features from multi-dimensional data, which not only ensures the accuracy and reliability of cognitive ability assessment, but also improves the efficiency of the entire assessment process.

[0061] Step S102 is explained. Specifically, Graph Attention Networks is a graph neural network architecture that introduces an attention mechanism when processing graph structured data. In a graph attention network, each node updates its own feature representation by aggregating feature information of its neighboring nodes. Unlike traditional graph neural networks (such as graph convolutional neural networks), the graph attention network does not simply average the features of neighboring nodes, but weights the features of neighboring nodes by using attention coefficients, allowing the model to decide which neighboring nodes should be paid attention to when updating the node representation. The graph attention layer is the basic unit of the graph attention network. Each layer updates the representation of each node in the graph by learning information from its neighboring nodes and then using the attention mechanism to weight this information.

[0062] In step S100, each procedural voice interaction task corresponds to a node in the graph. Its feature vector is the key feature extracted from each task through the feature extraction and selection process described above. A graph attention network is then used to process these nodes, updating the feature representation of each node and establishing relationships with other tasks.

[0063] In the graph attention layer (graph attention network), the features of the nodes need to undergo a learnable linear transformation, which is usually implemented by a weight matrix W. For each node v in the graph, its feature vector h v After linear transformation, the new feature representation is obtained:

[0064] h' v =W·h v

[0065] For each neighbor node j of node i, the original attention coefficient e is calculated by a small single-layer feedforward neural network ij Its input is the concatenation or element-wise multiplication of the feature representations of node i and node j. The specific calculation formula is as follows:

[0066]

[0067] in is a learnable weight vector, || means and LeakyReLU is an activation function, which is defined as follows:

[0068]

[0069] Here, α is a hyperparameter.

[0070] Then as Figure 4 As shown, the softmax function is used to normalize all the calculated attention coefficients to ensure that the sum of the weights of each node is 1:

[0071]

[0072] in Represents the set of neighbor nodes of node i.

[0073] Using the normalized attention coefficient α ij , we can get the new feature vector h' for each node i i , which is the feature vector h of all its neighbor nodes j j The weighted sum of , the whole process can be regarded as a feature aggregation process, which is calculated as follows:

[0074]

[0075] In order to stabilize the learning process and improve the expressiveness of features, it is preferred that the graph attention layer also adopts a multi-head attention mechanism. In the multi-head attention mechanism, all the above processes are repeated K times, and each "head" has its own independent weight matrix W k and the corresponding attention vector a k Specifically, K unique attention heads will perform the feature aggregation process and then complete the final output feature representation according to the following calculation formula:

[0076]

[0077] Among them, || represents the splicing operation, is the attention coefficient calculated by the kth attention head, W k is the weight matrix of the corresponding linear transformation.

[0078] However, performing multi-head attention on the final (prediction) layer of the network will result in the wrong output size, so in this case, no concatenation is used. Therefore, an average is often used in the final layer of the network, and the nonlinear transformation is delayed until the average calculation is completed, rather than performing the nonlinear transformation first and then the averaging operation.

[0079] Step S103 is described. Specifically, Figure 2 As shown, preferably, the global feature is represented as the arithmetic mean of the feature vectors of all nodes (average pooling layer).

[0080] The average pooling layer calculates the arithmetic mean of the feature vectors of all nodes (features of different cognitive tasks) obtained after the two graph attention layers, and obtains a global feature representation that reflects the characteristics of the entire graph. The calculation method is as follows:

[0081]

[0082] Among them H avg is the result obtained after average pooling, N represents the total number of nodes in the graph, h i is the eigenvector of node i.

[0083] After two graph attention layers, each node (features of different cognitive tasks) receives a feature representation weighted by the attention mechanism. The average pooling layer integrates these features, helping to capture commonalities between nodes and reducing the variance in feature representations caused by the attention mechanism. This approach provides a holistic representation of cognitive ability, capturing the average performance across all different cognitive tasks.

[0084] When assessing cognitive abilities, people are often more concerned with an individual's overall cognitive state. Using the average pooling layer, the model can generate a global score representing the overall cognitive level. While the graph attention layer may produce high-dimensional feature representations, the average pooling layer aggregates these features, reducing the dimensionality of the model output to match the desired output format.

[0085] When performing different cognitive tasks, participants may exhibit varying performance across tasks. These differences may stem from factors such as age, gender, education level, and cultural differences. These individual differences can increase the variability of evaluation results, impacting their accuracy and reliability. Therefore, it is essential to smooth the multidimensional feature vectors of the output nodes using an average pooling layer. This layer can smooth performance fluctuations caused by various non-cognitive factors, thereby reducing variability in evaluation results.

[0086] In environments with high effectiveness requirements, such as clinical environments, the efficiency of evaluation tools is also important. The average pooling layer can provide a fast and effective method to integrate the results of cognitive tasks, helping to complete the evaluation quickly and effectively.

[0087] It can be seen that the cognitive impairment detection method according to the first embodiment of the present invention achieves the following technical effects.

[0088] 1. Through a set of Chinese voice task processes, covering different tasks from pronunciation to free statement. These tasks systematically and comprehensively evaluate the user's multi-faceted cognitive abilities, including pronunciation control, memory, listening comprehension, visual processing, logical expression and narrative ability, etc. This process provides a structured method for the multi-dimensional evaluation of cognitive abilities. These tasks can specifically evaluate various fine-grained cognitive abilities including memory, attention, pronunciation, language comprehension, etc. Through theoretical analysis and experimental verification, the design of this process-based voice interaction task is closer to the actual language use context, can more accurately reflect the individual's cognitive state, and especially shows a relatively higher sensitivity in capturing subtle signals of early cognitive decline.

[0089] 2. By extracting speech and language features for each specific voice interaction task, we achieve the extraction and feature selection of user cognitive ability feature vectors. Compared with traditional feature extraction methods, this method can automatically capture key information related to cognitive ability assessment.

[0090] 3. The innovative use of a graph attention network allows the feature vectors of the nodes representing each task to be dynamically determined through learned attention weights rather than simply superimposed or concatenated. This reflects the relative importance of different cognitive tasks in the assessment of user cognitive abilities, automatically learns the relationships between different tasks, and assigns appropriate weights to each task. The graph attention layer is particularly helpful in automatically adjusting the model to maintain the accuracy and robustness of the assessment in practical applications with user differences and incomplete data. Using a graph attention network to dynamically weight features also makes the assessment results not only more accurate but also reflects more complex cognitive abilities, enabling more accurate assessments of users' cognitive abilities and the early detection of symptoms of neurodegenerative diseases.

[0091] 4. Combining speech recognition, natural language processing, and deep learning technologies enables a fine-grained assessment of cognitive abilities. This approach objectively and accurately captures fine-grained differences in users' performance when completing specific tasks, enabling a comprehensive assessment of cognitive abilities. This significantly overcomes the detailed limitations of existing technologies and improves the accuracy of cognitive ability assessments.

[0092] The cognitive impairment detection device of the second embodiment of the present invention includes: an acquisition module for acquiring information data about the user's cognitive status through multiple voice interaction tasks; an extraction module for performing feature extraction on the acquired information data based on machine learning to obtain the key features of the node corresponding to each voice interaction task, i.e., the feature vector; a processing module for processing the nodes through two graph attention networks, updating the feature representation of the key features of each node, and obtaining a new feature vector for each node; a detection module for calculating a global feature representation reflecting the characteristics of the entire graph through the new feature vector of each node, and evaluating cognitive ability based on the global feature representation, thereby detecting cognitive impairment. The cognitive impairment detection device corresponds to the cognitive impairment detection method of the first embodiment, so the various variations in the first embodiment are also applicable to the second embodiment and will not be repeated here.

[0093] As described above, the cognitive impairment detection device according to the second embodiment of the present invention solves the problems of inaccurate information recommended by the prior art, which leads to the emergence of information cocoons, by accurately calculating the user portrait. It improves the accuracy and real-time performance of information recommendations and provides users with more accurate decision support.

[0094] The third embodiment of the present invention provides a computer device, the internal structure of which can be shown as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database 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 network interface of the computer device is used to connect to an external device to exchange data with the external device. When the computer program is executed by the processor, it implements the cognitive impairment detection method involved in the first embodiment of the present invention.

[0095] Those skilled in the art will understand that Figure 5 The 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.

[0096] A fourth embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When a processor executes the computer program, the method for detecting cognitive impairment according to the first embodiment of the present invention is implemented.

[0097] A fifth embodiment of the present invention provides a computer program product, including a computer program. When a processor executes the computer program, the method for detecting cognitive impairment according to the first embodiment of the present invention is implemented.

[0098] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. 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 can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0099] Although some specific embodiments of the present invention have been described in detail through examples, those skilled in the art will understand that the above examples are for illustration only and are not intended to limit the scope of the present invention. Those skilled in the art will understand that the above embodiments may be modified or some technical features may be replaced with equivalents without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A method for detecting cognitive impairment, characterized in that: The following steps are involved: Acquisition step: obtaining information data about the user's cognitive status through multiple voice interaction tasks; The extraction step uses machine learning to extract features from the acquired information data to obtain the key features of the node corresponding to each voice interaction task, namely the feature vector; In the processing step, two graph attention networks are used to process the nodes, update the feature representation of the key features of each node, and obtain the new feature vector of each node; In the detection step, the new feature vector of each node is used to calculate the global feature representation that reflects the characteristics of the entire graph. The cognitive ability is evaluated based on the global feature representation, thereby detecting cognitive impairment.

2. The method for detecting cognitive impairment according to claim 1, wherein: Multiple voice interaction tasks include: vowel pronunciation, story reading, listening comprehension, picture description, process description, recent event recall, and free statement.

3. The method for detecting cognitive impairment according to claim 1, wherein: In the extraction step, feature extraction is performed on the acquired information data based on the support vector machine method.

4. The method for detecting cognitive impairment according to claim 3, wherein: Kernel functions are used for feature extraction in support vector machines.

5. The cognitive impairment detection method according to claim 1, wherein: A multi-head attention mechanism is adopted in the graph attention network.

6. The method for detecting cognitive impairment according to claim 1, It is characterized in that The global feature is expressed as the arithmetic mean of the feature vectors of all nodes.

7. A cognitive impairment detection device, characterized in that: include: An acquisition module is used to obtain information data about the user's cognitive status through multiple voice interaction tasks; The extraction module is used to extract features from the acquired information data based on machine learning to obtain the key features of the node corresponding to each voice interaction task, namely the feature vector; The processing module is used to process the nodes through two graph attention networks, update the feature representation of the key features of each node, and obtain a new feature vector for each node; The detection module is used to calculate a global feature representation reflecting the characteristics of the entire graph through the new feature vector of each node, evaluate cognitive ability based on the global feature representation, and thus detect cognitive impairment.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the cognitive impairment detection method according to any one of claims 1 to 6 are implemented.

9. 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 cognitive impairment detection method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the cognitive impairment detection method according to any one of claims 1 to 6 are implemented.

Images