CN114676249A - Data sample detection method and system for Alzheimer's disease - Google Patents

Abstract

The invention provides a method and a system for detecting data samples of Alzheimer's disease, comprising the following steps: based on the intelligent home equipment of the Internet of things, passive and active multi-task data acquisition is carried out; performing voice-to-text processing on voice data in the collected data, and authorizing text information to be transmitted at each intelligent home node to obtain authorized text data; after the authorized text data is verified, extracting text data characteristics and carrying out statement continuity analysis; and summarizing and aggregating the extracted features, classifying, and combining the sentence continuity analysis result to obtain a final detection result of the data sample aiming at the Alzheimer disease. The invention has lower dependency on equipment, and can be deployed efficiently and perform simple data collection; a safety mechanism is provided to ensure the safety of the whole process; high precision of data acquisition is guaranteed based on multi-task investigation and an effective feature extraction mechanism.

Description

Data sample detection method and system for Alzheimer's disease

technical field

The present invention relates to the technical field of medical devices, in particular, to a data sample detection method and system for Alzheimer's disease, and simultaneously provides a corresponding device and a computer-readable storage medium.

Background technique

Alzheimer's disease is a progressive neurodegenerative disease with an insidious onset. Clinically, it is characterized by generalized dementia manifestations such as memory impairment, aphasia, apraxia, agnosia, impairment of visuospatial skills, executive dysfunction, and personality and behavioral changes. For the collection and detection of physiological data related to Alzheimer's disease, there is always the need for professional and expensive equipment, and it needs to be carried out continuously in the hospital. Because the sick population is mainly concentrated in the middle-aged and the elderly, it is very inconvenient for the early examination. The irreversible characteristics of Alzheimer's disease have extremely adverse effects on life and health without early detection and prevention.

Since people with Alzheimer's disease will show slow speech, forgetting events and other characteristics in the early stage, this phenomenon can be expressed through speech and text. Therefore, machine learning methods can be used to analyze the collected data, find small changes, and distinguish between healthy and sick people. The current inspection methods mainly include manual inspection based on questionnaires and pathological inspection based on medical equipment. There is always room for improvement in the accuracy and cost of the manual inspection method for Alzheimer's disease based on the questionnaire (MMSE). Pathological examinations based on medical equipment, such as MRI and other means, may be difficult to detect in the early stage, and rely on expensive medical equipment, which is costly and may bring other hazards to the health of users.

Today, with the widespread deployment of massive distributed IoT devices in smart home scenarios (Amazon Alexa, Google home, etc.), the user experience has been greatly optimized, and many complex data collection problems have been solved. If the data collection and detection of Alzheimer's disease can be realized with the help of IoT technology, it will have great research value and application potential.

At present, there is no description or report of the technology similar to the present invention, and no similar materials at home and abroad have been collected.

SUMMARY OF THE INVENTION

Aiming at the above deficiencies in the prior art, the present invention provides a data sample detection method and system for Alzheimer's disease, as well as a corresponding device and a computer-readable storage medium.

According to one aspect of the present invention, there is provided a data sample detection method for Alzheimer's disease, comprising:

Passive and active multi-task data collection based on IoT smart home devices;

Perform voice-to-text processing on the voice data in the collected data, and authorize the text information to be transmitted at each smart home node to obtain the authorized text data;

After the authorized text data is verified, the text data features are extracted, and sentence coherence analysis is performed;

The extracted features are aggregated and classified, and combined with the sentence coherence analysis result, the final detection result of the Alzheimer's disease data sample is obtained.

Preferably, the passive data collection includes: daily command information data collection for smart home devices; the active data collection includes: voice and text data collection for set tasks.

Preferably, the voice data in the collected data is subjected to voice-to-text processing, and the text information that needs to be transmitted is authorized at each smart home node, and the authorized text data includes:

preprocessing the collected data;

The voice data in the preprocessed data is subjected to voice-to-text processing;

At each smart home node, sign and authorize the text information to be transmitted, and obtain the authorized text data.

Preferably, the preprocessing includes:

Perform filtering and denoising processing on the collected data;

Part of the collected data is pre-emphasized.

Preferably, the signature authorization is performed on the text information to be transmitted, and the authorized text data includes:

For the i-th smart home node, use the GroupGen polynomial time algorithm to generate a cyclic group G, the order of the cyclic group G is p, the generator is g, and the cyclic group G satisfies the DDH assumption;

Randomly select two elements a and b from [1, p], generate g ^a and g ^b and exchange them, and calculate to obtain g ^ab as the signature key;

Use the signature algorithm Sig( ) to sign the text information to be transmitted, and use the hash function Hash to map the text information to obtain the authorized text data Pre(Datai):

Pre(Data _i )=[Trans(Data _i ), Sig(g ^ab , Hash(Trans(Data _i )))]

Wherein, Pre(Data _i ) is the authorized text data of the ith smart home node, and Trans(Data _i ) is the text information transmitted by the ith smart home node.

Preferably, after the authorized text data is verified, the text data features are extracted, and sentence coherence analysis is performed, including:

A signature verification algorithm is used to verify whether the text data has been tampered with; if the data is tampered, the calculation result of the corresponding smart home edge node is directly discarded; if the data has not been tampered with, the corresponding feature extraction and statement coherence are performed analyze;

Using a hierarchical attention network model to extract features from words to sentences to discourse on the text data;

Sentence coherence analysis is performed on the text data based on Markov chains.

Preferably, the hierarchical attention network model includes: a word-level encoder layer, a word-level attention layer, a sentence-level encoder layer, and a sentence-level attention layer; wherein:

The word-level encoder layer is used to encode each word to obtain a latent vector;

The word-level attention layer is used to perform dot product calculation on the selected hidden vector to obtain the attention weight;

The sentence-level encoder layer is used to perform a weighted sum of the hidden vectors according to the obtained sequence of hidden vectors to obtain the vector of the sentence;

The sentence-level attention layer is used to perform dot product calculation on the vector of the selected sentence to obtain the attention weight of the sentence;

The vector of each sentence is weighted and averaged according to the attention weight of the sentence, and the final discourse vector is obtained and output, so as to realize feature extraction from words to sentences and then to the entire discourse.

Preferably, the sentence coherence analysis on the text data based on the Markov chain includes:

Based on the word frequency, multiple words with the highest frequency in the text data are extracted, and the obtained words are recorded as W _set [w ₁ , w ₂ ......, w _n ];

For any text data, cut according to the sentence, retrieve the sentence j that contains the words in W _set , and extract the corresponding words, which are recorded as

Use the existing text training data set of Alzheimer's disease patients to train the Markov chain to obtain the corresponding transition probability; according to the transition probability, compare the corresponding vocabulary of sentence j in the text data, if the transition probability Correct, the coherence score Score _i of the sentence j is increased by 1, otherwise do nothing;

The coherence scores of all sentences in the text data are normalized, divided according to the set threshold Th, the Score _i greater than the threshold value is modified to 1, and the Score _i less than the threshold value is modified to 0 to obtain the final result. Statement coherence analysis results.

Preferably, a lightweight neural network is used to classify the aggregated features to obtain a classification result; the classification result is combined with the sentence coherence analysis result to obtain a final data sample for Alzheimer's disease test results.

According to another aspect of the present invention, there is provided a data sample detection system for Alzheimer's disease, comprising:

A multi-source data acquisition module, which is based on IoT smart home devices for passive and active multi-task data acquisition;

a data preprocessing module, which performs voice-to-text processing on the voice data in the collected data, and authorizes the text information to be transmitted at each smart home node to obtain the authorized text data;

A feature extraction module, which extracts text data features after verifying the authorized text data, and performs statement coherence analysis;

A result detection module, which summarizes and aggregates the extracted features and performs classification, and combines the sentence coherence analysis results to obtain a final detection result for Alzheimer's disease data samples.

According to a third aspect of the present invention, there is provided a device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor can be used to execute any of the above-mentioned programs when the processor executes the program A method of, or, operating a system as described above.

According to a fourth aspect of the present invention, there is provided a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the program can be used to execute any one of the above-mentioned methods, or to run the above-mentioned method. system.

Due to the adoption of the above-mentioned technical solution, the present invention has the following beneficial effects compared with the prior art:

The data sample detection method and system for Alzheimer's disease provided by the present invention can distinguish whether a subject suffers from early Alzheimer's disease for the related data detection task of early Alzheimer's disease. The present invention only needs to use IoT devices widely deployed in smart homes to collect user data in multi-task scenarios. Since four tasks are defined, namely instruction task, description task, recall task, and writing task, including passive and active scenarios, and covering multi-source data of speech and text, there are more user features that can be used for follow-up The prediction and classification of Alzheimer's disease-related data is conducive to improving the accuracy and validity of the final data prediction results.

The data sample detection method and system for Alzheimer's disease provided by the present invention also adds data analysis on the difference between Alzheimer's disease patients and healthy people in feature extraction and final classification. In addition to the preprocessing of the collected data, a Markov chain-based semantic coherence analysis is added, and a hierarchical attention network suitable for this task is selected to extract the features of the text. The present invention selects a variety of classification models, and according to the data prediction results, can feedback whether the subject meets the data characteristics of Alzheimer's disease, and the feature collection and prediction of early Alzheimer's disease have any of the following or any number of contributions:

- Low dependence on equipment, only need to complete multi-task coordinated deployment based on IoT devices in smart home scenarios, that is, data collection can be performed, efficient deployment and simple data collection can be performed.

-Proposed a security mechanism to ensure the security of the entire process, and added a signature verification mechanism to prevent data from being tampered with during transmission.

- Based on multi-task inspection and effective feature extraction mechanism, it ensures the high precision of data collection. A hierarchical attention network is introduced for feature extraction of data, and Markov chain is used to fit semantic coherence to improve system performance.

Description of drawings

Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

FIG. 1 is a flowchart of a data sample detection method for Alzheimer's disease according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of components of a data sample detection system for Alzheimer's disease according to an embodiment of the present invention.

FIG. 3 is a working schematic diagram of a data sample detection system for Alzheimer's disease in a preferred embodiment of the present invention.

Detailed ways

The embodiments of the present invention are described in detail below: This embodiment is implemented on the premise of the technical solution of the present invention, and provides detailed implementation modes and specific operation processes. It should be pointed out that for those skilled in the art, without departing from the concept of the present invention, several modifications and improvements can be made, which all belong to the protection scope of the present invention.

FIG. 1 is a flowchart of a data sample detection method for Alzheimer's disease according to an embodiment of the present invention.

As shown in FIG. 1 , the data sample detection method for Alzheimer's disease provided by this embodiment may include the following steps:

S100, based on IoT smart home equipment, performs passive and active multi-task data collection;

S200, performing voice-to-text processing on the voice data in the collected data, and authorizing the text information to be transmitted at each smart home node to obtain the authorized text data;

S300, after verifying the authorized text data, extract the features of the text data, and perform sentence coherence analysis;

S400, the extracted features are aggregated and classified, and combined with the sentence coherence analysis result, a final detection result of the Alzheimer's disease data sample is obtained.

In a preferred embodiment of S100, passive data collection may include: daily command information data collection for smart home devices; active data collection may include: voice and text data collection for set tasks. In a specific application example, the set multi-tasks may include: description tasks, recall tasks, and writing tasks; further, passive and active multi-task data collection includes: instruction tasks, description tasks, and recall tasks. Voice data (the voice data will be processed as transcribed text in subsequent processing) and text data obtained under the writing task; among which, the voice data is collected by relying on the smart speakers and microphones of the smart home, and the text data is dependent on the intelligence of the smart home. Scanners and other equipment for auxiliary collection.

In a preferred embodiment of S200, voice-to-text processing is performed on the voice data in the collected data, and the text information to be transmitted is authorized at each smart home node, and the authorized text data can include the following steps :

S201, preprocessing the collected data;

S202, performing voice-to-text processing on the voice data in the preprocessed data;

S203, at each smart home node, perform signature authorization on the text information to be transmitted, and obtain authorized text data.

In a preferred embodiment of S201, the preprocessing may include the following:

Filter and denoise the collected data;

Pre-emphasis is performed on some of the collected data.

In a preferred embodiment of S203, in order to generate a signature key to ensure that the transmission data is not tampered with, perform signature authorization on the text information to be transmitted, and obtain the authorized text data, the following steps may be included:

S2031, for the i-th smart home node, use the GroupGen polynomial time algorithm to generate a cyclic group G, the order of the cyclic group G is p, the generator is g, and the cyclic group G satisfies the DDH assumption;

S2032, randomly select two elements a and b from [1, p], generate g ^a and g ^b and exchange them, and calculate and obtain g ^ab as a signature key;

S2033, use the signature algorithm Sig( ) to sign the text information to be transmitted, and use the hash function Hash to map the text information, and obtain the authorized text data Pre(Data _i ):

Pre(Data _i )=[Trans(Data _i ), Sig(g ^ab , Hash(Trans(Data _i )))]

Wherein, Pre(Data _i ) is the authorized text data of the ith smart home node, and Trans(Data _i ) is the text information transmitted by the ith smart home node. The text information Trans(Data _i ) transmitted by the i-th smart home node may be text information obtained by voice transcription, or may be text information directly obtained by a smart home device for completing a writing task.

In a preferred embodiment of S300, after the authorized text data is verified, the text data features are extracted, and sentence coherence analysis is performed, which may include the following steps:

S301, using a signature verification algorithm to verify whether the text data has been tampered with; if the data has been tampered with, directly discard the calculation result of the corresponding smart home edge node; if the data has not been tampered with, perform corresponding feature extraction and statement coherence analyze;

S302, using a hierarchical attention network model to extract features from words to sentences and then to discourses on the text data;

S303, perform sentence coherence analysis on the text data based on the Markov chain.

In a preferred embodiment of S302, the hierarchical attention network model includes: a word-level encoder layer, a word-level attention layer, a sentence-level encoder layer, and a sentence-level attention layer; wherein:

The word-level encoder layer is used to encode each word to obtain a latent vector;

The word-level attention layer is used to calculate the dot product of the selected latent vector to obtain the attention weight;

The sentence-level encoder layer is used to perform a weighted sum of the hidden vectors according to the sequence of the obtained hidden vectors to obtain the vector of the sentence;

The sentence-level attention layer is used to calculate the dot product of the vector of the selected sentence to obtain the attention weight of the sentence;

The vector of each sentence is weighted and averaged according to the attention weight of the sentence, and the final discourse vector representation v is obtained and output, and the feature extraction from words to sentences to the entire discourse is realized.

For the obtained discourse vector, the output can be obtained through the softmax layer classification of the fully connected layer.

In a preferred embodiment of S303, performing sentence coherence analysis on the text data based on the Markov chain may include the following steps:

S3031, based on the word frequency, extract a plurality of words with the highest frequency in the text data, and record the obtained words as W _set [w ₁ , w ₂ ......, w _n ];

S3032, for any text data, cut according to the sentence, retrieve the sentence j that contains the word in W _set , and extract the corresponding word, which is recorded as

S3033, use the existing text training data set of Alzheimer's patients to train the Markov chain to obtain the corresponding transition probability; compare the corresponding vocabulary of sentence j in the text data according to the transition probability, if the transition is correct, Increase the coherence score Score _i of sentence j by 1, otherwise do nothing;

S3034, normalize the coherence scores of all sentences in the text data, divide according to the set threshold Th, modify the Score _{i greater than the threshold to 1, and modify the Score i smaller} than the threshold to 0, to obtain the final Statement coherence analysis results.

In a preferred embodiment of S400, a lightweight neural network is used to classify the aggregated features to obtain a classification result; the classification result is combined with the sentence coherence analysis result to obtain the final Alzheimer's disease-specific The test results of the data sample. Among them, only when both results are data samples that meet normal conditions, the collected data samples are judged to be normal samples, otherwise they are diseased samples.

The data sample detection method for Alzheimer's disease provided by the above embodiments of the present invention meets the needs of multi-scenario tasks, and realizes highly reliable, safe, and highly effective data for Alzheimer's disease through the signature verification scheme. Sample home testing. Deploy in smart home scenarios, use massive IoT devices to collect, analyze, and process multi-task data, and use signature verification schemes during data transmission to ensure that data is not tampered with during transmission, effectively ensuring data transmission. safety. Finally, the machine learning model is used to classify and predict user data, which provides auxiliary data reference for the early detection of Alzheimer's disease.

A preferred embodiment of the present invention provides a data sample detection method for Alzheimer's disease. The method provided by this preferred embodiment can be deployed and used in a smart home, and includes the following steps:

In step 1, multi-task data collection is performed, and the voice can be collected by relying on the microphone inside the device of the smart home, including passive and active data collection. Passive collection refers to everyday commands to smart home devices, including wake words and conversations. Active is the remaining task defined by the present invention.

The second step is to perform data preprocessing. First, some denoising processing is performed on the speech, and some low-pass filters can be used for filtering. Thereafter, speech-to-text processing is performed on the collected data. At each smart home node, the information to be transmitted is signed to avoid being tampered with by attackers during the transmission process.

The third step is to perform feature extraction. This module uses a hierarchical attention network to extract features from the transcribed text, from words to sentences to discourse, to extract features of the entire discourse and prepare for final classification and detection. And use the Markov chain-based model to analyze the sentence coherence of the text, as a reference for the final classification.

The fourth step is to make an analysis decision. The medical data center collects the aggregated aggregated features, and uses lightweight neural networks and traditional simple classification models to classify and predict the results. Referring to the score of the previous coherence analysis, the final decision for Alzheimer's disease is obtained.

The specific scheme of the method provided by the preferred embodiment will be further described in detail below with reference to a specific application example.

表示指令任务的第i个用户的数据。第二项是主动式的对指定图片进行描述，数据记为

表示描述任务的第i个用户的数据。而另外两种数据分别针对于回忆任务和书写任务，记为

针对特定一个用户Useri，最终获得的数据是

In the method provided by this specific application example, commercial smart home IoT devices can be used for data collection. For N users, there are four tasks that require data collection. The first item is passive command collection, and the data is recorded as

Data representing the ith user of the command task. The second item is to actively describe the specified picture, and the data is recorded as

Represents data describing the ith user of the task. The other two kinds of data are respectively for the recall task and the writing task, denoted as

For a specific user Useri, the final data obtained is

In the preprocessing stage, it is time to filter and pre-emphasize some unclear audio to facilitate transcription into text. The method provided by this specific application example uses the open source package tool of speech-to-text to obtain the transcribed text Trans(Datai). In order to ensure the security and immutability of data transmission, a signature verification scheme is introduced to ensure data security. For the medical data center and the i-th smart home node, first use the GroupGen polynomial time algorithm to generate a cyclic group G, the order of the group is p, and the generator is g. The generated group needs to satisfy the DDH assumption (Decisional Diffie-Hellman Assumption). Randomly select two elements a, b from [1, p], generate g ^a and g ^b and exchange them, calculate and obtain g ^ab as the signature key, and use the signature algorithm Sig( ) to sign the collected data, And use the hash function Hash to map the data to obtain new data.

Pre(Data _i )=[Trans(Data _i ), Sig(g ^ab , Hash(Trans(Data _i )))]

Among them, Pre(Data _i ) is the authorized text data for the i-th user, which is the data after preprocessing, voice transcription text and signature verification, in preparation for feature extraction.

In the feature extraction process, the medical data center first needs to perform signature verification on the received data, and use a polynomial time verification algorithm to verify whether the data has been tampered with. If the data is tampered with, the calculation result of the corresponding smart home edge node will be directly discarded, and then the corresponding feature extraction will be performed.

The method provided by this specific application example performs feature extraction based on the difference in speech and text between healthy people and people with Alzheimer's disease. Specifically, the feature extraction model adopts a hierarchical attention network model, which consists of the following four parts: a word-level encoder, a word-level attention layer, a sentence-level encoder, and a sentence-level encoder. attention layer. After encoding each word, a latent vector is obtained, and the pre-selected vector is used for dot product calculation to obtain the attention weight. Then according to the sequence of hidden vectors, weighted sum is performed to obtain the vector of the sentence. The subsequent sentence-to-discourse design process is similar to the word-to-sentence process, and finally the corresponding feature vectors of the entire text are obtained.

Markov chains have since been introduced to explore the semantic coherence case, which is used to assist the final prediction results. Specifically, the 50 most frequently appearing words in Pre(Data _i ) are firstly extracted based on word frequency. The setting of 50 here is designed considering the size of the discourse and the amount of tasks. The obtained words are recorded as W _set [w ₁ , w ₂ ......, w ₅₀ ]. For the data of a certain user, cut it according to the sentence, retrieve the sentence containing the words in the W _set , and extract the corresponding vocabulary. If a sentence does not appear in the word in the W _set , it will not be recorded, otherwise it will be recorded. For example, for the jth sentence, if the sentence contains k words in the W _set , record as

Using the existing text data set of Alzheimer's patients to train the Markov chain, the corresponding transition probability can be obtained, that is, for a word that appears in the W _set , the next W _set face word should be what is. Through this rule, it is compared with the records in the currently collected samples. If the transfer is correct, the sample coherence score Score _i of the i-th user is increased by 1, and if it does not match, no action is taken. Finally, normalize the coherence scores of all samples of the user, and divide them according to the set threshold Th. The Score _i greater than the threshold is modified to 1, and the score i is less than the threshold is modified to 0 to assist the final classification result.

After feature extraction, a classification task is required to get the final Alzheimer's disease prediction result, using lightweight neural networks and traditional classification methods (such as random forests, k-neighbors, support vector machines, etc.) Classification, and finally select the best classification effect. In this specific application example, the collected data needs to be divided into a training set and a test set, which may be divided according to 7:3. Using a lightweight algorithm can achieve fast classification, and finally obtain a result of 0 or 1. 0 represents that the data sample of the subject is classified as a data sample that meets the conditions of Alzheimer's disease, and 1 represents the data sample of the subject. A data sample that is classified as eligible for a healthy population. The final result is obtained by referring to the analysis results of the previous Markov chain. If both results are judged to be 1, the final result output is 1, which means that the subject is a healthy population, and the rest are set as a sick population, which should be avoided as much as possible. Missed inspection.

FIG. 2 is a schematic diagram of components of a data sample detection system for Alzheimer's disease according to an embodiment of the present invention.

As shown in Figure 2, the data sample detection system for Alzheimer's disease provided by this embodiment may include the following modules:

A multi-source data acquisition module, which is based on IoT smart home devices for passive and active multi-task data acquisition;

A data preprocessing module, which performs voice-to-text processing on the voice data in the collected data, and authorizes the text information to be transmitted at each smart home node to obtain the authorized text data;

A feature extraction module, which verifies the authorized text data, extracts the text data features, and analyzes the sentence coherence;

A result detection module, which summarizes and aggregates the extracted features and classifies them, and combines the sentence coherence analysis results to obtain the final detection results for Alzheimer's disease data samples.

The working content of the data sample detection system for Alzheimer's disease provided by the above embodiments of the present invention will be further described below with reference to the accompanying drawings.

As shown in Figure 3, in a preferred embodiment of this embodiment:

Multi-source data acquisition module:

Data is collected for various tasks and submitted to the medical data center for processing and management.

In order to ensure the validity and diversity of the collected data, this module defines four kinds of tasks, which are passively collected instruction tasks and actively collected description tasks, recall tasks and writing tasks, and extensively collect multi-source data from users.

Data collection needs to rely on IoT devices that are widely deployed in smart home scenarios, which can easily collect data.

Data preprocessing module:

Contains speech-to-text processing of aggregated data and signing of information to be transmitted.

The voice-to-text open source package tool is used to process the collected voice data. At the same time, considering that the collected data may be clear and the number of channels is different, preprocessing is required.

A signature mechanism is added to sign the data at the data collection end, laying a solid foundation for subsequent verification of security.

Feature extraction module:

In the medical data center, it is necessary to verify the received data signature to determine whether it is from a legitimate user, and then use the hierarchical attention network to extract valid features, and perform semantic coherence analysis to obtain the final aggregated features.

Verify the collected data signatures to determine whether they come from legitimate users, and then perform feature extraction for valid content, select a network model suitable for this scenario, and extract appropriate features for the classification of the latter module.

The semantic coherence analysis based on Markov chain is added, and the coherence score is given to provide a reference for the final classification.

Result detection module:

In medical data centers, using neural networks and other simple classifier models to train and classify data can provide auxiliary data references for whether users are sick or not.

For the classification of the collected features, neural networks or other commonly used classification models, such as random forest, k-adjacent, support vector machine, etc., are used to synthesize the coherence score given by the previous module to give the final classification result. Provide help for data detection of early Alzheimer's disease.

It should be noted that the steps in the method provided by the present invention can be implemented by using corresponding modules, devices, units, etc. in the system, and those skilled in the art can refer to the technical solutions of the method to realize the composition of the system, that is, the implementation of the method. The example can be understood as a preferred example of the construction system, which will not be repeated here.

An embodiment of the present invention provides a device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, the processor can be used to execute any of the above methods of the present invention, Or, run the system of any of the above embodiments of the present invention.

An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the program can be used to execute any of the above-mentioned methods of the present invention, or to execute the above-mentioned implementation of the present invention. system of any of the examples.

Optionally, the memory is used to store the program; the memory may include volatile memory (English: volatile memory), such as random-access memory (English: random-access memory, abbreviation: RAM), such as static random-access memory ( English: static random-access memory, abbreviation: SRAM), double data rate synchronous dynamic random access memory (English: Double Data Rate Synchronous Dynamic Random Access Memory, abbreviation: DDR SDRAM), etc.; memory can also include non-volatile Memory (English: non-volatile memory), such as flash memory (English: flash memory). The memory is used to store computer programs (such as application programs, functional modules, etc. for implementing the above methods), computer instructions, etc., and the above computer programs, computer instructions, etc. can be stored in one or more memories in partitions. And the above-mentioned computer programs, computer instructions, data, etc. can be called by the processor.

The computer programs, computer instructions, etc. described above may be partitioned and stored in one or more memories. And the above-mentioned computer programs, computer instructions, data, etc. can be called by the processor.

The processor is configured to execute the computer program stored in the memory, so as to implement each step in the method involved in the above embodiments. For details, refer to the relevant descriptions in the foregoing method embodiments.

The processor and memory can be separate structures or integrated structures that are integrated together. When the processor and the memory are independent structures, the memory and the processor can be coupled and connected through a bus.

The data sample detection method and system for Alzheimer's disease provided by the above embodiments of the present invention consider the user's performance under various tasks, and take into account both active detection and passive detection, and use the smart home scene. It is convenient, reliable and low-cost to collect data using equipment widely deployed in China. The main principle is based on the differences in language, thinking, and text performance between the sick population and the healthy population. For example, the voice of the sick population has many pauses and blanks, and the logic is worse. These fine-grained differences can be implemented by the present invention. The characteristic analysis proposed in the example is obtained. At present, manual inspection based on questionnaires and pathological inspections based on medical equipment have been proposed, but there are still many defects and deficiencies, and it is difficult to meet the requirements for the detection and prevention of early Alzheimer's disease. At the same time, there is no security protection mechanism in the existing solutions. For users, voice text and other private data (such as age, gender), etc. may be leaked during the diagnosis process, and the data may be tampered with. A signature mechanism is introduced in the embodiment to ensure the reliability of the data source. The data sample detection method and system for Alzheimer's disease provided by the above embodiments of the present invention have the following features: low device dependence: data collection only needs to be performed by relying on IoT devices deployed in smart home scenarios; Mechanism: Ensure that users' health data is not leaked, and the data is provided safely; high precision and low cost: Compared with the existing Alzheimer's disease data detection solutions, the multi-task data collection is more multi-source, which is convenient It can be deployed at low cost while achieving better detection accuracy.

Matters not covered in the above embodiments of the present invention are known in the art.

The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above-mentioned specific embodiments, and those skilled in the art can make various variations or modifications within the scope of the claims, which do not affect the essential content of the present invention.

Claims (10)

1. A data sample detection method for Alzheimer's disease is characterized by comprising the following steps:

based on the intelligent home equipment of the Internet of things, passive and active multi-task data acquisition is carried out;

performing voice-to-text processing on voice data in the collected data, and authorizing text information to be transmitted at each intelligent home node to obtain authorized text data;

after the authorized text data is verified, extracting text data characteristics and carrying out statement continuity analysis;

and summarizing and aggregating the extracted features, classifying, and combining the sentence continuity analysis result to obtain a final detection result of the data sample aiming at the Alzheimer disease.

2. The method for data sample detection of alzheimer's disease as set forth in claim 1, wherein said passive data acquisition comprises: acquiring instruction information data of the intelligent household equipment in daily life; the active data acquisition comprises: voice and text data collection for a set task.

3. The method according to claim 1, wherein the step of performing speech-to-text processing on the speech data in the collected data, and authorizing the text information to be transmitted at each smart home node to obtain the authorized text data comprises:

preprocessing the acquired data;

performing voice-to-text processing on voice data in the preprocessed data;

and at each intelligent home node, performing signature authorization on the text information to be transmitted to obtain authorized text data.

4. The method for detecting data samples for Alzheimer's disease according to claim 3, further comprising any one or more of:

-said pre-treatment comprising:

carrying out filtering and denoising processing on the acquired data;

pre-emphasis processing is carried out on part of the collected data;

-said signing and authorizing the text information to be transmitted to obtain authorized text data, comprising:

for the ith intelligent home node, generating a cyclic group G by using a GroupGen polynomial time algorithm, wherein the order of the cyclic group G is p, the generator is G, and the cyclic group G meets the DDH assumption;

From [1, p ]]Randomly selecting two elements a and b to generate g^aAnd g^bAnd exchanging to obtain g^abAs a signing key;

signing the text information to be transmitted by using a signature algorithm Sig (-) and mapping the text information by using a Hash function Hash to obtain authorized text Data Pre (Data)_i)：

Pre(Data_i)＝[Trans(Data_i),Sig(g^ab,Hash(Trans(Data_i)))]

Wherein, Pre (Data)_i) Is the authorized text Data of the ith intelligent home node, Trans (Data)_i) And transmitting the text information for the ith intelligent home node.

5. The method for detecting data samples for alzheimer's disease as set forth in claim 1, wherein said verifying the authorized text data, extracting text data features, and performing sentence continuity analysis comprises:

verifying whether the text data is tampered by adopting a signature verification algorithm; if the data are tampered, directly discarding the current calculation result of the corresponding intelligent household edge node; if the data is not tampered, corresponding feature extraction and statement continuity analysis are carried out;

performing feature extraction from words to sentences to language fragments on the text data by adopting a layered attention network model;

and performing sentence consistency analysis on the text data based on the Markov chain.

6. The method for detecting data samples for Alzheimer's disease according to claim 5, further comprising any one or more of:

-the hierarchical attention network model comprises: a word level encoder layer, a word level attention layer, a sentence level encoder layer and a sentence level attention layer; wherein:

the word level encoder layer is used for encoding each word to obtain a hidden vector;

the word level attention layer is used for performing dot product calculation on the selected implicit vectors to obtain attention weight;

the sentence level encoder layer is used for carrying out weighted sum on the hidden vectors according to the obtained sequence of the hidden vectors to obtain the vectors of the sentences;

the sentence level attention layer is used for carrying out dot product calculation on the vector of the selected sentence to obtain the attention weight of the sentence;

carrying out weighted average on the vector of each sentence according to the attention weight of the sentence to obtain and output a final language piece vector, and realizing feature extraction from words to sentences to the whole language piece; -said sentence consistency analysis of said text data based on markov chains, comprising:

Extracting a plurality of words with highest occurrence frequency in the text data based on word frequency, and marking the obtained words as W_set[w₁,w₂……,w_n]；

For any text data, cutting according to sentences, and searching the text data in which W appears_setThe sentence j of the Chinese word is extracted and recorded as

Training a Markov chain by adopting a text training data set of an existing Alzheimer's disease patient to obtain a corresponding transition probability; comparing the corresponding vocabulary of the sentence j in the text data according to the transition probability, and if the transition is correct, scoring the continuity Score of the sentence j_iIncreasing 1, otherwise, not doing any operation;

normalizing the consistency scores of all sentences of the text data, dividing the sentences according to a set threshold Th, and scoring the sentences larger than the threshold_iModified to 1, Score less than threshold_iAnd modifying to be 0 to obtain a final sentence consistency analysis result.

7. The method for detecting the data samples of the alzheimer's disease as claimed in claim 1, wherein the aggregated features are classified by using a lightweight neural network to obtain a classification result; and combining the classification result with the sentence continuity analysis result to obtain a final detection result of the data sample aiming at the Alzheimer disease.

8. A data sample testing system for alzheimer's disease comprising:

the multi-source data acquisition module is used for carrying out passive and active multi-task data acquisition based on the Internet of things intelligent household equipment;

the data preprocessing module is used for carrying out voice-to-text processing on voice data in the collected data, authorizing text information to be transmitted at each intelligent home node and acquiring the authorized text data;

the feature extraction module is used for extracting the feature of the text data and carrying out statement continuity analysis after the authorized text data is verified;

and the result detection module is used for classifying the extracted features after summarizing and aggregating the features and obtaining a final detection result of the data sample aiming at the Alzheimer's disease by combining the sentence consistency analysis result.

9. An apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the program when executed by the processor is operable to perform the method of any one of claims 1 to 7 or to operate the system of claim 8.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is adapted to carry out the method of any one of claims 1 to 7 or to carry out the system of claim 8.

Images