Alzheimer's disease patient behavior monitoring method and system based on data analysis

Abstract

The application relates to the technical field of medical information processing, and discloses an Alzheimer's disease patient behavior monitoring method and system based on data analysis. The method comprises the following steps: collecting physiological data, behavior data and voice data of a monitoring target in a preset time interval; converting the voice data into text data, analyzing the voice data and the text data to determine a first degeneration level of the monitoring target; determining the actual occurrence frequency of each behavior type based on the behavior data; determining a standard frequency range of the behavior type based on the first degeneration level, comparing the standard frequency range with the actual occurrence frequency to determine a second degeneration level of the monitoring target; determining the diurnal activity ratio of the monitoring target based on the physiological data, and determining a third degeneration level based on the diurnal activity ratio; and comprehensively determining an overall degeneration level based on the first degeneration level, the second degeneration level and the third degeneration level. The application can collect multi-dimensional behavior data of a patient, thereby realizing accurate monitoring of the development of the patient's condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data processing, in particular to an Alzheimer's disease patient behavior monitoring method and system based on data analysis. BACKGROUND

[0002] The cognitive assessment method of Alzheimer's disease mainly depends on neuropsychological scales and imaging examinations, which has the limitations of strong invasiveness, high cost, and insufficient timeliness. In recent years, with the breakthrough progress of intelligent perception technology and multi-modal data analysis, significant achievements have been made in the field of non-invasive assessment.

[0003] For example, the Chinese patent document with publication number CN119324063A discloses an Alzheimer's disease early warning analysis method and system based on multi-Chinese syllable, which uses simple Chinese syllables for patient data collection and combines multi-syllable fusion analysis method to obtain an early warning scheme for Alzheimer's disease, which is beneficial to improve the accuracy of early warning analysis. For example, the Chinese patent document with publication number CN117831761A discloses a multi-dimensional Alzheimer's disease evaluation system and method, which integrates AI interactive question and answer health evaluation, integrated VR glasses game health evaluation, eye muscle electrical test, olfactory electrophysiological examination, and ultrasonic blood flow test, etc. Multi-dimensional evaluation methods comprehensively evaluate the cognitive ability, coordination and creativity of users, obtain multi-aspect data, and finally evaluate cognitive impairment by matching and analyzing these data results.

[0004] The first technology above uses single-dimensional speech analysis, which is difficult to capture the systematic degradation of patient's daily life function and cannot construct a dynamic degradation trajectory through continuous time series data. The second technology above integrates multiple detection methods, but lacks physiological rhythm quantification and grading verification mechanism between specific behavior types and speech degradation, resulting in insufficient behavior anomaly sensitivity. SUMMARY

[0005] To solve the problems raised in the background art, the present application provides an Alzheimer's disease patient behavior monitoring method and system based on data analysis.

[0006] To achieve the above-mentioned invention purpose, the present application provides an Alzheimer's disease patient behavior monitoring method based on data analysis, comprising:

[0007] Collecting speech data, behavior data and physiological data of the monitoring target within a preset time interval;

[0008] Converting the speech data into text data, analyzing the speech data and the text data to determine the first degradation level of the monitoring target;

[0009] setting a plurality of behavior types, determining an actual occurrence frequency of each of the behavior types based on the behavior data;

[0010] determining a standard frequency range of each of the behavior types based on the first degradation level, determining a second degradation level of the monitoring target by comparing the standard frequency range with the actual occurrence frequency;

[0011] determining a diurnal activity ratio of the monitoring target based on the physiological data, determining a third degradation level based on the diurnal activity ratio;

[0012] determining a total degradation level of the monitoring target by integrating the first degradation level, the second degradation level and the third degradation level.

[0013] Further, the first degradation level includes the following steps:

[0014] setting a standard pause range, splitting the speech data into a plurality of independent speeches based on the standard pause range, and generating corresponding independent text data based on the independent speeches;

[0015] setting a plurality of emotional words, obtaining standard text data by deleting the emotional words in the independent text data, extracting text features in the standard text data, clustering the standard text data based on the text features, obtaining a plurality of clusters, and counting a first number of the standard text data included in each of the clusters;

[0016] initially sorting the clusters from large to small based on the first number, obtaining inter-class similarity between each of the clusters, adjusting the initial sorting of the clusters according to the size of the inter-class similarity, obtaining a final sorting, determining a corresponding language usage rate based on the position of the cluster in the final sorting, screening the cluster with the language usage rate greater than a first threshold value, and taking the standard text data included therein as an analysis target, calculating a logical score of the analysis target, sorting the logical score based on the generation time of the independent speech to determine the change of the logical score, and determining a first thinking score based on the change.

[0017] determining language coherence based on the text data, correcting the first thinking score to obtain a second thinking score based on the language coherence, and determining the first degradation level based on the second thinking score.

[0018] Further, obtaining the final sorting includes the following steps:

[0019] The cluster group with the first sequence in the initial sorting is positioned as a first calibration cluster, the cluster groups with the inter-class similarity greater than a second threshold value are screened from the first calibration cluster, and the cluster groups are re-ordered from large to small according to the inter-class similarity after the first calibration cluster to obtain a first sorting part, and the cluster groups not included in the first sorting part are secondary clusters, the first largest secondary cluster is taken as a second calibration cluster, the secondary clusters with the inter-class similarity greater than the second threshold value are screened from the second calibration cluster, and the secondary clusters are re-ordered from large to small after the second calibration cluster to obtain a second sorting part, and the step is repeated until the division of all the cluster groups is completed, and each calibration cluster and sorting part is connected in turn according to the division order to obtain a final sorting.

[0020] Further, the calculation of the logical score includes the following steps:

[0021] The standard text data adjacent to the analysis target is obtained, the theme vector of the analysis target and the standard text data is extracted, the theme similarity is calculated based on the theme vector, the theme consistency score is determined based on the size of the theme similarity, the logical coherence probability between the analysis target and the adjacent standard text data is determined based on the deep learning model, the logical coherence score is determined based on the logical coherence probability, and the logical score is obtained by adding the theme consistency score and the logical coherence score.

[0022] Further, the determination of the first thinking score includes the following steps:

[0023] A numerical sequence is generated based on the sorted logical score, linear regression analysis is performed on the numerical sequence to determine the slope, statistical analysis is performed on the numerical sequence to obtain the standard deviation and the coefficient of variation, a score rule is set, the score rule includes the numerical combination of the slope, the standard deviation and the coefficient of variation and the corresponding score, and the actual obtained slope, standard deviation and coefficient of variation are matched with the score rule to determine the first thinking score.

[0024] Further, the correction of the first thinking score based on the language coherence includes the following steps:

[0025] The language coherence includes language pause frequency and thinking pause frequency, component analysis is performed on each of the independent text data, and the defective text therein is positioned, the defective text is the independent text data lacking sentence components, the defective text and the adjacent independent text data are respectively merged into test text, component analysis is performed on the test text, if the test text is no longer determined as the defective text, the test text is defined as a first text, and the number of occurrences of the first text is counted to obtain the language pause frequency;

[0026] locating the independent text data where the discourse word appears, if the number of the discourse word appearing therein exceeds a third threshold value, defining the independent text data as a second text, and counting the number of the second text to obtain the thinking pause frequency;

[0027] weighting and summing the language pause frequency and the thinking pause frequency to obtain a transition value, determining a correction coefficient based on the transition value, and correcting the first thinking score based on the correction coefficient.

[0028] Further, determining the second degradation level comprises the following steps:

[0029] setting the standard frequency range of each behavior type under different first degradation levels, if the actual appearance frequency of the behavior type is greater than the standard frequency range, determining the corresponding behavior type as a repetitive behavior, if the number of the repetitive behavior type exceeds a fourth threshold value, raising the first degradation level to obtain the second degradation level, otherwise, using the original first degradation level as the second degradation level.

[0030] Further, locating the cluster center of each cluster group, and taking the distance between the cluster centers as the inter-class similarity.

[0031] Further, taking the highest level among the first degradation level, the second degradation level and the third degradation level as the overall degradation level.

[0032] The application also provides an Alzheimer's disease patient behavior monitoring system based on data analysis, which is used to implement the Alzheimer's disease patient behavior monitoring method described above, and the system comprises:

[0033] a collection module, which collects voice data, behavior data and physiological data of a monitoring target within a preset time interval;

[0034] a first evaluation module, which converts the voice data into text data, and analyzes the voice data and the text data to determine a first degradation level of the monitoring target;

[0035] a second evaluation module, which sets multiple behavior types, determines the actual appearance frequency of each behavior type based on the behavior data, determines the standard frequency range of each behavior type based on the first degradation level, and compares the standard frequency range and the actual appearance frequency to determine a second degradation level of the monitoring target;

[0036] a third evaluation module, which determines the diurnal activity ratio of the monitoring target based on the physiological data, and determines a third degradation level based on the diurnal activity ratio.

[0037] a summary module, which determines an overall degeneration level of the monitoring target by integrating the first degeneration level, the second degeneration level and the third degeneration level.

[0038] Advantages:

[0039] The present application realizes comprehensive monitoring of the daily activities of a patient by collecting physiological data, behavioral data and voice data of the patient. Through analysis of the voice data and the text data, the language logic degeneration condition of the patient can be determined, and then by comparing the actual occurrence frequency of the behavioral data with the standard frequency range, the second degeneration level can be determined. In addition, the third degeneration level evaluation based on the diurnal activity ratio further enriches the dimension of the disease condition evaluation. Finally, by integrating the first, second and third degeneration levels, the overall degeneration level of the patient can be determined. This comprehensive evaluation method can more comprehensively reflect the development condition of the patient's disease. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 a schematic diagram of the steps of the Alzheimer's disease patient behavior monitoring method based on data analysis of the present application;

[0041] Figure 2 a schematic diagram of the principle of the initial sorting of the present application;

[0042] Figure 3 a schematic diagram of the principle of adjusting the initial sorting of the present application;

[0043] Figure 4 a schematic diagram of the structure of the Alzheimer's disease patient behavior monitoring system based on data analysis of the present application; DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0045] It can be understood that the terms "first", "second" and the like used in the present application can be used herein to describe various elements, but unless specifically stated, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of the present application, the first xx script can be referred to as the second xx script, and similarly, the second xx script can be referred to as the first xx script.

[0046] As shown in Figure 1 a kind of Alzheimer's disease patient behavior monitoring method based on data analysis, comprising:

[0047] S1: Collect voice data, behavior data and physiological data of the monitoring target in a preset time interval.

[0048] The preset time interval is set to two months in this embodiment, and in other embodiments, the preset time interval can also be one month or three months. The physiological data includes heart rate, wake-up time and sleep time, etc., which are obtained through a smart bracelet or a smart watch. The behavior data includes various action behaviors, such as the purchase behavior of various daily necessities (obtained through the purchase data under the predetermined account), or the daily cleaning and other living habit behaviors (obtained through home monitoring video combined with a deep recognition model). The voice data includes various daily conversation voices of the monitoring target, which can be obtained through a smart speaker or a recording device carried by the monitoring target. In particular, the initially obtained voice data can contain sounds other than the monitoring target, so after recording is completed, the voice belonging to the monitoring target is extracted as voice data through tone and timbre, which is convenient for subsequent analysis.

[0049] S2: Convert the voice data into text data, and analyze the voice data and the text data to determine a first degeneration level of the monitoring target.

[0050] The obtained voice data is converted into text data through voice recognition ASR, and then the voice data and the text data are analyzed to determine the language logic change of the monitoring target in two months. If the language logic degenerates, it indicates that the disease may have worsened. The first degeneration level is quantified in this embodiment, and the specific determination method of the first degeneration level is described in detail later.

[0051] S3: Set multiple behavior types, and determine the actual occurrence frequency of each behavior type based on the behavior data.

[0052] S4: Determine the standard frequency range of each behavior type based on the first degeneration level, and compare the standard frequency range with the actual occurrence frequency to determine a second degeneration level of the monitoring target.

[0053] The behavior types of this embodiment include various daily necessities purchase behaviors, cleaning behaviors and daily habit behaviors, such as clothes folding, drawer pulling, etc. The average value of the number of times of these behaviors in the preset time interval is taken as the actual occurrence frequency. For example, if there are 600 times of drawer pulling behaviors in 60 days of 2 months, the actual occurrence frequency is 600 / 60 = 10 times / day. The first degeneration level of this embodiment includes level 1, level 2 and level 3. The greater the level, the more serious the degeneration degree.

[0054] Establish a standard frequency range for each behavior type under each first degradation level, for example, under level 1, the number of times of buying yogurt is not more than 3 times / month, the number of times of cleaning every day is not more than 2 times / day, the number of times of pulling the drawer is not more than 12 times / day, etc., and other values can be freely set by those skilled in the art. After determining the first degradation level, it can be determined whether the actual frequency matches the standard frequency range, and then the second degradation level is determined according to the matching result.

[0055] S5: determining the circadian activity ratio of the monitoring target based on the physiological data, and determining the third degradation level based on the circadian activity ratio;

[0056] S6: determining the overall degradation level of the monitoring target by comprehensively determining the first degradation level, the second degradation level and the third degradation level.

[0057] Among them, the level with the highest level among the first degradation level, the second degradation level and the third degradation level is taken as the overall degradation level.

[0058] In the process of obtaining the third degradation level, first, the ratio of the waking time to the sleep time every day in the past two months is calculated, and the ratio of the two is taken as the circadian activity ratio of each day. If the circadian activity ratio of a day is above 1.5, the day is classified as mild, if it is between 1-1.5, the day is classified as moderate, and if it is less than 1, the day is classified as severe. The number of days of mild, moderate and severe in the past two months is counted, and the third degradation level includes level 1, level 2 and level 3. If the number of days of mild is the most, the third degradation level is determined as level 1, if the number of days of moderate is the most, the third degradation level is determined as level 2, and if the number of days of severe is the most, the third degradation level is determined as level 3.

[0059] In the process of determining the final degradation level, if the first degradation level is level 3, the second degradation level is level 1, and the third degradation level is level 2, the overall degradation level is determined as level 3.

[0060] The present application realizes comprehensive monitoring of the daily activities of the patient by collecting the physiological data, behavior data and voice data of the patient. Through the analysis of voice data and text data, the language logic degradation of the patient can be determined, and then the second degradation level can be determined by comparing the actual frequency of the behavior data with the standard frequency range. In addition, the third degradation level evaluation based on the circadian activity ratio further enriches the dimension of the disease assessment. Finally, by comprehensively determining the first, second and third degradation levels, the overall degradation level of the patient can be determined. This comprehensive evaluation method can more comprehensively reflect the development of the patient's condition.

[0061] It is particularly noted that the multi-dimensional behavior data of the patient can be summarized by the present application, so as to realize accurate monitoring of the development of the patient's condition.

[0062] Determining the first degradation level includes the following steps:

[0063] Set a standard pause range, split the speech data into multiple independent speech segments based on the standard pause range, and generate corresponding independent text data based on the independent speech segments.

[0064] In this embodiment, the standard pause range is set to greater than 800ms. For continuously occurring speech data, pause points are first detected and their durations are recorded. If the pause duration of one pause point falls within the standard pause range, the speech data is split into two independent speech segments using that pause point as the dividing point. If there are two such pause points, the speech data is split into three independent speech segments, and so on. Then, speech recognition technology is used to convert each independent speech segment into corresponding text data. Splitting the original speech data facilitates subsequent logical analysis.

[0065] By setting up various modal particles and removing them from the independent text data, standard text data is obtained. Text features are extracted from the standard text data, and the standard text data is clustered based on the text features to obtain multiple clusters. The number of standard text data included in each cluster is counted.

[0066] Interjections such as "that" and "um" are excluded because they interfere with the analysis process in this step. Text features include TF-IDF features and BERT vectors. The acquired standard text data is then clustered based on these features using algorithms such as k-means, DBSCAN, or hierarchical clustering. After clustering, standard text data with similar semantics are grouped into one cluster. The number of standard text data points included in each cluster is counted; a higher number indicates that the monitored target uses this semantic expression more frequently in daily life.

[0067] The clusters are initially sorted from largest to smallest based on the first quantity, and the inter-class similarity between each cluster is obtained. The initial sorting of the clusters is adjusted according to the magnitude of the inter-class similarity to obtain the final sorting. The language usage rate corresponding to the cluster is determined based on the position of the cluster in the final sorting. Clusters with language usage rates greater than a first threshold are selected, and the standard text data included in them are used as the analysis target. The logicality score of the analysis target is calculated. The logicality score is sorted based on the generation time of independent speech to determine the change of the logicality score. The first thinking score is determined based on the change.

[0068] In this embodiment, the cluster center of each cluster is located, and the distance between the cluster centers is used as the inter-cluster similarity.

[0069] like Figure 2 As shown, Figure 2The median value represents the magnitude of the first quantity. In the initial ranking, the cluster with the larger the first quantity, such as cluster A, ranks higher. If the two clusters have the same quantity, they are randomly ranked according to their order. After obtaining the initial ranking, the inter-class similarity between the two clusters is calculated, and the initial ranking of the clusters is readjusted based on the inter-class similarity. The specific adjustment method will be described later. In some embodiments, the inter-class similarity can also be the cosine similarity of the cluster centers, or the Mahalanobis distance between the cluster centers of one cluster and the other cluster.

[0070] In the final ranking, the higher the cluster ranks, the more frequently the monitored target uses expressions with this semantic meaning in daily life. To quantify this, this invention uses a first formula to calculate the language usage rate P of the i-th cluster. i The first formula is: P i =(Nn i ) / N, where n i Let represent the order value of the i-th cluster in the final sort, and N be the total number of clusters included in the final sort. For example, if there are 100 clusters in the final sort, and a certain cluster is ranked 10, then its language usage rate is (100-10) / 100 = 0.9.

[0071] The first threshold is set to 0.5, filtering standard text data within clusters with a language usage rate greater than 0.5, and defining these as the analysis target. Based on previous records, the analysis target is a single sentence, which is then subjected to logical analysis to determine its logicality score. A higher logicality score indicates more logically sound language, while a lower score indicates poorer logic. For example, "It's going to rain tomorrow, I'm going out to exercise" would receive a poor logicality score. When recording speech data, the generation time of the speech data is also recorded. When splitting the speech data, each independent speech item is assigned a generation time based on its generation time, which determines the order in which the independent text data was generated. By sorting the independent text data according to their chronological order, the logicality scores can be sorted, ultimately revealing the changes in the logicality scores.

[0072] If the logic score remains stable, it indicates that the patient's language ability has not deteriorated. If the logic score gradually decreases, it indicates that the patient's language ability is deteriorating. Finally, in order to quantify the changes, the first thought score is determined based on the specific changes. The higher the first thought score, the stronger the patient's verbal logic.

[0073] Language coherence is determined based on textual data. The first thinking score is then revised based on language coherence to obtain the second thinking score. The first degradation level is determined based on the second thinking score.

[0074] The language coherence includes a language pause frequency and a thinking pause frequency, and the higher the language pause frequency and the thinking pause frequency, the worse the language coherence of the patient. The first thinking score calculated before is corrected according to the language coherence, so as to obtain a second thinking score, and finally the first degradation level is determined according to the second thinking score.

[0075] The final ranking obtained in this embodiment includes the following steps:

[0076] The cluster group with the first order in the initial ranking is positioned as a first calibration cluster, the cluster groups with the inter-class similarity greater than the second threshold value are screened from the first calibration cluster, and the cluster groups are reordered from large to small after the first calibration cluster according to the inter-class similarity, to obtain a first ranking part, and the cluster groups not included in the first ranking part are secondary clusters, the first largest secondary cluster is taken as a second calibration cluster, the secondary clusters with the inter-class similarity greater than the second threshold value are screened from the second calibration cluster, and the secondary clusters are reordered from large to small after the second calibration cluster according to the inter-class similarity, to obtain a second ranking part, and the step is repeated until the division of all cluster groups is completed, and the final ranking is obtained by connecting the calibration clusters and the ranking parts in turn according to the division order.

[0077] In this embodiment, the second threshold value is set to 0.7, the cluster groups with the inter-class similarity greater than 0.7 from the first calibration cluster are screened, and the cluster groups are sorted from large to small according to the inter-class similarity with the first calibration cluster, as shown in Figure 3 The inter-class similarity of cluster group B and cluster group C with the first calibration cluster (cluster group A) is 0.9 and 0.8 respectively, so cluster group C is sorted after the first calibration cluster, and cluster group B is sorted after cluster group C. Similarly, the other cluster groups are sorted, and the first ranking part is obtained. For the remaining cluster groups, for the convenience of distinguishing description, they are named as secondary clusters, the first largest secondary cluster is named as a second calibration cluster, the inter-class similarity of each secondary cluster with the second calibration cluster is obtained, and the secondary clusters are screened based on the second threshold value, the secondary clusters screened are sorted based on the previous sorting principle, to obtain a second ranking part, and the process is repeated. Finally, the first ranking part is sorted after the first calibration cluster, the second calibration cluster is sorted after the first ranking part, and the second ranking part is sorted after the second calibration cluster, and so on, so as to obtain the final ranking.

[0078] The logic score calculated in this embodiment includes the following steps:

[0079] The standard text data adjacent to the analysis target is obtained and analyzed, the theme vector of the analysis target and the standard text data is extracted, the theme similarity is calculated based on the theme vector, the theme consistency score is determined based on the size of the theme similarity, the logical coherence probability between the analysis target and the adjacent standard text data is determined based on the deep learning model, the logical coherence score is determined based on the logical coherence probability, and the theme consistency score and the logical coherence score are added to obtain the logic score.

[0080] The standard text data adjacent to the analysis target includes both the standard text data after the analysis target and the standard text data before the analysis target. The topic vectors of the analysis target and the standard text data are extracted based on the BERT model, the cosine similarity between the topic vectors is calculated, and the one with the maximum cosine similarity is taken as the topic consistency score. For example, the cosine similarity between the analysis target and the standard text data after it is 0.8, and the cosine similarity between the analysis target and the standard text data before it is 0.6, then 0.8 is taken as the topic consistency score of the analysis target.

[0081] The deep learning model can be a GPT-4 model or a DeepseekR1-70B model. Since the essence of such a model is to calculate the probability of various statements appearing after the current statement based on the context, when calculating the logical coherence probability, the analysis target together with the adjacent standard text data can be input into the deep learning model to obtain the logical coherence probability. The model outputs the average value of the conditional probability of each token in the previously input text content. "Tomorrow it will rain, and I will go out for exercise" will obtain a lower logical coherence probability. Finally, the topic consistency score and the logical coherence score are added to obtain the logicality score.

[0082] The embodiment determines the first thinking score including the following steps:

[0083] A numerical sequence is generated based on the sorted logicality score, linear regression analysis is performed on the numerical sequence to determine the slope, statistical analysis is performed on the numerical sequence to obtain the standard deviation and coefficient of variation, a scoring rule is set, the scoring rule includes the numerical combination of the slope, the standard deviation, and the coefficient of variation and the corresponding score, and the actual obtained slope, standard deviation, and coefficient of variation are matched with the scoring rule to determine the first thinking score.

[0084] Generally, the cognitive degradation process of a patient is gradually declining, so a linear regression model is used to fit the numerical sequence to obtain a better fitting result. Specifically, the least squares method is used for fitting. After fitting, the slope of the fitting function is obtained, and then the standard deviation and the coefficient of variation of the numerical sequence are calculated. The coefficient of variation is the percentage of the standard deviation normalized to the average value, which serves to eliminate the dimension effect and facilitate the determination of volatility. Then the first thinking score is determined by establishing a scoring rule. The scoring rule is shown in Table 1.

[0085] Table 1

[0086]

[0087] The embodiment corrects the first thinking score based on language coherence including the following steps:

[0088] The language coherence includes a language pause frequency and a thinking pause frequency, component analysis is performed on each piece of independent text data, and defective text in which is located, the defective text is independent text data lacking in sentence components, the defective text is combined with adjacent independent text data into test text respectively, component analysis is performed on the test text, if the test text is no longer determined as defective text, the test text is defined as first text, and the number of first texts is counted to obtain the language pause frequency.

[0089] The language pause frequency refers to abnormal pauses occurring in the process of conversation, such as a sentence is not finished and a long pause is entered, in the process of dividing speech data, such a case can cause errors in division, so that a complete sentence is incorrectly divided into two sentences. The thinking pause frequency refers to word retrieval pauses occurring in the process of conversation, for example, “I want to buy that that, uh… that”, that is, a longer time is needed to think of the word that the person wants to express.

[0090] In order to identify the two pause conditions, the present application first performs component analysis on each piece of independent text data, obtains subject components, predicate components and object components in each piece of independent text data through component analysis, and if there is a component missing, for example, lacking an object, the piece of independent text data is defined as defective text. Then, the defective text is combined with the independent text data before and after it into test text, if the test text no longer appears component missing, it indicates that the appearance of the defective text is caused by abnormal pause, that is, an abnormal pause occurs, and the test text is defined as first text, which is convenient for subsequent statistical analysis, that is, the number of first texts is the language pause frequency.

[0091] The independent text data in which the interjection appears is located, if the number of interjections appearing in the independent text data exceeds a third threshold, the independent text data is defined as second text, and the number of second texts is counted to obtain the thinking pause frequency.

[0092] The language pause frequency and the thinking pause frequency are weighted and summed to obtain a transition value, a correction coefficient is determined based on the transition value, and the first thinking score is corrected based on the correction coefficient.

[0093] Based on the previous description, if the number of interjections in a piece of independent text data is greater than 3, it is defined as second text, and the number of second texts is defined as the thinking pause frequency. For detection of interjections, an interjection library can be pre-set to determine.

[0094] Before the weighted sum is performed, the weights of the language pause frequency and the thinking pause frequency are first set, for example, 0.7 and 0.3 respectively, if the language pause frequency and the thinking pause frequency are 30 and 20 respectively, then the weighted sum is 30*0.7+20*0.3=27. The corresponding relationship between the transition value and the correction coefficient is set, if the transition value is between 0-30, then the correction coefficient is 0.9, if it is between 30-60, then the correction coefficient is 0.7, other correction coefficients corresponding to the transition value can be freely set according to actual conditions. When the transition value is 27, the correction coefficient is set to 0.9, and the correction coefficient is multiplied by the first thinking score to correct it.

[0095] The embodiment determines the second degradation level, which includes the following steps:

[0096] The standard frequency range under different first degradation levels is set for each behavior type, if the actual occurrence frequency of the behavior type is greater than the standard frequency range, the corresponding behavior type is determined as a repeated behavior, if the number of repeated behavior types exceeds the fourth threshold, the first degradation level is increased to obtain the second degradation level, otherwise the original first degradation level is taken as the second degradation level.

[0097] As before, the behavior type includes various daily commodity purchase behaviors, cleaning behaviors and daily habit behaviors, the second degradation level of the embodiment includes 1 level, 2 level and 3 level, for example, for cleaning behavior A, the standard frequency range under the first degradation level of 1 level is 2-3 times / day, the standard frequency range under the first degradation level of 2 level is 4-5 times / day, and the standard frequency range under the first degradation level of 3 level is 6-7 times / day. If the first degradation level is 1, but the frequency of cleaning behavior A is 4 times / day, then cleaning behavior A is determined as a repeated behavior. If the number of repeated behaviors is too large, the first degradation level is increased by 1 as the second degradation level, if the first degradation level at this time is 3 level, then the second degradation level after promotion is 3 level.

[0098] As shown in Figure 4 The application also provides an Alzheimer's disease patient behavior monitoring system based on data analysis, which is used to realize the Alzheimer's disease patient behavior monitoring method described above, and the system includes:

[0099] The acquisition module acquires the voice data, behavior data and physiological data of the monitoring target in the preset time interval;

[0100] The first evaluation module converts the voice data into text data, analyzes the voice data and text data to determine the first degradation level of the monitoring target.

[0101] The second evaluation module sets multiple behavior types, determines an actual occurrence frequency of each behavior type based on the behavior data, determines a standard frequency range of each behavior type based on the first degradation level, and determines the second degradation level of the monitoring target by comparing the standard frequency range with the actual occurrence frequency.

[0102] The third evaluation module determines a day-night activity ratio of the monitoring target based on the physiological data, and determines the third degradation level based on the day-night activity ratio.

[0103] The summary module determines a general degradation level of the monitoring target by synthesizing the first degradation level, the second degradation level, and the third degradation level.

[0104] It should be understood that any combination of the technical features of the above-described embodiments can be made, and for the sake of brevity, all possible combinations of the technical features in the above-described embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered within the scope of the present disclosure.

[0105] The above-described is only a preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A data analysis-based method for monitoring the behavior of Alzheimer's disease patients, characterized in that, Collect and monitor voice, behavioral, and physiological data of the target within a preset time interval; The voice data is converted into text data, and the voice data and text data are analyzed to determine the first degradation level of the monitored target. Multiple behavior types are set, and the actual frequency of occurrence of each behavior type is determined based on the behavior data; Based on the first degradation level, a standard frequency range for each of the aforementioned behavior types is determined, and a second degradation level for the monitored target is determined by comparing the standard frequency range with the actual occurrence frequency. The diurnal activity ratio of the monitored target is determined based on the physiological data, and the third degradation level is determined based on the diurnal activity ratio. The overall degradation level of the monitored target is determined by combining the first degradation level, the second degradation level, and the third degradation level. Determining the first degradation level includes the following steps: Set a standard pause range, split the speech data into multiple independent speech segments based on the standard pause range, and generate corresponding independent text data based on the independent speech segments; Multiple modal particles are set, and the modal particles in the independent text data are deleted to obtain standard text data. Text features are extracted from the standard text data, and the standard text data is clustered based on the text features to obtain multiple clusters. The first number of standard text data included in each cluster is counted. The clusters are initially sorted from largest to smallest based on the first quantity, and the inter-class similarity between each cluster is obtained. The initial sorting of the clusters is adjusted according to the magnitude of the inter-class similarity to obtain the final sorting. The language usage rate corresponding to the position of the cluster in the final sorting is determined. Clusters with language usage rates greater than a first threshold are selected, and the standard text data included therein are used as analysis targets. The logical reasoning score of the analysis targets is calculated. The logical reasoning score is sorted based on the generation time of the independent speech to determine the change of the logical reasoning score. A first thinking score is determined based on the change. Based on the text data, language coherence is determined; based on the language coherence, the first thinking score is corrected to obtain a second thinking score; and based on the second thinking score, the first degradation level is determined.

2. The method according to claim 1, characterized in that, Obtaining the final sort involves the following steps: The clusters in the initial sorting are identified as the first labeled clusters. Clusters with inter-class similarity greater than a second threshold to the first labeled clusters are selected and reordered from largest to smallest according to inter-class similarity to the first labeled clusters to obtain the first sorted portion. Clusters not included in the first sorted portion are identified as secondary clusters. The secondary cluster with the largest number is identified as the second labeled cluster. Secondary clusters with inter-class similarity greater than the second threshold to the second labeled clusters are selected and reordered from largest to smallest according to inter-class similarity to the second labeled clusters to obtain the second sorted portion. This step is repeated until all clusters are divided. Each labeled cluster and sorted portion is connected sequentially according to the division order to obtain the final sort.

3. The method according to claim 1, characterized in that, Calculating the logicality score includes the following steps: The standard text data adjacent to the analysis target is obtained, topic vectors of the analysis target and the standard text data are extracted, topic similarity is calculated based on the topic vectors, topic consistency score is determined based on the magnitude of the topic similarity, logical coherence probability between the analysis target and the adjacent standard text data is determined based on a deep learning model, logical coherence score is determined based on the logical coherence probability, and the topic consistency score and the logical coherence score are added together to obtain the logicality score.

4. The method according to claim 1, characterized in that, Determining the first thinking score includes the following steps: A numerical sequence is generated based on the sorted logical scores. Linear regression analysis is performed on the numerical sequence to determine the slope. Statistical analysis is then performed on the numerical sequence to obtain the standard deviation and coefficient of variation. Scoring rules are set, including numerical combinations of the slope, standard deviation, and coefficient of variation, and corresponding scores. The actual obtained slope, standard deviation, and coefficient of variation are matched with the scoring rules to determine the first thinking score.

5. The method according to claim 1, characterized in that, The steps for revising the first thinking score based on the aforementioned language coherence are as follows: The language coherence includes the frequency of language pauses and the frequency of thought pauses. Component analysis is performed on each of the independent text data, and defective texts are located. Defective texts are independent text data that lack sentence components. The defective texts are merged with adjacent independent text data to form test texts. Component analysis is performed on the test texts. If the test texts are no longer identified as defective texts, they are defined as first texts. The number of occurrences of the first texts is counted to obtain the language pause frequency. The independent text data containing the interjection is located. If the number of interjections in the independent text data exceeds a third threshold, the independent text data is defined as the second text. The number of occurrences of the second text is counted to obtain the frequency of the thought pause. The frequency of language pauses and the frequency of thinking pauses are weighted and summed to obtain a transition value. A correction coefficient is determined based on the transition value, and the first thinking score is corrected based on the correction coefficient.

6. The method according to claim 1, characterized in that, Determining the second degradation level includes the following steps: For each behavior type, a standard frequency range is set under different first degradation levels. If the actual occurrence frequency of the behavior type is greater than the standard frequency range, the corresponding behavior type is determined to be a recurring behavior. If the number of recurring behaviors of the behavior type exceeds a fourth threshold, the first degradation level is increased to obtain the second degradation level. Otherwise, the original first degradation level is used as the second degradation level.

7. The method according to claim 1, characterized in that, Locate the cluster center of each cluster and use the distance between the cluster centers as the inter-cluster similarity.

8. The method according to claim 1, characterized in that, The highest level among the first degradation level, the second degradation level, and the third degradation level is taken as the overall degradation level.

9. A data analysis-based behavior monitoring system for Alzheimer's disease patients, used to implement the behavior monitoring method for Alzheimer's disease patients as described in any one of claims 1-8, characterized in that, The system includes: The data acquisition module collects voice data, behavioral data, and physiological data of the monitored target within a preset time interval. The first evaluation module converts the voice data into text data and analyzes the voice data and text data to determine the first degradation level of the monitored target. The second evaluation module sets multiple behavior types, determines the actual frequency of each behavior type based on the behavior data, determines the standard frequency range of each behavior type based on the first degradation level, and determines the second degradation level of the monitored target by comparing the standard frequency range with the actual frequency. The third evaluation module determines the diurnal activity ratio of the monitored target based on the physiological data, and determines the third degradation level based on the diurnal activity ratio; The aggregation module combines the first degradation level, the second degradation level, and the third degradation level to determine the overall degradation level of the monitored target.

Images