An assessment method, device, equipment and medium for risk of occult emergency in the elderly

Abstract

The present application relates to the field of intelligent medical auxiliary technology, and more particularly to an old age hidden emergency risk assessment method, device, equipment and medium, the method comprises: obtaining the voice information, swallowing video, physiological information and environmental information of the old patient;Based on the voice information, swallowing video, physiological information and environmental information, through feature extraction, get the voice feature vector, swallowing feature vector, interaction feature vector, and get the fusion feature, to predict the risk probability of various types of hidden emergency of each period, get the hidden emergency risk classification result of each period;Based on the hidden emergency risk classification result of each period, infer the risk state transition probability and posterior probability;Based on the posterior probability, draw the time series change curve of various types of hidden emergency risk;Based on the time series change curve of various types of hidden emergency risk, the old age hidden emergency risk is evaluated, and the accurate evaluation of cognitive impairment old age hidden emergency is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical assistance technology, and in particular to a method, device, equipment and medium for assessing the risk of hidden acute illnesses in the elderly. Background Technology

[0002] With the acceleration of population aging, the incidence of cognitive impairment and other insidious acute diseases in the elderly is constantly rising, becoming a medical problem that urgently needs to be solved.

[0003] Traditional methods for detecting cognitive impairment mainly rely on neuropsychological tests, imaging examinations, and biomarker detection. However, these methods are highly susceptible to subjective factors, costly, and complex to operate, making large-scale screening and long-term monitoring difficult. In recent years, the development of artificial intelligence technology has provided new ideas for the early screening and precise intervention of cognitive impairment. However, existing intelligent prediction models still suffer from insufficient generalization ability and cannot adapt to the real-world screening needs of large populations.

[0004] The application of speech recognition technology has offered new possibilities for the detection of cognitive impairment. Studies have shown that patients with cognitive impairment exhibit specific abnormalities in their speech expression, such as slowed speech rate and increased pauses. These features can serve as important indicators for identifying cognitive impairment. However, existing speech recognition systems still suffer from inaccurate recognition and misinterpretation of intent when handling complex contexts and technical terms, making it difficult to meet the needs of practical applications. Furthermore, abnormal swallowing function is also a common clinical manifestation in patients with cognitive impairment, but currently, there is a lack of effective objective assessment methods.

[0005] Therefore, how to make full use of speech acoustic features and swallowing function information to assess the hidden acute risks of elderly patients is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] In view of the above problems, the present invention provides a method, apparatus, device and medium for assessing the risk of latent acute illness in the elderly that overcomes or at least partially solves the above problems.

[0007] In a first aspect, the present invention provides a method for assessing the risk of latent acute illnesses in the elderly, comprising: Acquire voice information, swallowing video, physiological information, and environmental information of elderly patients; Based on the aforementioned voice information, swallowing video, physiological information, and surrounding environment information, feature extraction is performed to obtain voice feature vectors, swallowing feature vectors, and physiological-environment interaction feature vectors. Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, a fusion feature is obtained; Based on the fusion features, the probability of various hidden acute illnesses in each time period is predicted, and the classification results of hidden acute illnesses in each time period are obtained. Based on the risk classification results of hidden acute illnesses in each time period, the risk state transition probability and posterior probability are inferred. Based on the posterior probability, time-series variation curves of various hidden acute disease risks were plotted. Based on the time-series variation curves of various hidden acute disease risks and the age groups of elderly patients, individualized intervention thresholds for each age group were obtained. The risk of hidden acute illnesses in the elderly is assessed based on the individualized intervention threshold for each age group and the time-series change curves of the various types of hidden acute illness risks.

[0008] Preferably, based on the speech information, swallowing video, physiological information, and surrounding environment information, feature extraction is performed to obtain speech feature vectors, swallowing feature vectors, and physiological-environment interaction feature vectors, including: Based on the speech information, speech rate features, pitch features, and pause frequency features are extracted. Determine the individual speech baseline parameters for each of the speech rate features, pitch features, and pause frequency features; Based on the individual speech baseline parameters of each feature, the speech feature vector is obtained by mapping to the standard feature space. Based on the swallowing video, swallowing time-course features, swallowing maximum acceleration features, laryngeal movement angular velocity features, swallowing coordination index features, and swallowing relative intensity features are extracted. Based on the physiological information and surrounding environmental information, the interaction coefficient features between temperature and blood oxygen, the interaction coefficient features between light and heart rate, and the corrected physiological index features are extracted.

[0009] Preferably, based on the speech feature vector, swallowing feature vector, and interaction feature vector, a fused feature is obtained, including: Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, the abnormality score of each type of feature at each time period is determined. Based on the anomaly score of each feature in each time period, the attention weight of each feature in each time period is obtained. Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, the contribution of any type of feature to any risk is obtained through clinical data annotation, thus yielding a risk weight matrix. Based on the risk weight matrix, the speech feature vector, the swallowing feature vector, and the interaction feature vector, the risk dimension feature vector is obtained. Based on the risk dimension feature vector and the attention weight of each type of feature in each time period, a fused feature is obtained.

[0010] Preferably, based on the fusion features, the probability of various types of hidden acute illnesses in each time period is predicted to obtain the hidden acute illness risk classification results for each time period.

[0011] Based on the fusion features, a dynamic metric learning model is used to dynamically capture the abnormal trends of the fusion features that change over time. The dynamic metric learning model obtains the metric matrix according to the temporal Mahalanobis distance. Based on the aforementioned abnormal trends, the probability of various types of hidden acute illnesses in each time period is predicted, and the classification results of hidden acute illnesses in each time period are obtained.

[0012] Preferably, based on the risk classification results of concealed acute illnesses in each time period, the risk state transition probability and posterior probability are inferred, including: Hidden Markov models were used to predict the hidden states of various hidden acute diseases at different time periods, and the results of the hidden states of hidden acute diseases at different time periods were obtained. Based on the latent state results of the latent acute illnesses in each time period and the risk classification results of the latent acute illnesses in each time period, the risk state transition probability and posterior probability are inferred.

[0013] Preferably, based on the posterior probability, time-series variation curves of various types of hidden acute illness risks are plotted, including: Based on the posterior probability, the posterior high-risk probability time series of various types of hidden acute disease risks for each time period is obtained; A cubic spline interpolation fitting algorithm is used to construct a cubic polynomial between two adjacent time points, as follows:

[0014] in, For any hidden acute illness risk, For any time period, The coefficients are obtained by solving using the least squares method. For the first Hidden acute disease risks at time points The probability of; Based on the aforementioned cubic polynomial, time-series variation curves of various hidden acute disease risks were plotted.

[0015] Preferably, based on the time-series variation curves of various latent acute disease risks and the age grouping of elderly patients, individualized intervention thresholds for any age group are obtained, including: Based on age grouping of elderly patients, the basic threshold for the risk of various hidden acute illnesses for elderly patients in each age group was obtained; Based on the time-series variation curves of the various types of hidden acute illness risks, the slope of the risk curve is obtained; Based on the slope of the risk curve, a trend correction factor is obtained; Obtain the underlying diseases of elderly patients and determine the corresponding underlying disease correction factors; Based on the baseline thresholds, trend correction factors, and underlying disease correction factors for various latent acute diseases in elderly patients of each age group, individualized intervention thresholds for any age group are obtained.

[0016] Secondly, the present invention also provides an assessment device for the risk of latent acute illnesses in the elderly, comprising: The acquisition module is used to acquire the elderly patient's voice information, swallowing video, physiological information, and environmental information. The first obtaining module is used to obtain, through feature extraction, a speech feature vector, a swallowing feature vector, and an interaction feature vector between physiology and environment based on the speech information, swallowing video, physiological information, and surrounding environment information. The second obtaining module is used to obtain fused features based on the speech feature vector, swallowing feature vector, and interaction feature vector; The third module is used to predict the probability of various hidden acute illnesses in each time period based on the fusion features, and obtain the hidden acute illness risk classification results for each time period. The inference module is used to infer the risk state transition probability and posterior probability based on the hidden acute disease risk classification results for each time period. The plotting module is used to plot the time-series variation curves of various hidden acute disease risks based on the posterior probability. The fourth module is used to obtain individualized intervention thresholds for each age group based on the time-series change curves of various hidden acute disease risks and the age groups of elderly patients. The assessment module is used to assess the risk of hidden acute illnesses in the elderly based on the individualized intervention thresholds for each age group and the time-series change curves of the risks of various types of hidden acute illnesses.

[0017] Thirdly, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0018] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0019] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages: This invention provides a method for assessing the risk of hidden acute illnesses in the elderly, comprising: acquiring the elderly patient's voice information, swallowing video, physiological information, and environmental information; based on the voice information, swallowing video, physiological information, and surrounding environmental information, extracting voice feature vectors, swallowing feature vectors, and interaction feature vectors to obtain fused features; based on the fused features, predicting the probability of various hidden acute illnesses at different time periods to obtain the classification results of hidden acute illnesses at different time periods; based on the classification results of hidden acute illnesses at different time periods, inferring the risk state transition probability and posterior probability; based on the posterior probability, plotting the temporal variation curves of various hidden acute illness risks; based on the temporal variation curves of various hidden acute illness risks and the age grouping of the elderly patient, obtaining the individualized intervention threshold for each age group; based on the individualized intervention threshold for each age group and the temporal variation curves of various hidden acute illness risks, assessing the risk of hidden acute illnesses in the elderly, thereby achieving accurate assessment of hidden acute illnesses in the elderly with cognitive impairment. Attached Figure Description

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating the method for assessing the risk of latent acute illness in the elderly in an embodiment of the present invention is shown; Figure 2 A schematic diagram of the structure of the assessment device for the risk of hidden acute illness in the elderly in an embodiment of the present invention is shown; Figure 3 A schematic diagram of the structure of a computer device for implementing a method for assessing the risk of latent acute illnesses in the elderly, as shown in an embodiment of the present invention, is illustrated. Detailed Implementation

[0021] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0022] Example 1: Embodiments of the present invention provide a method for assessing the risk of latent acute illnesses in the elderly, such as... Figure 1 As shown, it includes: S101, acquires the elderly patient's voice information, swallowing video, physiological information and environmental information; S102, based on speech information, swallowing video, physiological information and surrounding environment information, through feature extraction, obtains speech feature vector, swallowing feature vector, and physiological and environmental interaction feature vector; S103, based on speech feature vector, swallowing feature vector, and interaction feature vector, the fused features are obtained; S104, based on fusion features, predicts the probability of various hidden acute illnesses in different time periods and obtains the classification results of hidden acute illnesses in different time periods; S105, Based on the risk classification results of hidden acute illnesses in each time period, infer the risk state transition probability and posterior probability; S106, Based on posterior probability, the time-series variation curves of various hidden acute disease risks are plotted; S107, based on the time-series change curves of various hidden acute disease risks and the age grouping of elderly patients, obtains individualized intervention thresholds for each age group; S108 assesses the risk of hidden acute illnesses in the elderly based on individualized intervention thresholds for each age group and time-series change curves of various hidden acute illness risks.

[0023] In S101, the elderly patient's voice information specifically refers to the language used in daily conversations, which can be acquired through a voice acquisition device installed on the elderly patient. This device can capture up to 5 minutes of conversation content.

[0024] The swallowing video specifically captures the swallowing behavior of elderly patients drinking warm water. This can be done using a high-definition camera, specifically by capturing three consecutive videos of each patient swallowing 50ml of warm water.

[0025] Physiological information can specifically include information on physiological states such as heart rate and blood oxygenation over a 10-minute period.

[0026] Environmental information can specifically include the collection of light intensity, temperature, and humidity, as well as the corresponding collection time of physiological information.

[0027] Next, the voice information, swallowing video, physiological information, and environmental information can be preprocessed separately.

[0028] Specifically, the speech data was denoised using wavelet thresholding. Laryngeal vibration signals were extracted from the swallowing video and denoised using Kalman filtering. Useless information was removed from both physiological and environmental data, and the speech data, the corresponding physiological information from the swallowing video, and the corresponding environmental information were mapped to the time axis to achieve time alignment.

[0029] Next, S102 is executed, based on speech information, swallowing video, physiological information and surrounding environment information, to obtain speech feature vector, swallowing feature vector and physiological and environmental interaction feature vector through feature extraction.

[0030] Specifically, based on speech information, speech rate features, pitch features, and pause frequency features are extracted; Determine the individual speech baseline parameters for each of the speech rate, pitch, and pause frequency features; Based on the individual speech baseline parameters of each feature, the speech feature vector is obtained by mapping to the standard feature space. Based on swallowing videos, swallowing time-course features, maximum swallowing acceleration features, laryngeal angular velocity features, and swallowing coordination index features were extracted. Based on physiological and environmental information, the interaction coefficients between temperature and blood oxygen, light intensity and heart rate, and corrected physiological indicators are extracted.

[0031] More specifically, the individual speech baseline parameters for each feature are obtained using the following formula:

[0032] It can be any one of the following features: speech rate, tone, and pause frequency. Baseline sample size For individual speech baseline parameters.

[0033] Next, the variance of individual speech features is calculated:

[0034] in, The variance of individual speech features. The standard deviation of an individual's speech features.

[0035] Then, based on the individual speech baseline parameters of each feature, they are mapped to the standard feature space to obtain the speech feature vector:

[0036] in, This is the global weight matrix, specifically obtained through training on a large-scale corpus. The mapping results for each speech feature are shown below. The individual adaptive bias is obtained through the following formula:

[0037] in, The speech-independent reference feature vectors for elderly patients are obtained from a corpus of normal populations. For the bias that is not yet determined, This represents the final, optimal result.

[0038] The resulting speech feature vector includes: the mapping results of speech rate, the mapping results of pitch, the mapping results of pause frequency, the speech rate variation coefficient, and the relative dynamic range of pitch.

[0039] Among them, the coefficient of variation of speech rate Pitch relative dynamic range .

[0040] Based on swallowing videos, swallowing time-course features, maximum swallowing acceleration features, laryngeal angular velocity features, and swallowing coordination index features were extracted.

[0041] Specifically, the analysis of swallowing videos, which integrates video optical flow analysis with laryngeal vibration signals, derives the biomechanical parameters of the swallowing action, overcoming the limitations of existing visual features.

[0042] Specifically, optical flow is used to extract the throat trajectory of video frames, thereby obtaining the pixel displacement trajectory of the throat ROI region. , The frame number, , This refers to the swallowing time.

[0043] Integrating the throat vibration signal, i.e., integrating the vibration acceleration, yields the velocity: , Used to iterate through 0~ The time point in the middle.

[0044] Based on this calculation, the swallowing time characteristics are:

[0045] Characteristics of maximum swallowing acceleration:

[0046] Characteristics of angular velocity of throat movement: , for Directional velocity, for Directional acceleration, for Directional velocity, for Directional acceleration is calculated using the central difference method.

[0047] Swallowing coordination index characteristics: ,in, The standard deviation of the angular velocity of the throat movement. This represents the average angular velocity of the throat movement. The closer the value is to 1, the better the swallowing coordination.

[0048] Relative intensity characteristics of swallowing:

[0049] in, For swallowing acceleration scale, For the swallowing time scale, It is a very small positive constant. Characteristics of the swallowing process, This refers to the relative intensity characteristics of swallowing.

[0050] This yields the swallowing feature vector described above.

[0051] Next, based on physiological and environmental information, we extracted the interaction coefficient features between temperature and blood oxygen, the interaction coefficient features between light and heart rate, and the corrected physiological indicators.

[0052] The physiological baseline includes: heart rate baseline: Baseline blood oxygen saturation: .

[0053] Environmental baseline: Ambient temperature baseline: Ambient light intensity baseline: .

[0054] Therefore, the interaction coefficient between temperature and blood oxygenation is characterized as follows:

[0055] This is actual blood oxygen saturation data. For the actual ambient temperature, use a very small positive constant. To avoid the denominator being 0.

[0056] Characteristics of the interaction coefficient between light exposure and heart rate:

[0057] in, For actual heart rate data, This refers to the actual ambient light intensity. This serves as a reference value for fixed ambient light intensity.

[0058] Corrected physiological characteristics:

[0059]

[0060] For the corrected heart rate data, This is the corrected blood oxygen saturation data.

[0061] The interaction coefficients between temperature and blood oxygen, light and heart rate, and corrected physiological indicators were extracted from these data.

[0062] Next, step S103 is executed to obtain fused features based on speech feature vectors, swallowing feature vectors, and interaction feature vectors.

[0063] Specifically, based on speech feature vectors, swallowing feature vectors, and interaction feature vectors, the abnormality score of each type of feature is determined at each time period; Based on the anomaly score of each feature class in each time period, the attention weight of each feature class in each time period is obtained. Based on speech feature vectors, swallowing feature vectors, and interaction feature vectors, the contribution of any type of feature to any risk is obtained through clinical data annotation, resulting in a risk weight matrix. Based on the risk weight matrix, speech feature vector, swallowing feature vector, and interaction feature vector, the risk dimension feature vector is obtained. Based on the risk dimension feature vector and the attention weight of each type of feature at each time period, the fused features are obtained.

[0064] First, the three main categories of features here include speech feature vectors, swallowing feature vectors, and interaction feature vectors. Each time period is specifically divided into nine time periods: resting preparation period, oral cavity initiation phase, oral cavity pushing phase, initial larynx elevation phase, peak larynx contraction phase, larynx descending phase, esophageal inlet opening phase, swallowing and breathing recovery phase, and swallowing completion and resting recovery phase. These nine time periods are derived from the motor analysis of the larynx, swallowing, and dialogue processes.

[0065] Each type of feature in each time period is Represents the speech feature vectors for each time period. This represents the swallowing feature vector for each time period. Represents the interaction feature vectors for each time period. Indicates the number of time periods.

[0066] Anomaly score for each feature type at each time period:

[0067] This represents the average of various characteristics across all time periods. Assign importance weights to various features. For the number of features, This represents the arithmetic mean of all time periods for each characteristic. represents the standard deviation of all time periods for each type of characteristic.

[0068] Next, based on the anomaly score of each feature class in each time period, the attention weight of each feature class in each time period is obtained:

[0069] in, For any given time period, The attention weights for each feature class at each time period.

[0070] Next, based on three risk categories—occult neurological emergencies, airway obstruction, and metabolic encephalopathy—the contribution of each feature to each risk is obtained through clinical data annotation, resulting in a risk weight matrix. Therefore, this risk weight matrix is: Where 12 represents the number of dimensions of the features (voice feature vectors have 3 dimensions, swallowing feature vectors have 5 dimensions, and interaction feature vectors have 4 dimensions), and 3 represents the number of risks. The contribution of any type of feature to any risk.

[0071] Next, the risk dimension feature vector is calculated, specifically:

[0072] in, The first in the matrix Column weight vector, This is the feature vector for the risk dimension.

[0073] Finally, based on the risk dimension feature vector and the attention weight of each type of feature at each time period, the fused feature is obtained:

[0074] Next, S104 is executed to predict the probability of various types of hidden acute illnesses in each time period based on the fusion features, and obtain the hidden acute illness risk classification results for each time period.

[0075] Specifically, based on the fusion features, a dynamic metric learning model is adopted to dynamically capture the abnormal trends of the fusion features that change over time. The dynamic metric learning model obtains the metric matrix according to the temporal Mahalanobis distance. Based on abnormal trends, the probability of various types of hidden acute illnesses in different time periods is predicted, and the risk classification results of hidden acute illnesses in different time periods are obtained.

[0076] Among them, the traditional Mahalanobis distance is improved into the temporal Mahalanobis distance:

[0077] The fusion features are the input. This outputs the risk classification results for hidden acute illnesses at different time periods. The risk classification results are obtained by training an SVM classifier. .

[0078] For dynamic metric matrix updates,

[0079] in, =0.005, for updating the step size. Indicates the preceding The mean of each sample.

[0080] Next, execute S105, and based on the risk classification results of hidden acute illnesses in each time period, infer the risk state transition probability and posterior probability.

[0081] Specifically, a hidden Markov model is used to predict the hidden state of various hidden acute disease risks in different time periods, and the hidden state results of hidden acute diseases in different time periods are obtained. Based on the latent state results of concealed acute illnesses in each time period and the risk classification results of concealed acute illnesses in each time period, the risk state transition probability and posterior probability are inferred.

[0082] The latent state results of latent acute illnesses here include: The latent state of insidious acute illness refers specifically to a potential pathological state that objectively exists within the body or at the physiological level in the elderly, is progressing, but has no typical external symptoms, cannot be detected by routine physical examinations, and cannot be detected by short-term observation. It is a hidden internal health state that cannot be directly observed.

[0083] Specifically, the latent state results of hidden acute conditions are predicted using a Hidden Markov Model (HMM).

[0084] Here, the model is trained by pre-labeling the latent state results of elderly patients and inputting observable data, thereby enabling the prediction of latent state results of hidden acute illnesses in elderly patients.

[0085] Next, we reason about the probability of risk state transition: First, introduce the time decay factor. Used to strengthen the weight of recent states and weaken the weight of distant states, risk state transition probability: , This represents the hidden state result from the previous time period.

[0086] in, The initial risk transfer probability was obtained based on clinical data. , , .

[0087] Posterior probability: , in, For the front Observations for each time period, This is the normalization constant.

[0088] Observation probability: .

[0089] After obtaining the posterior probability, execute S106 to plot the time-series change curves of various hidden acute disease risks based on the posterior probability.

[0090] Specifically, based on posterior probabilities, the posterior high-risk probability time series of various types of hidden acute disease risks are obtained for each time period; A cubic spline interpolation fitting algorithm is used to construct a cubic polynomial between two adjacent time points, as follows:

[0091] in, For any hidden acute illness risk, For any time period, The coefficients are obtained by solving using the least squares method. For the first Hidden acute disease risks at time points The probability of; Based on cubic polynomials, time-series variation curves of various hidden acute disease risks were plotted.

[0092] First, we obtain nine discrete posterior high-risk probability time series for each type of hidden acute disease risk, which are hidden neurological acute diseases, airway obstruction, and metabolic encephalopathy.

[0093] Construct a cubic polynomial between two time points:

[0094] The least squares method is used to solve for each coefficient. .

[0095] Finally, according to the swallowing time on the horizontal axis The vertical axis is the first Hidden acute disease risks at time points The probability was used to plot the time-series variation curves of various hidden acute illness risks. This resulted in three isolated, continuous, smooth risk evolution curves.

[0096] Next, S107 is executed, based on the time-series change curves of various hidden acute disease risks and the age groups of elderly patients, to obtain individualized intervention thresholds for each age group.

[0097] Specifically, based on the age grouping of elderly patients, the basic threshold for the risk of various hidden acute illnesses for elderly patients in each age group is obtained; Based on the time-series variation curves of various hidden acute disease risks, the slope of the risk curve is obtained; The trend correction factor is obtained based on the slope of the risk curve; Obtain the underlying diseases of elderly patients and determine the corresponding underlying disease correction factors; Based on the baseline thresholds, trend correction factors, and underlying disease correction factors for various latent acute conditions in elderly patients of different age groups, individualized intervention thresholds for any age group are obtained.

[0098] First, elderly patients were grouped by age: 60-70 years old, 70-80 years old, and over 80 years old. Then, corresponding baseline thresholds for various types of latent acute illnesses were assigned to each age group, as shown in the table below:

[0099] Among them, R1 is a latent neurological emergency, R2 is airway obstruction, and R3 is metabolic encephalopathy.

[0100] The slope of the risk curve is Thus, the trend correction factor is obtained. When the risk curve is on an upward trend, then .

[0101] Based on the underlying diseases of elderly patients, corresponding underlying disease correction factors are determined. For example, if an elderly patient has hypertension or diabetes, the corresponding underlying disease correction factor is determined. Elderly patients with a history of stroke have corresponding baseline correction factors. When elderly patients have no underlying diseases, the corresponding baseline correction factor is... .

[0102] The individualized intervention threshold for any age group can be obtained using the following formula:

[0103] The baseline thresholds for the risk of various latent acute illnesses in elderly patients of different age groups were established. As a trend correction factor, Basic correction factor, For any age group, the individualized intervention threshold is set.

[0104] Finally, S108 was performed to assess the risk of latent acute illnesses in the elderly based on individualized intervention thresholds for each age group and time-series curves showing the changes in the risk of various latent acute illnesses.

[0105] By comparing the time-series variation curves of various latent acute disease risks with the individualized intervention thresholds for each age group obtained in S107, the risk level of latent acute diseases in the elderly can be obtained. When an elderly person is diagnosed with a hidden acute illness and deemed to be at low risk, the recommended measure is to collect data at three different time points every three days; if... To determine that hidden acute illnesses in the elderly are of medium risk, the measures taken include collecting data at four different times each day; if The elderly were identified as having hidden acute conditions, which is a high-risk condition. This triggered an emergency procedure, increasing the number of water and swallowing samples to five, and providing real-time medical alerts.

[0106] Through accurate risk monitoring as described above, effective care can be provided to elderly patients with cognitive impairment.

[0107] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages: This invention provides a method for assessing the risk of latent acute illnesses in the elderly, comprising: acquiring the elderly patient's voice information, swallowing video, physiological information, and environmental information; based on the voice information, swallowing video, physiological information, and surrounding environmental information, obtaining voice feature vectors, swallowing feature vectors, and interaction feature vectors through feature extraction to obtain fused features; based on the fused features, predicting the probability of various latent acute illnesses at different time periods to obtain latent acute illness risk classification results at different time periods; based on the latent acute illness risk classification results at different time periods, inferring the risk state transition probability and posterior probability; based on the posterior probability, plotting the temporal variation curves of various latent acute illness risks; based on the temporal variation curves of various latent acute illness risks and the age grouping of the elderly patient, obtaining individualized intervention thresholds for each age group; based on the individualized intervention thresholds for each age group and the temporal variation curves of various latent acute illness risks, assessing the risk of latent acute illnesses in the elderly, thereby achieving accurate assessment of latent acute illnesses in the elderly with cognitive impairment.

[0108] Example 2: Based on the same inventive concept, embodiments of the present invention provide an assessment device for the risk of latent acute illnesses in the elderly, such as... Figure 2 As shown, it includes: The acquisition module 201 is used to acquire the elderly patient's voice information, swallowing video, physiological information, and environmental information. The first obtaining module 202 is used to obtain, through feature extraction, a speech feature vector, a swallowing feature vector, and an interaction feature vector between physiology and environment based on the speech information, swallowing video, physiological information, and surrounding environment information. The second obtaining module 203 is used to obtain fused features based on the speech feature vector, swallowing feature vector, and interaction feature vector; The third module 204 is used to predict the probability of various hidden acute illnesses in each time period based on the fusion features, and obtain the hidden acute illness risk classification results for each time period. Inference module 205 is used to infer the risk state transition probability and posterior probability based on the hidden acute disease risk classification results of each time period; The plotting module 206 is used to plot the time-series change curves of various hidden acute disease risks based on the posterior probability. The fourth module 207 is used to obtain individualized intervention thresholds for each age group based on the time-series change curves of various hidden acute disease risks and the age grouping of elderly patients. The assessment module 208 is used to assess the risk of hidden acute illnesses in the elderly based on the individualized intervention threshold for each age group and the time-series change curves of the various types of hidden acute illness risks.

[0109] In one alternative implementation, the first receiving module 202 is configured to: Based on the speech information, speech rate features, pitch features, and pause frequency features are extracted. Determine the individual speech baseline parameters for each of the speech rate features, pitch features, and pause frequency features; Based on the individual speech baseline parameters of each feature, the speech feature vector is obtained by mapping to the standard feature space. Based on the swallowing video, swallowing time-course features, swallowing maximum acceleration features, laryngeal movement angular velocity features, swallowing coordination index features, and swallowing relative intensity features are extracted. Based on the physiological information and surrounding environmental information, the interaction coefficient features between temperature and blood oxygen, the interaction coefficient features between light and heart rate, and the corrected physiological index features are extracted.

[0110] In one alternative implementation, the second obtaining module 203 is configured to: Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, the abnormality score of each type of feature at each time period is determined. Based on the anomaly score of each feature in each time period, the attention weight of each feature in each time period is obtained. Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, the contribution of any type of feature to any risk is obtained through clinical data annotation, thus yielding a risk weight matrix. Based on the risk weight matrix, the speech feature vector, the swallowing feature vector, and the interaction feature vector, the risk dimension feature vector is obtained. Based on the risk dimension feature vector and the attention weight of each type of feature in each time period, a fused feature is obtained.

[0111] In one alternative implementation, the third obtaining module 204 is used for: Based on the fusion features, a dynamic metric learning model is used to dynamically capture the abnormal trends of the fusion features that change over time. The dynamic metric learning model obtains the metric matrix according to the temporal Mahalanobis distance. Based on the aforementioned abnormal trends, the probability of various types of hidden acute illnesses in each time period is predicted, and the risk classification results of hidden acute illnesses in each time period are obtained.

[0112] In one alternative implementation, the inference module 205 is configured to: Hidden Markov models were used to predict the hidden states of various hidden acute diseases at different time periods, and the results of the hidden states of hidden acute diseases at different time periods were obtained. Based on the latent state results of the latent acute illnesses in each time period and the risk classification results of the latent acute illnesses in each time period, the risk state transition probability and posterior probability are inferred.

[0113] In one alternative implementation, the drawing module 206 is used for: Based on the posterior probability, the posterior high-risk probability time series of various types of hidden acute disease risks for each time period is obtained; A cubic spline interpolation fitting algorithm is used to construct a cubic polynomial between two adjacent time points, as follows:

[0114] in, For any hidden acute illness risk, For any time period, The coefficients are obtained by solving using the least squares method. For the first Hidden acute disease risks at time points The probability of; Based on the aforementioned cubic polynomial, time-series variation curves of various hidden acute disease risks were plotted.

[0115] In one alternative implementation, the fourth obtaining module 207 is used for: Based on age grouping of elderly patients, the basic threshold for the risk of various hidden acute illnesses for elderly patients in each age group was obtained; Based on the time-series variation curves of the various types of hidden acute illness risks, the slope of the risk curve is obtained; Based on the slope of the risk curve, a trend correction factor is obtained; Obtain the underlying diseases of elderly patients and determine the corresponding underlying disease correction factors; Based on the baseline thresholds, trend correction factors, and underlying disease correction factors for various latent acute diseases in elderly patients of each age group, individualized intervention thresholds for any age group are obtained.

[0116] Example 3: Based on the same inventive concept, embodiments of the present invention provide a computer device, such as... Figure 3As shown, it includes a memory 304, a processor 302, and a computer program stored in the memory 304 and executable on the processor 302. When the processor 302 executes the program, it implements the steps of the above-described method for assessing the risk of hidden acute illnesses in the elderly.

[0117] Among them, Figure 3 In this document, a bus architecture (represented by bus 300) is used. Bus 300 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 306 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0118] Example 4: Based on the same inventive concept, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for assessing the risk of hidden acute illnesses in the elderly.

[0119] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0120] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0121] Similarly, it should be understood that, in order to simplify the invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are explicitly recited in each embodiment. Rather, as reflected in each embodiment, inventive aspects lie in fewer than all features of the single foregoing disclosed embodiment. Therefore, the claims, following the detailed description, are hereby expressly incorporated into this detailed description, wherein each claim itself is a separate embodiment of the invention.

[0122] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0123] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the specific implementation, any of the claimed embodiments can be used in any combination.

[0124] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the assessment device for the risk of hidden acute illnesses in the elderly, or the computer device according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0125] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

Claims

1. A method for assessing the risk of latent acute illnesses in the elderly, characterized in that, include: Acquire voice information, swallowing video, physiological information, and environmental information of elderly patients; Based on the aforementioned voice information, swallowing video, physiological information, and surrounding environment information, feature extraction is performed to obtain voice feature vectors, swallowing feature vectors, and physiological-environment interaction feature vectors. Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, a fusion feature is obtained; Based on the fusion features, the probability of various hidden acute illnesses in each time period is predicted, and the classification results of hidden acute illnesses in each time period are obtained. Based on the risk classification results of hidden acute illnesses in each time period, the risk state transition probability and posterior probability are inferred. Based on the posterior probability, time-series variation curves of various hidden acute disease risks were plotted. Based on the time-series variation curves of various hidden acute disease risks and the age groups of elderly patients, individualized intervention thresholds for each age group were obtained. The risk of hidden acute illnesses in the elderly is assessed based on the individualized intervention threshold for each age group and the time-series change curves of the various types of hidden acute illness risks.

2. The method as described in claim 1, characterized in that, Based on the aforementioned speech information, swallowing video, physiological information, and surrounding environmental information, feature extraction is performed to obtain speech feature vectors, swallowing feature vectors, and physiological-environment interaction feature vectors, including: Based on the speech information, speech rate features, pitch features, and pause frequency features are extracted. Determine the individual speech baseline parameters for each of the speech rate features, pitch features, and pause frequency features; Based on the individual speech baseline parameters of each feature, the speech feature vector is obtained by mapping to the standard feature space. Based on the swallowing video, swallowing time-course features, swallowing maximum acceleration features, laryngeal movement angular velocity features, swallowing coordination index features, and swallowing relative intensity features are extracted. Based on the physiological information and surrounding environmental information, the interaction coefficient features between temperature and blood oxygen, the interaction coefficient features between light and heart rate, and the corrected physiological index features are extracted.

3. The method as described in claim 1, characterized in that, Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, fused features are obtained, including: Based on the speech feature vector, swallowing feature vector, and interaction feature vector, the abnormality score of each type of feature is determined at each time period. Based on the anomaly score of each feature in each time period, the attention weight of each feature in each time period is obtained. Based on the aforementioned speech feature vector, swallowing feature vector, and interaction feature vector, the contribution of any type of feature to any risk is obtained through clinical data annotation, thus yielding a risk weight matrix. Based on the risk weight matrix, the speech feature vector, the swallowing feature vector, and the interaction feature vector, the risk dimension feature vector is obtained. Based on the risk dimension feature vector and the attention weight of each type of feature in each time period, a fused feature is obtained.

4. The method as described in claim 1, characterized in that, Based on the fusion features, the probability of various hidden acute illnesses in each time period is predicted, and the classification results of hidden acute illnesses in each time period are obtained. Based on the fusion features, a dynamic metric learning model is used to dynamically capture the abnormal trends of the fusion features that change over time. The dynamic metric learning model obtains the metric matrix according to the temporal Mahalanobis distance. Based on the aforementioned abnormal trends, the probability of various types of hidden acute illnesses in each time period is predicted, and the risk classification results of hidden acute illnesses in each time period are obtained.

5. The method as described in claim 1, characterized in that, Based on the risk classification results of concealed acute illnesses in each time period, the risk state transition probability and posterior probability are inferred, including: Hidden Markov models were used to predict the hidden states of various hidden acute diseases at different time periods, and the results of the hidden states of hidden acute diseases at different time periods were obtained. Based on the latent state results of the latent acute illnesses in each time period and the risk classification results of the latent acute illnesses in each time period, the risk state transition probability and posterior probability are inferred.

6. The method as described in claim 1, characterized in that, Based on the posterior probability, time-series variation curves of various types of hidden acute illness risks were plotted, including: Based on the posterior probability, the posterior high-risk probability time series of various types of hidden acute disease risks for each time period is obtained; A cubic spline interpolation fitting algorithm is used to construct a cubic polynomial between two adjacent time points, as follows: ； in, For any hidden acute illness risk, For any time period, The coefficients are obtained by solving using the least squares method. For the first Hidden acute disease risks at time points The probability of; Based on the aforementioned cubic polynomial, time-series variation curves of various hidden acute disease risks were plotted.

7. The method as described in claim 1, characterized in that, Based on the time-series variation curves of various latent acute disease risks and age groupings of elderly patients, individualized intervention thresholds for any age group were obtained, including: Based on age grouping of elderly patients, the basic threshold for the risk of various hidden acute illnesses for elderly patients in each age group was obtained; Based on the time-series variation curves of the various types of hidden acute illness risks, the slope of the risk curve is obtained; Based on the slope of the risk curve, a trend correction factor is obtained; Obtain the underlying diseases of elderly patients and determine the corresponding underlying disease correction factors; Based on the baseline thresholds, trend correction factors, and underlying disease correction factors for various latent acute diseases in elderly patients of each age group, individualized intervention thresholds for any age group are obtained.

8. A device for assessing the risk of latent acute illnesses in the elderly, characterized in that, include: The acquisition module is used to acquire the elderly patient's voice information, swallowing video, physiological information, and environmental information. The first obtaining module is used to obtain, through feature extraction, a speech feature vector, a swallowing feature vector, and an interaction feature vector between physiology and environment based on the speech information, swallowing video, physiological information, and surrounding environment information. The second obtaining module is used to obtain fused features based on the speech feature vector, swallowing feature vector, and interaction feature vector; The third module is used to predict the probability of various hidden acute illnesses in each time period based on the fusion features, and obtain the hidden acute illness risk classification results for each time period. The inference module is used to infer the risk state transition probability and posterior probability based on the hidden acute disease risk classification results for each time period. The plotting module is used to plot the time-series variation curves of various hidden acute disease risks based on the posterior probability. The fourth module is used to obtain individualized intervention thresholds for each age group based on the time-series change curves of various hidden acute disease risks and the age groups of elderly patients. The assessment module is used to assess the risk of hidden acute illnesses in the elderly based on the individualized intervention thresholds for each age group and the time-series change curves of the risks of various types of hidden acute illnesses.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images