A Deep Learning-Based Prediction Method for Adolescent Anxiety

Abstract

This invention relates to the field of medical data processing technology, specifically to a deep learning-based method for predicting adolescent anxiety. The method includes: quantifying gut microbiota characteristics, anxiety behavior characteristics, and environmental stress characteristics; calculating time-series weighting coefficients; integrating gut microbiota characteristics, anxiety behavior characteristics, environmental stress characteristics, and time-series weighting coefficients into a model; and outputting the prediction results. This invention fills the gap in dynamic biomarker modeling in existing anxiety disorder models by integrating research on the human gut microbiota with existing behavioral and environmental factors. Through deep learning, it seeks the long-term relationship between adolescent gut microbiota and anxiety symptoms, addressing the subjectivity, invasiveness, privacy risks, cost, and delays in existing anxiety disorder diagnosis, thus enabling early prevention. It also provides causal evidence for the "gut-brain axis" theory in adolescents, promoting the development of psychomicrobiome research.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and more specifically to a method for predicting adolescent anxiety based on deep learning. Background Technology

[0002] The brain and gastrointestinal tract communicate and interact in a complex two-way manner through the brain-gut axis (BGA). The brain-gut axis is a neuro-endocrine-immune network composed of the central nervous system (CNS), the autonomic nervous system (ANS), the enteric nervous system (ENS), and the hypothalamic-pituitary-adrenal axis (HPA). Through the brain-gut axis, the gastrointestinal tract is synergistically regulated by various levels of the central nervous system, the enteric nervous system, and the hypothalamic-pituitary-adrenal axis. Internal and external stimuli, as well as psychological and emotional factors, can all affect gastrointestinal function. Simultaneously, gastrointestinal activity participates in the regulation of central sensation, emotion, and behavior.

[0003] In recent years, mental health issues among adolescents have become increasingly prominent, with the incidence of mental illnesses, especially anxiety disorders, on the rise. Current research primarily attributes anxiety disorders to behavioral factors (personal factors, such as sleep disturbances, social withdrawal, academic avoidance, irritability, self-doubt, and physical discomfort) and environmental factors (external factors, such as academic pressure, family conflict, peer relationships, and parent-child communication). However, the brain-gut axis theory reveals that adolescent anxiety disorders may also be influenced by the gastrointestinal tract, with this influence stemming primarily from various gut microbiota. Current technologies lack in-depth research in this area; therefore, a deep learning-based predictive method for adolescent anxiety is needed to explore the relationships between gut microbiota, environmental factors, and behavioral factors to identify causal connections and ultimately achieve prediction. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a deep learning-based method for predicting adolescent anxiety by exploring the relationship between gut microbiota, environmental factors and behavioral factors, finding causal relationships, and thus making predictions.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A deep learning-based method for predicting adolescent anxiety includes:

[0007] Quantification of gut microbiota characteristics, calculation of relative abundance and diversity index of microbiota;

[0008] = ; =- ;

[0009] Let be the relative abundance of the k-th bacterial species in the j-th adolescent sample. The sequencing count of the kth bacterial species in this sample; This is the total sequencing count of all core bacterial communities in the sample. Let K be the microbial diversity index of the j-th adolescent sample; K is the number of microbial species.

[0010] Quantifying anxiety behavior characteristics, and calculating the total anxiety behavior score of the j-th adolescent. ;

[0011] Quantifying environmental stress characteristics and calculating the total environmental stress score of the j-th adolescent. ;

[0012] Calculate the time-series weighting coefficients. = , Let t be the temporal weight of the j-th adolescent sample at time t; t is the time point, and T is the total number of time steps. is the time decay coefficient, e is the natural constant, and s is the time decay coefficient starting from 1 and going sequentially up to T;

[0013] Feature fusion = , They are respectively The value at time t;

[0014] The model predicts anxiety levels using an LSTM+Attention deep learning model, with the fused features as input. ;when A value <30 indicates low risk; a value ≤30 indicates low risk. <60 indicates medium risk. A value of 60 or higher indicates a high risk level.

[0015] Preferably, the gut microbiota information is obtained by periodically collecting fecal samples from adolescents and performing metagenomic sequencing.

[0016] Preferably, the total score of anxiety behavior of the j-th adolescent is calculated. , = ;

[0017] in, Let L be the total score of anxiety behavior for the j-th adolescent sample, ranging from 0 to 100; L is the number of types of anxiety behavior. The weight of the l-th type of anxiety behavior, The score is the level score of the l-th type of anxiety behavior in this sample.

[0018] Preferably, the total environmental stress score of the j-th adolescent is calculated. , = ;

[0019] in, The total environmental stress score of the j-th adolescent sample ranges from 0 to 100; P represents the number of different types of environmental stress. Let p be the weight of the p-th type of environmental pressure; This represents the level score of the p-th type of environmental pressure in this sample.

[0020] Preferably, the weights and ratings of the anxiety behaviors are set based on clinical data;

[0021] The weights and levels of environmental pressure are set manually.

[0022] Preferably, some of the data on anxiety behaviors and environmental stress are obtained through monitoring equipment.

[0023] Preferably, the deep learning model further includes a mean squared error loss function, Loss;

[0024] Loss= Where M is the total number of samples; T is the total number of time steps; Let represent the true anxiety level of the j-th sample at time t.

[0025] Preferably, It is obtained by physicians through clinical scale evaluation.

[0026] Preferably, =0.2.

[0027] Preferably, when the output predicted anxiety level is medium or high risk, the corresponding adolescent is retested manually.

[0028] The beneficial effects of this invention are as follows: by integrating research on the human gut microbiota with existing behavior and environment, it overcomes the limitations of a single data source, combines gut microbiota abundance and species with anxiety behavior and environmental stress data, fills the gap in dynamic modeling of biomarkers in existing anxiety disorder models, and uses deep learning to find the long-term relationship between adolescent gut microbiota and anxiety symptoms. This solves the problems of subjectivity, invasiveness, privacy risks, cost, and lag in the diagnosis of anxiety disorders in existing technologies. Schools can use the model regularly to predict anxiety in students, achieving early prevention. This provides a new method for predicting adolescent anxiety that is objective, non-invasive, low-cost, easy to promote, and has an early warning effect. It also provides causal evidence for the "gut-brain axis" theory in adolescents and promotes the development of mental microbiome. By using temporal weights and dynamic fusion, it is possible to capture short-term fluctuations in adolescent anxiety and improve sensitivity. By using a time decay coefficient, it can be adjusted according to the age group, gender, or individual of adolescents to adapt to changes in different situations and accommodate a wider range of adolescents. By using LSTM+Attention deep learning, LSTM processes temporal data and Attention focuses on key time points, which can better capture one or more of the core behaviors, environments, and microbiota corresponding to the time points when anxiety occurs, thus ensuring predictive effectiveness. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a deep learning-based method for predicting adolescent anxiety, which is a specific embodiment of the present invention. Detailed Implementation

[0030] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0031] Please refer to Figure 1 A deep learning-based method for predicting adolescent anxiety includes:

[0032] Quantification of gut microbiota characteristics, calculation of relative abundance and diversity index of microbiota;

[0033] = ; =- ;

[0034] Let be the relative abundance of the k-th bacterial species in the j-th adolescent sample. The sequencing count of the kth bacterial species in this sample; This is the total sequencing count of all core bacterial communities in the sample. Let K be the microbial diversity index of the j-th adolescent sample; K is the number of microbial species.

[0035] Quantifying anxiety behavior characteristics, and calculating the total anxiety behavior score of the j-th adolescent. ;

[0036] Quantifying environmental stress characteristics and calculating the total environmental stress score of the j-th adolescent. ;

[0037] Calculate the time-series weighting coefficients. = , Let t be the temporal weight of the j-th adolescent sample at time t; t is the time point, and T is the total number of time steps. is the time decay coefficient, e is the natural constant, and s is the time decay coefficient starting from 1 and going sequentially up to T;

[0038] Feature fusion = , They are respectively The value at time t;

[0039] The model predicts anxiety levels using an LSTM+Attention deep learning model, with the fused features as input. ;when A value <30 indicates low risk; a value ≤30 indicates low risk. <60 indicates medium risk. A value of 60 or higher indicates a high risk level.

[0040] As described above, by integrating research on the human gut microbiota with existing behavioral and environmental data, the limitations of single data sources are overcome. By combining gut microbiota abundance and species with anxiety behavior and environmental stress data, the gap in dynamic biomarker modeling in existing anxiety disorder models is filled. Through deep learning, the long-term relationship between adolescent gut microbiota and anxiety symptoms is explored, solving the problems of subjectivity, invasiveness, privacy risks, cost, and lag in existing anxiety disorder diagnosis technologies. Schools can regularly use the model to predict anxiety in students, achieving early prevention. This provides a new method for predicting adolescent anxiety that is objective, non-invasive, low-cost, easy to promote, and has early warning effects. It also provides causal evidence for the "gut-brain axis" theory in adolescents and promotes the development of psychomicrobiome. By using temporal weights and dynamic fusion, it is possible to capture short-term fluctuations in adolescent anxiety and improve sensitivity. By using a time decay coefficient, it can be adjusted according to the age group, gender, or individual of adolescents to adapt to changes in different situations and accommodate a wider range of adolescents. By using LSTM+Attention deep learning, LSTM processes temporal data and Attention focuses on key time points, which can better capture one or more of the core behaviors, environments, and microbiota corresponding to the time points when anxiety occurs, thus ensuring predictive effectiveness.

[0041] Furthermore, gut microbiota characteristics were quantified by obtaining microbiota information through metagenomic sequencing of fecal samples collected from adolescents on a regular basis.

[0042] As can be seen from the above description, non-invasive testing can be achieved by collecting fecal samples, thus avoiding the invasion of privacy.

[0043] Furthermore, the total score of anxiety behavior of the j-th adolescent was statistically analyzed. , = ;

[0044] in, Let L be the total score of anxiety behavior for the j-th adolescent sample, ranging from 0 to 100; L is the number of types of anxiety behavior. The weight of the l-th type of anxiety behavior, The score is the level score of the l-th type of anxiety behavior in this sample.

[0045] Furthermore, the total environmental stress score of the j-th adolescent was statistically analyzed. , = ;

[0046] in, The total environmental stress score of the j-th adolescent sample ranges from 0 to 100; P represents the number of different types of environmental stress. Let p be the weight of the p-th type of environmental pressure; This represents the level score of the p-th type of environmental pressure in this sample.

[0047] Furthermore, the weights and level scores of the anxiety behaviors are set based on clinical data;

[0048] The weights and levels of environmental pressure are set manually.

[0049] As described above, by quantifying weights and ranking scores, anxiety behaviors include: sleep disorders, social withdrawal, academic avoidance, irritability, self-deprecation, and physical discomfort; Hypothesis: sleep disorders =0.25, social withdrawal =0.25, weighted sum of 1; sleep disorder =5 points, social withdrawal =10 points, converting anxious behaviors into quantitative scores to quantify the contribution of the behavior to anxiety and avoid misjudgment based on a single behavior. Types of environmental stress, such as academic pressure, family conflict, peer relationships, parent-child communication, school bullying, and environmental pollution, are similar to anxious behaviors and are all calibrated and statistically analyzed.

[0050] Furthermore, some of the data on anxiety behaviors and environmental stress were obtained through monitoring equipment.

[0051] As can be seen from the above description, many anxiety behaviors, such as sleep disorders and irritability, can be monitored by data obtained through monitoring devices. Sleep disorders can be monitored by sleep monitoring bracelets, and irritability can be monitored by heart rate monitoring bracelets; environmental pollution can also be monitored by local environmental monitoring stations.

[0052] Furthermore, deep learning models also include the mean squared error loss function, Loss;

[0053] Loss= Where M is the total number of samples; T is the total number of time steps; Let represent the true anxiety level of the j-th sample at time t.

[0054] As described above, Loss is the overall loss of the model. The smaller the loss value, the more accurate the prediction. By minimizing the error between the predicted value and the true value, the model parameters are optimized, such as the hidden layer weights of LSTM and the attention coefficient of Attention, to ensure prediction accuracy.

[0055] Furthermore, It is obtained by physicians through clinical scale evaluation.

[0056] Furthermore, =0.2.

[0057] As can be seen from the above description, =0.2, meaning that recent data has a higher weight, which is consistent with the short-term fluctuation characteristics of anxiety. That is, many anxieties have a decay period, and the duration of anxiety is generally 1-3 months.

[0058] Furthermore, when the output predicted anxiety level is medium or high risk, the corresponding adolescent is retested manually.

[0059] As can be seen from the above description, manual retesting can determine the predictive effect. The model can only predict and provides a probabilistic conclusion, not a decisive result.

[0060] Example 1

[0061] A deep learning-based method for predicting adolescent anxiety includes:

[0062] Quantification of gut microbiota characteristics

[0063] Calculate the relative abundance of the bacterial community: = ;

[0064] in, Let be the relative abundance of the k-th bacterial species in the j-th adolescent sample, with a value ranging from 0 to 1. The sequencing count of the k-th bacterial species in the sample is obtained by sequencing the stool sample collected from the adolescent, for example, using a home sampling kit, which avoids invasiveness to the adolescent and protects their privacy. The stool sample is sequenced using metagenomic sequencing, and the bacterial count is obtained using methods such as 16S rRNA sequencing. The total sequencing count of all core microbial communities in the sample can be the total number of microbial communities detected in the stool, or it can be a selection of certain microbial communities associated with anxiety. The microbial communities selected as needed are called K, and then the total count is summed. By converting the absolute count of microbial communities into a proportional value, the difference in sequencing depth between samples is eliminated, which facilitates cross-sectional comparison of the microbial community structure of different adolescents.

[0065] Calculate the Shannon index (microbial diversity index): =- ;

[0066] in, Let K be the microbial diversity index of the j-th adolescent sample; the larger the value, the higher the microbial diversity. K represents the number of microbial species. The base-2 logarithm is used to standardize the scope of diversity calculations;

[0067] Quantifying anxiety behaviors, which include: sleep disturbances, social withdrawal, academic avoidance, irritability, self-deprecation, and physical discomfort. In reality, it may include more than just these six, and additional or fewer behaviors can be added as needed. The total anxiety behavior score of the j-th adolescent is calculated. ; = ;

[0068] in, Let L be the total score of anxiety behavior for the j-th adolescent sample, ranging from 0 to 100; L is the number of types of anxiety behavior. The weight of the l-th type of anxiety behavior, This represents the grade score of the l-th type of anxiety behavior in this sample. The weights and grade scores of the anxiety behaviors are based on clinical data; for example, sleep disorders. =0.25, social withdrawal =0.25, academic avoidance =0.1, irritable =0.15, self-denial =0.1, feeling unwell =0.15, the weight sum can be 1, the specific weight can also be adjusted according to the actual situation, age, region, gender, etc.; Since there are 6 anxiety behaviors, the highest level score for each anxiety behavior is 16 points (for ease of calculation, it can be adjusted to 5 items, 20 points each). The level scores are none = 0 points, mild = 4 points, moderate = 8 points, high = 12 points, severe = 16 points. If there are 5 behaviors, the level scores are none = 0 points, mild = 5 points, moderate = 10 points, high = 15 points, severe = 20 points. It can be adjusted as needed or for ease of calculation, as long as the total score does not exceed 100 points. The same applies to environmental stress characteristics.

[0069] Environmental stress characteristics are quantified. Environmental stress includes: academic stress, family conflict, peer relationships, parent-child communication, school bullying, and environmental pollution; in reality, it may include more than just these six, and additional or fewer factors can be added as needed; the total environmental stress score of the j-th adolescent is calculated. ; = ;

[0070] in, The total environmental stress score of the j-th adolescent sample ranges from 0 to 100; P represents the number of different types of environmental stress. Let p be the weight of the p-th type of environmental pressure; This represents the level score of the p-th type of environmental stress in this sample. The weights and level scores of the environmental stresses are set manually. For example, academic stress. =0.1, family conflict =0.15, peer relationships =0.1, Parent-child communication =0.15, school bullying =0.4, environmental pollution =0.1, the weight sum can be 1, the specific weight can also be adjusted according to the actual situation, age, region, gender, etc.; Since there are 6 environmental stresses, the highest level score for each environmental stress is 16 points (for ease of calculation, it can be adjusted to 5 items, 20 points each). The level scores are none = 0 points, mild = 4 points, moderate = 8 points, high = 12 points, severe = 16 points. If there are 5 behaviors, the level scores are none = 0 points, mild = 5 points, moderate = 10 points, high = 15 points, severe = 20 points. It can be adjusted as needed or for ease of calculation, as long as the total score does not exceed 100 points.

[0071] The aforementioned anxiety behaviors and environmental stress data can be obtained as needed, allowing for self-assessment by adolescents or assessment by parents. Other data should be obtained through devices whenever possible to ensure sufficient objectivity. For example, sleep disorders can be assessed using sleep monitoring bracelets, and irritability can be assessed using heart rate monitoring bracelets; environmental pollution data can also be obtained through local environmental monitoring stations.

[0072] Calculate the time-series weighting coefficients. = , Let t be the temporal weight of the j-th adolescent sample at time t; t is the time point, and T is the total number of time steps. The time decay coefficient, =0.2 (a value of 0.2 means that recent data has a higher weight, which is consistent with the short-term fluctuation characteristics of anxiety, that is, many anxieties have a decay period, and the duration of anxiety is generally 1-3 months). Of course, it can also be adjusted according to the actual situation to meet the design requirements; e is the natural constant, and s is the number of iterations from 1 to T.

[0073] Feature fusion = , They are respectively The value at time t;

[0074] The model prediction uses an LSTM+Attention deep learning model, which includes a mean squared error loss function (Loss); Loss = Where M is the total number of samples; T is the total number of time steps; Let represent the true anxiety level of the j-th sample at time t. The data is obtained by physicians through clinical scale evaluations. Loss is the overall loss of the model; the smaller the loss value, the more accurate the prediction. By minimizing the error between the predicted value and the true value, the model parameters are optimized, such as the hidden layer weights of LSTM and the attention coefficient of Attention, to ensure prediction accuracy.

[0075] The fused features are used as input, and the model outputs a prediction of the anxiety level. The output value is 0-100; when A value <30 indicates low risk; a value ≤30 indicates low risk. <60 indicates medium risk. A score of ≥60 indicates high risk. Continuous predicted values ​​are transformed into intuitive risk levels, facilitating early intervention by schools and families. When the predicted anxiety level is medium or high risk, the adolescent is manually retested.

[0076] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A deep learning-based method for predicting adolescent anxiety, characterized in that, include: Quantification of gut microbiota characteristics, calculation of relative abundance and diversity index of microbiota; = ； =- ； Let be the relative abundance of the k-th bacterial species in the j-th adolescent sample. The sequencing count of the kth bacterial species in this sample; This is the total sequencing count of all core bacterial communities in the sample. Let K be the microbial diversity index of the j-th adolescent sample; K is the number of microbial species. Quantifying anxiety behavior characteristics, and calculating the total anxiety behavior score of the j-th adolescent. , = ; in, Let L be the total score of anxiety behavior for the j-th adolescent sample, ranging from 0 to 100; L is the number of types of anxiety behavior. The weight of the l-th type of anxiety behavior, The score is the level score for the l-th type of anxiety behavior in this sample; Quantifying environmental stress characteristics and calculating the total environmental stress score of the j-th adolescent. , = ; in, The total environmental stress score of the j-th adolescent sample ranges from 0 to 100; P represents the number of different types of environmental stress. Let p be the weight of the p-th type of environmental pressure; This represents the level score of the p-th type of environmental pressure in this sample. Calculate the time-series weighting coefficients. = , Let t be the temporal weight of the j-th adolescent sample at time t; t is the time point, and T is the total number of time steps. is the time decay coefficient, e is the natural constant, and s is the time decay coefficient starting from 1 and going sequentially up to T; Feature fusion = , They are respectively The value at time t; The model predicts using an LSTM+Attention deep learning model, with the fused features... As input, the model outputs a prediction of anxiety level. ;when A value <30 indicates low risk; a value ≤30 indicates low risk. <60 indicates medium risk. A value of 60 or higher indicates a high risk level.

2. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, When quantifying gut microbiota characteristics, microbiota information is obtained by periodically collecting fecal samples from adolescents and performing metagenomic sequencing.

3. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, The weights and level scores of the anxiety behaviors were determined based on clinical data. The weights and levels of environmental pressure are set manually.

4. The method for predicting adolescent anxiety based on deep learning according to claim 3, characterized in that, Some of the data on anxiety behaviors and environmental stress were obtained through monitoring equipment.

5. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, Deep learning models also include the mean squared error loss function, Loss; Loss= Where M is the total number of samples; T is the total number of time steps; Let represent the true anxiety level of the j-th sample at time t.

6. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, It is obtained by physicians through clinical scale evaluation.

7. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, =0.2。 8. The method for predicting adolescent anxiety based on deep learning according to claim 1, characterized in that, When the predicted anxiety level is medium or high risk, the corresponding adolescent is retested manually.

Images