Artificial intelligence-based intelligent assessment system for thyroid function abnormalities

Abstract

The application relates to the technical field of medical data analysis, in particular to an intelligent thyroid function abnormality evaluation system based on artificial intelligence. In a data acquisition module, the growth and behavior parameter time series data of a new-born to be measured and historical new-borns are collected; because the initial signs of the new-borns are similar and are not affected by genetic factors such as genetic diseases, the subsequent growth trends are similar, so in a relationship model fitting module, the growth parameter time series variation of the historical samples is calculated, a dynamic relationship model of the growth parameter and the behavior parameter is fitted, and the quantitative correlation between the behavior and the growth parameter variation is realized; a characteristic parameter screening module substitutes the behavior parameter of the sample to be measured into the model, predicts the growth parameter theoretical value, compares the actual data, and screens out characteristic parameters sensitive to the thyroid function, so that the interference of redundant data is reduced; and a risk prediction module trains a neural network based on samples with known thyroid risk scores in a key characteristic parameter space, focuses on high-correlation variables, and improves the risk score accuracy.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, specifically to an intelligent assessment system for thyroid dysfunction based on artificial intelligence. Background Technology

[0002] In the first few days to several weeks after birth, some healthy full-term infants may experience a transient increase in thyroid-stimulating hormone (TSH) levels. This is because their hypothalamic-pituitary-thyroid axis (HPT axis) is not yet fully mature, and due to residual maternal hormones and the effects of adapting to the external environment, they may also experience this phenomenon. Thyroid hormones play a crucial role in the growth and development of newborns (especially the development of the nervous system). If not diagnosed and treated promptly, it can lead to irreversible sequelae such as intellectual disability, growth retardation, and hearing impairment.

[0003] Current technology often uses the TSH level in a newborn's heel prick blood sample to compare with a preset diagnostic threshold to make a diagnosis. However, elevated TSH is sometimes not true hypothyroidism, but a physiological compensatory response. Therefore, some samples need to be retested after a period of time. Furthermore, thyroid dysfunction is usually related to the newborn's growth status. So, if the TSH level is relied on alone when warning of thyroid dysfunction in newborns, it may be difficult to obtain an accurate assessment of the abnormality in newborn samples with elevated TSH levels that have not reached the preset diagnostic threshold while waiting for retesting, and may even delay treatment. Summary of the Invention

[0004] To address the issue that elevated TSH levels are sometimes not true hypothyroidism but rather a physiological compensatory response, requiring some samples to be retested after a period of time, and because thyroid dysfunction is often related to a newborn's growth status, relying solely on TSH levels for early warning of neonatal thyroid dysfunction may lead to inaccurate assessment of abnormalities in newborns with elevated TSH levels that have not reached the preset diagnostic threshold, potentially delaying treatment. The purpose of this invention is to provide an artificial intelligence-based intelligent assessment system for thyroid dysfunction. The specific technical solution adopted is as follows: This invention proposes an intelligent assessment system for thyroid dysfunction based on artificial intelligence, the system comprising: The data acquisition module is used to acquire sample data, which includes test samples of newborns to be tested and historical samples of historical newborns. Each sample data contains time-series data of multiple growth parameters and time-series data of multiple behavioral parameters. The relational model fitting module is used to calculate the time series data of the change of each growth parameter based on the time series data of each growth parameter in historical samples; and to fit the relational model corresponding to each behavior parameter based on the correlation between the change of each behavior parameter and the change of the growth parameter in historical samples. The feature parameter filtering module is used to substitute the values ​​of the behavioral parameters of the test sample into the corresponding relational model and combine them with the numerical features of the growth parameters to obtain the predicted time series data of each growth parameter of the test sample; based on the difference change characteristics between the time series data of the growth parameters of the test sample and the predicted time series data, feature parameters are filtered from all growth parameters. The risk prediction module is used to numerically train a neural network based on training samples with known thyroid risk scores and feature parameters, thereby predicting the risk of the test samples.

[0005] Furthermore, the method for obtaining the time series data of the change includes: In each historical sample, for any two adjacent time points, the difference between the value of the growth parameter at the later time point and the value of the growth parameter at the previous time point is taken as the change at the later time point, thus obtaining the time series data of the change of each growth parameter.

[0006] Furthermore, the method for obtaining the relational model includes: The types of behavioral parameters include feeding amount, crying duration, and sleep duration; A relational model is constructed based on the correlation between behavioral parameters and growth parameters. The relational model is as follows: in, Indicates the amount of food fed; Indicates the duration of crying; Indicates sleep duration; Characteristic coefficients representing feeding amount; Characteristic coefficients representing the duration of crying; Characteristic coefficients representing sleep duration; This represents the change in the k-th growth parameter; Obtain the relation array for each historical sample at each time point. The relation array includes: feeding amount, crying duration, sleep duration, and the change in the kth growth parameter. The relation model is fitted using the relation array of all historical samples under each growth parameter and the least squares method to obtain all feature coefficients, thereby obtaining the relation model corresponding to each growth parameter.

[0007] Furthermore, the method for obtaining the predicted time series data includes: In the sample to be tested, obtain the relation array at each time point, including: feeding amount, crying duration, and sleep duration. Substitute the relation array at each time point into the relation model corresponding to each behavioral parameter to obtain the change at each time point under each growth parameter. The value at the first time step in the time series data of each growth parameter of the test sample is taken as the target value. The sum of the target value and the change in each growth parameter at the next adjacent time step is taken as the new target value at the next adjacent time step. This process is repeated until the target values ​​at all time steps are obtained. The target values ​​at all time steps are then fitted according to the time series to obtain the predicted time series data of the test sample under each growth parameter.

[0008] Furthermore, the method for obtaining the feature parameters includes: The absolute value of the difference between the time series data of the test sample under each growth parameter and the predicted time series data at the same time point is used as the growth bias factor at each time point. Analyze the differences in growth deviation factors between adjacent time points to determine the quantitative anomaly factors; By combining the variation and fluctuation characteristics of the quantity anomaly factor and the growth deviation factor, the possible anomaly indicators of the sample under each growth parameter are obtained. Growth parameters whose abnormality probability index exceeds the preset abnormality threshold are used as feature parameters.

[0009] Furthermore, the method for obtaining the quantity anomaly factor includes: At each two adjacent time points, the difference between the growth deviation factor of the next time point and the previous time point is calculated as the deviation growth factor. Among all the deviation growth factors at adjacent time points, the ratio of the number of deviation growth factors with positive values ​​to the total number of deviation growth factors is taken as the quantity anomaly factor.

[0010] Furthermore, the method for obtaining the possible anomaly indicators includes: The variance of the absolute values ​​of all deviation growth factors is taken as the volatility factor. The sum of the volatility factor and the quantity anomaly factor is normalized and used as the possible anomaly index of the sample under each growth parameter.

[0011] Furthermore, the process of numerically training a neural network based on training samples with known thyroid risk scores under feature parameters to predict the risk of the test sample includes: The training samples are used as the input to the neural network based on the values ​​of the feature parameters, and the thyroid risk scores of the training samples are used as the output of the neural network to obtain a trained neural network. The value of the feature parameters of the test sample at the current time is used as the input of the trained neural network, and the output is the thyroid risk score of the test sample. A thyroid risk warning is issued when the thyroid risk score of the sample to be tested is greater than or equal to the preset risk threshold.

[0012] Furthermore, the thyroid risk score is a normalized value, with a preset risk threshold of 0.65.

[0013] Furthermore, the neural network is a CNN.

[0014] The present invention has the following beneficial effects: By collecting time-series data on growth and behavioral parameters of the newborn to be tested and historical newborns, a data foundation can be provided for constructing an assessment system covering multiple physiological and behavioral dimensions, as well as for subsequent abnormality warnings. Since newborns generally have similar initial physical characteristics during their growth process, such as weight, length, head circumference, and gestational age, their subsequent growth trends are roughly similar in the absence of genetic diseases or other genetic factors. Therefore, for the newborn to be tested, the changes in growth parameters can be predicted based on the growth parameters of historical newborns. This facilitates comparison of predicted data with actual data in subsequent modules, thereby identifying potential abnormalities in the growth trend of the newborn to be tested. Therefore, in the relational model fitting module, the time-series changes of each growth parameter in the historical samples are calculated, and a dynamic relational model between it and the behavioral parameters is fitted to characterize the behavioral-growth correlation characteristics of the newborn population, realizing a quantitative correlation between behavioral parameters and changes in growth parameters. Furthermore, in the feature parameter selection module, the behavioral parameters of the test samples are substituted into the relational model to predict their theoretical growth parameters. The differences in time-series data between the predicted and actual growth parameters are calculated, and feature parameters sensitive to thyroid function are selected, reducing redundant data interference and improving model efficiency. Finally, in the risk prediction module, based on training samples with known thyroid risk scores, a neural network is trained in the selected key feature parameter space, focusing on variables highly correlated with risk and avoiding interference from irrelevant parameters, significantly improving the accuracy of risk scoring for the test samples. Attached Figure Description

[0015] Figure 1 This is a system block diagram of an artificial intelligence-based intelligent assessment system for thyroid dysfunction provided in one embodiment of the present invention; Figure 2 This is a flowchart of a method for obtaining feature parameters according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the system structure of an artificial intelligence-based intelligent assessment system for thyroid dysfunction provided in one embodiment of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent assessment system for thyroid dysfunction based on artificial intelligence provided by this invention.

[0017] Please see Figure 1 The diagram illustrates a system block diagram of an artificial intelligence-based intelligent assessment system for thyroid dysfunction provided by an embodiment of the present invention. The system includes: a data acquisition module 101, a relationship model fitting module 102, a feature parameter screening module 103, and a risk prediction module 104.

[0018] The data acquisition module 101 is used to acquire sample data, which includes test samples of newborns to be tested and historical samples of historical newborns. Each sample data contains time-series data of multiple growth parameters and time-series data of multiple behavioral parameters.

[0019] Thyroid hormones play a crucial role in the growth and development of newborns, primarily regulating their metabolism, promoting intellectual development, and maintaining the cardiovascular system. Newborns are in a peak growth period, and if there is a risk of diagnosis, timely intervention is necessary to avoid serious consequences. However, newborns' thyroid function is usually not fully developed, and initial thyroid function tests are often relatively high. Administering medication directly to newborns with unclear thyroid function may cause additional side effects, so repeated retesting is necessary. During the waiting period for retesting, the newborn's condition should be closely monitored. This period typically lasts from a few days to over ten days. If the risk of dynamic thyroid function abnormalities can be assessed during this waiting period, timely intervention can be initiated for newborns with abnormal thyroid function.

[0020] For current newborns, TSH levels are obtained through heel prick blood sampling. A commonly used diagnostic range for elevated TSH levels, insufficient for a definitive diagnosis, is greater than 4 mIU / L but less than 20 mIU / L. Therefore, newborns with TSH levels within this range are considered as candidates for testing. Healthy newborns with the same initial growth status as the candidate are considered as historical newborns, meaning those with identical initial growth parameters. In this embodiment, growth parameters can be set as weight, length, head circumference, gestational age, etc. Growth parameter data can be obtained through electronic medical records or health records. Thus, the waiting period between the candidate and follow-up examinations can assess the difference in growth performance between the candidate and normal historical newborns, thereby determining the risk of thyroid dysfunction.

[0021] Given that newborns with the same initial growth state generally have similar subsequent growth trends in the absence of genetic diseases or other genetic factors, after excluding congenital factors, the environmental factors that cause differences in growth and development mainly include differences in energy intake, energy consumption, and energy conversion. Representative behavioral activities include feeding amount, crying duration, and sleep duration. Therefore, the behavioral parameters in this embodiment of the invention include feeding amount, crying duration, and sleep duration, which can also be recorded and obtained through electronic medical records or health records.

[0022] In this embodiment of the invention, the time-series data is collected daily, and the duration is set to the waiting time for re-examination of the newborn to be tested. The specific duration can be adjusted according to the implementation scenario. Thus, all sample data can be obtained, including the test samples of the newborn to be tested and the historical samples of historical newborns. Each sample data contains time-series data of three behavioral parameters and four growth parameters. The values ​​of the behavioral parameters in the time-series data of behavioral parameters and the values ​​of the growth parameters in the time-series data of growth parameters should be normalized values. The specific normalization method is a known technique, and maximum and minimum value normalization can be used. The maximum and minimum values ​​can be selected from the database based on the values ​​of the behavioral parameters and growth parameters in the time-series data.

[0023] It should be noted that the collection and acquisition of newborn personal information data in this embodiment of the invention are all authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0024] The relational model fitting module 102 is used to calculate the time series data of the change of each growth parameter based on the time series data of each growth parameter in the historical samples; and to fit the relational model corresponding to each behavior parameter based on the correlation between the change of each behavior parameter and the change of the growth parameter in the historical samples.

[0025] By assessing the differences in symptoms between newborns awaiting re-examination and those of normal historical newborns during the waiting period, the risk of thyroid dysfunction can be obtained. Newborns develop rapidly and still show significant growth changes during the waiting period, such as weight gain and height increase. Normally, newborns with similar growth parameters at birth exhibit similar growth characteristics during the same growth cycle. Therefore, differential analysis between newborns awaiting re-examination and normal historical newborns can reveal the differences in various growth parameters suggestive of thyroid dysfunction.

[0026] Furthermore, if the current newborn shows significant differences in growth changes compared to newborns in the same historical period during continuous monitoring, these differences may reflect underlying diseases that the current newborn may have. Therefore, in this module, the time series data of the change in each growth parameter is first calculated based on the time series data of each growth parameter in the historical sample. This time series data characterizes the ideal growth change characteristics that the newborn under test should have under normal circumstances.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining time series data of change includes: In each historical sample, for any two adjacent time points, the difference between the value of the growth parameter at the later time point and the value of the growth parameter at the previous time point is taken as the change at the later time point, thus obtaining the time series data of the change of each growth parameter. The time series data of the change of growth parameters characterizes the trend and characteristics of each growth parameter of the newborn under normal conditions without thyroid dysfunction.

[0028] During growth and development, the environmental factors causing differences in growth mainly include differences in energy intake, energy consumption, and energy conversion. Specifically, newborns' growth hormone is mainly secreted in pulses during deep sleep. Therefore, the length of sleep is the main conversion process of the remaining energy intake of newborns, and the longer the sleep duration and the better the quality, the more growth hormone is generally produced, and the higher the increase in the newborn's growth parameters. In addition, the amount of food a newborn is fed represents the amount of energy input, while the duration of crying represents the amount of energy consumption. When the amount of energy input from feeding is higher than the energy consumed by crying, the newborn's accumulated growth will be greater. Therefore, in this module, a growth model between energy change and growth change can be established based on the correlation between changes in behavioral parameters and growth parameters in historical samples. The model parameters are then fitted and solved to obtain the relationship model corresponding to each behavioral parameter, which can be used to more intuitively reflect the data relationship between the characteristics of changes in behavioral parameters and growth parameters.

[0029] Preferably, in one embodiment of the present invention, the method for obtaining the relation model includes: Based on the aforementioned logical analysis, it is known that sleep duration, a behavioral parameter, directly affects the growth parameters, and the increase in sleep duration and growth parameters should show a positive correlation. Meanwhile, feeding amount and crying duration, behavioral parameters, show an inverse logical relationship with the increase in growth parameters, and there is a partial offsetting effect between the two. Therefore, based on these characteristics, a relational model is constructed to illustrate the correlation between behavioral parameters and growth parameters. The relational model is as follows: in, Indicates the amount of food fed; Indicates the duration of crying; Indicates sleep duration; Characteristic coefficients representing feeding amount; Characteristic coefficients representing the duration of crying; Characteristic coefficients representing sleep duration; This represents the change in the k-th growth parameter.

[0030] In the relational model, This indicates the portion of energy input exceeding energy consumption. A larger value suggests a more significant increase in growth parameters. Therefore, this indicator is correlated with the contribution of sleep duration. Multiply them to obtain the final change in growth parameters.

[0031] Then, the relational array for each historical sample at each time point (in this embodiment of the invention, time is in days) is obtained. The relational array includes: feeding amount, crying duration, sleep duration, and the change in the k-th growth parameter. It should be noted that in this embodiment of the invention, the change in each growth parameter at the first time point can be set to 0.

[0032] Based on the relational arrays of all historical samples under each growth parameter and the least squares method, the aforementioned relational model is fitted to obtain all the characteristic coefficients. This allows us to obtain a complete relational model for each growth parameter.

[0033] It should be noted that the least squares method is a well-known technique, and the specific process will not be elaborated here.

[0034] Thus, each growth parameter corresponds to a relational model.

[0035] The feature parameter filtering module 103 is used to substitute the values ​​of the behavioral parameters of the test sample into the corresponding relational model and combine them with the numerical features of the growth parameters to obtain the predicted time series data of each growth parameter of the test sample; based on the difference change characteristics between the time series data of the growth parameters of the test sample and the predicted time series data, feature parameters are filtered from all growth parameters.

[0036] In this module, based on the relationship model corresponding to each growth parameter and the behavioral and growth parameters of the sample to be tested, the numerical change characteristics that each growth parameter of the sample to be tested should have can be predicted, and the predicted time series data of each growth parameter of the sample to be tested can be obtained, which can be used to characterize the numerical change characteristics that each growth parameter of the newborn to be tested should have under ideal conditions.

[0037] Preferably, in one embodiment of the present invention, the method for obtaining predicted time series data includes: In the sample to be tested, obtain the relation array at each time point, including: feeding amount, crying duration, and sleep duration. Substitute the relation array at each time point into the relation model corresponding to each behavioral parameter to obtain the change at each time point under each growth parameter.

[0038] Then, the value at the first time step in the time series data of each growth parameter of the test sample is taken as the target value. The sum of the target value and the change in each growth parameter at the next adjacent time step is taken as the new target value at the next adjacent time step. This process is repeated until the target values ​​at all time steps are obtained. Then, the target values ​​at all time steps are fitted according to the time series to obtain the predicted time series data of the test sample under each growth parameter.

[0039] To illustrate the aforementioned process, let's take weight as an example of a growth parameter. The actual time-series data of the weight of the sample to be tested are, for example, 0.2, 0.26, and 0.33. The changes at each time point are calculated by the relational model to be 0, 0.05, and 0.08. In obtaining the predicted time-series data of weight, we first take the weight at the first time point, which is 0.2, as the target value. Then, we take the sum of 0.2 and the change at the second time point, which is 0.05, as the predicted weight at the second time point, which is also the new target value. Finally, we take the sum of the new target value 0.25 and the change at the third time point, which is 0.08, as the predicted weight at the third time point, which is 0.33. Thus, we obtain the predicted time-series data of the sample to be tested under the growth parameter of weight: 0.2, 0.25, and 0.33.

[0040] At this point, the predicted time series data of the test sample under each growth parameter can be obtained.

[0041] Because hypothyroidism is a common type of thyroid dysfunction in newborns, and thyroid hormones produced by the thyroid gland are crucial for the function of growth hormone, a difference between the actual and predicted values ​​of growth parameters indicates a discrepancy between the newborn's expected and actual growth, which may be caused by thyroid dysfunction. However, in normal newborns, as thyroid function matures, growth parameters gradually approach normal levels, which is the ideal situation. Therefore, in this module, after obtaining the predicted time-series data for each growth parameter of the sample, feature parameters—that is, growth parameters with abnormal characteristics—are selected from all growth parameters based on the changes in the difference between the time-series data and the predicted time-series data. These parameters can more significantly indicate the possible abnormal growth status of the newborn.

[0042] Preferably, in one embodiment of the present invention, the method for obtaining the feature parameters includes: Please see Figure 2 The diagram illustrates a method flowchart for obtaining feature parameters according to an embodiment of the present invention, which includes the following steps: Step S301: Compare the differences between the time series data of the test sample under each growth parameter and the predicted time series data at the same time point, and determine the growth deviation factor corresponding to each time point.

[0043] The time series data of the test sample under each growth parameter represents the actual growth changes, while the corresponding predicted time series data represents the growth changes that should be present under ideal conditions, that is, when the sample is healthy and disease-free. Therefore, the absolute value of the difference between the time series data of the test sample under each growth parameter and the predicted time series data at the same moment is used as the growth deviation factor at each moment. The larger the growth deviation factor, the more obvious the deviation between the actual growth data and the ideal growth data is at that moment, and the newborn under test is highly likely to have hypothyroidism at that moment.

[0044] Step S302: Analyze the differences in growth deviation factors between adjacent time points to determine the quantitative anomaly factors.

[0045] If the newborn being tested has normal growth, then even if there is a difference between the actual value and the predicted value of the growth index, it will gradually approach the predicted value over time. In other words, the growth deviation factor of the test sample at each time point will gradually decrease over time. Conversely, if the growth deviation factor gradually increases over time, then the possibility of the test sample having an abnormal growth state will be higher.

[0046] Therefore, based on the aforementioned logic, at each two adjacent time points, the difference between the growth deviation factor of the later time point and the previous time point is calculated as the deviation growth factor. If the deviation growth factor is positive, it means that the growth deviation factor is also increasing with time, which is regarded as the true value of the growth parameter deviating more and more from the ideal value over time. Therefore, among all the deviation growth factors at adjacent time points, the ratio of the number of deviation growth factors with positive values ​​to the total number of deviation growth factors is taken as the quantity anomaly factor. The larger the quantity anomaly factor, the higher the probability of an anomaly of this growth parameter in the sample to be tested.

[0047] Step S303: Combine the variation and fluctuation characteristics of the quantity anomaly factor and the growth deviation factor to obtain the possible anomaly indicators of the sample under each growth parameter.

[0048] Furthermore, as the thyroid function of newborns gradually develops and matures, under normal circumstances, even if the development rate is slower than normal, the difference between the corresponding growth parameters and the normally predicted growth parameters should gradually decrease. At this time, if the fluctuation range of the growth deviation factor corresponding to a certain growth parameter of the test sample is large, it indicates that the growth parameter of the newborn under test has a high risk of abnormality.

[0049] Therefore, the variance of the absolute values ​​of all deviation growth factors of the test sample under a certain growth parameter is calculated as the fluctuation factor. The larger the fluctuation factor, the more obvious the fluctuation of the deviation growth factor, and the higher the probability of anomaly of that growth parameter. Based on the logic in step S302, it is known that the larger the value of the quantity anomaly factor, the greater the probability of anomaly of the corresponding growth parameter. Therefore, the normalized value of the sum of the fluctuation factor and the quantity anomaly factor is used as the anomaly probability index of the test sample under each growth parameter. The larger the anomaly probability index, the more significant the anomaly state of that growth parameter in the test sample. Normalization is a technique well known to those skilled in the art. The normalization function can be linear normalization or standard normalization, such as maximum and minimum value normalization. The maximum and minimum values ​​can be obtained from a large amount of historical sample data.

[0050] Step S304: Select characteristic parameters from all growth parameters of the sample to be tested based on the anomaly probability index.

[0051] Based on the analysis in step S303, it can be seen that the larger the abnormality index of a certain growth parameter, the more significant the abnormal state of that growth parameter. Therefore, among all growth parameters of the sample to be tested, the growth parameters with an abnormality index greater than the preset abnormality threshold are taken as feature parameters.

[0052] It should be noted that in this embodiment of the invention, the preset abnormal threshold is 0.7. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0053] The risk prediction module 104 is used to numerically train a neural network based on training samples with known thyroid risk scores and feature parameters, thereby predicting the risk of the test sample.

[0054] Based on the aforementioned module, growth parameters that can significantly characterize abnormal growth status in the sample to be tested can be obtained, namely, feature parameters. Feature parameters can clearly reflect the risk of hypothyroidism in the newborn to be tested. In order to further realize artificial intelligence recognition, in this module of the present invention, a neural network can be used to learn the values ​​of the feature parameters of training samples with known thyroid risk, thereby obtaining a trained neural network for risk prediction of the sample to be tested.

[0055] Preferably, in one embodiment of the present invention, a neural network is numerically trained based on training samples with known risks under feature parameters to predict the risk of the test sample, including: The training samples are used as input to the neural network based on their feature parameters, and their thyroid risk scores are used as output. This results in a trained neural network. The test sample's feature parameter values ​​at the current time are then used as input to the trained neural network to output its thyroid risk score. Finally, a thyroid risk warning is triggered when the test sample's thyroid risk score is greater than or equal to a preset risk threshold.

[0056] It should be noted that, in this embodiment of the present invention, the value of the thyroid risk score should be a normalized value. Therefore, the preset risk threshold can be 0.65. The specific value can be adjusted according to the implementation scenario and is not limited here. The neural network in this embodiment of the present invention can be a CNN. In other embodiments of the present invention, other types of neural networks can also be used and are not limited here. The training process of the neural network is a well-known technology and will not be described in detail here.

[0057] Unless otherwise specified, the normalized values ​​in the embodiments of this invention are generally in the range of 0 to 1.

[0058] In summary, by collecting time-series data on growth and behavioral parameters of the newborn to be tested and historical newborns, a data foundation can be provided for constructing an assessment system covering multiple physiological and behavioral dimensions, as well as for subsequent abnormality warnings. Since newborns generally have similar initial physical characteristics during their growth process, such as weight, length, head circumference, and gestational age, their subsequent growth trends are roughly similar in the absence of genetic diseases or other gene factors. Therefore, for the newborn to be tested, the changes in growth parameters can be predicted based on the growth parameters of historical newborns. This facilitates comparison of predicted data with actual data in subsequent modules, thereby identifying potential abnormalities in the growth trend of the newborn to be tested. Therefore, in the relational model fitting module, the time-series changes of each growth parameter in the historical samples are calculated, and a dynamic relational model between it and the behavioral parameters is fitted to characterize the behavioral-growth correlation characteristics of the newborn population, realizing a quantitative correlation between behavioral parameters and changes in growth parameters. Furthermore, in the feature parameter selection module, the behavioral parameters of the test samples are substituted into the relational model to predict their theoretical growth parameters. The differences in time-series data between the predicted and actual growth parameters are calculated, and feature parameters sensitive to thyroid function are selected, reducing redundant data interference and improving model efficiency. Finally, in the risk prediction module, based on training samples with known thyroid risk scores, a neural network is trained in the selected key feature parameter space, focusing on variables highly correlated with risk and avoiding interference from irrelevant parameters, significantly improving the accuracy of risk scoring for the test samples.

[0059] An embodiment of the present invention provides an intelligent assessment system for thyroid dysfunction based on artificial intelligence, comprising a processor, a memory, a bus, and a communication interface, wherein the processor, the communication interface, and the memory are connected via the bus; wherein the memory may include high-speed random access memory, the bus may be an ISA bus, PCI bus, or EISA bus, etc., and the processor may be an integrated circuit chip with signal processing capabilities; the memory stores at least one instruction, at least one program, code set, or instruction set, and when the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor, it implements the steps in the various modules of the aforementioned embodiments.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent assessment system for thyroid dysfunction based on artificial intelligence, characterized in that, The system includes: The data acquisition module is used to acquire sample data, which includes test samples of newborns to be tested and historical samples of historical newborns. Each sample data contains time-series data of multiple growth parameters and time-series data of multiple behavioral parameters. The relational model fitting module is used to calculate the time series data of the change of each growth parameter based on the time series data of each growth parameter in historical samples; and to fit the relational model corresponding to each behavior parameter based on the correlation between the change of each behavior parameter and the change of the growth parameter in historical samples. The feature parameter filtering module is used to substitute the values ​​of the behavioral parameters of the test sample into the corresponding relational model and combine them with the numerical features of the growth parameters to obtain the predicted time series data of each growth parameter of the test sample; based on the difference change characteristics between the time series data of the growth parameters of the test sample and the predicted time series data, feature parameters are filtered from all growth parameters. The risk prediction module is used to numerically train a neural network based on training samples with known thyroid risk scores and feature parameters, thereby predicting the risk of the test samples.

2. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the time series data of the change includes: In each historical sample, for any two adjacent time points, the difference between the value of the growth parameter at the later time point and the value of the growth parameter at the previous time point is taken as the change at the later time point, thus obtaining the time series data of the change of each growth parameter.

3. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the relational model includes: The types of behavioral parameters include feeding amount, crying duration, and sleep duration; A relational model is constructed based on the correlation between behavioral parameters and growth parameters. The relational model is as follows: in, Indicates the amount of food fed; Indicates the duration of crying; Indicates sleep duration; Characteristic coefficients representing feeding amount; Characteristic coefficients representing the duration of crying; Characteristic coefficients representing sleep duration; This represents the change in the k-th growth parameter; Obtain the relation array for each historical sample at each time point. The relation array includes: feeding amount, crying duration, sleep duration, and the change in the kth growth parameter. The relation model is fitted using the relation array of all historical samples under each growth parameter and the least squares method to obtain all feature coefficients, thereby obtaining the relation model corresponding to each growth parameter.

4. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the predicted time series data includes: In the sample to be tested, obtain the relation array at each time point, including: feeding amount, crying duration, and sleep duration. Substitute the relation array at each time point into the relation model corresponding to each behavioral parameter to obtain the change at each time point under each growth parameter. The value at the first time step in the time series data of each growth parameter of the test sample is taken as the target value. The sum of the target value and the change in each growth parameter at the next adjacent time step is taken as the new target value at the next adjacent time step. This process is repeated until the target values ​​at all time steps are obtained. The target values ​​at all time steps are then fitted according to the time series to obtain the predicted time series data of the test sample under each growth parameter.

5. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the feature parameters includes: The absolute value of the difference between the time series data of the test sample under each growth parameter and the predicted time series data at the same time point is used as the growth bias factor at each time point. Analyze the differences in growth deviation factors between adjacent time points to determine the quantitative anomaly factors; By combining the variation and fluctuation characteristics of the quantity anomaly factor and the growth deviation factor, the possible anomaly indicators of the sample under each growth parameter are obtained. Growth parameters whose abnormality probability index exceeds the preset abnormality threshold are used as feature parameters.

6. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 5, characterized in that, The method for obtaining the quantity anomaly factor includes: At each two adjacent time points, the difference between the growth deviation factor of the next time point and the previous time point is calculated as the deviation growth factor. Among all the deviation growth factors at adjacent time points, the ratio of the number of deviation growth factors with positive values ​​to the total number of deviation growth factors is taken as the quantity anomaly factor.

7. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 5, characterized in that, The methods for obtaining the possible indicators of anomalies include: The variance of the absolute values ​​of all deviation growth factors is taken as the volatility factor. The sum of the volatility factor and the quantity anomaly factor is normalized and used as the possible anomaly index of the sample under each growth parameter.

8. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 1, characterized in that, The method of numerically training a neural network based on training samples with known thyroid risk scores under feature parameters to predict the risk of test samples includes: The training samples are used as the input to the neural network based on the values ​​of the feature parameters, and the thyroid risk scores of the training samples are used as the output of the neural network to obtain a trained neural network. The value of the feature parameters of the test sample at the current time is used as the input of the trained neural network, and the output is the thyroid risk score of the test sample. A thyroid risk warning is issued when the thyroid risk score of the sample to be tested is greater than or equal to the preset risk threshold.

9. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 8, characterized in that, The thyroid risk score is a normalized value, with a preset risk threshold of 0.

65.

10. The intelligent assessment system for thyroid dysfunction based on artificial intelligence according to claim 8, characterized in that, The neural network is a CNN.

Images