Periodontal disease-high blood sugar bidirectional feature analysis system based on deep learning

Abstract

The present application relates to the technical field of clinical medicine and pathophysiology, and in particular to a periodontal disease-high blood sugar bidirectional feature analysis system based on deep learning. The system obtains an abnormal depth constant through oral case data, extracts an instantaneous blood sugar concentration value at a preset time step, and calculates free blood sugar increment by calculating the difference between the reference blood sugar concentration; the single-step blood sugar metabolism ratio is obtained by weighting the abnormal depth constant and the free blood sugar increment, the cumulative blood sugar metabolism load is obtained by adjusting the unmetabolized blood sugar concentration of the previous time step; the concentration range value and the continuous step number of the blood sugar concentration cycle are extracted, the reference range parameter and the reference continuous step number are calculated based on the multi-layer nested loop, the cumulative blood sugar metabolism load is adjusted, and the associated feature value is obtained by combining the abnormal depth constant; the metabolic decompensation feature vector is obtained by reducing the dimension of the time axis. The problem that periodontal abnormal static data and blood sugar dynamic data cannot be fused into a unified joint feature due to different dimensions and units is solved.

Description

Technical Field

[0001] This invention relates to the fields of clinical medicine and pathophysiology, specifically to a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia. Background Technology

[0002] There is a clear bidirectional promoting relationship between periodontitis and hyperglycemia. Local inflammation of periodontal tissues releases inflammatory factors into the bloodstream, exacerbating insulin resistance. Long-term hyperglycemia or frequent large fluctuations in blood sugar can damage microvascular endothelium and accelerate the breakdown of periodontal supporting tissues. In routine medical data collection, dental probing data shows low-frequency, static spatial examination constants, while continuous blood glucose monitoring data from endocrinology shows high-frequency, dynamic one-dimensional time series.

[0003] Existing early warning systems for processing cross-departmental disease data typically employ independent and parallel evaluation logic. Blood glucose concentration assessment often uses a threshold truncation method, which disrupts the complete cycle of continuous fluctuations and fails to extract closed-loop alternating fluctuation features. Furthermore, existing systems lack spatiotemporal alignment and fusion mechanisms for heterogeneous data, failing to synchronously map static local periodontal abnormality constants as real-time weights to high-frequency dynamic blood glucose time series. Due to these data dimensional mismatches and conflicting units, existing systems struggle to transform cross-organ interactions into joint features at a unified scale. Summary of the Invention

[0004] To address the technical challenges of dimensional mismatch and conflicting units in cross-departmental disease data, this invention aims to provide a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia. The specific technical solution adopted is as follows: This invention proposes a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia, the system comprising: The basic calculation module is used to read dental case data, obtain the probing depth and bleeding record of each valid examination site, and calculate the abnormal depth constant based on the probing depth and bleeding record; extract the instantaneous blood glucose concentration value based on the preset time step, and calculate the free blood glucose increment at each time step by the difference between the instantaneous blood glucose concentration value and the baseline blood glucose concentration. The metabolic accumulation module is used to extract the basal metabolic ratio constant from the collected medical cross dataset; and to weight the basal metabolic ratio constant according to the abnormality depth constant and the free blood glucose increment to obtain the single-step blood glucose metabolism ratio; and to adjust the unmetabolized blood glucose concentration of the previous time step based on the single-step blood glucose metabolism ratio and the free blood glucose increment of each time step to obtain the cumulative blood glucose metabolic load of each time step. The correlation module is used to sort the cumulative glucose metabolic load by time step to obtain each glucose concentration rise and fall cycle and its corresponding concentration range value and duration step; based on the concentration range value and duration step of the multi-layer nested glucose concentration rise and fall cycle, the baseline range parameter and baseline duration step of the nested time step are calculated; the cumulative glucose metabolic load is adjusted according to the baseline range parameter and baseline duration step, and then the correlation feature value is calculated with the anomaly depth constant. The dimensionality reduction module is used to sort the associated feature values ​​according to the time step, traverse the sorted associated feature values ​​and perform time axis dimensionality reduction to obtain the metabolic decompensation feature vector.

[0005] Furthermore, the method for obtaining the abnormal depth constant includes: Read dental case data, extract the probing depth and bleeding records of all valid examination sites based on the case data, and convert the bleeding records into Boolean values; Calculate the product of the probing depth and the Boolean value of the bleeding record for each valid examination site, and then take the mean value based on the number of valid examination sites to obtain the abnormal depth constant.

[0006] Furthermore, the method for obtaining the basal metabolic rate constant includes: The collected medical cross-dataset was divided into a training set and a validation set. The model was trained on the training set based on a grid search algorithm, and its performance was evaluated on the validation set. The constant with the best performance on the validation set was selected as the basal metabolic ratio constant.

[0007] Furthermore, the method for obtaining the single-step glucose metabolism ratio includes: Based on the medical cross-dataset and grid search algorithm, the parameters were iteratively optimized to obtain the periodontal abnormality decay coefficient and the blood glucose concentration decay coefficient. Calculate the product of the periodontal abnormality attenuation coefficient and the abnormality depth constant, and standardize the product to obtain the periodontal abnormality weight factor. This product is negatively correlated with the periodontal abnormality weight factor. Calculate the product of the blood glucose concentration decay coefficient and the free blood glucose increment, standardize the product to obtain the blood glucose weighting factor, and the product is negatively correlated with the blood glucose weighting factor. The basal metabolic rate constant is obtained by multiplying and weighting the periodontal abnormality weight factor and the blood glucose weight factor.

[0008] Furthermore, the method for obtaining the cumulative glucose metabolic load includes: The proportion of unmetabolized blood glucose in a single step is calculated based on the proportion of blood glucose metabolism in each time step. The product of the proportion of unmetabolized blood glucose in a single step and the concentration of unmetabolized blood glucose in the previous time step is calculated and added to the corresponding increase in free blood glucose to obtain the theoretical cumulative blood glucose metabolic load. The theoretical cumulative glucose metabolic load is compared with the preset upper limit constant of unmetabolized glucose concentration, and the minimum value is taken as the cumulative glucose metabolic load at each time step.

[0009] Furthermore, the method for obtaining the concentration range value and the number of duration steps includes: The accumulated glucose metabolic load is sorted according to the time step order to obtain the unmetabolized glucose concentration sequence. The rainflow counting algorithm is used to search the unmetabolized glucose concentration sequence to extract all glucose concentration rise and fall cycles. Calculate the absolute value of the difference between the minimum and maximum values ​​of the cumulative glucose metabolic load for each glucose concentration rise and fall cycle to obtain the corresponding concentration range value; Calculate the difference between the termination time step and the start time step for each blood glucose concentration rise / fall cycle to obtain the corresponding duration step.

[0010] Furthermore, after obtaining the concentration range and the number of duration steps, the process also includes: Extract the starting time step of each blood glucose concentration rise and fall cycle. If the starting time step is within the previous preset number of time step intervals, the corresponding blood glucose concentration rise and fall cycle is removed.

[0011] Furthermore, the method for obtaining the benchmark range parameter and the benchmark duration step count includes: For time steps with multiple nesting layers, the baseline range parameter of the nested time step is calculated based on the concentration range value of the multiple nested blood glucose concentration rise and fall cycles and the number of nesting layers. Extract the maximum duration of the multi-layered nested blood glucose concentration rise and fall cycle as the baseline duration of the nested time steps.

[0012] Furthermore, the method for obtaining the associated feature values ​​includes: The abnormal amplification factor was obtained by iteratively optimizing the parameters based on a medical cross-dataset and a grid search algorithm. The periodontal abnormality factor was calculated based on the abnormal magnification factor and the abnormal depth constant. The baseline range parameter and baseline duration are standardized. Based on the standardized baseline range parameter and baseline duration, the cyclic fluctuation term is calculated. The product of the cyclic fluctuation term and the cumulative glucose metabolic load at the corresponding time step is calculated to obtain the glucose concentration factor. The product of the periodontal abnormality factor and the blood glucose concentration factor is calculated to obtain the associated feature value.

[0013] Furthermore, the method for obtaining the metabolic decompensation feature vector includes: The associated feature values ​​are sorted according to the time step order to obtain the associated feature value sequence. The associated feature value sequence is then input into a gated recurrent unit network and subjected to time axis dimensionality reduction to output a metabolic decompensation feature vector.

[0014] The present invention has the following beneficial effects: This invention obtains an abnormal depth constant by reading dental case data, extracts instantaneous blood glucose concentration based on a preset time step, and calculates the free blood glucose increment by comparing it with the baseline blood glucose concentration. The basal metabolic ratio constant is weighted using the abnormal depth constant and the free blood glucose increment to obtain the single-step blood glucose metabolism ratio. The accumulated blood glucose metabolic load is obtained by iteratively adjusting the unmetabolized blood glucose concentration of the previous time step, establishing a preliminary correspondence between static spatial data and dynamic temporal data, thus solving the problem of dimensional mismatch in heterogeneous data. By extracting the concentration range and duration of blood glucose concentration cycles, and calculating the baseline range parameter and baseline duration based on multi-layered nested cycles, the closed-loop alternating characteristics of blood glucose fluctuations are fully preserved. The accumulated blood glucose metabolic load is adjusted according to the baseline range parameter and baseline duration, and combined with the abnormal depth constant to obtain associated feature values, further eliminating dimensional conflicts and transforming cross-organ interactions into joint features of a unified scale. Finally, by performing time-axis dimensionality reduction on the associated feature values, a metabolic decompensation feature vector is obtained, which can accurately output the bidirectional coupling strength, solving the problem of difficulty in transforming cross-organ interactions into joint features of a unified scale. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The flowchart illustrates a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia, according to an embodiment of the present invention. The method includes: The basic calculation module 101 is used to read dental case data, obtain the probing depth and bleeding record of each valid examination site, and calculate the abnormal depth constant based on the probing depth and bleeding record; extract the instantaneous blood glucose concentration value based on the preset time step, and calculate the free blood glucose increment at each time step by the difference between the instantaneous blood glucose concentration value and the baseline blood glucose concentration.

[0021] In clinical dental examinations, periodontal probing records two key indicators for each valid examination site: probing depth and bleeding record. The system obtains dental case data through a medical data interface, including the probing depth and bleeding record for each valid examination site. Probing depth reflects the depth of the periodontal pocket and can indirectly indicate the degree of damage to the periodontal supporting tissues, while the bleeding record can directly indicate whether there is abnormal activity at the site. If only probing depth is used, abnormal activity will be ignored; if only bleeding record is used, the severity of tissue damage cannot be reflected. Therefore, combining the two can characterize the patient's overall periodontal abnormal load. An abnormal depth constant is calculated based on the probing depth and bleeding record. The larger the abnormal depth constant, the more active bleeding and the deeper the periodontal pocket.

[0022] In continuous glucose monitoring, an absolute value of instantaneous blood glucose concentration is recorded at each preset time step. However, the baseline levels of fasting blood glucose concentration vary among different patients. Physiologically, only excess glucose exceeding the individual's normal baseline level will place an additional burden on the metabolic system. Therefore, the instantaneous blood glucose concentration value is extracted based on the preset time step, and the free blood glucose increment at each time step is obtained by calculating the difference between the instantaneous blood glucose concentration value and the baseline blood glucose concentration. The free blood glucose increment eliminates the patient's baseline blood glucose difference and reflects the blood glucose load exceeding the individual's normal baseline level at each time step.

[0023] It should be noted that, in order to prevent the calculated free blood glucose increment from being negative when the instantaneous blood glucose concentration is less than the baseline blood glucose concentration, thus affecting subsequent calculations, after calculating the difference between the instantaneous blood glucose concentration and the baseline blood glucose concentration, a non-negative truncation is performed by comparing it with 0, and the maximum value of the comparison result is taken as the free blood glucose increment.

[0024] In one specific implementation of this invention, the time step is set to 15 minutes. The sampling interval of commonly used continuous blood glucose monitoring devices in clinical practice is generally 15 minutes, which can balance time resolution and computational complexity.

[0025] In one specific implementation of this invention, the baseline blood glucose concentration is the 5th percentile extracted from the patient's continuous blood glucose monitoring data. This represents the patient's normal value under hypoglycemic conditions and can effectively eliminate the patient's baseline blood glucose differences.

[0026] The metabolic accumulation module 102 is used to extract the basal metabolic ratio constant from the collected medical cross dataset; and to weight the basal metabolic ratio constant according to the abnormality depth constant and the free blood glucose increment to obtain the single-step blood glucose metabolism ratio; and to adjust the unmetabolized blood glucose concentration of the previous time step based on the single-step blood glucose metabolism ratio and the free blood glucose increment of each time step to obtain the cumulative blood glucose metabolic load of each time step.

[0027] Under normal physiological conditions, the human body can metabolize a certain proportion of blood glucose in each time step. Since the basal metabolic capacity varies among different individuals or under different pathological conditions, it is necessary to first collect a medical cross dataset containing periodontal and blood glucose information and extract a basal metabolic ratio constant. The basal metabolic ratio constant can reflect the proportion of blood glucose that the human body can metabolize in each time step when there are no periodontal abnormalities and blood glucose is ideal.

[0028] Because periodontal abnormalities release inflammatory factors that inhibit systemic glucose metabolism, and excessively high glucose concentrations can also generate negative feedback and reduce metabolic efficiency, it is necessary to use the abnormality depth constant and free glucose increment at the current time step to weight and correct the basal metabolic ratio constant to obtain the single-step glucose metabolism ratio, which reflects the proportion of glucose that can actually be metabolized at each time step under the current periodontal abnormality and glucose level.

[0029] Because glucose metabolism is a continuous process, unmetabolized glucose from the previous time step remains and adds to the newly added glucose in the current time step, forming a new cumulative load. To reflect the cumulative effect of glucose in the body, the concentration of unmetabolized glucose left over from the previous time step is dynamically adjusted based on the proportion of glucose metabolism in each time step and the increase in free glucose in that time step, to obtain the cumulative glucose metabolic load of the current time step. The cumulative glucose metabolic load represents the total glucose load that has not been metabolized in the body up to the current time step, which integrates historical backlog and current addition, and is regulated by real-time metabolism.

[0030] The association module 103 is used to sort the cumulative blood glucose metabolic load according to the time step to obtain each blood glucose concentration rise and fall cycle and the corresponding concentration range value and duration step; based on the concentration range value and duration step of the multi-layer nested blood glucose concentration rise and fall cycle, the baseline range parameter and baseline duration step of the nested time step are calculated; the cumulative blood glucose metabolic load is adjusted according to the baseline range parameter and baseline duration step, and then the association feature value is calculated with the anomaly depth constant.

[0031] Repeated fluctuations in blood glucose levels cause additional damage to blood vessels and periodontal tissues. The greater the amplitude and duration of these fluctuations, the more significant the harm. Therefore, this invention sorts the accumulated glucose metabolic load according to time steps and identifies each cycle of blood glucose concentration fluctuations based on the sorting results. A cycle of fluctuations is a complete period of time from a local trough to a local peak and then back to the same trough. For each cycle of blood glucose concentration fluctuations, the system records the concentration range and the number of duration steps. The concentration range reflects the severity of fluctuations in a single cycle, and the number of duration steps represents the duration of the cycle. Together, they constitute the basic damage characteristics of blood glucose concentration fluctuations.

[0032] In actual blood glucose concentration fluctuations, there is a phenomenon of large-scale nested small-scale cycles. For the same time step, it may be under multiple blood glucose concentration fluctuation cycles at different levels. If only the concentration range value and duration of a single cycle are used, the contribution of other nested cycles will be lost. Based on the concentration range value and duration of multiple nested blood glucose concentration fluctuation cycles, a comprehensive baseline range parameter and baseline duration of each nested time step are calculated. Both quantify the overall blood glucose concentration fluctuation environment in which the time step is located, reflecting the dominant role of large-scale long-cycle cycles while taking into account the superimposed influence of small-scale high-frequency cycles within.

[0033] Considering the impact of blood glucose concentration fluctuations, the cumulative blood glucose metabolic load is adjusted based on the baseline range parameter and the baseline duration step count. Since the abnormal depth constant reflects the abnormal characteristics of the patient's local area, it will amplify the metabolic harm caused by blood glucose concentration fluctuations. Therefore, the adjusted value is combined with the abnormal depth constant to calculate the associated feature value. The associated feature value represents the real-time intensity of the bidirectional coupling effect between periodontal abnormalities and blood glucose at the current time step.

[0034] The dimensionality reduction module 104 is used to sort the associated feature values ​​according to the time step, traverse the sorted associated feature values ​​to perform time axis dimensionality reduction, and obtain the metabolic decompensation feature vector.

[0035] The associated feature values ​​are sorted according to the time step order. The length of the sorted sequence is equal to the total number of time steps, usually several thousand points. If it is directly used as the output for analysis or processing, it will face the problem of excessive data dimensionality. Therefore, it is necessary to reduce the dimensionality of the information. By traversing the sorted associated feature values ​​and reducing the dimensionality through the time axis, a metabolic decompensation feature vector is obtained. The metabolic decompensation feature vector has a stable dimension and does not depend on the number of time steps. It reflects the coupling strength between periodontal abnormalities and hyperglycemia. The larger the overall amplitude, the stronger the periodontal abnormality suppresses the ability of blood glucose metabolism, and the greater the impact of blood glucose concentration changes on periodontal tissues. That is, the stronger the bidirectional malignant coupling between the two physiological systems.

[0036] In summary, this system obtains anomaly depth constants by reading dental case data, extracts instantaneous blood glucose concentrations based on preset time steps, and calculates free blood glucose increments by comparing the instantaneous blood glucose concentration with a baseline blood glucose concentration. It then extracts a basal metabolic rate constant using a medical cross-dataset, weights it with the anomaly depth constant and free blood glucose increment to obtain the single-step blood glucose metabolism rate, and adjusts the unmetabolized blood glucose concentration from the previous time step to obtain the cumulative blood glucose metabolic load for each time step. This concentration is then sorted by time step to extract the concentration range and duration of blood glucose concentration cycles. Based on multi-layered nested cycles, the system calculates the baseline range parameters and baseline duration of nested time steps, adjusts the cumulative blood glucose metabolic load, and combines it with the anomaly depth constant to obtain associated feature values. Finally, the associated feature values ​​are sorted by time step and dimensionality reduced to obtain a metabolic decompensation feature vector. This solves the problem that periodontal abnormal static data and blood glucose dynamic data cannot be integrated into a unified joint feature due to differences in dimension and scale.

[0037] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the abnormal depth constant includes: Read dental case data and extract probing depth and bleeding records for all valid examination sites. Probing depth is a continuous value that reflects the degree of periodontal tissue damage. Bleeding records are a binary judgment made by clinicians, divided into bleeding and no bleeding. In order to perform mathematical operations with probing depth, bleeding records are converted into Boolean values, assigning a value of 1 to sites with bleeding and a value of 0 to sites without bleeding.

[0038] For each valid examination site, the product of the corresponding probing depth and the Boolean value of the bleeding record is calculated. When bleeding exists at the site, the product equals the probing depth, indicating that the abnormal load at that site is proportional to the probing depth. Otherwise, the product is 0, indicating that there is no active abnormality at the site and it is not included in the abnormal load. After calculating the product of all valid examination sites, the system sums all the products and divides them by the number of valid examination sites to obtain the mean value. This eliminates the influence of different numbers of valid examination sites among different patients and yields the abnormal depth constant, which represents the average probing depth of sites with active bleeding among all valid examination sites in the entire mouth, reflecting the severity and range of activity of periodontal abnormalities.

[0039] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the basal metabolic rate constant includes: Because ideal metabolic capacity varies among different patients and under different pathological conditions, it is necessary to extract a basal metabolic rate constant to represent the proportion of blood glucose that the body can metabolize in each time step when there are no periodontal abnormalities and blood glucose is ideal. A medical cross-dataset containing periodontal and blood glucose information is collected and divided into a training set and a validation set. On the training set, a grid search algorithm is used for iterative parameter optimization. A set of candidate numerical ranges is pre-defined for the hyperparameters to be determined. By iteratively traversing various combinations of values ​​within the candidate range, the prediction accuracy under different parameter combinations is compared using the cross-entropy loss function. The parameter combination that minimizes the cross-entropy loss function is extracted and solidified as a hyperparameter constant, thus obtaining the basal metabolic rate constant. Preferably, the basal metabolic rate constant ranges from 0.15 to 0.3. The grid search algorithm and cross-entropy loss function are techniques well-known to those skilled in the art, and their specific details are not elaborated upon.

[0040] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the single-step glucose metabolism ratio includes: The suppression of metabolic capacity by periodontal abnormalities and transient hyperglycemia is usually non-linear, and the suppression effect gradually increases with the degree of abnormality or blood glucose concentration. It is suitable to use an exponential function to calculate the abnormality depth constant and free blood glucose increment. However, both the abnormality depth constant and the free blood glucose increment have physical dimensions and cannot be directly exponentially calculated. Therefore, a grid search algorithm is used to simultaneously calibrate the periodontal abnormality attenuation coefficient and the blood glucose concentration attenuation coefficient, which are inversely related to the dimensions of the abnormality depth constant and the free blood glucose increment, from the medical cross-dataset. The periodontal abnormality attenuation coefficient and the blood glucose concentration attenuation coefficient are fixed in the range of greater than 0 and less than 1, which can eliminate the dimension problem. These two attenuation coefficients are fixed in the same parameter iterative optimization process as the aforementioned basal metabolic ratio constant. The specific process is not described in detail.

[0041] The product of the periodontal abnormality attenuation coefficient and the abnormality depth constant is calculated. This product is a dimensionless value. Taking the negative of this product and applying an exponential function, the periodontal abnormality weighting factor is obtained. This product is negatively correlated with the periodontal abnormality weighting factor. Similarly, the blood glucose weighting factor is calculated using the blood glucose concentration attenuation coefficient and the free blood glucose increment. The basal metabolic rate constant is multiplied by the periodontal abnormality weighting factor and the blood glucose weighting factor to obtain the single-step blood glucose metabolism ratio. The single-step blood glucose metabolism ratio is less than or equal to the basal metabolic rate constant and decreases as periodontal abnormalities worsen or blood glucose levels rise, reflecting the suppression of metabolic capacity by pathological factors.

[0042] A negative correlation indicates that as the product of the periodontal abnormality attenuation coefficient and the abnormality depth constant increases, the corresponding periodontal abnormality weighting factor decreases. This negative correlation can be achieved in different forms, such as an exponential form, where the weighting factor is... x is the product, and the weighting factor strictly decreases as the product increases; there is also a reciprocal form, where the weighting factor is... x is the product, and it also satisfies the condition that the larger the product, the smaller the weight factor; both can achieve the result that when the product increases, the corresponding weight factor decreases.

[0043] It should be noted that, in order to prevent the calculation result of the single-step glucose metabolism ratio from being absolutely zero and affecting subsequent calculations, a minimum protection constant of 0.01 is preset. The minimum protection constant is compared with the above calculation result, and the maximum value is taken as the single-step glucose metabolism ratio.

[0044] In one specific implementation of this invention, the formula for obtaining the single-step glucose metabolism ratio is expressed as: ; in, is the minimum protection constant, and the max function is used to compare the minimum protection constant with the calculated result and take the maximum value; t is the index of the time step; It is the basal metabolic rate constant; the exp function is used for negative correlation processing; It is the periodontal abnormal attenuation coefficient. It is the anomalous depth constant. It is a weighting factor for periodontal abnormalities; It is the blood glucose concentration attenuation coefficient. It is the increase in free blood glucose at time step t. It is a blood glucose weighting factor; It represents the proportion of single-step blood glucose metabolism at time step t.

[0045] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the cumulative glucose metabolic load includes: Blood glucose levels in the human body are a continuous accumulation process. Unmetabolized blood glucose from the previous time step remains and, combined with the increase in free blood glucose in the current time step, forms a new load. Therefore, the cumulative blood glucose concentration needs to be calculated based on each time step. The unmetabolized blood glucose ratio at each time step is calculated, and the product of this ratio and the unmetabolized blood glucose concentration from the previous time step represents the unmetabolized blood glucose concentration at the current time step. This product is then added to the corresponding increase in free blood glucose to obtain the theoretical cumulative blood glucose metabolic load. However, the blood glucose concentration in the human body does not increase indefinitely. To prevent the value from diverging due to unlimited growth caused by continuous accumulation of blood glucose, a pre-set upper limit constant for the unmetabolized blood glucose concentration is used. The theoretical cumulative blood glucose metabolic load is compared with this upper limit constant, and the minimum value is taken as the cumulative blood glucose metabolic load for the corresponding time step.

[0046] In one specific implementation of this invention, the formula for obtaining the cumulative glucose metabolic load is expressed as: ; in, The preset upper limit constant of unmetabolized blood glucose concentration is set to 35 mmol / L, which covers the highest blood glucose range in clinical practice and ensures the effectiveness of the cutoff trigger. The min function compares the upper limit constant of unmetabolized blood glucose concentration with the theoretical cumulative blood glucose metabolic load and takes the minimum value. It is the cumulative glucose metabolic load at time step t-1, representing the unmetabolized glucose concentration at the previous time step; It is the percentage of unmetabolized blood glucose in a single step, reflecting how much blood glucose remains in the previous time step that cannot be metabolized and continues to accumulate; It is the cumulative glucose metabolic load at time step t.

[0047] It should be noted that at the initial moment, i.e., at time step t=0, since the system does not have the previous accumulated glucose metabolic load, the system forcibly assigns the accumulated glucose metabolic load at the initial moment to 0.

[0048] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the concentration range value and the number of duration steps includes: Repeated fluctuations in blood glucose concentration have additional damaging effects on blood vessels and periodontal tissues. Therefore, the accumulated blood glucose metabolic load is first sorted according to the time step to obtain a sequence of unmetabolized blood glucose concentrations that change over time. This sequence is then input into a rainflow counting algorithm to automatically search for and output all complete blood glucose concentration fluctuation cycles. The rainflow counting algorithm is a well-known technique in the art, and its specific details will not be elaborated upon.

[0049] For each blood glucose concentration rise and fall cycle, the system extracts the maximum and minimum values ​​of the cumulative blood glucose metabolic load within the cycle, and calculates the absolute value of the difference between the maximum and minimum values ​​to obtain the concentration range value of the corresponding blood glucose concentration rise and fall cycle, reflecting the severity of the cycle fluctuation; the system extracts the start time step and end time step of the cycle, calculates the difference between the end time step and the start time step, and obtains the number of steps of the corresponding blood glucose concentration rise and fall cycle, reflecting the duration of the cycle.

[0050] Preferably, in some specific implementations of the embodiments of the present invention, after obtaining the concentration range value and the number of duration steps, the method further includes: Because the initial cumulative glucose metabolic load is set to 0, but in real physiological conditions, patients already have a certain amount of blood glucose in their bodies before monitoring, the rise and fall of the cumulative glucose metabolic load starting from zero is not a real physiological fluctuation, but a pseudo-cycle introduced by the initial error. The blood glucose concentration rise and fall cycle during this period needs to be removed to prevent it from affecting subsequent analysis.

[0051] The system extracts the starting time step of each blood glucose concentration rise and fall cycle. If the starting time step is within the previous preset number of time step intervals, the blood glucose concentration rise and fall cycle is determined to be a pseudo cycle and is removed from the valid set of blood glucose concentration rise and fall cycles.

[0052] In one specific implementation of this invention, the preset number is 96, and 96 time steps correspond to 24 hours. In real medical scenarios, after 24 hours, the effect of setting the blood glucose concentration to 0 on fluctuations can be basically eliminated.

[0053] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the benchmark range parameter and the benchmark duration step number includes: For a nested time step, obtain the concentration range value and nesting level of all blood glucose concentration rise and fall cycles covering the time step. Based on the concentration range value and nesting level, a comprehensive baseline range parameter is calculated using the root mean square, which is the baseline range parameter of the nested time step, reflecting the overall blood glucose concentration fluctuation range of the time step. At the same time, extract the maximum duration step number from all blood glucose concentration rise and fall cycles covering the time step, as the baseline duration step number of the time step, reflecting the overall blood glucose concentration fluctuation duration of the time step.

[0054] It should be noted that for time steps without blood glucose concentration rise and fall cycle coverage or whose corresponding cycles are excluded, the corresponding baseline range parameter and baseline duration step number are directly assigned to 0.

[0055] In one specific implementation of this invention, the formula for obtaining the benchmark range parameter is expressed as: ; Where k is the index of the nested blood glucose concentration rise and fall cycle; M is the nesting level of the nested time step; and m is the index number of the M nested blood glucose concentration rise and fall cycles. It is the concentration range value corresponding to the m-th nested blood glucose concentration rise and fall cycle; It is the baseline range parameter of the nested time steps.

[0056] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the associated feature values ​​includes: The aforementioned grid search algorithm is used to identify the amplification factor from the medical cross-dataset and solidify it into a positive real number greater than 0 and less than 1. The specific process is the same as the aforementioned parameter iterative optimization, and will not be elaborated in detail. The amplification factor is used to amplify the degree of influence of periodontal abnormalities on overall metabolism. In order to eliminate dimensions, the system presets a maximum constant for periodontal abnormalities and normalizes the maximum value of the abnormal depth constant. The normalized abnormal depth constant is multiplied by the amplification factor. In order to prevent the product from being 0 due to the patient not having periodontal abnormalities, the product is added to 1 to obtain the periodontal abnormality factor.

[0057] To eliminate the dimensions of the baseline range parameter and baseline duration step count, the system presets a unit concentration range reference constant and a unit time step reference constant. The baseline range parameter and baseline duration step count are then standardized. The product of the standardized baseline range parameter and baseline duration step count represents a comprehensive measurement of the magnitude and duration of blood glucose concentration fluctuations. To prevent numerical anomalies caused by extreme cyclic fluctuations, this product is nonlinearly compressed using a logarithmic function. To prevent the calculation from being meaningless without blood glucose concentration fluctuations, the product is summed with 1, and then the logarithmic function is applied to obtain the nonlinearly compressed product. To avoid the result being 0 when there are no blood glucose concentration fluctuations, affecting subsequent blood glucose concentration factor calculations, the nonlinearly compressed result is summed with 1 to obtain the cyclic fluctuation term. The cumulative blood glucose metabolic load is standardized using a preset unit cumulative load. The product of the cyclic fluctuation term and the standardized cumulative blood glucose metabolic load at the corresponding time step is calculated to obtain the blood glucose concentration factor. The blood glucose concentration factor reflects the equivalent blood glucose load after suppressing blood glucose concentration fluctuations, including both the steady-state accumulation of blood glucose and the additional effect of cyclic fluctuations.

[0058] The product of the periodontal abnormality factor and the blood glucose concentration factor is calculated to obtain the correlation feature value. The correlation feature value reflects the bidirectional coupling strength between periodontal abnormality and hyperglycemia. The larger the value, the more significant the malignant interaction between periodontal abnormality and hyperglycemia.

[0059] In one specific implementation of this invention, the formula for obtaining the associated feature value is expressed as follows: ; in, It is the cumulative glucose metabolic load at time step t. The preset cumulative load is set to 1 mmol / L. It is the standardized cumulative glucose metabolic load; It is the baseline range parameter of the nested time steps. The preset unit concentration range reference constant is set to 1 mmol / L. It is the standardized baseline range parameter; It is the baseline duration of the nested time steps. This is the preset unit time step reference constant, set to 1. It is the standardized baseline duration step count; It is a cyclic fluctuation term; It is an abnormal magnification factor; This is the preset maximum constant for periodontal abnormalities, set to 10mm, which can cover the degree of abnormality for the vast majority of patients. If a very small number of patients have an abnormality depth constant that exceeds the maximum constant for periodontal abnormalities, the system will set the normalization result to 1. It is a periodontal abnormality factor; It is the correlation feature value at time step t.

[0060] Preferably, in some specific implementations of the embodiments of the present invention, the method for obtaining the metabolic decompensation feature vector includes: The associated feature values ​​are sorted according to the time step order to obtain the associated feature value sequence. The associated feature value sequence is input into the gated recurrent unit network. During the training phase, the mean square error reconstruction loss of the associated feature value sequence is used as the training objective. After training, only the encoder part is retained for time axis dimensionality reduction. The high-frequency time-series fluctuation values ​​are reduced in dimensionality along the time axis and mapped into a condensed static feature tensor. A fixed-length metabolic decompensation feature vector is output. The gated recurrent unit network is a technical means well known to those skilled in the art, and the specific details are not elaborated.

[0061] The dimension of the metabolic decompensation feature vector is fixed and does not depend on the number of original time steps. Each component together carries information about the bidirectional coupling between periodontal abnormalities and blood glucose concentration throughout the monitoring process. The larger the overall amplitude of the feature vector, the stronger the suppression of blood glucose metabolism by periodontal abnormalities, the greater the impact of blood glucose fluctuations on periodontal tissues, and the more significant the bidirectional malignant coupling between the two physiological systems.

[0062] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia, characterized in that, The system includes: The basic calculation module is used to read dental case data, obtain the probing depth and bleeding record of each valid examination site, and calculate the abnormal depth constant based on the probing depth and bleeding record; extract the instantaneous blood glucose concentration value based on the preset time step, and calculate the free blood glucose increment at each time step by the difference between the instantaneous blood glucose concentration value and the baseline blood glucose concentration. The metabolic accumulation module is used to extract the basal metabolic ratio constant from the collected medical cross dataset; and to weight the basal metabolic ratio constant according to the abnormality depth constant and the free blood glucose increment to obtain the single-step blood glucose metabolism ratio; and to adjust the unmetabolized blood glucose concentration of the previous time step based on the single-step blood glucose metabolism ratio and the free blood glucose increment of each time step to obtain the cumulative blood glucose metabolic load of each time step. The correlation module is used to sort the cumulative glucose metabolic load by time step to obtain each glucose concentration rise and fall cycle and its corresponding concentration range value and duration step; based on the concentration range value and duration step of the multi-layer nested glucose concentration rise and fall cycle, the baseline range parameter and baseline duration step of the nested time step are calculated; the cumulative glucose metabolic load is adjusted according to the baseline range parameter and baseline duration step, and then the correlation feature value is calculated with the anomaly depth constant. The dimensionality reduction module is used to sort the associated feature values ​​according to the time step, traverse the sorted associated feature values ​​and perform time axis dimensionality reduction to obtain the metabolic decompensation feature vector.

2. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the abnormal depth constant includes: Read dental case data, extract the probing depth and bleeding records of all valid examination sites based on the case data, and convert the bleeding records into Boolean values; Calculate the product of the probing depth and the Boolean value of the bleeding record for each valid examination site, and then take the mean value based on the number of valid examination sites to obtain the abnormal depth constant.

3. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the basal metabolic rate constant includes: The collected medical cross-dataset was divided into a training set and a validation set. The model was trained on the training set based on a grid search algorithm, and its performance was evaluated on the validation set. The constant with the best performance on the validation set was selected as the basal metabolic ratio constant.

4. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the single-step glucose metabolism ratio includes: Based on the medical cross-dataset and grid search algorithm, the parameters were iteratively optimized to obtain the periodontal abnormality decay coefficient and the blood glucose concentration decay coefficient. Calculate the product of the periodontal abnormality attenuation coefficient and the abnormality depth constant, and standardize the product to obtain the periodontal abnormality weight factor. This product is negatively correlated with the periodontal abnormality weight factor. Calculate the product of the blood glucose concentration decay coefficient and the free blood glucose increment, standardize the product to obtain the blood glucose weighting factor, and the product is negatively correlated with the blood glucose weighting factor. The basal metabolic rate constant is obtained by multiplying and weighting the periodontal abnormality weight factor and the blood glucose weight factor.

5. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the cumulative glucose metabolic load includes: The proportion of unmetabolized blood glucose in a single step is calculated based on the proportion of blood glucose metabolism in each time step. The product of the proportion of unmetabolized blood glucose in a single step and the concentration of unmetabolized blood glucose in the previous time step is calculated and added to the corresponding increase in free blood glucose to obtain the theoretical cumulative blood glucose metabolic load. The theoretical cumulative glucose metabolic load is compared with the preset upper limit constant of unmetabolized glucose concentration, and the minimum value is taken as the cumulative glucose metabolic load at each time step.

6. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The methods for obtaining the concentration range value and the number of duration steps include: The accumulated glucose metabolic load is sorted according to the time step order to obtain the unmetabolized glucose concentration sequence. The rainflow counting algorithm is used to search the unmetabolized glucose concentration sequence to extract all glucose concentration rise and fall cycles. Calculate the absolute value of the difference between the minimum and maximum values ​​of the cumulative glucose metabolic load for each glucose concentration rise and fall cycle to obtain the corresponding concentration range value; Calculate the difference between the termination time step and the start time step for each blood glucose concentration rise / fall cycle to obtain the corresponding duration step.

7. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 6, characterized in that, After obtaining the concentration range and the number of duration steps, the following is also included: Extract the starting time step of each blood glucose concentration rise and fall cycle. If the starting time step is within the previous preset number of time step intervals, the corresponding blood glucose concentration rise and fall cycle is removed.

8. The deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The methods for obtaining the benchmark range parameter and the benchmark duration step number include: For time steps with multiple nesting layers, the baseline range parameter of the nested time step is calculated based on the concentration range value of the multiple nested blood glucose concentration rise and fall cycles and the number of nesting layers. Extract the maximum duration of the multi-layered nested blood glucose concentration rise and fall cycle as the baseline duration of the nested time steps.

9. A deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the associated feature values ​​includes: The abnormal amplification factor was obtained by iteratively optimizing the parameters based on a medical cross-dataset and a grid search algorithm. The periodontal abnormality factor was calculated based on the abnormal magnification factor and the abnormal depth constant. The baseline range parameter and baseline duration are standardized. Based on the standardized baseline range parameter and baseline duration, the cyclic fluctuation term is calculated. The product of the cyclic fluctuation term and the cumulative glucose metabolic load at the corresponding time step is calculated to obtain the glucose concentration factor. The product of the periodontal abnormality factor and the blood glucose concentration factor is calculated to obtain the associated feature value.

10. A deep learning-based bidirectional feature analysis system for periodontal disease and hyperglycemia according to claim 1, characterized in that, The method for obtaining the metabolic decompensation feature vector includes: The associated feature values ​​are sorted according to the time step order to obtain the associated feature value sequence. The associated feature value sequence is then input into a gated recurrent unit network and subjected to time axis dimensionality reduction to output a metabolic decompensation feature vector.

Images