Periodontal tooth health monitoring biosensor system

Abstract

The invention relates to the technical field of biosensors, and discloses a periodontal tooth health monitoring biosensor system. The system comprises a signal acquisition module, a feature extraction module, a core screening module and an anomaly judgment module. The signal acquisition module is used for acquiring multi-dimensional signals from a plurality of heterogeneous sensing terminals and integrating the multi-dimensional signals into a health monitoring signal database; a feature extraction module analyzes the database to obtain a health monitoring feature matrix, determines a correlation graph of each dimension and constructs a process correlation graph; a core screening module extracts health feature vectors, determines health index sensitive entropies, and screens out a health core feature sequence in combination with the process association map; and the abnormity judgment module carries out health abnormity judgment according to the sequence. According to the system, by integrating multi-dimensional signals, mining feature association and screening core features, precise monitoring and anomaly judgment of periodontal tooth health are achieved.

Description

Technical Field

[0001] This invention relates to the field of biosensor technology, specifically to a biosensor system for monitoring periodontal and dental health. Background Technology

[0002] Periodontal disease and dental caries are common problems in oral health. Their pathogenesis is complex, and early symptoms are often subtle and easily overlooked, leading to disease progression. Traditional periodontal and dental health monitoring mainly relies on clinical oral examinations, such as periodontal probing and X-rays. These methods have certain limitations. Periodontal probing depends on the dentist's experience, and different dentists may have different interpretations. It also cannot continuously monitor the dynamic changes in periodontal tissues. While X-rays can provide structural information about teeth and alveolar bone, they are invasive procedures with potential radiation exposure risks, and they are also difficult to detect early, subtle lesions.

[0003] With the development of sensing technology, some oral health monitoring devices based on single sensors have emerged, such as sensors for detecting saliva pH and sensors for monitoring dental plaque. However, these single sensors can only acquire information from a specific dimension and cannot comprehensively reflect the overall condition of periodontal and dental health. Periodontal and dental health is a complex system involving the interaction of multiple factors, including the inflammatory response of periodontal tissues, the degree of mineralization of dental hard tissues, and changes in saliva composition. Monitoring data from a single dimension is insufficient to accurately determine the health status.

[0004] Furthermore, existing monitoring systems also have shortcomings in data processing. The acquired multidimensional monitoring data is often scattered and lacks an effective integration mechanism, leading to the neglect of correlations between data points. Simultaneously, the inherent relationships between features across different dimensions are not fully considered during feature extraction and analysis, resulting in a lack of representativeness in the selected health features and affecting the accuracy of anomaly detection.

[0005] In practical applications, due to the lack of systematic integration and in-depth analysis of multidimensional data, existing monitoring methods struggle to provide early warnings for periodontal and dental diseases. Many patients only seek medical attention when obvious symptoms appear, by which time the condition has already progressed to a more serious stage, increasing both the difficulty and cost of treatment and causing significant suffering and financial burden. Therefore, developing a monitoring system capable of integrating multidimensional sensor data, accurately extracting health characteristics, and effectively identifying anomalies has become an urgent problem to be solved in the field of oral health monitoring. Summary of the Invention

[0006] The purpose of this invention is to provide a biosensor system for monitoring periodontal and dental health, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a periodontal and dental health monitoring biosensor system, the system comprising:

[0008] The signal acquisition module is used to acquire multi-dimensional signals related to periodontal and dental health from multiple heterogeneous sensing terminals, and to integrate the multi-dimensional signals to obtain a health monitoring signal database.

[0009] The feature extraction module is used to perform feature parsing on the health monitoring signal database to obtain a health monitoring feature matrix, determine the dimensional correlation diagram of each dimension in the health monitoring feature matrix, and construct a process correlation diagram related to periodontal and dental health through the dimensional correlation diagrams of each dimension.

[0010] The core screening module is used to extract health feature vectors of each dimension from the health monitoring feature matrix, determine the health indicator sensitivity entropy of each dimension based on the corresponding health feature vectors, and screen out the health core feature sequence based on all health indicator sensitivity entropies and the process association graph.

[0011] The anomaly detection module is used to detect anomalies in periodontal and dental health based on the core health feature sequence.

[0012] Preferably, the multidimensional signals related to periodontal and dental health include oral bioelectric signals, saliva composition signals, periodontal tissue pressure signals, tooth surface temperature signals, and occlusal stress signals.

[0013] Preferably, the integration of the multidimensional signals to obtain a health monitoring signal database specifically includes:

[0014] The multidimensional signal is preprocessed to obtain a preprocessed multidimensional signal;

[0015] Feature mapping is performed on the preprocessed multidimensional signals to obtain health monitoring signals in various dimensions;

[0016] A health monitoring signal database is constructed using health monitoring signals from all dimensions.

[0017] Preferably, the process of performing feature analysis on the health monitoring signal database to obtain a health monitoring feature matrix specifically includes:

[0018] Feature extraction is performed on the health monitoring signals of each dimension in the health monitoring signal database to obtain health feature vectors for each dimension.

[0019] A health monitoring feature matrix is ​​constructed based on health feature vectors of all dimensions.

[0020] Preferably, determining the dimensional correlation graph for each dimension in the health monitoring feature matrix specifically includes:

[0021] Select one dimension from all dimensions of the health monitoring feature matrix and obtain the health feature vector of the selected dimension;

[0022] Determine the feature correlation between each health-related feature in the selected dimension's health feature vector;

[0023] Based on the feature correlation between various health-related features, a dimensional correlation diagram of the selected dimension is constructed, thereby obtaining the dimensional correlation diagram of each dimension in the health monitoring feature matrix.

[0024] Preferably, constructing a process association map related to periodontal and dental health through dimensional association diagrams of various dimensions specifically includes:

[0025] Identify key health-related features for each dimension, and then determine the feature correlation between these key health-related features;

[0026] By connecting the dimensional correlation diagrams of each dimension based on the feature correlation degree between each key health-related feature, a process correlation diagram related to periodontal and dental health is obtained.

[0027] Preferably, determining the sensitivity entropy of health indicators in each dimension based on the corresponding health feature vector specifically includes:

[0028] Obtain the calibration factor for each dimension;

[0029] For each dimension of the health feature vector, obtain the weights corresponding to each health-related feature in the health feature vector;

[0030] The health indicator sensitivity entropy of the corresponding dimension of the health feature vector is determined by the weights and calibration factors of each health-related feature, thereby obtaining the health indicator sensitivity entropy of each dimension.

[0031] Preferably, the selection of core health feature sequences based on the sensitivity entropy of all health indicators and the process correlation graph specifically includes:

[0032] For each health-related feature in the health monitoring feature matrix, determine the health index sensitivity entropy of the dimension in which the health-related feature is located.

[0033] Extract all feature correlation degrees corresponding to the health-related features from the process correlation graph;

[0034] The health core entropy of the health-related feature is determined by the health index sensitivity entropy of the dimension in which the health-related feature is located and the correlation degree of all corresponding features, thereby obtaining the health core entropy of each health-related feature in the health monitoring feature matrix.

[0035] All core health features are selected based on the core entropy of each health-related feature.

[0036] Construct a sequence of core health features based on all core health features.

[0037] Preferably, the abnormality judgment of periodontal and dental health based on the health core feature sequence involves inputting the health core feature sequence into an abnormality recognition model for detection, thereby obtaining an abnormality judgment report of periodontal and dental health.

[0038] Preferably, the anomaly detection model is a convolutional neural network model.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] This patented biosensor system for periodontal and dental health monitoring acquires and integrates multidimensional signals from multiple heterogeneous sensing terminals through a signal acquisition module, constructing a health monitoring signal database. This solves the problem of incomplete data from traditional single-sensor monitoring. Traditional monitoring methods rely on single-dimensional information and cannot comprehensively reflect the health status of periodontal and dental tissues. In contrast, this system integrates multidimensional signals covering multiple aspects such as periodontal tissue inflammation, the state of hard dental tissues, and saliva composition, enabling subsequent analysis to be based on a more comprehensive data foundation.

[0041] The feature extraction module analyzes the health monitoring signal database to construct a process correlation graph, enabling in-depth exploration of the intrinsic connections between features across different dimensions. Previous monitoring systems often analyzed data in isolation when processing multidimensional data, neglecting their interactions. This process correlation graph clearly presents the correlation paths and influence relationships between various health indicators, providing a basis for accurately grasping the overall state of periodontal and dental health. This mining of data correlations helps uncover potential health problems hidden behind individual data points, leading to a deeper understanding of health status.

[0042] The core screening module uses health indicator sensitivity entropy and process correlation graphs to select core health feature sequences, improving the representativeness and relevance of features. Traditional feature screening methods often rely on experience or simple statistical analysis, easily including redundant or irrelevant features, affecting the efficiency and accuracy of subsequent judgments. Health indicator sensitivity entropy, however, quantifies the sensitivity of each dimension of features to health status. Combined with the correlation strength between features in the process correlation graph, the selected core feature sequences accurately reflect key health information, reducing interference from irrelevant information.

[0043] The anomaly detection module identifies abnormalities based on a core health feature sequence, enabling precise identification of periodontal and dental health abnormalities. Because this core feature sequence contains key information most representative of changes in health status, anomaly detection based on it can more sensitively capture early, subtle lesions. Compared to traditional methods relying on subjective doctor judgment or single-indicator monitoring, this system's anomaly detection is more objective and accurate, allowing for timely detection of abnormalities in the early stages of disease. This provides patients with earlier treatment opportunities, reduces the risk of disease progression, and minimizes the adverse consequences of misdiagnosis or missed diagnosis. Attached Figure Description

[0044] Figure 1 This is a schematic diagram illustrating the working principle of the periodontal and dental health monitoring biosensor system described in this invention.

[0045] Figure 2 A flowchart for multidimensional signal integration;

[0046] Figure 3 A flowchart for constructing a dimension association graph;

[0047] Figure 4 A flowchart for calculating the sensitivity entropy of health indicators;

[0048] Figure 5 A flowchart for screening core health feature sequences. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figures 1-5 This invention provides a periodontal and dental health monitoring biosensor system, comprising: a signal acquisition module, a feature extraction module, a core screening module, and an anomaly detection module. The specific implementation steps are as follows:

[0051] The signal acquisition module operates as follows: it acquires multidimensional signals related to periodontal and dental health from multiple heterogeneous sensing terminals. These signals originate from different types of sensors and cover various physiological and physical parameters related to periodontal and dental health. After acquiring the multidimensional signals, they are integrated and processed. Through operations such as unified data format conversion, time axis alignment, and removal of redundant information, the scattered signal data is summarized and organized to ultimately form a health monitoring signal database. This database centrally stores various health monitoring signals that have undergone preliminary processing.

[0052] The feature extraction module operates as follows: It performs feature analysis on the data in the health monitoring signal database. By analyzing parameters such as waveform, amplitude, and frequency of signals from different dimensions within the database, representative feature information is extracted, resulting in a health monitoring feature matrix. After obtaining this matrix, the correlation between features within each dimension is analyzed to determine the dimensional correlation graph for each dimension. Subsequently, based on the dimensional correlation graphs of each dimension, the interaction and influence paths of features between different dimensions are identified, constructing a process correlation graph related to periodontal and dental health. This graph visually presents the dynamic correlation process of each feature in health monitoring.

[0053] The core screening module operates as follows: Health feature vectors for each dimension are extracted from the health monitoring feature matrix. Each vector contains key feature parameters related to health status in the corresponding dimension. Based on these health feature vectors, the sensitivity entropy of health indicators for each dimension is calculated. This sensitivity entropy reflects the degree to which the corresponding dimension's features are sensitive to changes in health status. Combining the sensitivity entropies of health indicators across all dimensions with the process correlation graph, features are screened, retaining those that significantly impact health status and occupy key positions in the correlation graph, forming a core health feature sequence.

[0054] The operation process of the anomaly detection module is as follows: taking the core health feature sequence as input, the module uses a preset detection algorithm to analyze the feature data in the sequence, compares it with the feature benchmark under normal health conditions, identifies abnormal features that deviate from the benchmark, thereby completing the anomaly detection of periodontal and dental health and outputting the detection result.

[0055] Example 1:

[0056] Multidimensional signals related to periodontal and dental health include oral bioelectrical signals, salivary composition signals, periodontal tissue pressure signals, tooth surface temperature signals, and occlusal stress signals. Oral bioelectrical signals are acquired using a flexible electrode sensor implanted inside the gum line. This sensor, made of biocompatible material, can directly contact periodontal tissue to capture the neural electrical activity of the periodontal ligament and gingival tissue, and the signal is expressed as voltage fluctuations over time. Saliva composition signals are acquired by a micro-biochemical sensor chip integrated inside the denture. This chip contains multiple specific detection units capable of detecting lysozyme concentration, lactate dehydrogenase activity, pH value, and the concentration of metabolites from microorganisms such as Streptococcus mutans in saliva. The signal output includes the concentration values ​​and change curves of each component. Periodontal tissue pressure signals are acquired using a thin-film pressure sensor attached to the inner wall of the periodontal pocket. The sensor is less than 0.1 mm thick and can generate resistance changes with the deformation of periodontal tissue, which are then converted into electrical signals corresponding to the pressure magnitude. The sensor records pressure changes during chewing, swallowing, and resting states. The tooth surface temperature signal is acquired using an array of infrared temperature sensors mounted on an intraoral mucosal occlusal device. This allows for simultaneous detection of temperature distribution across multiple tooth surfaces, outputting real-time temperature values ​​and temperature field distribution images for each detection point. Occlusal stress signals are obtained using miniature piezoresistive sensors embedded in the occlusal surface of the tooth crown. These sensors are distributed in key areas of occlusal contact and can sense the magnitude and distribution of stress at different contact points during occlusion. The signal is represented as a dynamic curve of stress value change with occlusal action.

[0057] The process of integrating multidimensional signals to obtain a health monitoring signal database is as follows: First, the multidimensional signals are preprocessed. For oral bioelectrical signals, bandpass filtering is used to remove 50Hz power frequency interference and high-frequency electromyographic noise, retaining the effective signal in the 0.5-30Hz frequency band. Baseline correction is then performed, calculating the signal baseline value using a sliding window to eliminate baseline drift caused by electrode contact instability. For saliva component signals, a moving average method is used for data smoothing to reduce random errors during detection. Simultaneously, the 3σ criterion is used to identify and remove outliers; when a detected value deviates from the mean by more than three times the standard deviation, it is considered an outlier and replaced by interpolation with adjacent data. For periodontal tissue pressure signals, wavelet thresholding is used to decompose the signal into five layers of wavelet coefficients. High-frequency coefficients are thresholded and the signal is reconstructed to remove environmental vibration interference. Then, standardization is performed to map the signal values ​​to the 0-1 range, eliminating the influence of initial biases from different sensors. For tooth surface temperature signals, median filtering is used to remove salt-and-pepper noise while retaining information on temperature abrupt changes. A temperature calibration algorithm converts the raw voltage values ​​output by the sensor into Celsius temperature values, ensuring consistency in temperature values ​​across different detection points. For occlusal stress signals, low-pass filtering is used to eliminate high-frequency vibration noise while retaining the trend information of stress changes. Stress distribution uniformity correction is applied, adjusting the signal values ​​according to the sensor position weights to make stress signals from different locations comparable.

[0058] The preprocessed multidimensional signals enter the feature mapping stage. For oral bioelectrical signals, the processed voltage signals are converted into feature parameters, including peak voltage, peak interval time, signal amplitude change rate, and energy proportion of specific frequency bands. These parameters correspond to the intensity, frequency, and rhythm characteristics of neural electrical activity, respectively. For salivary component signals, the concentrations of each component are converted into standardized indices. For example, the lysozyme activity index is mapped to a value of 0-100 according to the concentration range, pH value is converted into an acid-base deviation index, and microbial metabolite concentration is converted into a microbial imbalance risk index. For periodontal tissue pressure signals, the pressure values ​​are converted into feature parameters such as pressure level, pressure duration proportion, and pressure change rate. The pressure level is divided into 5 intervals according to the pressure magnitude, and the pressure duration proportion is the ratio of the duration of a certain pressure level to the total monitoring time. For tooth surface temperature signals, they are converted into feature parameters such as average temperature value, the difference between the highest and lowest temperatures, the number of abnormal temperature points, and the slope of the temperature change trend. Abnormal temperature points are defined as detection points that exceed the normal body temperature range (36.5-37.5℃). For the occlusal stress signal, it is converted into characteristic parameters such as maximum stress value, stress distribution standard deviation, occlusal symmetry index and stress change amplitude. Among them, the occlusal symmetry index reflects the symmetry of the occlusal stress of corresponding teeth in the upper and lower jaws.

[0059] Finally, a health monitoring signal database is constructed. Health monitoring signals from various dimensions are stored in time series, with each data record including a collection timestamp, signal dimension identifier, feature parameter name, and corresponding parameter value. The database employs a distributed storage structure, dividing data into blocks by day, and supports data queries by time range, signal dimension, and feature parameter type. A data index table is also established to record the collection time range, data volume, and integrity identifier for each dimension of signal, facilitating rapid data location and retrieval. The database also includes a data update mechanism, receiving newly collected health monitoring signals in real time, automatically performing format conversion and storage to ensure the timeliness and continuity of the data in the database. Through the above process, a multi-dimensional, multi-feature health monitoring signal database is formed, providing foundational data for subsequent feature extraction and health status analysis.

[0060] Example 2:

[0061] The process of performing feature analysis on the health monitoring signal database to obtain the health monitoring feature matrix is ​​as follows: Feature extraction operations are performed on the health monitoring signals of each dimension in the health monitoring signal database. For oral bioelectrical signals, time-domain features are extracted from the preprocessed signals, including the number of peaks per unit time, the average interval between peaks, the duration of each peak, the difference between the maximum and minimum amplitude of the signal, and the total energy of the signal within a 1-minute time period. Simultaneously, the waveform change characteristics of the signal are analyzed, extracting the rising and falling slopes of the waveform, as well as the frequency of specific waveform patterns (such as spike waves and slow waves). For saliva component signals, the arithmetic mean, median, and mode of the concentration of each component are calculated within the monitoring period. The number of times the concentration value exceeds the reference range is counted, as well as the trend of concentration changes (increasing, decreasing, or remaining stable) during continuous monitoring.

[0062] For periodontal tissue pressure signals, the peak value, time to reach the peak, duration of the peak value, and rate of pressure rise from baseline to peak and rate of pressure fall from peak to baseline are extracted for each pressure change. Simultaneously, the number of pressure fluctuations per unit time and the percentage of time the pressure value is in different ranges (e.g., low, medium, high pressure) are calculated. For tooth surface temperature signals, the average temperature, maximum temperature, minimum temperature, and the difference between the maximum and minimum temperatures at each monitoring point are extracted. The number of monitoring points with temperatures exceeding the normal range and their duration are statistically analyzed, along with the slope of the temperature-time curve and the differences in temperature distribution across different tooth surfaces. For occlusal stress signals, the maximum stress value, minimum stress value, average stress value, and uniformity of stress distribution are extracted for each occlusal process (achieved by calculating the standard deviation of stress values). The frequency of reaching the maximum occlusal stress, the duration of each occlusion, and the interval between two occlusions are also recorded.

[0063] Through the above operations, the health monitoring signal for each dimension is transformed into a set of health feature vectors containing multiple parameters. All health feature vectors for all dimensions are arranged column-wise to form a health monitoring feature matrix. The number of rows in the matrix corresponds to the number of monitoring samples (e.g., monitoring data at different time points), the number of columns corresponds to the sum of feature parameters for all dimensions, and each element represents the specific value of a monitoring sample for a particular feature parameter.

[0064] The process of determining the dimensional correlation diagram for each dimension in the health monitoring feature matrix is ​​as follows: Select any dimension from all dimensions of the health monitoring feature matrix (e.g., the saliva component signal dimension) and obtain the health feature vector for that dimension (including parameters such as mean concentration, number of concentration fluctuations, and trend). Calculate the correlation degree between the various health-related features in this vector. For example, analyze the relationship between mean concentration and the number of concentration fluctuations (e.g., does a higher mean concentration lead to more fluctuations), and the correlation between the concentration change trend and the number of times it exceeds the normal range (e.g., does a decreasing trend accompany more instances of exceeding the range). The correlation degree is calculated by analyzing the numerical change trends of the two feature parameters across all monitoring samples. If both parameters rise or fall simultaneously more frequently, the correlation degree is high; if the trends are opposite or there is no obvious pattern, the correlation degree is low.

[0065] Based on the calculated feature correlations, a dimensional correlation graph is constructed for each dimension. In the graph, circular nodes represent various health-related features, and the size of the nodes is determined by the importance of the feature in the dimension (e.g., nodes for mean concentration are larger than nodes for single fluctuation values). Nodes are connected by line segments, with the thickness of the line segment representing the degree of feature correlation; the higher the correlation, the thicker the line segment. For feature pairs with correlations below a set threshold, no connecting lines are drawn. Following the same method, other dimensions in the health monitoring feature matrix (such as oral bioelectrical signal dimension, periodontal tissue pressure signal dimension, etc.) are processed sequentially. The correlations between features within each dimension are calculated, and corresponding dimensional correlation graphs are constructed, ultimately yielding the dimensional correlation graph for each dimension in the health monitoring feature matrix.

[0066] Example 3:

[0067] When constructing a process correlation map related to periodontal and dental health using dimensional correlation maps of various dimensions, the first step is to identify key health-related features from the dimensional correlation maps of each dimension. The identification of key health-related features is based on the connection strength and frequency of occurrence of features in the dimensional correlation maps. If a feature has a thick connection segment with other features within a dimension (i.e., high correlation), and shows a significant correlation in most monitored samples, then it is identified as a key health-related feature. For example, in the correlation map of the periodontal tissue pressure signal dimension, the connection segment between pressure peak and pressure duration is the thickest, and shows a strong correlation in more than 80% of the samples; therefore, these two features are identified as key health-related features for this dimension.

[0068] After identifying the key health-related features across all dimensions, the correlation between these key features across different dimensions is calculated. For example, the correlation between the peak pressure in the periodontal tissue pressure signal dimension and the mean lysozyme concentration in the salivary component signal dimension is calculated by analyzing the numerical trends of both over a continuous monitoring period. If the mean lysozyme concentration also increases within 24 hours of a pressure peak increase in more than 60% of the total monitoring times, a high correlation is considered to exist; if this proportion is less than 30%, the correlation is considered low.

[0069] A process flow diagram is formed by connecting the dimensional relationship diagrams of various dimensions based on the correlation of key cross-dimensional features. Only connections between features with a correlation higher than 50% are retained to ensure the simplicity and effectiveness of the diagram. In the diagram, nodes of different dimensions are distinguished by different colors (e.g., nodes of the periodontal tissue pressure signal dimension are blue, and nodes of the salivary component signal dimension are green). Cross-dimensional connections are dashed lines, while connections within the same dimension are solid lines, thus clearly presenting the relationship types between features.

[0070] When determining the sensitivity entropy of health indicators for each dimension based on the corresponding health feature vector, a calibration factor is first set for each dimension. The numerical range of the calibration factor is 0.1-1.0. Among them, the calibration factor for the saliva component signal dimension is set to 0.9 because it directly reflects the biochemical changes of the oral microenvironment; the calibration factor for the tooth surface temperature signal dimension is set to 0.3 because it is greatly affected by the environment; the calibration factors for other dimensions are determined based on the stability of clinical data, with the periodontal tissue pressure signal dimension set to 0.7, the oral bioelectric signal dimension set to 0.6, and the occlusal stress signal dimension set to 0.5.

[0071] For each dimension of the health feature vector, the weights are determined by analyzing the correlation between the feature parameters and known health states. For example, in the salivary component signal dimension, the mean lysozyme concentration has a relatively high correlation with periodontitis, so its weight is set to 0.3; while the pH value has a relatively low correlation, so its weight is set to 0.1. The sum of the weights of all features is 1.0, achieved through normalization (e.g., if a dimension has 5 features with initial weights of 3, 2, 2, 2, and 1, the normalized weights are 0.3, 0.2, 0.2, 0.2, and 0.1).

[0072] The sensitivity entropy of health indicators is calculated using the following formula:

[0073]

[0074] Where S represents the sensitivity entropy of health indicators, w i Let represent the weight of the i-th health-related feature in this dimension, k represent the calibration factor for this dimension, n represent the number of features in the health feature vector of this dimension, and ln represent the natural logarithm function. Using this formula, the weighted value of each feature (the product of the weight and the calibration factor) is converted into an entropy value, ultimately yielding the sensitivity entropy of health indicators for each dimension.

[0075] Example 4:

[0076] When selecting the core health feature sequence based on the sensitivity entropy of all health indicators and the process correlation graph, it is necessary to extract each health-related feature from the health monitoring feature matrix one by one and clarify its corresponding dimension. For example, the mean lysozyme concentration feature in the saliva component signal dimension has its corresponding health indicator sensitivity entropy determined through previous calculations. This value is directly related to the sensitivity of this dimension to changes in health status.

[0077] In the process correlation graph, for each health-related feature, its correlation degree with all other features is extracted. Taking the pressure peak feature in the periodontal tissue pressure signal dimension as an example, it is necessary to find the correlation degree between this feature and the pressure duration feature within the same dimension, as well as the correlation degree with the mean lysozyme concentration feature in the salivary composition signal dimension and the highest temperature feature in the tooth surface temperature signal dimension across dimensions. These correlation degrees are reflected by the properties of the connecting lines in the graph; the thicker the line, the higher the correlation degree.

[0078] The core health entropy of a health-related feature is calculated by combining its health index sensitivity entropy with the correlation degrees of all extracted features. The calculation involves multiplying the health index sensitivity entropy by each correlation degree separately, then summing all the products. For example, if the health index sensitivity entropy of a feature's dimension is 0.7, and its correlation degrees with three other features are 0.6, 0.5, and 0.3, the calculation would be 0.7 × 0.6 + 0.7 × 0.5 + 0.7 × 0.3, and the final value is the core health entropy of that feature. This method is used to calculate the core health entropy of all health-related features in the health monitoring feature matrix.

[0079] After calculating the health core entropy for all health-related features, a health core entropy threshold is set. This threshold is determined based on the importance distribution of features in the clinical health data. For example, in a matrix containing 100 health-related features, the minimum health core entropy of the top 30% of features is selected as the threshold. The health core entropy of each feature is compared with this threshold, and features with core entropy higher than the threshold are selected as health core features.

[0080] The selected core health features need to be sorted according to their association paths in the process association graph to form a sequence of core health features. The sorting starts with basic features directly related to periodontal and dental health, such as pH value in the saliva composition signal dimension, as it directly reflects the oral acid-base environment and is used as the starting feature of the sequence. Subsequently, based on the association direction between features in the graph, features with the highest correlation are added sequentially. For example, after pH value, the mean lysozyme concentration feature, which has the highest correlation with it, is added; then, the peak periodontal tissue pressure feature, which has the highest correlation with the mean lysozyme concentration feature, is added, and so on. If a feature is associated with multiple features already added to the sequence, its position is determined by the sum of the correlation scores, with features having higher sums of correlation scores added to the sequence first.

[0081] During the sorting process, it is necessary to ensure that the features in the sequence can cover the main association paths in the process association map to avoid missing key association links. For example, although the maximum stress value feature in the occlusal stress signal dimension has a low direct correlation with the starting feature, it has a high correlation with the temperature change trend feature in the tooth surface temperature signal dimension, and this temperature feature is already in the sequence. Therefore, the maximum stress value feature is added to the sequence to ensure the integrity of the association path.

[0082] Example 5:

[0083] When identifying anomalies in periodontal and dental health based on a core health feature sequence, this sequence must be input into the anomaly detection model. The core health feature sequence contains key feature parameters across multiple dimensions, such as the average concentration of lysozyme in saliva, peak periodontal tissue pressure, highest tooth surface temperature, and maximum occlusal stress. These parameters are arranged in a specific order to form an input data structure that the model can recognize. After receiving this sequence, the anomaly detection model analyzes each feature parameter layer by layer.

[0084] The input layer of the anomaly detection model converts the core health feature sequence into a matrix form. The number of rows in the matrix corresponds to the number of features, and the number of columns corresponds to the length of the time series of the monitored samples. For example, if the sequence contains 10 core features, and each feature contains monitoring data for 24 consecutive hours, then the input matrix is ​​a 10-row, 24-column numerical matrix, where each element represents the specific value of a feature at a certain moment.

[0085] Following the input layer is a convolutional layer, which contains multiple convolutional kernels of different sizes. Each kernel scans the input matrix using a sliding window to extract local feature patterns. For example, a 3×3 kernel can capture feature change trends within three adjacent time points, while a 5×5 kernel can identify feature association patterns over a longer time range. During convolution, the kernel multiplies and sums the values ​​of corresponding regions in the matrix to generate feature maps, each corresponding to a feature extraction pattern.

[0086] The feature maps output from the convolutional layers enter the pooling layers. These layers reduce the amount of data through downsampling while preserving key feature information. A common pooling method is max pooling, which selects the maximum value from a region of the feature map as the representative value for that region. For example, when max pooling a 2×2 region, the maximum value among the four values ​​within that region is taken, reducing the size of the feature map to one-quarter of its original size, effectively reducing the complexity of subsequent calculations.

[0087] The feature map processed by the pooling layer is flattened into a one-dimensional vector and input into the fully connected layer. The fully connected layer contains multiple neurons, each connected to all neurons in the previous layer, and performs non-linear transformations on the input vector using weight parameters. These transformations integrate feature information extracted by different convolutional kernels, establishing a mapping relationship between features and health status categories. For example, combining lysozyme concentration changes with periodontal tissue pressure features can help determine if there are any inflammation-related abnormal patterns.

[0088] The output of the fully connected layer is fed into the output layer, which uses a softmax activation function to convert the result into a probability distribution for different types of health abnormalities. For example, the output might show a probability of 0.6 for "gingivitis," 0.3 for "periodontitis," and 0.1 for "normal." Based on the probability distribution, the model determines the most likely type of health abnormality and generates an abnormality report.

[0089] The anomaly identification report comprises several parts. First, it lists the identified abnormal features, such as "mean lysozyme concentration exceeding the normal range by 30%" and "persistently elevated peak periodontal pressure." Second, it explains the correlation between these abnormal features, such as "the increase in lysozyme concentration and the increase in peak periodontal pressure show a synchronous trend." Finally, it provides the possible types of health problems and specific numerical values ​​of related features, offering reference information for subsequent clinical diagnosis. The entire anomaly identification process utilizes multi-level model processing to achieve a systematic assessment of periodontal and dental health.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A periodontal and dental health monitoring biosensor system, characterized in that, The biosensor system includes: The signal acquisition module is used to acquire multi-dimensional signals related to periodontal and dental health from multiple heterogeneous sensing terminals, and to integrate the multi-dimensional signals to obtain a health monitoring signal database. The feature extraction module is used to perform feature parsing on the health monitoring signal database to obtain a health monitoring feature matrix, determine the dimensional correlation diagram of each dimension in the health monitoring feature matrix, and construct a process correlation diagram related to periodontal and dental health through the dimensional correlation diagrams of each dimension. The core screening module is used to extract health feature vectors of each dimension from the health monitoring feature matrix, determine the health indicator sensitivity entropy of each dimension based on the corresponding health feature vectors, and screen out the health core feature sequence based on all health indicator sensitivity entropies and the process association graph. The anomaly detection module is used to detect anomalies in periodontal and dental health based on the core health feature sequence.

2. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The multidimensional signals related to periodontal and dental health include oral bioelectric signals, saliva composition signals, periodontal tissue pressure signals, tooth surface temperature signals, and occlusal stress signals.

3. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The health monitoring signal database obtained by integrating the multidimensional signals specifically includes: The multidimensional signal is preprocessed to obtain a preprocessed multidimensional signal; Feature mapping is performed on the preprocessed multidimensional signals to obtain health monitoring signals in various dimensions; A health monitoring signal database is constructed using health monitoring signals from all dimensions.

4. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The health monitoring signal database is analyzed to obtain a health monitoring feature matrix, which specifically includes: Feature extraction is performed on the health monitoring signals of each dimension in the health monitoring signal database to obtain health feature vectors for each dimension. A health monitoring feature matrix is ​​constructed based on health feature vectors of all dimensions.

5. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, Determining the dimensional correlation graph for each dimension in the health monitoring feature matrix specifically includes: Select one dimension from all dimensions of the health monitoring feature matrix and obtain the health feature vector of the selected dimension; Determine the feature correlation between each health-related feature in the selected dimension's health feature vector; Based on the feature correlation between various health-related features, a dimensional correlation diagram of the selected dimension is constructed, thereby obtaining the dimensional correlation diagram of each dimension in the health monitoring feature matrix.

6. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The process atlas related to periodontal and dental health is constructed by using dimensional relationship diagrams across various dimensions, specifically including: Identify key health-related features for each dimension, and then determine the feature correlation between these key health-related features; By connecting the dimensional correlation diagrams of each dimension based on the feature correlation degree between each key health-related feature, a process correlation diagram related to periodontal and dental health is obtained.

7. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The sensitivity entropy of health indicators in each dimension is determined based on the corresponding health feature vector, specifically including: Obtain the calibration factor for each dimension; For each dimension of the health feature vector, obtain the weights corresponding to each health-related feature in the health feature vector; The health indicator sensitivity entropy of the corresponding dimension of the health feature vector is determined by the weights and calibration factors of each health-related feature, thereby obtaining the health indicator sensitivity entropy of each dimension.

8. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, Based on the sensitivity entropy of all health indicators and the process correlation graph, the core health feature sequence was selected, specifically including: For each health-related feature in the health monitoring feature matrix, determine the health index sensitivity entropy of the dimension in which the health-related feature is located. Extract all feature correlation degrees corresponding to the health-related features from the process correlation graph; The health core entropy of the health-related feature is determined by the health index sensitivity entropy of the dimension in which the health-related feature is located and the correlation degree of all corresponding features, thereby obtaining the health core entropy of each health-related feature in the health monitoring feature matrix. All core health features are selected based on the core entropy of each health-related feature. Construct a sequence of core health features based on all core health features.

9. The periodontal and dental health monitoring biosensor system as described in claim 1, characterized in that, The abnormality identification of periodontal and dental health based on the aforementioned core health feature sequence involves inputting the core health feature sequence into an abnormality recognition model for detection, thereby obtaining an abnormality identification report of periodontal and dental health.

10. The periodontal and dental health monitoring biosensor system as described in claim 9, characterized in that, The anomaly detection model is a convolutional neural network model.

Images