Search

CN-121971041-A - Night tooth grinding intelligent monitoring method based on biting force sensing and time sequence analysis

CN121971041ACN 121971041 ACN121971041 ACN 121971041ACN-121971041-A

Abstract

The invention relates to the technical field of intelligent biological behavior monitoring, and particularly discloses a night tooth grinding intelligent monitoring method based on bite force sensing and time sequence analysis, which comprises the steps of synchronously collecting multichannel bite force data and physiological signals of a user at night; the method comprises the steps of carrying out deep time sequence mode analysis on biting force data, extracting space-time characteristic vectors, inputting the characteristic vectors and physiological signals into a personalized sleep stage self-adaptive calibration model to generate a calibrated molar behavior-sleep stage association matrix, inputting the matrix into a pre-trained multi-task risk assessment model, and outputting a personalized molar risk assessment report which quantitatively predicts specific clinical risks such as tooth wear, joint dysfunction and the like. According to the invention, accurate association of tooth grinding behavior and sleep stages is realized under the condition of no professional electroencephalogram equipment, and the paradigm upgrading from description phenomenon to early warning clinical risk is completed, so that an effective tool is provided for household accurate sleep health management.

Inventors

  • Yang Shenying

Assignees

  • 绍兴市口腔医院

Dates

Publication Date
20260505
Application Date
20260126

Claims (10)

  1. 1. The intelligent night tooth grinding monitoring method based on biting force sensing and time sequence analysis is characterized by comprising the following steps of: Acquiring multi-channel time sequence engagement force data and synchronous physiological signals of a user during night sleep; Carrying out time sequence mode analysis on the multichannel time sequence biting force data, and extracting time sequence mode feature vectors representing dynamic mechanical features of tooth grinding behaviors; Inputting the time sequence pattern feature vector and the synchronous physiological signals into a personalized sleep stage self-adaptive calibration model to generate a calibrated molar behavior-sleep stage association matrix; inputting the calibrated molar behavior-sleep stage correlation matrix into a pre-trained multi-task risk assessment model to generate a molar risk assessment report.
  2. 2. The intelligent night tooth grinding monitoring method based on engagement force sensing and time series analysis according to claim 1, wherein the acquiring multichannel time series engagement force data and synchronous physiological signals during night sleep of the user specifically comprises: synchronously collecting an engagement force original signal flow and a physiological original signal flow through a wearable monitoring device; The method comprises the steps of aligning an engagement force original signal flow and a physiological original signal flow to the same time axis based on a unified time stamp sequence, carrying out low-pass filtering on the engagement force signal, carrying out sensor sensitivity calibration, and outputting calibrated multichannel time sequence engagement force data; and packaging the multichannel time sequence engagement force data and the synchronous physiological signals into data packets according to a preset time window, and transmitting the data packets to adjacent gateway equipment or mobile terminals for subsequent analysis or temporarily storing the data packets in a local memory of the equipment through low-power consumption Bluetooth.
  3. 3. The intelligent night tooth grinding monitoring method based on bite force sensing and time sequence analysis according to claim 2, wherein the step of synchronously collecting the bite force raw signal stream and the physiological raw signal stream by the wearable monitoring device comprises the following steps: continuously acquiring pressure analog signals of all teeth positions at a first sampling frequency by using a flexible occlusion force sensor array integrated in the oral cavity protector to form an occlusion force original signal flow; Synchronously acquiring a photoelectric volume pulse wave signal and a triaxial acceleration signal at a second sampling frequency by using a physiological signal sensing module integrated in a wrist strap or a chest patch to jointly form a physiological original signal stream; And a microcontroller in the wearable monitoring equipment prints a unified time stamp sequence from the same high-precision real-time clock on all original signal streams.
  4. 4. The intelligent night tooth grinding monitoring method based on engagement force sensing and time sequence analysis according to claim 1, wherein the multi-channel time sequence engagement force data is subjected to time sequence pattern analysis, and time sequence pattern feature vectors representing dynamic mechanical features of tooth grinding behaviors are extracted, and specifically comprise the following steps: segmenting the multi-channel time series bite force data by sliding time windows, calculating bite force amplitudes of all channels in each time window, and marking the time windows as potential tooth grinding active segments when the resultant force amplitudes exceed a dynamic baseline threshold value adaptive to the quiet bite state of a user; Extracting multi-dimensional time sequence characteristics from the data of each potential tooth grinding movable section in parallel; the multi-dimensional time sequence feature extracted from each potential tooth grinding movable section is spliced and standardized to form primary feature vectors of the potential tooth grinding movable sections, and the primary feature vectors of all the movable sections in the same sleeping period are arranged according to a time sequence order to be assembled into a finally output time sequence pattern feature vector used for representing the dynamic pattern of the night tooth grinding behavior.
  5. 5. The intelligent night tooth monitor method based on bite force sensing and time series analysis according to claim 4, wherein for each of the potential tooth grinding activity segments data, multi-dimensional time series features are extracted in parallel, comprising in particular: extracting space distribution time sequence characteristics, and calculating space coordinates of a dentition biting force central point at each sampling moment in a potential tooth grinding movable section, so as to extract the moving total path length, the average moving speed and the residence time duty ratio of a central point track in a tooth grinding area; Extracting time domain rhythm characteristics, identifying local peak points of biting force in a potential tooth grinding active section, analyzing time interval sequences of the peak points, calculating variation coefficients of the peak points to quantify regularity of rhythm, and extracting main oscillation frequency of signals among the peak points; And (3) extracting mechanical load time sequence characteristics, counting the proportion of the accumulated time of the biting force data of each channel exceeding a preset physiological load threshold value in the total duration of the active segment in the potential tooth grinding active segment, and calculating the average slope of the rising edge and the falling edge of the resultant force waveform.
  6. 6. The intelligent night tooth grinding monitoring method based on engagement force sensing and time sequence analysis according to claim 1, wherein the time sequence pattern feature vector and the synchronous physiological signal are input into a personalized sleep stage self-adaptive calibration model to generate a calibrated tooth grinding behavior-sleep stage association matrix, and the method specifically comprises the following steps: processing the synchronous physiological signals, extracting time domain and frequency domain features of heart rate variability from the photoelectric volume pulse wave signals, extracting body movement intensity and body position change features from the triaxial acceleration signals, inputting the extracted features into a light sleep stage initial judgment model, and outputting an initial sleep stage probability sequence; calculating a pattern confidence score for each potential tooth grinding event characterized in the time sequence pattern feature vector according to specific spatial distribution, time domain rhythm and mechanical load characteristics of the potential tooth grinding event; And the preliminary sleep stage probability sequence, the occurrence time of the tooth grinding event and the mode confidence coefficient score are input into a core calibration module of the personalized sleep stage self-adaptive calibration model together, and the calibration module performs iterative learning and matching to generate a final calibrated tooth grinding behavior-sleep stage association matrix.
  7. 7. The intelligent night tooth monitor method based on bite force perception and time series analysis according to claim 6, wherein the workflow of the personalized sleep stage adaptive calibration model specifically comprises: In the initial stage, the model searches a law of stable co-occurrence of a high-confidence molar event and a specific sleep stage, and the law is used as a personalized benchmark of a user; For events occurring in sleep stages that do not match the personalized benchmarks, the model will arbitrate in conjunction with its pattern confidence score, low confidence events being de-weighted or marked as pending, high confidence events triggering fine-tuning of the current sleep stage probability; After multiple rounds of iterative calibration, the model outputs a final estimated sleep stage label and a calibrated associated intensity value for each molar event, integrates all events and their calibrated labels with the intensity values in a timeline, and generates a final calibrated molar behavior-sleep stage correlation matrix.
  8. 8. The intelligent night-time molar monitoring method based on bite force perception and time series analysis according to claim 1, wherein the calibrated molar behavior-sleep stage correlation matrix is input into a pre-trained multi-task risk assessment model to generate a molar risk assessment report, specifically comprising: Analyzing the calibrated molar behavior-sleep stage correlation matrix, and extracting an enhanced feature set for risk assessment from the enhanced feature set, wherein the enhanced feature set not only comprises an estimated sleep stage of each molar event in the correlation matrix and a calibrated correlation intensity value, but also comprises derivative features calculated based on event clusters; inputting the reinforced feature set into the pre-trained multi-task risk assessment model, and outputting quantitative indexes of each clinical risk; summarizing and interpretively processing the output result of the multitasking risk assessment model to generate a structured personalized molar risk assessment report.
  9. 9. The intelligent night tooth monitoring method based on engagement force sensing and time series analysis according to claim 8, wherein the pretrained multitask risk assessment model adopts a hierarchical adaptive architecture, and specifically comprises the following modules connected in sequence: The shared feature depth analysis module is used for receiving the enhanced feature set, utilizing the features with time sequence dependency relationship in the two-way gating circulation unit network depth analysis feature set, and outputting a high-dimensional shared depth feature tensor which is fused with the time sequence and the mode information of the night tooth grinding activity; the multi-task special feature branching module receives the shared depth feature tensor, utilizes three independent sub-network branches connected in parallel, learns and extracts special feature representations most relevant to specific risk assessment tasks of the branches on the basis of shared features, and predicts different specific clinical risks respectively; And the personalized adaptation and output module is used for receiving the special characteristic representation of each branch, the built-in personalized parameter modulation layer is used for receiving the baseline physiological parameter of the user as the condition input during model reasoning, dynamically fine-adjusting the weight of each branch and outputting the quantitative index of each clinical risk.
  10. 10. The intelligent night tooth monitoring method based on engagement force sensing and time series analysis according to claim 9, wherein the three independent subnetwork branches in the multi-task dedicated feature branch module specifically comprise: The structural damage risk branch consists of a multi-layer perceptron, and the last layer is a Softmax function and is used for outputting the risk probability of abnormal abrasion or prosthesis and implant overload and the main affected tooth position prediction about different tooth positions. A dysfunction risk branch, consisting of a feed-forward network containing self-attention mechanisms, is used to capture complex combinations of features highly correlated with muscle-joint dysfunction and output a risk level for inducing or aggravating temporomandibular joint dysfunction or morning masticatory muscle fatigue. The sleep quality associated risk branch is composed of a one-dimensional convolution network and is specially used for extracting local mode characteristics related to sleep stage fragmentation and outputting potential influence evaluation scores of molar activities on the sleep stage fragmentation degree.

Description

Night tooth grinding intelligent monitoring method based on biting force sensing and time sequence analysis Technical Field The invention relates to the technical field of intelligent biological behavior monitoring, in particular to an intelligent night tooth grinding monitoring method based on biting force sensing and time sequence analysis. Background Clinical monitoring of bruxism at night has long faced the dilemma that home scenarios and professional accuracy are difficult to combine. The prior art mainly comprises two types, namely household monitoring equipment based on electromyographic EMG or sound threshold judgment, such as CN106455989A, which is convenient to wear, but can only provide limited tooth grinding event frequency and intensity information, and signals are easy to interfere and have insufficient specificity. The other is polysomnography PSG laboratory monitoring, which can provide accurate sleep stage and event correlation as a gold standard, but has the disadvantages of expensive equipment, complex operation and difficult popularization. In-depth analysis finds that the prior art has two neglected but critical cold door technical problems, namely, one of the two neglected but critical cold door technical problems is completely lack of the recognition capability of sleep physiological stages when a tooth grinding event occurs. The clinical significance of molar occurrence during the light sleep phase is quite different from that during the deep sleep phase, and the lack of this information leads to a significant reduction in the value of home monitoring. Secondly, the analysis dimension is limited to the detection of events that occur, lacking a risk prediction model directly linked to long-term clinical consequences such as tooth wear, temporomandibular joint disorders. The inability of existing methods to answer this molar pattern may lead to a clinical core concern of what specific risk. Therefore, there is a need in the industry for an innovative technical solution that not only accurately detects a bruxism event, but also correlates it with a sleep physiological background, and ultimately achieves personalized clinical risk assessment in a home non-invasive environment. Disclosure of Invention The invention aims to provide a night tooth grinding intelligent monitoring method based on biting force perception and time sequence analysis, which aims to solve the problem that the household night tooth grinding monitoring technology in the prior art cannot accurately correlate tooth grinding behaviors with sleep physiological stages and predict clinical risks under the condition of no professional brain electrical equipment. In order to solve the technical problems, the invention specifically provides the following technical scheme: the intelligent night tooth grinding monitoring method based on biting force sensing and time sequence analysis comprises the following steps: S1, acquiring multichannel time sequence biting force data and synchronous physiological signals of a user during night sleep; S2, carrying out time sequence mode analysis on the multichannel time sequence biting force data, and extracting time sequence mode feature vectors representing dynamic mechanical features of tooth grinding behaviors; S3, inputting the time sequence pattern feature vector and the synchronous physiological signals into a personalized sleep stage self-adaptive calibration model to generate a calibrated molar behavior-sleep stage association matrix; S4, inputting the calibrated molar behavior-sleep stage correlation matrix into a pre-trained multi-task risk assessment model, and generating a molar risk assessment report. As a preferred embodiment of the present invention, the S1 specifically includes: s11, synchronously collecting an engagement force original signal flow and a physiological original signal flow through a wearable monitoring device; S12, aligning an occlusion force original signal flow and a physiological original signal flow to the same time axis based on a unified time stamp sequence, carrying out low-pass filtering on the occlusion force signal to remove high-frequency noise, recommending the cut-off frequency to be 30Hz, carrying out sensor sensitivity calibration, and outputting calibrated multichannel time sequence occlusion force data; S13, packaging the multichannel time sequence engagement force data and the synchronous physiological signals into data packets according to a preset time window, and transmitting the data packets to adjacent gateway equipment or mobile terminals for subsequent analysis or temporarily storing the data packets in a local memory of the equipment through low-power consumption Bluetooth. As a preferred embodiment of the present invention, the step S11 specifically includes: S111, continuously acquiring pressure analog signals of all teeth positions at a first sampling frequency by using a flexible occlusion force sensor array integrated in the or