Buccal intelligent health monitoring equipment

Abstract

The invention discloses a buccal intelligent health monitoring device which comprises a brace-shaped buccal device, a storage bin with ultraviolet disinfection and wireless charging functions and a matched software system, the buccal device is made of non-toxic medical-grade materials, and a multi-lead wet electrode, a PPG sensor, a respiration monitoring sensor group, an oral environment monitoring sensor group and the like are arranged in the buccal device. Multi-dimensional physiological parameters such as electrocardio, heart rate, breath, blood pressure and oral health are collected and subjected to pre-amplification, filtering and analog-to-digital conversion, and then data are transmitted through low-power-consumption Bluetooth / Wi-Fi. The software system is combined with LSTM and CNN algorithms to realize anomaly detection and pressure and fatigue evaluation, and supports graded early warning and emergency rescue calling. The device is comfortable to wear, adapts to daily and sports scenes, is accurate and stable in signal acquisition, is suitable for long-term health monitoring of cardiovascular patients, old people and other groups, and has the functions of daily health management and clinical data support.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of biomedical engineering and smart wearable devices, specifically to a mouth-held smart health monitoring device. Background Technology

[0002] With cardiovascular diseases and other chronic diseases becoming major threats to human health, early monitoring and prevention are crucial for reducing morbidity and mortality. Traditional ECG monitoring devices are bulky and poorly portable, making it difficult to meet the needs of long-term real-time monitoring. Existing wearable devices (such as wristbands and patches) mostly use dry electrode designs, which have problems such as poor skin contact, susceptibility to motion interference, and insufficient wearing comfort, resulting in limited monitoring accuracy and stability.

[0003] The oral cavity, with its stable temperature and humidity, proximity to the heart, and abundant subcutaneous blood vessels, is an ideal site for collecting physiological signals. However, existing health monitoring devices for the oral cavity are extremely rare, and most are for laboratory research. There is a lack of mature products that combine portability, comfort, and multi-parameter monitoring capabilities, failing to meet the widespread demand for routine health monitoring. Therefore, there is an urgent need to develop an intelligent health monitoring device that is adapted to the oral environment, provides accurate signal acquisition, and is easy to wear. Summary of the Invention

[0004] The purpose of this invention is to provide an oral-type intelligent health monitoring device to address the problem that there are very few existing health monitoring devices for the oral cavity area, and most of them are for laboratory research.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A mouth-held intelligent health monitoring device includes a mouth-held device, a mouth-held device storage compartment, and a supporting software system, with the following specific structure: Oral device: Design: It adopts the form of a portable dental brace. The sealed shell is made of non-toxic and harmless resin or ceramic material. The front end is machined to fit the tooth contour to fix the device and prevent accidental swallowing. The rear end is an integrated installation cavity to accommodate various functional modules. Power supply module: Built-in lithium-ion rechargeable battery and wireless charging receiver coil. The wireless charging receiver coil is electrically connected to the battery to realize the wireless charging function. Communication module: integrates a low-power Bluetooth (BLE) or Wi-Fi module to establish a wireless data transmission channel with external smart devices through a built-in antenna; Sensing devices: include various types of sensors, all of which are electrically connected to the processing module via wires. ECG monitoring component: 2-6 multi-lead wet electrodes, evenly arranged inside the device in positions corresponding to the areas rich in subcutaneous blood vessels on the gums, tongue, and inside the mouth, and used in conjunction with conductive gel; Heart rate monitoring component: Photoplethysmography (PPG) sensor, integrated at the back end of the device; Breathing monitoring component: Composed of an airflow sensor, a pressure sensor and an infrared gas sensor, all arranged at the rear of the mouthpiece. The sensing ends of the airflow sensor and the pressure sensor face the airflow channel of the nasal cavity or oral cavity, and the detection end of the infrared gas sensor corresponds to the path of exhaled gas. Oral environment monitoring components include a miniature infrared temperature sensor, a capacitive or resistive humidity sensor, a solid pH sensor, a multispectral imaging sensor, and a fluorescence imaging sensor, which are respectively arranged in the corresponding areas where the mouthpiece contacts the oral tissue. The lenses of the multispectral imaging sensor and the fluorescence imaging sensor are directed towards the tongue, pharynx, tonsils, and gum areas. Attitude monitoring component: Composed of a high-sensitivity electronic gyroscope and an acceleration detection unit, with optional addition of a GPS positioning module, which is fixedly connected to the main body of the mouthpiece; Processing module: It has a built-in preamplifier, filter circuit and analog-to-digital converter (ADC). The input of the preamplifier is connected to each sensor through wires, and the output is connected to the input of the ADC through the filter circuit. The output of the ADC is connected to the signal of the wireless communication module. It is used for signal amplification, filtering and analog-to-digital conversion processing.

[0006] Oral device storage compartment: It has a cavity for receiving mouthpieces, and contains an ultraviolet disinfection unit, a power supply unit, and a wireless charging transmitting coil; The ultraviolet disinfection unit covers the inside of the storage cavity for disinfection and sterilization during equipment storage. The wireless charging transmitting coil corresponds to the wireless charging receiving coil of the mouthpiece. The power supply unit supplies power to the ultraviolet disinfection unit and the wireless charging transmitting coil, enabling wireless charging of the mouthpiece.

[0007] Supporting software system: This includes computer platform software and mobile platform software, which establish data interaction with the wireless communication module; Built-in data storage module, signal analysis module and early warning and rescue module; The data storage module is used to store the raw data collected by the sensor and the processed data; The signal analysis module interacts with the data storage module to analyze and process physiological signals; The early warning and rescue module can be linked with the GPS positioning module to output location information and trigger a call command.

[0008] The present invention has the following beneficial effects: High accuracy: The use of multi-lead wet electrodes to collect ECG signals, combined with the moist environment of the oral cavity to reduce contact resistance and motion interference, results in signal acquisition accuracy superior to traditional dry electrode devices; the multi-sensor data fusion and optimized signal processing flow further improve the reliability of monitoring data.

[0009] Comfortable to wear: The brace-like shape adapts to the oral cavity structure, the medical-grade materials are non-irritating, and it can be worn for a long time without affecting daily activities (such as speaking, eating, and exercising), solving the problem of discomfort caused by traditional wearable devices.

[0010] High adaptability to various scenarios: The sealed design of the device can adapt to various sports scenarios such as running and swimming, without problems such as electrode displacement or failure when exposed to water. The storage compartment has the functions of disinfection, charging and storage, improving the convenience of use.

[0011] Comprehensive functions: It integrates multi-dimensional monitoring functions such as ECG, heart rate, respiration, blood pressure, and oral health, and combines intelligent algorithms to realize value-added services such as abnormal warning, stress assessment, and fatigue detection, meeting the dual needs of daily health monitoring and clinical data support.

[0012] Timely response: Supports real-time data transmission and local analysis, automatically triggers tiered early warnings in abnormal situations, and enables emergency rescue calls in conjunction with GPS positioning, providing safety guarantees for high-risk groups. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the arrangement of internal electrodes and sensors in the oral cavity device of the present invention; Figure 3 This is a schematic diagram of the interface and charging principle between the mouthpiece and the storage compartment of the present invention; Figure 4 This is a flowchart of the signal acquisition and processing process of the oral device of the present invention; Figure 5 This is a logic block diagram of the present invention. Detailed Implementation

[0014] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0015] like Figures 1-5As shown, the oral-type intelligent health monitoring device of the present invention consists of two main parts: an oral device and a storage compartment. The oral device directly contacts the user's mouth and performs the core function of health monitoring, while the storage compartment provides storage and power support for the oral device. Multiple wet electrodes are integrated inside the oral device, evenly distributed in different parts of the device, especially in areas contacting the gums, palate, and the skin inside the mouth. The main purpose of the wet electrode design is to continuously collect the user's electrophysiological signals through close contact with the soft tissue inside the mouth. The wet electrodes have high sensitivity and can effectively capture weak bioelectrical signals generated by organs such as the heart, such as electrocardiogram signals. These signals can provide important information for subsequent health status analysis.

[0016] Each wet electrode is connected to the signal processing module inside the oral device via specially designed wires. This module is responsible for the initial processing of the acquired electrical signals to ensure signal quality and accuracy. The signal processing module consists of several functional circuits, including a preamplifier, a filtering circuit, and an analog-to-digital converter. First, the preamplifier amplifies the weak bioelectrical signals acquired by the wet electrodes, bringing the signal amplitude to a level suitable for further processing. Then, the signal passes through the filtering circuit, which primarily removes interference noise and high-frequency components, resulting in a purer bioelectrical signal. Finally, the filtered analog signal is converted into a digital signal by the analog-to-digital converter for subsequent data transmission and processing.

[0017] The signal processing module is designed with various challenges that may be encountered when working inside the oral cavity, such as weak signal strength and high noise interference. Through optimized amplification, filtering, and analog-to-digital conversion, the module effectively improves the signal's anti-interference capability and ensures that the acquired electrophysiological signals are sufficiently accurate for further health data analysis and anomaly detection. Furthermore, the in-mouth device connects to external devices such as smartphones via a wireless communication module, transmitting processed signal data to the cloud in real time for further analysis and evaluation in medical and health applications.

[0018] ECG signal acquisition principle: like Figure 2As shown, the wet electrodes of this invention are connected to the signal processing module via wires inside the oral device. The wet electrodes are the core components for electrophysiological signal acquisition. They continuously collect human electrophysiological signals through direct contact with the skin inside the oral cavity, particularly the gums, tongue surface, and the mucous membrane area inside the mouth. Because the skin inside the oral cavity is relatively moist, the contact resistance between the wet electrodes and the skin is low, thus enabling more sensitive detection of human bioelectrical signals. The acquisition of electrophysiological signals is based on the classical principle of bioelectricity, which reflects the electrical activity of the heart and other organs by detecting changes in the potential across the cell membrane. The heart generates a certain current when it beats, which is transmitted to the body surface and converted into an electrical signal by the wet electrodes.

[0019] During signal acquisition, the wet electrode is responsible for converting the weak bioelectrical signal into a measurable voltage signal. Typically, the voltage of the electrocardiogram (ECG) signal induced on the human body surface is very weak, generally in the millivolt range. Therefore, it needs to be initially amplified by a dedicated preamplifier to enhance the signal to a level suitable for subsequent processing. The preamplifier is a crucial link in the signal processing chain; it needs to possess high sensitivity and low noise characteristics to ensure that the acquired signal is not contaminated by environmental noise or noise from the equipment itself during amplification. In this way, the signal amplified by the preamplifier reaches a detectable range, facilitating further analysis.

[0020] To reduce the impact of environmental noise and other interference factors, the amplified signal needs to be processed by a filter. The filter design is mainly aimed at noise of different frequencies, especially high-frequency noise and power supply frequency interference (such as 50Hz or 60Hz power grid interference signals). Through precise filtering, unwanted high-frequency components can be effectively removed, while retaining low-frequency signals that are of important diagnostic significance for cardiac activity. The filtered signal is purer and can reflect the true electrical activity of the heart, thereby improving the accuracy of the electrocardiogram.

[0021] After amplification and filtering, the electrocardiogram (ECG) signal is converted from an analog signal to a digital signal using an analog-to-digital converter (ADC). An ADC converts continuously changing analog voltage signals into discrete digital signals so that computers or other digital devices can process and store this data. The digitized ECG signal can then be recorded and analyzed to generate classic electrocardiogram (ECG) waveforms. Typical ECG waveforms include the P wave, QRS complex, and T wave, each reflecting the electrical activity of different parts of the heart. The P wave represents atrial depolarization, the QRS complex represents ventricular depolarization, and the T wave is associated with ventricular repolarization. The shape, size, and time intervals of these waveforms provide important diagnostic information for heart health.

[0022] The collected electrocardiogram (ECG) signal data is transmitted in real time to external devices, such as smartphones or computers, via a wireless communication module. Through these devices, users can view their ECG data in real time, and doctors can remotely monitor the user's ECG status for health assessments and anomaly detection. For example, conditions such as tachycardia, bradycardia, and arrhythmias can be detected promptly through analysis of the ECG waveform, allowing for further medical intervention based on the analysis results.

[0023] Principle of respiratory signal acquisition: The rear component of the mouthpiece is equipped with airflow and pressure sensors, primarily used to detect the user's respiratory rate, airflow velocity, and airflow pressure. The airflow sensor monitors the user's respiratory rate and pattern in real time by detecting changes in airflow within the nasal or oral cavity. Each time the user inhales and exhales, the airflow sensor senses changes in the speed and flow of air passing through it, accurately measuring the respiratory rate per minute and identifying any abnormalities such as excessively rapid, slow, or irregular breathing. Furthermore, the pressure sensor in the device provides additional information about airflow intensity by sensing changes in the pressure generated by the airflow. This data contributes to a more comprehensive understanding of the user's respiratory status.

[0024] To monitor the carbon dioxide (CO2) content in exhaled breath, the device also integrates an infrared gas sensor. The infrared gas sensor operates based on the principle of infrared light absorption. Different gas molecules have different absorption characteristics under specific wavelengths of infrared light; CO2 molecules can absorb infrared light of a specific wavelength. The sensor determines the CO2 concentration by emitting infrared light and measuring the amount of light absorbed after the exhaled breath passes through it. This non-contact detection technology has high sensitivity and stability, enabling real-time monitoring of changes in carbon dioxide levels in the user's breath, providing a key indicator of respiratory health. By comprehensively monitoring airflow, pressure, and CO2 concentration, it can help identify respiratory problems such as hypoventilation and hyperventilation.

[0025] Oral environment sensor: This mouthpiece incorporates miniature infrared temperature sensors, capacitive or resistive humidity sensors, and solid-state pH sensors to accurately monitor changes in temperature, humidity, and pH within the oral cavity. These sensors detect minute temperature fluctuations in the mouth, identifying early signs of localized inflammation or infection. For example, when inflammation occurs in a part of the mouth, the local temperature typically rises; the infrared temperature sensor can detect these minute temperature changes in real time. The humidity sensor measures the water vapor content in the air to determine the moisture level in the mouth, thus identifying early symptoms of dry mouth. Dry mouth can lead to oral infections, increased tooth decay, and other oral discomfort. The device also integrates a solid-state pH sensor to monitor changes in the pH of oral fluids. An acidic environment is often associated with enamel erosion and tooth decay; pH ​​monitoring helps identify whether the user is at risk to their dental health. Through comprehensive data analysis, these sensors can effectively predict and prevent oral health problems.

[0026] The mouthpiece also incorporates miniature multispectral imaging sensors and fluorescence imaging sensors to observe the condition of tissues inside the oral cavity. These imaging sensors utilize visible light and fluorescence reflection technology to efficiently capture the condition of areas such as the tongue, throat, tonsils, and gums. Through multispectral imaging, the device can observe the color, thickness, and coverage of the tongue coating, assessing the user's digestive function or nutritional status. For the throat and tonsils, fluorescence imaging can identify redness or swelling of the throat mucosa, and whether the tonsils are enlarged or inflamed. These symptoms may be related to respiratory infections, tonsillitis, and other diseases. Simultaneously, the device can also detect gum bleeding or inflammation through imaging technology; these phenomena are often associated with periodontal disease or gingivitis. Combining multispectral and fluorescence imaging data, the device provides a comprehensive assessment of oral health, helping to identify early lesions or potential health problems and offering timely prevention or treatment recommendations to users.

[0027] Data transmission and processing: like Figure 3 As shown, the oral device is equipped with an advanced wireless communication module. This module is responsible for transmitting the collected bioelectrical signals and other monitoring data to external devices in real time, such as smartphones, tablets, or other terminal devices that support Bluetooth and Wi-Fi connectivity. This wireless communication module uses Bluetooth Low Energy (BLE) or Wi-Fi technology, which minimizes power consumption while ensuring data transmission stability and reliability, thereby extending the device's battery life. BLE is particularly suitable for health monitoring scenarios requiring long-term, continuous data transmission, allowing users to obtain continuous health data feedback without frequent charging.

[0028] When a mobile phone or other terminal device receives bioelectrical signals from the mouthpiece, the accompanying application further processes and analyzes them. The application integrates multiple analysis modules, including an electrocardiogram (ECG) signal analysis module, a heart rate variability (HRV) analysis module, and a health status assessment module. The ECG analysis module can analyze the transmitted ECG signals in real time, generate a complete ECG, and detect possible abnormalities such as arrhythmias, bradycardia, or tachycardia. The HRV analysis module helps users assess stress levels and fatigue by analyzing changes in heart rate intervals. The health status assessment module combines multiple bioparameters to generate personalized health suggestions and warnings, alerting users to potential health risks or physical conditions requiring attention. Users can check their health status at any time through the application and take appropriate actions based on the analysis results, such as rest, exercise, or seeking medical attention.

[0029] Signal processing and analysis: like Figure 4 As shown, after receiving multi-dimensional physiological signals such as ECG, respiration, and oral environment transmitted in real time by the in-mouth device, the application first performs systematic preprocessing on these signals. This step is crucial to ensuring the accuracy and reliability of the data. The preprocessing process includes several key steps: First, the application uses a noise reduction algorithm to filter noise from the acquired signals, removing invalid signals caused by electromagnetic interference, environmental factors, or device noise. Next, baseline drift correction is performed, a common problem in processing biological signals such as ECGs. Baseline drift is caused by factors such as respiration and exercise, so correction can obtain more stable and accurate signals. Subsequently, the application extracts the heartbeat signal, identifying and separating the heartbeat waveform to extract essential data such as heart rate and blood pressure.

[0030] The preprocessed signals are further analyzed, and the application extracts key physiological parameters using specific algorithms. These parameters include, but are not limited to, heart rate, blood pressure, heart rate variability, respiratory rate, and exhaled oxygen content. Real-time changes in heart rate can reflect the user's cardiac health; excessively high or low heart rates may indicate potential cardiovascular problems. Blood pressure data estimated through changes in heart intervals helps users understand their blood pressure fluctuations. Heart rate variability analysis can identify the user's stress level and fatigue status, while respiratory rate is closely related to lung function; abnormal respiratory rate may indicate respiratory diseases. Changes in exhaled oxygen content are related to respiratory efficiency and metabolic level.

[0031] The analyzed physiological signals are converted into visual charts, allowing users to intuitively understand their health status. The application displays these continuous time-series signals as different curves, including electrocardiograms, blood pressure changes, heart rate changes, respiratory rate, and exhaled oxygen content. These charts are displayed in different data windows within the application, allowing users to view changes in various physiological parameters in real time and easily track their health trends.

[0032] Furthermore, the application integrates a built-in abnormal parameter detection algorithm, capable of automatically detecting abnormal data in various physiological indicators. For example, if an arrhythmia is detected in the electrocardiogram (ECG) or blood pressure fluctuates excessively, the system will issue a health warning based on a set threshold. After receiving the warning, users can consult a doctor or take preventative measures. Through this process, the application not only provides users with a comprehensive display of health data but also promptly identifies potential health problems, offering personalized health management services.

[0033] Built-in health indicator detection function in the application: Built-in applications on mobile phones or computers have data collection, display, storage, and local analysis capabilities.

[0034] Continuous monitoring of electrocardiogram (ECG) status can capture the heart's electrical activity in real time, helping to analyze and assess the user's cardiac health, specifically aiding in the analysis of the following physical indicators: Algorithm for detecting cardiac abnormalities: Real-time analysis of electrocardiogram and blood pressure data using time-series anomaly detection technology can detect cardiac abnormalities such as tachycardia, bradycardia, and arrhythmias (e.g., atrial fibrillation, ventricular fibrillation). Simultaneously, by establishing a memory mechanism to analyze heart rate variability, it can further monitor stress levels and fatigue.

[0035] Wavelet transform was used to detect the R waves (peaks in the QRS complex) in the electrocardiogram, and the position of each R wave was marked. The time interval between consecutive R waves was calculated to obtain the RR interval sequence.

[0036] A baseline model of heart rate is established based on historical health data or normal electrocardiogram data from a recent period. A sliding window approach can be used to statistically analyze the RR intervals over a period of time to determine the normal heart rate range and fluctuation characteristics, and this data is then fused with blood pressure data. Furthermore, a Long Short-Term Memory (LSTM) network algorithm is used to analyze the temporal characteristics of normal heart rate and blood pressure sequences, and the algorithm's memory mechanism is utilized to continuously record and analyze the RR interval sequences within a certain time window, capturing the trends and fluctuation patterns of heart rate and blood pressure.

[0037] The system analyzes the current RR interval in real time and compares it with baseline heart rate and blood pressure. If the RR interval continues to shorten and exceeds a predetermined threshold (too fast heart rate), or continues to lengthen and exceeds a predetermined threshold (too slow heart rate), an alarm for tachycardia or bradycardia is triggered. If the blood pressure is too high or too low, an alarm for abnormal blood pressure is triggered.

[0038] By detecting sudden changes or irregular fluctuations in the heartbeat interval, atrial fibrillation, ventricular fibrillation, and other arrhythmic states can be identified. The LSTM algorithm is used to analyze newly acquired heartbeat data sequences and compare them with baseline heart rate characteristics. When abnormal timing features are observed, a heart rate abnormality alarm is triggered.

[0039] The system calculates indicators such as the standard deviation and root mean square difference of heart rate intervals to analyze heart rate velocity (HRV). Changes in HRV reflect the regulatory capacity of the autonomic nervous system. Low HRV is usually associated with high stress levels; the system assesses the user's stress state by detecting the downward trend in HRV. Fatigue is typically manifested as a decrease in HRV and changes in heart rate (e.g., a high heart rate with small fluctuations). Through continuous HRV analysis, fatigue can be identified and alerts can be issued.

[0040] Respiratory abnormality detection algorithm: By detecting the oxygen and carbon dioxide levels in exhaled breath, it is possible to monitor and assess a user's respiratory function, metabolic rate, lung function, and cardiopulmonary health in real time. These sensors provide crucial data support for clinical diagnosis and personal health monitoring.

[0041] Infrared gas sensors are used to collect real-time data on the concentration of oxygen and carbon dioxide in exhaled air.

[0042] Record time-series data, with each time point including oxygen concentration (O2), carbon dioxide concentration (CO2), and possible auxiliary data such as respiratory rate.

[0043] By calculating a user's respiratory quotient, which is the ratio of carbon dioxide (CO2) to oxygen (O2) within the same time period, the levels of CO2 and O2 contained in the exhaled air can be determined.

[0044] The LSTM algorithm is used to detect abnormal breathing, monitor the oxygen and carbon dioxide content in exhaled air, establish baseline data of gas content under healthy conditions, and compare the time-series characteristics with the measured data during wear, so as to realize real-time monitoring of user breathing, abnormal detection and alarm mechanism.

[0045] Monitoring CO2 concentration can indirectly assess blood acid-base balance. A high CO2 concentration (RQ) in exhaled CO2 may indicate a high metabolic rate or be related to acidosis, while a low O2 concentration may indicate insufficient oxygen utilization or be related to alkalosis. This triggers an alarm for abnormal exhaled gases.

[0046] By monitoring and analyzing users' breathing levels in real time, it provides real-time health status feedback, including breathing efficiency, lung function status, metabolic rate, and blood acid-base balance, and monitors symptoms such as stress, fatigue levels, and sleep apnea. Furthermore, it predicts future health risks based on historical data, which is crucial for the early detection of respiratory diseases such as chronic obstructive pulmonary disease (COPD) and asthma.

[0047] Oral health detection algorithm: High-resolution images of the inside of the oral cavity are acquired using optical or multispectral imaging sensors. The data includes images of areas such as the tongue, gums, pharynx, and tonsils.

[0048] Multispectral imaging or fluorescence imaging is used to acquire images in different spectral bands to capture more biological information.

[0049] The collected image data is labeled, including marking diseased areas such as dental caries, periodontal disease, oral cancer, inflammation, and infection.

[0050] Data preprocessing: Image enhancement: Applying image enhancement techniques (such as histogram equalization and contrast stretching) to improve image quality.

[0051] Image segmentation: If the amount of data is large, the image can be divided into smaller blocks so that the neural network can process it.

[0052] Normalization: Normalize the image pixel values ​​to the range of 0-1 to ensure the stability and convergence speed of the neural network model.

[0053] Constructing a detection model: Convolutional Neural Network (CNN): Design a deep convolutional neural network to extract spatial features from images. CNN layers extract local features, such as edges, textures, and shapes, layer by layer, and finally output a feature vector through fully connected layers.

[0054] Multi-task learning (MTL): Combining the framework of multi-task learning, separate output layers are set for tasks such as caries detection, periodontal disease detection, cancer detection, and inflammation detection, while sharing the underlying features.

[0055] Attention mechanism: Introducing an attention mechanism into CNN enhances the focus on lesion areas and improves detection accuracy.

[0056] A CNN model is trained using input oral cavity images, extracting high-dimensional features through convolutional layers. These features represent the morphological and structural information of various parts of the oral cavity. A softmax layer outputs the probability of different lesion categories to determine the presence of lesions such as dental caries, periodontal disease, and oral cancer in the input image. A regression algorithm quantifies the severity of lesions such as inflammation and infection, assessing the severity of the lesions by outputting continuous values. An attention mechanism automatically focuses on abnormal regions in the image, such as dental caries, swollen tonsils, and hyperplastic tissue, further improving the accuracy of lesion detection. Shared CNN feature representations are used for joint analysis of different health indicators. Multiple output heads handle different tasks, such as dental caries, periodontal disease, cancer, and inflammation, and the model improves overall performance during training by sharing information.

[0057] Finally, the trained lesion detection model is deployed in the application on the terminal device to achieve real-time detection of oral health status.

[0058] For RGB images, bilateral filtering or nonlocal mean filtering is used to suppress noise while preserving edge details. For multispectral images, spectral smoothing (such as Savitzky-Golay or wavelet denoising) is performed to reduce high-spectral noise. For bands with missing values ​​or NaN values, interpolation (spectral interpolation or neighborhood-based spatial interpolation) or direct removal of pixels / bands with a large number of missing values ​​is used. Considering spectral drift under different imaging conditions, spectral response normalization and spectral standardization steps are added. The difference between the mean and the mean of each band is divided by the standard deviation, and the original values ​​are retained so that the model can learn absolute reflectance information.

[0059] To ensure stable neural network training, spatial and photometric preprocessing is required: the image is cropped or divided into images of a fixed size of 256×256, while ensuring that each image retains complete spectral channels; the images are normalized at the pixel level; and to avoid distribution drift, standardized parameters or training set statistics are saved on the training / validation / test sets respectively.

[0060] Data augmentation must support both spatial and spectral augmentation. Spatial augmentation includes random rotation, translation, mirroring, scaling, random cropping, and elastic deformation; spectral augmentation includes brightness / contrast jitter, Gamma transformation, and random Gaussian noise superposition. Multispectral-specific augmentations also include simulated banddropout, spectral shifting (adding a small shift to each band), and spectral aliasing (randomly linearly combining several neighboring bands). These augmentations improve the model's robustness to sensor differences and spectral drift. To address class imbalance (e.g., extreme imbalance between oral cancer and common inflammation samples), a hybrid strategy of "minority class oversampling + major class undersampling" is used for sampling. Finally, three sets are generated: training, validation, and testing sets, segmented by patient / consultation batch to avoid data leakage (different images from the same patient do not cross sets).

[0061] The symptom detection is based on an end-to-end deep network—a dual-branch attention-enhanced multi-task oral health detection network, which we name Oropharynx-MultiSpectralNet (OMS-Net for short).

[0062] The overall network structure is as follows: a dual-stream encoder extracts low-to-high-level features from RGB and multispectral signals respectively. The encoder internally combines spectral / channel attention and spatial attention modules to enhance key band and spatial lesion signals. The features from the two encoders are fused at multiple scales through a cross-modal fusion module (including feature alignment, channel-wise weighting, and a lightweight Transformer layer). The fused shared representation is fed into multiple task heads (MTLs): a pixel-level segmentation head, a lesion classification head (multi-label binary classification), and a degree regression head. During training, a weighted multi-task loss is used, combined with uncertain weights or dynamic weight adjustment.

[0063] Input and preprocessing interface: The model input is a concatenated tensor. Here, 3 represents the RGB channels, and B represents the number of multispectral bands. However, internally, the model does not directly treat all channels as a single-stream input; instead, it first divides them into RGB branch inputs. MS branch input The two branches are processed in parallel to utilize pre-trained weights and different convolution strategies.

[0064] The RGB branch (the backbone adopts the ResNet-50 style for easy use of ImageNet pre-trained weights): The network starts with a 7×7 convolution with stride 2 (conv1) + BN + ReLU + 3×3 max pooling, and then enters 4 stages (stage1..4) which are Bottleneckblocks, outputting multi-scale features F_rgb^1 (1 / 4 resolution), F_rgb^2 (1 / 8), F_rgb^3 (1 / 16), and F_rgb^4 (1 / 32). Each stage is followed by a lightweight spatial attention module to highlight possible lesion structures (such as the edge of a cavity, the outline of a swollen tonsil).

[0065] Multi-Spectral (MS) Branch: Since multi-spectral channels typically have fewer channels but contain important spectral dimensions, a set of spectral coupling layers is used first: the first layer is a 1×1 spatial convolution + 1D spectral convolution (or an initial block of 3D convolution) to mix spectral channels and preserve spectral information. Specifically, it can be implemented as: MS-Conv0: Conv3D(kernel=(3,3,3), stride=(1,2,2) or first channel-wise 1D convoverbands), followed by a set of shallow 2D DBottleneck-like blocks (3 blocks) to generate multi-scale features F_ms^1..F_ms^4, corresponding to the scale of the RGB branch. This structure is called an MS stage. Meanwhile, Spectral attention (SpectralSE) is introduced in each MSstage: global average pooling is first performed to obtain the response vector of each band in the spatial dimension, and then a small MLP (DownSample>ReLU->UpSample>Sigmoid) is used to generate per-band weights, which are multiplied back into the channel to highlight the bands that are more valuable for diagnosis (such as near-infrared enhancement of inflammatory signals).

[0066] Cross-modal attention and multi-scale fusion: Before fusion, F_rgb^i and F_ms^i at corresponding scales are aligned in terms of channel number, and 1×1 convolutions are used to map the features to a unified number of channels C. Then, a three-stage fusion is adopted: First, local complementary fusion (concat+conv): at each scale, the two features are concatenated and then compressed and activated by 3×3 convolutions to produce F_fuse^i; second, cross-modal attention (Cross-Attention): F_fuse^i is input into a lightweight Transformer layer, which can model the long-range dependency between RGB and MS and enhance spectral-spatial correlation; third, multi-scale adaptive fusion: adaptive weights are recalculated for the output of each scale to determine the contribution of each scale in the final upsampling reconstruction.

[0067] Shared Decoder and Feature Pyramid: The fused multi-scale features are passed through a top-down decoder (similar to U-Net / FPN structure), progressively upsampling from the high-scale F_fuse^4 and skipping connections with the low-scale F_fuse^3...F_fuse^1. A spatial attention module is applied after each upsampling to preserve target boundary information. The decoder output generates a high-resolution shared feature map. .

[0068] The Multitasking Output Header (MTL) splits several parallel headers from F_shared: Pixel-level segmentation head: Uses three 3×3 convolutions + BN + ReLU, followed by upsampling to the original resolution, and finally uses a 1×1 convolution to output pixel-level binary or multi-class segmentation maps (outputting multi-channel masks according to lesion type). DiceLoss + BCE is used for loss to balance imbalance and boundary quality.

[0069] The multi-label classification head outputs a multi-label probability vector for the entire image / detection box (dental caries, periodontal disease, oral cancer, infection, etc.). It is implemented by performing global average pooling (GAP) on F_shared, passing it through two fully connected layers, and then outputting a sigmoid multi-label probability. FocalLoss or weighted BCE is used for the loss to handle class imbalance.

[0070] Severity Regression Header: For indicators requiring quantification, such as inflammation / infection, the decoder performs a gap operation on the F_shared value and then appends several FC values ​​to output a continuous value (normalized to [0,1]). The loss is calculated using MSE or MAE, and the units are converted to those of clinical / scoring standards.

[0071] Attention to detail: The entire network simultaneously uses spectral attention, channel attention, spatial attention (emphasizing lesion location), and a cross-modal Transformer (for long-range spectral-spatial dependence). This allows for the selection of key spectral information in the band dimension and precise localization of lesion morphology in the spatial dimension.

[0072] Loss and Training Strategies: The overall loss is defined as L = λ_segL_seg + λ_clsL_cls + λ_regL_reg + λ_auxL_aux, where each λ can be automatically balanced across different tasks using uncertainty weighting or dynamic weight adjustment. For extremely imbalanced classification tasks, class-balanced loss is used; for segmentation, a Dice+BCE composite loss is used; if finite sample learning is deployed, semi-supervised loss (pseudo-label consistency) or contrastive learning assistance (comparing positive / negative samples at the feature layer) can be added.

[0073] Training settings: Input image size 256×256, batch size depends on GPU memory; AdamW optimizer (initial learning rate 1e-4), learning rate using cosine annealing; training for 50–200 epochs (depending on data volume), freezing the first two layers of the RGB backbone early in training to stabilize the training (if ImageNet pre-training is used); dropout (0.1–0.3) and data augmentation are added to avoid overfitting. For the multispectral branch, if pre-trained weights are lacking, pre-training can be performed separately using a self-supervised spectral reconstruction task or contrastive learning, followed by joint fine-tuning.

[0074] Inference and Post-processing: The segmentation mask output by the model is thresholded, morphologically processed, and filtered for small connected components to remove noise. If patient-level diagnosis is required, the final conclusion is obtained based on multi-image voting or weighted fusion of overlapping regions. To deploy on mobile terminals / edge devices, a lightweight EfficientNet-Lite student network is distilled using knowledge distillation, while retaining the key attention submodule to meet latency and power consumption requirements.

[0075] Complete data flow description (end-to-end): Data acquisition > Denoising and spectral smoothing > Missing value handling and band selection > Spatial + spectral data augmentation > Patient-based segmentation of training / validation / test sets > Feeding into the network: Parallel encoding of RGB and MS branches (spectral addition / channel attention) > Multi-scale cross-modal fusion (concat + Transformer) > Decoder upsampling and skipping > Generation of shared features > Parallel output of multi-task heads (segmentation / classification / regression / detection) > Loss calculation and reverse update > Post-inference processing and clinical output (probability / mask / severity score) > Deployment and acceleration (pruning, quantization, distillation).

[0076] Network hierarchy and characteristic description table

[0077] System architecture and expansion: The oral-type intelligent health monitoring device of this invention adopts a modular design, and each sensor and processing module can be expanded and upgraded as needed. For example, in future versions, a blood oxygen saturation monitoring module can be added to achieve more comprehensive health monitoring.

[0078] The system's wireless communication module supports multi-device connection, allowing users to monitor multiple physiological indicators simultaneously. The data can be uploaded to a cloud server for big data analysis and remote medical services.

[0079] This invention provides a mouth-held intelligent health monitoring device suitable for users who need long-term monitoring of electrocardiograms, heart rates, respiration, and heart sounds, such as high-risk cardiovascular patients, athletes, and the elderly. Due to the device's high wearing comfort, users can monitor their health status anytime in daily life, promptly detect potential health problems, and prevent them from escalating.

[0080] Compared to existing portable or wearable devices, the mouth-held smart health monitoring device proposed in this embodiment has the following characteristics: Accuracy: Using multi-lead wet electrodes to collect ECG data provides better fit compared to dry electrodes used in common portable devices, and is less affected by factors such as the tightness of the wearer's clothing or skin perspiration.

[0081] Stability: The design of the mouth-held device can adapt to various usage scenarios. For example, it will not cause inaccurate data collection due to electrode displacement or water contact during running or swimming.

[0082] Portability: The design of the in-mouth device makes it highly portable, allowing users to monitor their health without affecting their daily activities.

[0083] Multifunctionality: It integrates ECG monitoring, heart rate monitoring, respiration monitoring and heart sound monitoring functions to provide comprehensive health monitoring services.

[0084] Intelligence: The data terminal integrates a neural network-based intelligent health status detection algorithm, which can realize localized functions such as heart rate abnormality, breathing abnormality and oral health check, and automatically trigger warning and alarm functions according to abnormal conditions.

[0085] Real-time performance: Through the wireless communication module, data can be transmitted to the data terminal and big data platform in real time, allowing users to view health data and receive health advice at any time.

[0086] Comfort: Made of medical-grade materials, it is comfortable to wear and is unlikely to cause allergic reactions or discomfort.

[0087] This invention relates to a mouth-mounted health monitoring device that combines biomedical engineering and modern communication technology, featuring high precision, low power consumption, and a superior user experience. It can provide daily health monitoring for ordinary users and also offer clinical monitoring data support for various medical institutions, including those practicing traditional Chinese medicine and Western medicine, playing a significant role in the early warning and prevention of cardiovascular diseases.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A mouth-held intelligent health monitoring device, characterized in that, This includes an oral cavity device, an oral cavity device storage compartment, and supporting software systems; The mouthpiece is in the form of a portable dental brace. The sealed shell is made of non-toxic and harmless resin or ceramic material. The front end is fitted to the contour of the teeth to prevent accidental swallowing. The rear end integrates a sensing device and a processing module. It also contains a lithium-ion rechargeable battery, a wireless charging receiving coil, and a wireless communication module. The oral device storage compartment has a built-in ultraviolet disinfection unit, a power supply unit, and a wireless charging transmitting coil; the software system includes computer platform software and mobile platform software, and establishes data interaction with the wireless communication module.

2. The oral-type intelligent health monitoring device according to claim 1, characterized in that, The sensing device includes 2-6 multi-lead wet electrodes, which are evenly arranged inside the mouthpiece and correspond to the areas of the gums, tongue and oral cavity rich in subcutaneous blood vessels. The wet electrodes are electrically connected to the processing module through wires and are used with conductive gel to reduce contact resistance.

3. The oral-type intelligent health monitoring device according to claim 1, characterized in that, The sensing device includes a photoplethysmography sensor, which is integrated at the rear end of the oral device and electrically connected to the processing module. It measures heart rate by detecting changes in blood flow within the oral cavity.

4. The oral-type intelligent health monitoring device according to claim 1, characterized in that, The sensing device includes a respiratory monitoring sensor group, which consists of an airflow sensor, a pressure sensor, and an infrared gas sensor, all of which are arranged at the rear of the mouthpiece. The sensing ends of the airflow sensor and the pressure sensor face the airflow channel of the nasal cavity or oral cavity, and the detection end of the infrared gas sensor corresponds to the path of exhaled gas.

5. The oral-type intelligent health monitoring device according to claim 4, characterized in that, The sensing device includes an oral environment monitoring sensor group, which includes a miniature infrared temperature sensor, a capacitive or resistive humidity sensor, a solid pH sensor, a multispectral imaging sensor, and a fluorescence imaging sensor. All of these are built into the corresponding areas where the mouthpiece contacts the oral tissue. The lenses of the multispectral imaging sensor and the fluorescence imaging sensor are directed toward the tongue, pharynx, tonsils, and gum areas.

6. The oral-type intelligent health monitoring device according to claim 1, characterized in that, The wireless communication module adopts a low-power Bluetooth or Wi-Fi module, which is integrated inside the mouthpiece and establishes a wireless data transmission channel with an external smartphone or tablet through an antenna.

7. The oral-type intelligent health monitoring device according to claim 5, characterized in that, The sensing device includes an attitude monitoring sensor, which consists of a high-sensitivity electronic gyroscope and an acceleration detection unit, and is integrated inside the mouthpiece. A GPS positioning module can be optionally added. The GPS positioning module is fixedly connected to the main body of the mouthpiece and electrically connected to the processing module.

8. The oral-type intelligent health monitoring device according to claim 7, characterized in that, The processing module has a built-in preamplifier, a filter circuit, and an analog-to-digital converter. The input of the preamplifier is connected to each sensor via wires, and the output is connected to the input of the analog-to-digital converter via the filter circuit. The output of the analog-to-digital converter is connected to the wireless communication module.

9. The oral-type intelligent health monitoring device according to claim 5, characterized in that, The software system has a built-in data storage module, a signal analysis module, and an early warning and rescue module. The data storage module is used to store the raw data and processed data collected by the sensor. The signal analysis module interacts with the data storage module. The early warning and rescue module can be linked with the GPS positioning module to output location information and trigger a call command.

10. The oral-type intelligent health monitoring device according to claim 1, characterized in that, The mouthpiece storage compartment is equipped with a cavity adapted to the mouthpiece. The irradiation range of the ultraviolet disinfection unit covers the inside of the cavity. The wireless charging transmitting coil corresponds to the wireless charging receiving coil of the mouthpiece. The power supply unit supplies power to the ultraviolet disinfection unit and the wireless charging transmitting coil.

Images