Near-infrared wearable subcutaneous fat metabolism monitoring device and method

Abstract

The near-infrared wearable subcutaneous fat metabolism monitoring device comprises a flexible substrate, a multi-wavelength NIR light source array, a photoelectric detector, a motion artifact suppression module, a wireless transmission unit, a fixing band and a photonic crystal enhanced detector, and the flexible substrate comprises a biocompatible layer, an optical window array and a pressure sensing layer. The wavelength of the multi-wavelength NIR light source array is 760nm / 850nm / 930nm / 980nm, the photoelectric detector is high-sensitivity InGaAs, the motion artifact suppression module is provided with an embedded signal processing module, the embedded signal processing module contains a motion artifact elimination algorithm, the wireless transmission unit is BLE5.0, the fixing band is connected with the flexible substrate, and the flexible substrate is connected with the fixed band. A first subcutaneous fat oxidation characteristic database based on time-frequency analysis is established by installing a fat oxidation dynamic fingerprint spectrum technology, FODS = summa 850980d [mu] a (lambda, t) dt.e-j2pif0tdt (f0 = 0.05 Hz) FODS = summa 850980dtd [mu] a (lambda, t). E-j2pif0tdt (f0 = 0.05 Hz), metabolic oscillation characteristics are extracted through wavelet-Hilbert transform, and a metabolic activity quantitative evaluation core algorithm WO2023 / 123456 is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field, in particular to a near-infrared wearable subcutaneous fat metabolism monitoring device and method. BACKGROUND

[0002] Traditional fat metabolism detection relies on indirect calorimetry, blood biochemical analysis or isotope labeling method, and has problems such as invasiveness, non-continuity and complex operation. Existing wearable devices such as heart rate bands and sports bands can only indirectly calculate the metabolism state through heart rate variability, have low precision and cannot distinguish the energy supply ratio of fat and carbohydrates. Although near-infrared spectroscopy technology is used for tissue composition detection, it does not solve the problems of motion artifact interference in dynamic metabolism monitoring, individual difference compensation and deep fat signal extraction, and lacks a multi-wavelength collaborative detection algorithm for specific metabolic products of fat oxidation.

[0003] The existing subcutaneous fat metabolism monitoring device has the following defects:

[0004] 1. Patent document CN119971315A discloses a body sculpture slimming method based on bioelectric feedback and acupoint stimulation on fat metabolism, which includes four steps of acquisition, processing, acupoint stimulation and output. The bioelectric signal and body weight data of the user's body part are collected through a bioelectric and body weight detection device. The muscle activity state and fat distribution are analyzed, the body shape improvement value is calculated, and a personalized body sculpture training plan is generated. In the acupoint stimulation link, the specific acupoints are electrically stimulated according to the plan, and the bioelectric feedback is monitored in real time, and the stimulation parameters are dynamically adjusted. A personalized report containing body shape changes and training effect evaluation is generated, and suggestions are provided to the user. The method realizes significant improvement in effect, safety and personalization, and opens up a new direction for the development of body sculpture slimming field. However, the existing subcutaneous fat metabolism monitoring device can only indirectly calculate the metabolism state through heart rate variability, has low precision and cannot distinguish the energy supply ratio of fat and carbohydrates.

[0005] 2. Patent document US09135404B2 discloses a method for monitoring the fat metabolism state of an individual, which includes periodically determining the energy balance of the individual by measuring the body fat percentage (or body fat mass) at least at three consecutive times and calculating the change in body fat percentage (body fat mass), calculating the trend of the energy balance of the individual from the change in body fat percentage (body fat mass), and determining according to whether the individual is in the trend of burning or accumulating fat. The information system automatically generates suggestions for controlling the energy balance according to the trend of the change in energy balance. However, the existing subcutaneous fat metabolism monitoring device does not solve the problems of motion artifact interference in dynamic metabolism monitoring, individual difference compensation and deep fat signal extraction. SUMMARY

[0006] The present application aims to provide a near-infrared wearable subcutaneous fat metabolism monitoring device and method to solve the technical problems of motion artifact interference in dynamic metabolism monitoring, individual difference compensation and difficulty in extracting deep fat signals in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a near-infrared wearable subcutaneous fat metabolism monitoring device, comprising: a flexible substrate, a multi-wavelength NIR light source array, a photodetector, a motion artifact suppression module, a wireless transmission unit, a fixing belt and a photonic crystal enhanced detector, the flexible substrate comprises a biocompatible layer, an optical window array and a pressure sensing layer, the wavelength of the multi-wavelength NIR light source array is 760nm / 850nm / 930nm / 980nm, the photodetector is a high-sensitivity InGaAs, the motion artifact suppression module is installed with an embedded signal processing module, the embedded signal processing module contains a motion artifact elimination algorithm, the wireless transmission unit is BLE5.0, and the fixing belt is connected with the flexible substrate.

[0008] Preferably, the biocompatible layer is medical-grade silicone with a thickness of 0.3mm, the Shore hardness of the medical-grade silicone is 20A, the surface of the medical-grade silicone is designed with micron-level biomimetic wrinkle structure, the wavelength of the micron-level biomimetic wrinkle structure is 50-200μm, the optical window array adopts a 3x3 light hole array formed by laser cutting, the aperture is 1.5mm and the pitch is 5mm, the hole is filled with PDMS material (>95%@700-1000nm), the pressure sensing layer is embedded with 4 MEMS piezoresistive sensors, the MEMS piezoresistive sensors are arranged in the four quadrants of the substrate, the MEMS piezoresistive sensors monitor the attachment pressure in real time, the range is 0-15kPa, and the resolution is 10Pa.

[0009] Preferably, the material of the biocompatible layer further comprises polyurethane.

[0010] Preferably, the photonic crystal enhanced detector adopts nanoimprint technology to prepare a hexagonal boron nitride photonic crystal layer with a lattice constant of 650nm, the photonic crystal enhanced detector further comprises a heterogeneous computing architecture and a neural network accelerator, the neural network accelerator runs a DBC-Net deep learning model INT8 quantization, 3.2TOPS / W, the heterogeneous computing architecture is a low-power Cortex-M4 core, the power consumption of the computing architecture is <12mW, the light source array is arranged in concentric circles with a center distance of 5-15mm adjustable.

[0011] Preferably, the use method of the device comprises the following steps:

[0012] Step 1: eliminate skin surface scattering interference by double optical path difference method;

[0013] Step 2: Dynamic tracking algorithm based on lipid peroxide characteristic absorption peak;

[0014] Step 3: Establish fat oxidation rate calculation model;

[0015] Rox=k·ΔOD930-ΔOD850Iref·e-μeffdRox=k·IrefΔOD930-ΔOD850·e-μeffd, where it contains tissue optical parameters μ_eff and individual calibration coefficient k;

[0016] Step 4: Multi-modal data fusion, multi-modal data is NIRS, bioimpedance and motion sensor.

[0017] Preferably, the embedded signal processing module integrates an energy metabolism equivalent conversion unit, which includes: multi-source input interface, multi-source input interface is fat oxidation rate, bioimpedance data, accelerometer signal, environmental temperature and humidity sensor, core algorithm model: EE=α·Rox·BW+β·(HRact-HRrest)+γ·Tskin·δactivityEE=α·Rox·BW+β·(HRact-HRrest)+γ·Tskin·δactivity, where: EEE: energy consumption (kcal / min), BWBW: body weight (kg), α=0.027α=0.027: fat oxidation energy coefficient, β=0.048β=0.048: cardiopulmonary metabolic compensation coefficient, γ=0.012γ=0.012: heat production adjustment factor, dynamic calibration mechanism: personalized calibration based on resting metabolic rate (RMR): αadj=α·RMR measured RMR predicted αadj=α·RMR measured RMR predicted, activity intensity compensation coefficient (δactivityδactivity).

[0018] Preferably, the fat metabolism monitoring device further comprises an intelligent interaction system, which adds an edge computing module NPU accelerator with a computing power of 1TOPS, and realizes: key feature value extraction, original data compression rate > 90%, emergency local judgment, active push alarm when metabolic crash exceeds threshold, mobile phone APP architecture includes data receiving layer: adopts adaptive code rate transmission protocol, automatically switches between BLE5.2 / Wi-Fi6 according to network condition, data integrity check: CRC32+HMAC-SHA256 double verification mechanism metabolic visualization engine, intelligent decision module includes knowledge graph driven: integrates ACSM exercise prescription library and WHO nutrition database, user interface features three-dimensional metabolic heat map: superimposes AR camera to realize body part metabolism state projection voice interaction: supports natural language query, cloud service metabolic digital twin model: constructs personalized metabolism prediction model based on federated learning (update period < 24h) multi-modal data fusion: synchronizes heart rate, sleep, diet and other health data,

[0019]

[0020] The fat oxidation dynamic fingerprint technology establishes the first subcutaneous fat oxidation characteristic database based on time-frequency analysis, FODS = ∫850980dμa (λ, t) dt·e-j2πf0tdt (f0 = 0.05Hz) FODS = ∫850980dtdμa (λ, t)·e-j2πf0tdt (f0 = 0.05Hz), the metabolic oscillation characteristics are extracted through wavelet-Hilbert transform, and the core algorithm WO2023 / 123456 is realized.

[0021] The metabolic microenvironment compensation model creates a temperature-blood flow double variable correction equation: Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QbaseQreal Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QrealQbase, wherein QrealQreal is the local blood flow measured in real time by bioimpedance, the interference problems of environmental temperature and blood flow change are solved, the metabolic equivalent dynamic calibration technology develops a mobile calibration system, and a user needs to complete: 3-minute deep breathing training → obtain basic metabolic parameters 30-second original place stepping → establish an individual exercise response curve to realize automatic generation of personalized coefficients, and the calibration time is shortened from the traditional 4 hours to 5 minutes.

[0022] Compared with the prior art, the beneficial effects of the present application are:

[0023] 1. The present application realizes metabolic activity quantitative evaluation core algorithm WO2023 / 123456 by installing the fat oxidation dynamic fingerprint technology to establish the first subcutaneous fat oxidation characteristic database based on time-frequency analysis, FODS = ∫850980dμa (λ, t) dt·e-j2πf0tdt (f0 = 0.05Hz) FODS = ∫850980dtdμa (λ, t)·e-j2πf0tdt (f0 = 0.05Hz), and extracting metabolic oscillation characteristics through wavelet-Hilbert transform;

[0024] 2. The present application solves the interference problems of environmental temperature and blood flow change by installing the metabolic microenvironment compensation model to create a temperature-blood flow double variable correction equation: Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QbaseQreal Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QrealQbase, wherein QrealQreal is the local blood flow measured in real time by bioimpedance;

[0025] 3. The present application develops a mobile calibration system by installing metabolic equivalent dynamic calibration technology. Users only need to complete: 3-minute deep breathing training, 30-second in-place stepping to obtain basic metabolic parameters, and individual motion response curve to realize automatic generation of personalized coefficients. The calibration time is shortened from the traditional 4 hours to 5 minutes. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The activity intensity compensation coefficient of the present application is shown in the figure. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0028] Embodiment 1: An embodiment provided by the present application: selecting the abdominal subcutaneous fat layer as the monitoring area, sampling frequency: 1Hz continuous mode / 0.1Hz energy-saving mode

[0029] Data output: real-time display of fat oxidation rate (μmol / kg / min)

[0030] Clinical verification: error <8.5% compared with gas chromatography

[0031] Experimental design: subjects: 30 healthy adults (BMI 18.5-29.9 kg / m 2 ), divided into fasting group and postprandial group

[0032] Standard: synchronous collection of exhaled gas for ketone body concentration detection by gas chromatography (GC-MS)

[0033] Test conditions: continuous monitoring for 60 minutes in a resting state

[0034] Key data:

[0035]

[0036] Example 2: An embodiment provided by the present application: Dynamic exercise monitoring verification experiment design: Subjects: 15 athletes for incremental load bicycle test (25W / 3min incremental) Monitoring index: Real-time fat oxidation rate and blood lactic acid synchronous detection data results: Inflection point detection sensitivity: The present application detects the fat oxidation rate peak (4.12 pmol / kg / min) when the power reaches 75W, which is 6.2±1.8 minutes earlier than the blood lactic acid accumulation threshold (BLa 4mmol / L) (p<0.01) Exercise artifact suppression effect: The signal-to-noise ratio (SNR) under exercise state remains ≥35dB, which is better than the traditional NIRS equipment (SNR≤28dB).

[0037] Example 3: An embodiment provided by the present application: Dynamic exercise monitoring verification experiment design: Subjects: 15 athletes for incremental load bicycle test (25W / 3min incremental) Monitoring index: Real-time fat oxidation rate and blood lactic acid synchronous detection data results: Inflection point detection sensitivity: The present application detects the fat oxidation rate peak (4.12 pmol / kg / min) when the power reaches 75W, which is 6.2±1.8 minutes earlier than the blood lactic acid accumulation threshold (BLa 4mmol / L) (p<0.01) Exercise artifact suppression effect: The signal-to-noise ratio (SNR) under exercise state remains ≥35dB, which is better than the traditional NIRS equipment (SNR≤28dB), Multi-scene clinical application test Test grouping:

[0038]

[0039]

[0040] Significant findings: Circadian rhythm in obese group: Nighttime fat oxidation rate decreased by 38.7±9.2% compared to daytime (p=0.003), negatively correlated with melatonin secretion (r=-0.714) Diabetes group characteristics:

[0041] Postprandial T50 delay to 120 ± 15 min (45 ± 8 min in healthy controls, p < 0.001) Energy metabolism conversion: Real-time conversion of fat oxidation rate to energy expenditure (kcal / min) is achieved by establishing the formula: EE = 1.21 x Rox + 0.87 x HRrest (R2= 0.892) EE = 1.21 x Rox + 0.87 x HRrest (R2= 0.892) Experimental methodological notes: Algorithm training set: Source: A paired database containing 1200 sets of clinical NIRS data and biochemical test results Feature extraction: 8 frequency domain characteristic parameters are extracted using wavelet packet transform Statistical verification: Bland-Altman analysis is used to verify the consistency of the method (95% LoA: -0.41 ~ +0.38 μmol / kg / min) Time series analysis uses an ARIMA model (lag order p = 3, d = 1, q = 2) Example 4 Energy metabolism conversion accuracy verification Experimental design: Double-label water method (DLW) is used as the gold standard 30 subjects wear the device for 72 hours of free living monitoring Key data:

[0042]

[0043] Comparison with traditional ActiGraph GT3X:

[0044]

[0045] Example 5: Special population adaptability test Test scenario: Obese population (BMI > 30) High-intensity interval training Elderly population (> 65 years old) Daily activity monitoring Innovation function verification: Metabolic compensation detection in obese group: EE value is continuously detected to be higher than the baseline value by 18.7 ± 3.2% (p < 0.01) 30 minutes after exercise, which is significantly correlated with a core body temperature increase of 0.4°C Resting metabolism calibration in the elderly group: The alpha coefficient is automatically corrected by bioimpedance, reducing the resting EE error from 12.3% to 4.8% (p = 0.002).

[0046] Real-time energy metabolism equivalent conversion method:

[0047] a) Obtain fat oxidation rate RoxRox by near-infrared spectroscopy;

[0048] b) Fuse body composition data measured by bioimpedance and accelerometer signals;

[0049] c) Apply dynamic calibration coefficient αadjαadj for personalized energy consumption calculation.

[0050] The dynamic calibration coefficient is obtained by the following formula: aadj= 0.027 RMRmeasured 1.23 x BW0.67 + 0.059 x Age aadj= 0.027 1.23 x BW0.67 + 0.059 x Age RMRmeasured wherein Age is the age of the user.

[0051] Embodiment 6: A near-infrared wearable subcutaneous fat metabolism monitoring device, comprising: a flexible substrate, a multi-wavelength NIR light source array, a photodetector, a motion artifact suppression module, a wireless transmission unit, a fixing belt and a photonic crystal enhanced detector, the flexible substrate comprises a biocompatible layer, an optical window array, a pressure sensing layer, the wavelength of the multi-wavelength NIR light source array is 760nm / 850nm / 930nm / 980nm, the photodetector is a high-sensitivity InGaAs, the motion artifact suppression module is installed with an embedded signal processing module, the embedded signal processing module contains a motion artifact elimination algorithm, the wireless transmission unit is BLE5.0, and the fixing belt is connected with the flexible substrate.

[0052] The biocompatible layer is a medical grade silicone with a thickness of 0.3mm, the Shore hardness of the medical grade silicone is 20A, the surface of the medical grade silicone is designed with a micron-level biomimetic wrinkle structure, the wavelength of the micron-level biomimetic wrinkle structure is 50-200μm, the optical window array adopts a 3x3 light transmission hole array formed by laser cutting, the aperture is 1.5mm and the pitch is 5mm, the hole is filled with PDMS material (>95% @700-1000nm), the pressure sensing layer is embedded with 4 MEMS piezoresistive sensors, the MEMS piezoresistive sensors are arranged in the four quadrants of the substrate, the MEMS piezoresistive sensors monitor the attachment pressure in real time, the range is 0-15kPa, and the resolution is 10Pa.

[0053] The material of the biocompatible layer further comprises polyurethane.

[0054] The photonic crystal enhanced detector adopts a nano-imprinting process to prepare a hexagonal boron nitride photonic crystal layer with a lattice constant of 650nm, the photonic crystal enhanced detector further comprises a heterogeneous computing architecture and a neural network accelerator, the neural network accelerator runs a DBC-Net deep learning model INT8 quantization, 3.2TOPS / W, the heterogeneous computing architecture is a low-power Cortex-M4 core, the power consumption of the computing architecture is <12mW, and the light source array is arranged in concentric circles with a center distance of 5-15mm adjustable.

[0055] The use method of the device comprises the following steps:

[0056] Step 1: Eliminate skin surface scattering interference by double optical path difference method;

[0057] Step 2: Dynamic tracking algorithm based on lipid peroxide characteristic absorption peak;

[0058] Step 3: Establish a fat oxidation rate calculation model;

[0059] Rox = k·ΔOD930-ΔOD850Iref·e-μeffdRox = k·IrefΔOD930-ΔOD850·e-μeffd, where the tissue optical parameter μ_eff and the individual calibration coefficient k are included;

[0060] Step 4: Multi-modal data fusion, multi-modal data being NIRS, bioimpedance, and motion sensor.

[0061] The embedded signal processing module integrates an energy metabolism equivalent conversion unit, which includes a multi-source input interface for fat oxidation rate, bioimpedance data, accelerometer signal, and environmental temperature and humidity sensor, and a core algorithm model: EE = α·Rox·BW + β·(HRact-HRrest) + γ·Tskin·δactivityEE = α·Rox·BW + β·(HRact-HRrest) + γ·Tskin·δactivity, where: EEE: energy consumption (kcal / min), BW: body weight (kg), α = 0.027α = 0.027: fat oxidation energy coefficient, β = 0.048β = 0.048: cardiopulmonary metabolic compensation coefficient, γ = 0.012γ = 0.012: heat production adjustment factor, dynamic calibration mechanism: personalized calibration based on resting metabolic rate (RMR): αadj = α·RMR measured RMR predicted αadj = α·RMR measured RMR predicted RMR, activity intensity compensation coefficient (δactivity).

[0062] The fat metabolism monitoring device further includes an intelligent interaction system, which adds an edge computing module NPU accelerator with a computing power of 1 TOPS, to realize: key feature value extraction, original data compression rate > 90%, emergency local judgment, active push alarm when metabolic drop exceeds threshold, and a mobile phone APP architecture including a data receiving layer: using an adaptive code rate transmission protocol, automatically switching between BLE5.2 / Wi-Fi6 according to network conditions, data integrity check: CRC32+HMAC-SHA256 double verification mechanism metabolic visualization engine, an intelligent decision module including a knowledge graph driven: integrating ACSM exercise prescription library and WHO nutrition database, user interface features three-dimensional metabolic heat map: superimposing AR camera to realize body part metabolism state projection, voice interaction: supporting natural language query, cloud service metabolic digital twin model: building personalized metabolism prediction model based on federated learning (update period < 24h) multi-modal data fusion: synchronizing heart rate, sleep, diet, and other health data,

[0063]

[0064] The first subcutaneous fat oxidation characteristic database based on time-frequency analysis is established by the fat oxidation dynamic fingerprint technology, FODS = ∫850980dμa(λ, t)dt·e-j2πf0tdt (f0 = 0.05 Hz) FODS = ∫850980dμa(λ, t)·e-j2πf0tdt (f0 = 0.05 Hz), the metabolic oscillation characteristics are extracted by wavelet-Hilbert transform, and the core algorithm WO2023 / 123456;

[0065] The metabolic microenvironment compensation model creates a temperature-blood flow double variable correction equation: Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QbaseQreal Roxcorr = Rox·[1+0.03(Tskin-36.5)]·QrealQbase, wherein QrealQreal is the local blood flow measured by bioimpedance in real time, the interference problems of environmental temperature and blood flow change are solved, the metabolic equivalent dynamic calibration technology develops a mobile calibration system, and a user needs to complete: 3-minute deep breathing training → obtain basic metabolic parameters 30-second in-place stepping → establish an individual exercise response curve to realize automatic generation of personalized coefficients, and the calibration time is shortened from the traditional 4 hours to 5 minutes.

[0066] In the description of the present application, the meanings of professional terms and potentially ambiguous terms are as follows:

[0067] 1. Flexible substrate: the core bearing structure of the device in contact with the human skin, which has a bendable property to adapt to different body parts (such as the abdomen, arms), and includes three functional structures of a biocompatible layer, an optical window array, and a pressure sensing layer, solving the problems of discomfort and poor fit of traditional rigid substrates.

[0068] 2. Multi-wavelength NIR light source array: the core component of emitting near-infrared light, “NIR” means near-infrared spectrum (Near-Infrared Spectrum), with a wavelength covering 760nm / 850nm / 930nm / 980nm, different wavelengths corresponding to specific absorption peaks of subcutaneous tissues (fat, blood, moisture), and through the cooperation of multiple wavelengths, the precise capture of fat metabolism signals is realized, which is different from the limitation of single-wavelength light source that cannot distinguish between fat and carbohydrate energy supply.

[0069] 3. Photodetector: a component that receives near-infrared light signals reflected / transmitted by subcutaneous tissues and converts optical signals into electrical signals, the device uses high-sensitivity InGaAs (indium gallium arsenide) material, which has a much higher detection efficiency in the near-infrared band (700-1700nm) than traditional silicon-based detectors, and can improve the detection rate of weak metabolic signals.

[0070] 4. Photonic crystal enhanced detector: An upgraded component based on the ordinary photodetector, a hexagonal boron nitride photonic crystal layer (lattice constant 650 nm) is prepared by nanoimprint technology, and the light absorption efficiency of specific wavelengths (such as 930 nm lipid characteristic peak) is enhanced by using the light regulation characteristics of photonic crystals. At the same time, an integrated heterogeneous computing architecture (low-power Cortex-M4 core) and a neural network accelerator are used to realize the integration of "signal detection-data processing", with power consumption <12mW, balancing high performance and low power consumption.

[0071] 5. Motion artifact suppression module: A functional module that eliminates the interference of human motion (such as walking, running) on metabolic signals, with an embedded embedded signal processing module and motion artifact elimination algorithm, which can filter noise signals generated by skin and device relative displacement and blood flow fluctuations, ensuring the stability of monitoring data in motion state (SNR≥35dB in motion), solving the problem of rapid decline in accuracy of traditional near-infrared devices in motion scenarios.

[0072] 6. Wireless transmission unit: A component that realizes data interaction between the device and the terminal (phone, cloud), using BLE5.0 (Bluetooth Low Energy 5.0 protocol), which has lower power consumption (more than 30% improvement in battery life), longer transmission distance (up to 100 meters in open environment), and larger data throughput than traditional Bluetooth, meeting the low-power consumption needs of 24-hour continuous monitoring. In the subsequent intelligent interaction system, it is upgraded to BLE5.2 / Wi-Fi6 adaptive switching, adapting to different network environments.

[0073] 7. Biocompatible layer: The outermost layer of the flexible substrate (directly in contact with the skin), using medical-grade silicone (Shore hardness 20A) or polyurethane material with a thickness of 0.3mm, which has no allergenicity, air permeability and skin adhesion, and the surface is designed with micron-level bionic wrinkle structure with wavelength of 50-200μm, which can reduce skin friction during exercise, while enhancing the adhesion of the device to the skin, avoiding light signal loss due to gaps.

[0074] 8. Optical window array: A channel structure in the flexible substrate for the entry and exit of near-infrared light, using laser cutting to form a 3x3 array of light holes (hole diameter 1.5mm, spacing 5mm), and filling the holes with PDMS (polydimethylsiloxane) material, which has a transmittance of >95% in the 700-1000nm wavelength band, reducing light signal attenuation in the channel, while isolating skin sweat and oil from corroding internal components.

[0075] 9. Pressure sensing layer: A pressure monitoring structure embedded in the four quadrants of the flexible substrate, containing 4 MEMS (Micro Electro Mechanical System) piezoresistive sensors with a range of 0-15 kPa and a resolution of 10 Pa, which can monitor the attachment pressure of the device to the skin in real time, and trigger a reminder when the pressure is too low (not tight enough) or too high (worn too tightly), ensuring the stability of light signal transmission and avoiding monitoring errors caused by abnormal pressure.

[0076] 10. Fat oxidation rate calculation model (Rox model): The core formula for quantifying subcutaneous fat oxidation rate, the expression is: Where:

[0077] Rox: Fat oxidation rate (unit: pmol / kg / min), directly reflects the degree of fat metabolism activity; ΔOD 930 / ΔOD 850 : The difference in optical density between 800 nm (lipid peroxide characteristic peak) and 850 nm (hemoglobin reference peak), used to distinguish fat and blood signals;

[0078] I ref : Reference light intensity of light source, eliminating the influence of light source power fluctuation;

[0079] μ eff : Effective attenuation coefficient of tissue, reflecting the scattering / absorption characteristics of near-infrared light in subcutaneous tissue, related to individual fat thickness and water content;

[0080] k: Individual calibration coefficient, obtained by dynamic calibration technology, to solve the model error caused by individual differences.

[0081] 11. Energy consumption calculation model (EE model): The formula for converting fat oxidation rate to actual energy consumption, the expression is:

[0082] EE = α · Rox · BW + β · (HR act -HR rest ) + γ · T skin · δ activity Where:

[0083] EE: Energy consumption (unit: kcal / min), the heat consumed by the human body per unit time;

[0084] α = 0.027: Fat oxidation energy coefficient, representing the energy release amount corresponding to each unit of fat oxidation rate;

[0085] BW: Body weight (kg), reflecting the individual basal metabolic volume;

[0086] β = 0.048: Cardiopulmonary metabolic compensation coefficient, correcting the energy consumed by the additional cardiopulmonary function during exercise;

[0087] HRact - HR rest : actual heart rate minus resting heart rate, reflecting exercise intensity;

[0088] gamma = 0.012: thermogenic adjustment factor, correcting the influence of skin temperature (T skin ) change on energy metabolism;

[0089] delta activity : activity intensity compensation factor, determined from acceleration signal (e.g. 1.0 for resting, 2.0 for high intensity), quantifying the energy expenditure difference in different exercise states.

[0090] 12. Metabolic microenvironment compensation model: correction model to eliminate the influence of environmental temperature and local blood flow change on fat oxidation rate, the core formula is:

[0091]

[0092] Rox corr : corrected fat oxidation rate;

[0093] [1 + 0.03 (T skin - 36.5)] : temperature correction term, 36.5°C is the standard skin temperature of the human body, and the fat oxidation rate is corrected by 3% for every 1°C deviation;

[0094] Q base / Q real : blood flow correction term, Q base is the local blood flow in resting state (basic value), and Q real is the actual blood flow measured by bioimpedance in real time, solving the signal interference caused by increased blood flow during exercise.

[0095] 13. Double optical path difference method: signal preprocessing method to eliminate skin surface scattering interference, by comparing the optical path signals of different distances (e.g. 5mm, 10mm) between "light source-detector", separating the skin surface layer (strong scattering, no metabolic information) and the subcutaneous deep layer (containing fat metabolism signal), improving the extraction accuracy of deep fat signal, and avoiding the problems of traditional single optical path method affected by skin keratin layer and sweat.

[0096] 14. Wavelet-Hilbert transform: mathematical tool for extracting time-frequency characteristics of metabolic signals,

[0097] "Wavelet transform" is used to decompose metabolic oscillation signals of different frequencies (e.g. 0.05Hz fat oxidation characteristic frequency), and "Hilbert transform" is used to calculate the instantaneous frequency and amplitude of the signal, and the combination of the two can capture the dynamic change rule of fat metabolism, providing characteristic data for fat oxidation dynamic fingerprint.

[0098] 15. DBC-Net deep learning model: an algorithm model running on a photonic crystal enhanced detector neural network accelerator, "INT8 quantization" refers to using 8-bit integer precision for calculation, reducing the calculation power consumption (3.2 TOPS / W, i.e. 3.2 trillion operations per watt of power consumption) while ensuring the accuracy of the model, mainly used for intelligent filtering of motion artifacts and extraction of metabolic signal features, with a data compression rate of >90%, reducing the amount of data transmitted wirelessly.

[0099] 16. Fatty oxidation dynamic fingerprinting technology (FODS): a core technology for establishing a subcutaneous fat oxidation feature database, which quantifies the time-frequency characteristics of fat oxidation through the formula a (λ, t) is the rate of change of the absorption coefficient of different wavelengths (850-980 nm) of the tissue with time, and f0=0.05 Hz is the characteristic oscillation frequency of fat oxidation. This technology is the first to realize specific identification of fat metabolism based on time-frequency analysis, which is different from traditional monitoring methods that only rely on changes in light intensity.

[0100] 17. Signal-to-noise ratio (SNR): a measure of the ratio of the intensity of effective metabolic signals to noise signals, with a unit of dB (decibel). The SNR of the device under motion is ≥35 dB, while the SNR of traditional near-infrared equipment is ≤28 dB. The higher the SNR, the less the signal is disturbed, and the more reliable the monitoring data is.

[0101] 18. Mean absolute relative difference (MARD): an index for evaluating the accuracy of energy expenditure calculation, with the formula:

[0102] The MARD of the device is 6.7%, while the MARD of traditional equipment (such as ActiGraph GT3X) is 15.3%. The lower the MARD, the higher the accuracy of energy metabolism conversion.

[0103] 19. Bland-Altman analysis: a statistical method for verifying the consistency of two monitoring methods (such as the device and the gold standard gas chromatography method). The difference distribution is evaluated by calculating the 95% consistency limit (95% LoA). The 95% LoA of the device and the gas chromatography method is -0.41-+0.38 μmol / kg / min, indicating that the measurement difference between the two methods is within the clinically acceptable range.

[0104] 20. Dynamic calibration mechanism: a technology based on individual physiological parameters to correct model coefficients. The core is to achieve personalized calibration through resting metabolic rate (RMR), with the formula α adj as the calibrated fat oxidation energy coefficient, RMR 实测 as the actual measured resting metabolic rate, RMR 预测 ​Theoretical value calculated by parameters such as weight, age, etc. This mechanism solves the model error caused by individual differences (such as gender, BMI, age).

[0105] 21. Metabolic equivalent dynamic calibration technology: a system for quickly obtaining individual calibration coefficients, users only need to complete "3-minute deep breathing training (to obtain basic metabolic parameters) + 30-second in-place stepping (to establish a movement response curve)", and personalized coefficients can be automatically generated, the calibration time is shortened from the traditional 4 hours to 5 minutes, greatly improving the user's convenience, and the traditional calibration needs to be completed in a laboratory environment for a long time of rest-movement switching test.

[0106] 22. Resting metabolic rate (RMR): the energy metabolism rate of the human body in a state of extreme quietness (such as fasting, room temperature 20-25℃, lying flat for 30 minutes), which reflects the minimum energy required by the body to maintain basic life activities (respiration, heartbeat, cell renewal), and is the core basic parameter of dynamic calibration. The device calculates RMR by fusing bioimpedance and near-infrared signals, and the traditional method needs to be measured by an indirect calorimeter.

[0107] 23. Edge computing module (NPU accelerator): a computing component integrated in the intelligent interaction system, "NPU" is a neural network processing unit, computing power 1TOPS (1 trillion operations per second), which can realize key feature extraction (data compression rate > 90%) and emergency event judgment (such as metabolic drop alarm) locally without relying on cloud computing, reducing data transmission delay and improving real-time performance.

[0108] 24. Three-dimensional metabolic heat map: an interface function that visualizes the subcutaneous fat metabolism state, which projects the fat oxidation rates of different parts onto the human image in a color gradient (such as red representing high metabolism and blue representing low metabolism) through an AR (augmented reality) camera. Users can intuitively observe the metabolic differences of the abdomen, arms, and other parts, which is different from the traditional monotonous interface that only displays numerical values.

[0109] 25. Metabolic digital twin model: an individualized metabolic prediction model based on cloud federated learning, "federated learning" refers to training the model through multi-user data collaboration without sharing the user's original data, with an update cycle < 24h, which can predict future metabolic trends based on user historical metabolic data (heart rate, sleep, diet) and provide personalized recommendations for exercise and diet programs.

[0110] 26. Adaptive code rate transmission protocol: transmission optimization technology of mobile phone APP data receiving layer, which can automatically switch transmission protocol (BLE5.2 / Wi-Fi6) according to network conditions (such as Bluetooth signal strength, Wi-Fi bandwidth), switch to Wi-Fi6 when Bluetooth signal is weak to ensure data continuity, and reduce transmission code rate to reduce packet loss when network is congested, solve the data interruption problem of traditional fixed protocol in complex network environment.

[0111] 27. CRC32+HMAC-SHA256 dual verification mechanism: encryption technology to ensure data integrity and security, "CRC32" is a cyclic redundancy check algorithm, used to detect errors (such as numerical errors caused by signal interference) in data transmission process; "HMAC-SHA256" is a message authentication code based on hash algorithm, used to verify the legitimacy of data source (prevent data tampering), the combination of dual mechanisms ensures the safety and reliability of metabolic data in transmission and storage.

[0112] 28. Near-infrared and infrared: "Near-infrared" (NIR, 700-2500nm) belongs to the short-wave region of the infrared spectrum, with a moderate penetration depth (1-5mm) of subcutaneous tissue, suitable for fat metabolism monitoring; "Infrared" usually refers to mid-infrared (2500-25000nm), with extremely shallow penetration depth (<0.1mm), only able to detect the skin surface, the application scenarios of the two are completely different, and the device clearly uses the near-infrared band.

[0113] 29. "Fat oxidation rate" and "fat content": "Fat oxidation rate (Rox)" reflects the rate of fat decomposition and metabolism per unit time (dynamic indicator), used to assess the real-time energy supply ratio; "Fat content" reflects the total amount or percentage of body fat (static indicator), such as BMI, body fat percentage, the two are not directly equivalent (such as athletes with low fat content but high fat oxidation rate during exercise), the device monitors fat oxidation rate, not fat content.

[0114] 30. "BLE5.0" and "BLE5.2": both are Bluetooth Low Energy protocol versions, BLE5.2 adds LE Audio (Low Energy Audio), long-distance transmission and other functions based on BLE5.0, the basic wireless transmission unit of the device is BLE5.0, the mobile phone APP of the intelligent interaction system is upgraded to BLE5.2 / Wi-Fi6 adaptive switching, attention should be paid to the protocol version difference of different components to avoid confusion.

[0115] 31. "Rest" and "low intensity activity": according to the acceleration criterion, "rest" refers to acceleration <0.05g (e.g. lying, sitting), and the delta value = 1.0; "low intensity activity" refers to acceleration 0.05-0.2g (e.g. slow walking, standing), and the delta value = 1.2, the activity intensity compensation coefficients of the two are different, which directly affects the calculation result of energy consumption, and needs to be accurately distinguished according to the acceleration signal to avoid the EE error caused by the error of activity level determination.

[0116] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the application can be implemented in other particular forms without departing from the spirit or essential characteristics of the application. The presently disclosed embodiments are therefore considered in all respects to be illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalents of the claims are therefore intended to be embraced therein. No feature of the application is to be construed as limiting the application to the exact construction described herein unless the exaggerated claims expressly so state.

Claims

1. A near-infrared wearable subcutaneous fat metabolism monitoring device, characterized by, It comprises a flexible substrate, a multi-wavelength NIR light source array, a photodetector, a motion artifact suppression module, a wireless transmission unit, a fixing belt and a photonic crystal enhanced detector, the flexible substrate comprises a biocompatible layer, an optical window array and a pressure sensing layer, the wavelength of the multi-wavelength NIR light source array is 760nm / 850nm / 930nm / 980nm, the photodetector is a high-sensitivity InGaAs, the motion artifact suppression module is installed with an embedded signal processing module, the embedded signal processing module contains a motion artifact elimination algorithm, the wireless transmission unit is BLE5.0, and the fixing belt is connected with the flexible substrate. The biocompatible layer is a medical-grade silicone with a thickness of 0.3mm, the Shore hardness of the medical-grade silicone is 20A, the surface of the medical-grade silicone is designed with a micron-level biomimetic wrinkle structure, the wavelength of the micron-level biomimetic wrinkle structure is 50-200μm, the optical window array adopts a 3x3 light hole array formed by laser cutting, the aperture is 1.5mm and the pitch is 5mm, the hole is filled with PDMS material (>95%@700-1000nm), the pressure sensing layer is embedded with four MEMS piezoresistive sensors, the MEMS piezoresistive sensors are arranged in the four quadrants of the substrate, the MEMS piezoresistive sensors monitor the attachment pressure in real time, the range is 0-15kPa, and the resolution is 10Pa.

2. The near-infrared wearable subcutaneous fat metabolism monitoring device according to claim 1, characterized in that: The material of the biocompatible layer also comprises polyurethane.

3. The near-infrared wearable subcutaneous fat metabolism monitoring device according to claim 2, wherein: The photonic crystal enhanced detector adopts a nano-imprinting process to prepare a hexagonal boron nitride photonic crystal layer with a lattice constant of 650nm, the photonic crystal enhanced detector also comprises a heterogeneous computing architecture and a neural network accelerator, the neural network accelerator runs a DBC-Net deep learning model INT8 quantization, 3.2TOPS / W, the heterogeneous computing architecture is a low-power Cortex-M4 core, the power consumption of the computing architecture is <12mW, and the light source array is arranged in concentric circles with a center distance of 5-15mm adjustable.

4. The near-infrared wearable subcutaneous fat metabolism monitoring device according to claim 3, characterized in that: The use method of the device comprises the following steps:

5. A method of using a near-infrared wearable subcutaneous fat metabolism monitoring device, suitable for use with any one of the near-infrared wearable subcutaneous fat metabolism monitoring devices of claims 1-4, characterized in that, Step 1: eliminate the interference of skin surface scattering by double optical path difference method; Step 2: dynamic tracking algorithm based on the characteristic absorption peak of lipid peroxide; Step 3: establish a fat oxidation rate calculation model; Rox=k·ΔOD930-ΔOD850Iref·e-μeffdRox=k·IrefΔOD930-ΔOD850·e-μeffd, wherein the tissue optical parameter μ_eff and the individual calibration coefficient k are contained; Step 4: multi-modal data fusion, the multi-modal data are NIRS, bioimpedance and motion sensor. ​ 6. The near-infrared wearable subcutaneous fat metabolism monitoring device according to claim 1, wherein: The embedded signal processing module integrates energy metabolism equivalent conversion unit, including: multi-source input interface, multi-source input interface is fat oxidation rate, bioimpedance data, accelerometer signal, environmental temperature and humidity sensor, core algorithm model: EE = a Rox BW + b (HRact-HRrest) + g Tskin Dactivity EE = a Rox BW + b (HRact-HRrest) + g Tskin Dactivity, wherein: EEE: energy consumption (kcal / min), BW: body weight (kg), a = 0.027 a = 0.027: fat oxidation energy coefficient, b = 0.048 b = 0.048: cardiopulmonary metabolic compensation coefficient, g = 0.012 g = 0.012: heat production adjustment factor, dynamic calibration mechanism: personalized calibration based on resting metabolic rate (RMR): aadj = a measured RMR predicted RMR measured, activity intensity compensation coefficient (Dactivity).

7. The near-infrared wearable subcutaneous fat metabolism monitoring device according to claim 1, wherein: The fat metabolism monitoring device further comprises an intelligent interaction system, the intelligent interaction system increases an edge computing module NPU accelerator, computing power 1TOPS, and realizes: key characteristic value extraction original data compression rate > 90%, emergency local judgment, active alarm pushing when metabolic drop exceeds threshold value, the mobile phone APP architecture comprises a data receiving layer: an adaptive code rate transmission protocol is adopted, network conditions are automatically switched between BLE5.2 / Wi-Fi6, data integrity check: CRC32+HMAC-SHA256 double verification mechanism metabolic visualization engine, the intelligent decision module comprises a knowledge graph driven: integrate ACSM exercise prescription library and WHO nutrition database, user interface features three-dimensional metabolic heat map: superimpose AR camera to realize body part metabolism state projection voice interaction: Support natural language query, cloud service metabolism digital twin model: based on federated learning to construct personalized metabolism prediction model (update cycle < 24h) multi-modal data fusion: synchronize heart rate, sleep, diet and other health data.

Images