Alveolar bone noninvasive detection method based on spatial offset Raman spectrum technology

Abstract

The invention discloses an alveolar bone noninvasive detection method based on a spatial offset Raman spectrum technology. The method comprises the following steps: S1, selecting a first near-infrared excitation wavelength and a second near-infrared excitation wavelength to form a multi-wavelength tunable excitation light beam; s2, synchronously acquiring a first group of original Raman spectrum signals under a plurality of spatial offset distances; s3, acquiring a second group of original Raman spectrum signals by using an acousto-optic composite modulation Raman enhancement mechanism; s4, inputting the soft tissue optical parameter set into a joint inversion equation; s5, obtaining a deep alveolar bone Raman characteristic spectrum data set; s6, training an alveolar bone tissue health recognition model; and S7, based on the alveolar bone tissue health recognition model, performing health state judgment on the deep alveolar bone Raman characteristic spectrum data set acquired in real time, and generating an alveolar bone health assessment result. The early screening and dynamic monitoring capabilities of alveolar bone diseases are greatly improved, and the clinical working efficiency of doctors and the doctor-seeing experience of patients are remarkably optimized.

Description

Technical Field

[0001] The present invention relates to the field of dental technology, and particularly to a non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy technology. Background Art

[0002] With the development of non-invasive detection technology, the medical detection field has been continuously exploring new means to replace traditional imaging methods with radiation risks and invasiveness. As the core component of the tooth support structure, the health status of the alveolar bone is directly related to the stability of the tooth body, the prognosis of implantation, and the early detection of oral systemic diseases. However, at present, the detection of the health status of the alveolar bone in clinical practice still mainly relies on X-ray imaging and cone beam computed tomography radiology methods. Although they have certain resolution and the ability to present anatomical structures, there are a series of insurmountable defects in practical applications.

[0003] First of all, the existing radiological detection means have a certain degree of risk of ionizing radiation exposure, are not suitable for sensitive populations such as children and pregnant women, and cannot meet the needs of frequent monitoring or dynamic assessment of high-risk patients. Secondly, it is difficult to obtain the true molecular structure information of bone tissue by relying on indirect indicators such as image recognition of alveolar bone density and height for such methods, and the detection sensitivity for early bone changes, micro-absorption or metabolic abnormalities is relatively low. Moreover, traditional CT and X-ray technologies have insufficient ability to distinguish between soft tissues and hard tissues, and are easily interfered by individual differences such as the thickness differences of the gingiva and mucosa, reducing the accuracy of evaluation.

[0004] In recent years, spatially offset Raman spectroscopy technology has received attention due to its ability to obtain spectral information deep in tissues. Some studies have attempted to apply it to bone tissue detection. However, the existing SORS applications mostly focus on shallow bone models or small animal experiments, and have not yet solved the problem of effectively extracting Raman signals of deep bone tissue under the condition of being wrapped by strongly scattering soft tissues in the complex tissue environment of the human oral cavity.

[0005] In summary, there is an urgent need to propose a new non-invasive detection method that adapts to the complex anatomical structure of the oral cavity and can efficiently identify the molecular characteristics of the alveolar bone, so as to achieve high-precision, dynamic and safe evaluation of the health status of the alveolar bone. Summary of the Invention

[0006] An object of the present invention is to propose a non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy technology. The present invention greatly improves the ability of early screening and dynamic monitoring of alveolar bone diseases, and significantly optimizes the clinical work efficiency of doctors and the medical experience of patients.

[0007] A non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy technology according to an embodiment of the present invention includes the following steps:

[0008] S1. Select the first near-infrared excitation wavelength and the second near-infrared excitation wavelength to form a dual-laser module, output a multi-wavelength tunable excitation beam through an electrically controlled switching method, couple the multi-wavelength tunable excitation beam into an improved spatially offset Raman detection device through a coaxial-annular dual-channel composite optical path, and vertically irradiate the corresponding area of the alveolar bone on the oral mucosa surface of the subject;

[0009] S2. Set an annular lateral collection area in the improved spatially offset Raman detection device, use a microelectromechanical system mirror to perform zoning control on the annular lateral collection area, and synchronously collect the first set of original Raman spectral signals at multiple spatial offset distances;

[0010] S3. During the irradiation of the multi-wavelength tunable excitation beam, synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone, perform frequency shift and phase modulation on the multi-wavelength tunable excitation beam through an acousto-optic modulator, and collect the second set of original Raman spectral signals using the acousto-optic composite modulation Raman enhancement mechanism;

[0011] S4. Use a short-pulse broadband fluorescence excitation light source to obtain a fluorescence image of the gingival soft tissue, estimate a set of soft tissue optical parameters based on the fluorescence image of the gingival soft tissue, and input the set of soft tissue optical parameters into a joint inversion equation;

[0012] S5. Input the first set of original Raman spectral signals and the second set of original Raman spectral signals together with the set of soft tissue optical parameters into the joint inversion equation, perform bone-soft separation calculation, and obtain a deep alveolar bone Raman characteristic spectral data set;

[0013] S6. Reorganize the deep alveolar bone Raman characteristic spectral data set into a spectral depth map stack according to the multi-wavelength dimension and the multi-depth dimension, input it into a convolutional neural network model for feature extraction learning, and train an alveolar bone tissue health recognition model;

[0014] S7. Based on the alveolar bone tissue health recognition model, perform a health status discrimination on the deep alveolar bone Raman characteristic spectral data set collected in real time, generate an alveolar bone health assessment result, and output the alveolar bone health assessment result to a display terminal in real time.

[0015] Optionally, S1 includes the following steps:

[0016] S11. Select the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2 in the excitation light source. The first near-infrared excitation wavelength and the second near-infrared excitation wavelength are set to 785 - 800 nm and 830 - 850 nm respectively. An electrically controlled switching method is used to achieve the switching of different excitation wavelengths in the excitation light source, and a multi-wavelength tunable excitation beam that can output different excitation wavelengths is constructed. The output powers of the multi-wavelength tunable excitation beam are respectively calibrated as the output power at the first near-infrared excitation wavelength and the output power at the second near-infrared excitation wavelength, ensuring that both of these output powers are greater than zero and do not exceed the maximum safe power threshold for laser irradiation of human tissues in the oral cavity. The maximum safe power threshold for laser irradiation of human tissues in the oral cavity is set to 5 mW;

[0017] S12. Through an electrically controlled wavelength switching driver, the excitation time window of the first near-infrared excitation wavelength and the excitation time window of the second near-infrared excitation wavelength are respectively set, so that the first near-infrared excitation wavelength and the second near-infrared excitation wavelength are sequentially output within a single detection cycle. The excitation time window of the first near-infrared excitation wavelength and the excitation time window of the second near-infrared excitation wavelength do not overlap in time, and the overall detection cycle is defined as the sum of these two excitation time windows, ensuring that the excitation signals at different excitation wavelengths can be clearly distinguished in time;

[0018] S13. The multi-wavelength tunable excitation beam is sequentially coupled into a coaxial-ring dual-channel composite optical path structure. The coaxial-ring dual-channel composite optical path structure includes a central optical axis incident path and an annular concentric collection channel. The central optical axis incident path is used for the primary incidence of the multi-wavelength tunable excitation beam. The inner diameter and outer diameter of the annular concentric collection channel are set to ensure that the multi-wavelength tunable excitation beam uniformly irradiates the oral mucosa surface. The ratio of the difference between the outer diameter and the inner diameter of the annular concentric collection channel to the inner diameter does not exceed the maximum allowable radial non-uniformity factor, and the value range of the maximum allowable radial non-uniformity factor is 0.1 to 0.3;

[0019] S14. The multi-wavelength tunable excitation beam modulated by the coaxial-ring dual-channel composite optical path is vertically irradiated onto the oral mucosa surface of the subject to be examined, and the incident angle θ in = 90° is set to form a directional light field irradiation area. The size of the directional light field irradiation area (L x , L y ), where L x = 5 mm, L y = 5 mm. The center of the irradiation area is aligned with the center coordinates (x0, y0) of the alveolar bone target area, so as to satisfy the following spatial matching constraint: |(x c - x0, y c - y0)| ≤ ∈, where L x represents the width of the laser irradiation area in the horizontal direction, L yrepresents the length of the laser irradiation area in the vertical direction, (x c , y c ) is the center coordinate of the laser irradiation area, ∈ is the positioning error threshold, and ∈ ≤ 0.5 mm.

[0020] Optionally, S2 includes the following steps:

[0021] S21. Set an annular lateral collection area around the periphery of the improved spatially offset Raman detection device. The annular lateral collection area is arranged symmetrically around the central optical axis and is composed of several Raman signal collection and detection units. The number of Raman signal collection and detection units is set to N d , and the spatial offset distance of each Raman signal collection and detection unit relative to the central excitation optical axis is set to where d i represents the spatial offset distance of the i-th Raman signal collection and detection unit, i = 1, 2,..., N d , N d ≥ 3;

[0022] S22. Embed a microelectromechanical system mirror assembly between the Raman signal collection and detection units in the annular lateral collection area. The microelectromechanical system mirror assembly is used to partition and switch the Raman signal paths with different spatial offset distances and guide them, realizing the switching control of the detection paths at different spatial offset distances. The working state of the mirror is adjusted through a control voltage signal , where represents the control voltage of the microelectromechanical system mirror corresponding to the i-th spatial offset channel;

[0023] S23. Synchronously regulate the control voltage signal with the excitation period of the multi-wavelength tunable excitation beam of the excitation laser. Perform laser irradiation and collect the corresponding Raman scattering signals at each spatial offset distance d i to construct an original Raman spectral signal group where represents the Raman spectral signal intensity obtained at the spatial coordinate (x, y) on the oral mucosa surface under the conditions of excitation wavelength λ k and spatial offset distance d i , λ k ∈ λ1, λ2;

[0024] S24. Perform sampling uniformity verification on each Raman spectral signal in the original Raman spectral signal group R1. Set the target sampling power threshold P target and the actual channel sampling power P di . If The Raman spectrum signal is then recorded as a valid sampling result, where δ is the allowable sampling power deviation tolerance, and δ ≤ 0.3 mW;

[0025] S25. Reorganize all the valid Raman spectrum signals that pass the sampling uniformity verification according to the spatial offset distance d i into the first set of original Raman spectrum signal sets where represents the original data set of Raman signals obtained under spatial offset control and without any bone-soft tissue separation or denoising processing.

[0026] Optionally, the improved spatial offset Raman detection device includes the following structures:

[0027] Set an annular detection structure as the main collection area of Raman signals. A central laser incident channel is set at the center of the annular detection structure. The central laser incident channel is used to transmit a multi-wavelength tunable excitation beam to the oral mucosa surface of the object to be examined. The optical axis of the central laser incident channel is defined as the central optical axis. The annular detection structure is symmetrically arranged around the central optical axis, and the outer diameter is set as r out , and the inner diameter is set as r in , satisfying (r out -r in ) / r in ≤ α, where α is the allowable maximum radial non-uniformity factor;

[0028] Set N d equidistant Raman signal collection and detection units within the annular detection structure. The distance between the center of each Raman signal collection and detection unit and the central optical axis is set as the spatial offset distance d i , where in millimeters, i = 1, 2,..., N d , N d ≥ 3, for synchronously collecting Raman signals at different spatial offset distances;

[0029] Each Raman signal collection and detection unit is connected to a set of band-pass filter modules. The central transmission wavelength of the band-pass filter module is set as λ R , and the spectral bandwidth of the band-pass filter module is Δλ, where λ R represents the center wavelength of the target Raman signal, and Δλ represents the allowable wavelength offset range. The band-pass filter module is used to shield the excitation laser reflection interference and only transmit the wavelength corresponding to the target Raman signal;

[0030] A microelectromechanical system mirror assembly is set at the front end of each Raman signal collection and detection unit. The microelectromechanical system mirror assembly receives the corresponding control voltage signal to achieve fine adjustment of the optical path angle and control the exit angle of the microelectromechanical system mirror assembly Form a cross focal region with the incident angle θ of the central laser incident path in to enhance the collection coupling efficiency of the deep Raman signal of the alveolar bone;

[0031] The output light beams of all Raman signal collection and detection units are introduced into the Raman signal spectrum acquisition module through an optical fiber bundle. The Raman signal spectrum acquisition module includes a grating spectrometer and a high-sensitivity CCD detector, and the CCD detector records the corresponding Raman signal intensities at multiple spatial offset distances to form an original Raman spectrum signal group.

[0032] Optionally, S3 includes the following steps:

[0033] S31. During the irradiation of the oral mucosa surface of the subject by the multi-wavelength tunable excitation light beam, synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone, and set the frequency f u and ultrasonic flux I LIFU of the low-intensity focused ultrasound, where f u represents the central frequency of the focused ultrasound, and I LIFU represents the power per unit area of the low-intensity focused ultrasound;

[0034] S32. Set an acousto-optic modulator in the incident path of the multi-wavelength tunable excitation light beam to perform frequency shift and phase modulation on the excitation lights of the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2, and set the modulation frequency to f m and the modulation amplitude to φ m , where f m represents the frequency shift applied by the acousto-optic modulator, and φ m represents the phase shift angle after modulation;

[0035] S33. Make the modulated excitation light output by the acousto-optic modulator completely coincide with the low-intensity focused ultrasound irradiation area in space and time, so that the modulated excitation light acts on the tissue interface modulated by the low-intensity focused ultrasound, inducing a change in the microscopic elastic modulus between the alveolar bone and the soft tissue covering it, and forming a Raman scattering response enhancement region;

[0036] S34. Synchronously collect the Raman scattering signals under the condition of the modulated excitation light on each spatial offset distance d i channel in the Raman scattering response enhancement region, and construct a set of modulated Raman spectrum signals where represents the Raman spectrum signal intensity obtained under the action of acousto-optic modulation at the spatial coordinates (x, y) under the conditions of the excitation wavelength λ k and the spatial offset distance d i , and λ k ∈λ1,λ2;

[0037] S35. Frequency-lock extraction is performed on all Raman spectroscopy signals in the modulated Raman spectroscopy signal set R2 to extract the coherent Raman component synchronized with the modulation frequency f m After removing the non-modulated background signal, the frequency-locked Raman components are uniformly recombined into the second set of original Raman spectroscopy signal sets wherein represents the original data set of Raman spectroscopy signals collected based on the acousto-optic composite modulation mechanism and not yet subjected to bone-soft tissue separation and noise reduction.

[0038] Optionally, S4 includes the following steps:

[0039] S41. A short-pulse broadband fluorescence excitation light source with a central excitation wavelength of 405 - 440 nm and a pulse width of τ p is used to irradiate the gingival soft tissue on the surface of the oral mucosa of the subject to be examined, and the time-integrated fluorescence original image F F (x, y) is collected and recorded; raw (x, y);

[0040] S42. Dark current subtraction and flat-field correction are performed on the collected fluorescence original image F raw (x, y) to obtain the corrected fluorescence image F cal (x, y), and the region of interest of the gingival soft tissue is selected in the corrected fluorescence image;

[0041] S43. Calculate the average fluorescence intensity within the region of interest of the gingival soft tissue and combine the average fluorescence intensity with the light power per unit area I F and the equivalent optical path d of the oral mucosa ref to estimate the absorption coefficient μ a,soft of the gingival soft tissue;

[0042] S44. Calculate the effective attenuation coefficient of the gingival soft tissue by combining the absorption coefficient of the gingival soft tissue and the reference scattering coefficient;

[0043] S45. Construct a set of soft tissue optical parameters Θ soft , and the set of soft tissue optical parameters includes the absorption coefficient of the gingival soft tissue and the effective attenuation coefficient of the gingival soft tissue.

[0044] Optionally, S5 includes the following steps:

[0045] S51. Perform spatial coordinate registration on the first set of original Raman spectroscopy signal sets and the second set of original Raman spectroscopy signal sets to obtain the spatially registered combined data set of Raman spectroscopy signals

[0046] S52. Combine the spatially registered Raman spectral signal combined dataset D match and the soft tissue optical parameter set and input them into the bone-soft tissue joint inversion equation for bone-soft tissue separation calculation;

[0047] S53. Based on the solution results of the bone-soft tissue joint inversion equation, extract and reconstruct the Raman spectral components representing the characteristics of the deep alveolar bone to form a deep alveolar bone Raman characteristic spectral dataset wherein, represents the Raman characteristic spectral signal intensity of the alveolar bone tissue obtained at the spatial coordinates (x, y) on the oral mucosa surface under the conditions of the excitation wavelength λ k and the spatial offset distance d i after bone-soft tissue separation calculation.

[0048] Optionally, S52 includes the following steps:

[0049] S521. Introduce the spatially registered Raman spectral signal combined dataset and the soft tissue optical parameter set into the bone-soft tissue joint inversion equation as input variables to establish the following optimization model:

[0050]

[0051] wherein, represents the set of Raman characteristic signals of the alveolar bone to be solved, is the loss function, which is used to measure the residual error between the combined signal and the theoretically inverted signal under the soft tissue model;

[0052] S522. The loss function contains two parts of residual terms, namely the basic Raman channel residual term and the modulated Raman channel residual term:

[0053]

[0054] wherein, and are the theoretical Raman signals derived from the current estimated Raman characteristic signals of the alveolar bone and the soft tissue optical parameters through the optical propagation model, respectively, and w1 and w2 are the residual weighting coefficients of the basic channel and the modulated channel;

[0055] S523. In each iteration process, the soft tissue optical parameter set Θ soft acts on the estimated Raman characteristic signal R bone of the alveolar bone to calculate the simulated spectral value;

[0056] S524. Repeat the optical propagation simulation and loss function evaluation process in step S523, and continuously adjust the Raman characteristic signal of the alveolar bone through gradient descent to make the loss function Converge to the global minimum or meet the convergence threshold condition, and finally complete the bone-soft tissue separation solution process.

[0057] Optionally, S6 includes the following steps:

[0058] S61. Organize each alveolar bone Raman characteristic spectral signal in the deep alveolar bone Raman characteristic spectral dataset in three dimensions according to the excitation wavelength dimension, spatial offset distance dimension, and spatial coordinate dimension to construct a spectral-depth joint distribution map stack T bone (x, y, λ k , d i ), and the constructed spectral-depth joint distribution map stack is used to describe the alveolar bone Raman characteristic signal intensity distribution under two-dimensional spatial coordinate positions, different excitation wavelengths, and different spatial offset distances. The spectral-depth joint distribution map stack represents the alveolar bone Raman characteristic signal under the spatial position (x, y), excitation wavelength λ k and spatial offset distance d i conditions;

[0059] S62. Take the spectral-depth joint distribution map stack as the input and input it into the convolutional neural network model for processing. In the convolutional neural network model, perform end-to-end auto-encoding processing on the input spectral-depth joint distribution map stack through continuous convolution calculations, activation function applications, and downsampling operations. The encoding result is a multi-scale spatial-spectral joint feature F bone ;

[0060] S63. During the feature extraction process, introduce a supervised training strategy. Evaluate the difference between the output result of the convolutional neural network model and the corresponding alveolar bone health status label by constructing an objective loss function. The objective loss function is used to guide parameter updates, and the goal is to minimize this loss value. The input of the convolutional neural network model is the spectral-depth joint distribution map stack, and the output is the prediction result. The prediction result is compared with the pre-defined set of health status labels to measure the classification error;

[0061] S64. Update the parameter set in the convolutional neural network model through the backpropagation algorithm. Recalculate the difference between the predicted output and the health status label in each iteration, and update the network parameters according to the gradient direction of the objective loss function; repeat the process of feature extraction, prediction, and parameter update until the objective loss function converges or meets the preset early stopping condition; when the objective loss function reaches the minimum or meets the training termination condition, obtain the trained alveolar bone tissue health recognition model

[0062] Optionally, S7 includes the following steps:

[0063] S71. Input the deep alveolar bone Raman characteristic spectrum data set into the alveolar bone tissue health recognition model for classification and inference to obtain the health discrimination label at the corresponding coordinate region (x, y). Form a set of alveolar bone health discrimination labels that includes all detection points.

[0064] S72. The set of alveolar bone health discrimination labels is matched and analyzed with the pre-set alveolar bone health status evaluation rules, and the alveolar bone health assessment result is output. The alveolar bone health status evaluation rules are classified and judged according to the average Raman characteristic intensity, the integrity of the Raman peak structure, and the consistency index between offset channels:

[0065] The alveolar bone health status is Grade I (normal): If the average intensity of the alveolar bone Raman signal and the signal-to-noise ratio SNR of the main peak peak ≥S1, and at the same time the standard deviation σ between the spatial offset channels d ≤δ1;

[0066] The alveolar bone health status is Grade II (mild absorption): If the average intensity T2 of the alveolar bone Raman signal ≤R bone <T1, or SNR peak <S1, but still satisfies σ d ≤δ2;

[0067] The alveolar bone health status is Grade III (moderate to severe absorption): If the average intensity R of the alveolar bone Raman signal bone <T2, or σ d >δ2;

[0068] Among them, R bone is the average intensity of the alveolar bone Raman signal, SNR peak is the signal-to-noise ratio of the main Raman peak, σ d is the signal standard deviation between the spatial offset channels, and T1, T2, S1, δ1, and δ2 are the evaluation thresholds built into the model;

[0069] S74. Visualize and output the alveolar bone health assessment result determined according to the evaluation rules in real time at the corresponding spatial coordinates (x, y) to the display terminal, construct a two-dimensional visualization map, and the output content includes the health status classification label, the comprehensive risk rating identifier, and the numerical range of the key feature parameters corresponding to each detection point.

[0070] The beneficial effects of the present invention are:

[0071] (1) The present invention adopts a multi-wavelength spatially offset Raman spectroscopy detection device and an acousto-optic modulation synchronous enhancement mechanism. It uses a MEMS controllable offset probe array to synchronously collect Raman signals at different depths, and constructs a combined bone and soft tissue inversion equation to couple and invert multi-channel Raman signals and a set of soft tissue optical parameters. A combined optimization loss function is introduced in the signal separation link, significantly improving the separation accuracy of Raman characteristic signals of deep alveolar bone and effectively suppressing the interference of soft tissue and background noise.

[0072] (2) The present invention reorganizes the deep alveolar bone Raman characteristic spectral data set into a high-order tensor according to the multi-wavelength and multi-offset depth dimensions and inputs it into a convolutional neural network model to end-to-end realize the automatic extraction of spatial-spectral joint features and the recognition of health status. Compared with existing traditional methods based on simple statistical features or one-dimensional spectral analysis, the proposed convolutional neural network can mine the distribution law of the molecular structure of alveolar bone at multiple scales and multiple spatial levels, realizing the automatic discrimination of multiple categories such as healthy, mild absorption, and moderate to severe absorption.

[0073] (3) The hierarchical output mechanism proposed by the present invention based on a discriminant label set and a health status evaluation rule can visually output the health status results of each detection point in real time in the form of a two-dimensional distribution heat map to the terminal. The health assessment process combines a multi-parameter threshold adaptive rule to realize the intelligent fusion determination of key indicators such as the average intensity of alveolar bone Raman signals, the signal-to-noise ratio of the main peak, and the inter-channel consistency, significantly improving the accuracy and transparency of risk grading. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0075] Figure 1 is a flowchart of a non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy technology proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0077] Refer to Figure 1 , a non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy technology, includes the following steps:

[0078] S1. Select the first near-infrared excitation wavelength and the second near-infrared excitation wavelength to form a dual-laser module, output a multi-wavelength tunable excitation beam through an electrically controlled switching method, couple the multi-wavelength tunable excitation beam into an improved spatially offset Raman detection device through a coaxial-ring dual-channel composite optical path, and vertically irradiate the corresponding area of the alveolar bone on the oral mucosa surface of the subject;

[0079] S2. Set an annular lateral collection area in the improved spatially offset Raman detection device, use a microelectromechanical system mirror to perform partition control on the annular lateral collection area, and synchronously collect the first set of original Raman spectral signals at multiple spatial offset distances;

[0080] S3. During the irradiation of the multi-wavelength tunable excitation beam, synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone, and perform frequency shift and phase modulation on the multi-wavelength tunable excitation beam through an acousto-optic modulator, and collect the second set of original Raman spectral signals by using the acousto-optic composite modulation Raman enhancement mechanism;

[0081] S4. Use a short-pulse broadband fluorescence excitation light source to obtain a fluorescence image of the gingival soft tissue, estimate a set of soft tissue optical parameters based on the fluorescence image of the gingival soft tissue, and input the set of soft tissue optical parameters into a joint inversion equation;

[0082] S5. Input the first set of original Raman spectral signals and the second set of original Raman spectral signals together with the set of soft tissue optical parameters into the joint inversion equation, perform bone-soft separation calculation, and obtain a deep alveolar bone Raman characteristic spectral data set;

[0083] S6. Reorganize the deep alveolar bone Raman characteristic spectral data set into a spectral depth map stack according to the multi-wavelength dimension and the multi-depth dimension, input it into a convolutional neural network model for feature extraction learning, and train an alveolar bone tissue health recognition model;

[0084] S7. Based on the alveolar bone tissue health recognition model, discriminate the health status of the deep alveolar bone Raman characteristic spectral data set collected in real time, generate an alveolar bone health assessment result, and output the alveolar bone health assessment result to a display terminal in real time.

[0085] In this embodiment, S1 includes the following steps:

[0086] S11. Select the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2 in the excitation light source. The first near-infrared excitation wavelength and the second near-infrared excitation wavelength are set to 785 - 800 nanometers and 830 - 850 nanometers respectively. An electrically controlled switching method is used to achieve the switching of different excitation wavelengths in the excitation light source, and a multi-wavelength tunable excitation beam that can output different excitation wavelengths is constructed. The output powers of the multi-wavelength tunable excitation beam are respectively calibrated as the output power at the first near-infrared excitation wavelength and the output power at the second near-infrared excitation wavelength, ensuring that both of these output powers are greater than zero and do not exceed the maximum safe power threshold for laser irradiation of human tissues in the oral cavity. The maximum safe power threshold for laser irradiation of human tissues in the oral cavity is set to 5 milliwatts;

[0087] S12. Through an electrically controlled wavelength switching driver, set the excitation time window of the first near-infrared excitation wavelength and the excitation time window of the second near-infrared excitation wavelength respectively, so that the first near-infrared excitation wavelength and the second near-infrared excitation wavelength are output in sequence within a single detection cycle. The excitation time window of the first near-infrared excitation wavelength and the excitation time window of the second near-infrared excitation wavelength do not overlap in time, and the overall detection cycle is defined as the sum of these two excitation time windows, ensuring that the excitation signals at different excitation wavelengths can be clearly distinguished in time;

[0088] S13. Couple the multi-wavelength tunable excitation beam into a coaxial-ring dual-channel composite optical path structure in sequence. The coaxial-ring dual-channel composite optical path structure includes a central optical axis incident path and an annular concentric collection channel. The central optical axis incident path is used for the primary incidence of the multi-wavelength tunable excitation beam. The inner diameter and outer diameter of the annular concentric collection channel are set to ensure that the multi-wavelength tunable excitation beam uniformly irradiates the oral mucosa surface. The ratio of the difference between the outer diameter and the inner diameter of the annular concentric collection channel to the inner diameter does not exceed the maximum allowable radial non-uniformity factor, and the value range of the maximum allowable radial non-uniformity factor is 0.1 to 0.3;

[0089] S14. Vertically irradiate the multi-wavelength tunable excitation beam modulated by the coaxial-ring dual-channel composite optical path onto the oral mucosa surface of the subject to be examined, and set the incident angle θ in = 90°, forming a directional light field irradiation area. The size of the directional light field irradiation area is (L x , L y ), where L x = 5mm, L y = 5mm. The center of the irradiation area is aligned with the center coordinates (x0, y0) of the alveolar bone target area, so as to satisfy the following spatial matching constraint: |(x c - x0, y c - y0)| ≤ ∈, where L x represents the width of the laser irradiation area in the horizontal direction, and L yrepresents the length of the laser irradiation area in the vertical direction, (x c , y c ) is the center coordinate of the laser irradiation area, ∈ is the positioning error threshold, and ∈ ≤ 0.5 mm.

[0090] In this embodiment, S2 includes the following steps:

[0091] S21. Set an annular lateral collection area around the periphery of the improved spatially offset Raman detection device. The annular lateral collection area is arranged symmetrically around the central optical axis and is composed of several Raman signal collection and detection units. The number of Raman signal collection and detection units is set to N d , and the spatial offset distance of each Raman signal collection and detection unit relative to the central excitation optical axis is set to where d i represents the spatial offset distance of the i-th Raman signal collection and detection unit, i = 1, 2,..., N d , and N d ≥ 3;

[0092] S22. Embed a microelectromechanical system mirror assembly between each Raman signal collection and detection unit in the annular lateral collection area. The microelectromechanical system mirror assembly is used to partition and switch the Raman signal paths with different spatial offset distances and guide them, realizing the switching control of the detection path under different spatial offset distances. The working state of the mirror is adjusted through a control voltage signal , where represents the control voltage of the microelectromechanical system mirror corresponding to the i-th spatial offset channel;

[0093] S23. Synchronously regulate the control voltage signal with the excitation period of the multi-wavelength tunable excitation beam of the excitation laser. Perform laser irradiation and collect the corresponding Raman scattering signals at each spatial offset distance d i to construct an original Raman spectral signal group where represents the Raman spectral signal intensity obtained at the spatial coordinates (x, y) on the oral mucosa surface under the conditions of excitation wavelength λ k and spatial offset distance d i , and λ k ∈ λ1, λ2;

[0094] S24. Perform sampling uniformity verification on each Raman spectral signal in the original Raman spectral signal group R1. Set the target sampling power threshold P target and the actual channel sampling power P di . If |P di - P targetIf |≤δ, the Raman spectroscopy signal is recorded as a valid sampling result, where δ is the allowable sampling power deviation tolerance and δ ≤ 0.3 mW;

[0095] S25. Reorganize all valid Raman spectroscopy signals that pass the sampling uniformity verification according to the spatial offset distance d i into the first set of original Raman spectroscopy signal sets where represents the set of original Raman signal data obtained under spatial offset control and without any bone-soft separation or denoising processing.

[0096] In this embodiment, the improved spatial offset Raman detection device includes the following structures:

[0097] Set an annular detection structure as the main collection area of Raman signals. A central laser incident channel is set at the center of the annular detection structure. The central laser incident channel is used to transmit a multi-wavelength tunable excitation beam to the oral mucosa surface of the subject to be examined. The optical axis of the central laser incident channel is defined as the central optical axis. The annular detection structure is symmetrically arranged around the central optical axis, and the outer diameter is set to r out , and the inner diameter is set to r in , satisfying (r out -r in ) / r in ≤α, where α is the allowable maximum radial non-uniformity factor;

[0098] ]>Set N d equidistant Raman signal collection and detection units in the annular detection structure. The distance between the center of each Raman signal collection and detection unit and the central optical axis is set as the spatial offset distance d i , where the unit is millimeter, i = 1, 2,..., N d , N d ≥3, for synchronously collecting Raman signals at different spatial offset distances;

[0099] Each Raman signal collection and detection unit is connected to a set of band-pass filter modules. The central transmission wavelength of the band-pass filter modules is set to λ R , and the spectral bandwidth of the band-pass filter modules is Δλ. Where λ R represents the central wavelength of the target Raman signal, and Δλ represents the allowable wavelength offset range. The band-pass filter modules are used to shield the excitation laser reflection interference and only transmit the wavelengths corresponding to the target Raman signals;

[0100] A microelectromechanical system mirror assembly is set at the front end of each Raman signal collection and detection unit. The microelectromechanical system mirror assembly receives the corresponding control voltage signal to achieve fine adjustment of the optical path angle and control the exit angle of the microelectromechanical system mirror assembly Form a cross focal area with the incident angle θ of the central laser incident path in to enhance the collection coupling efficiency of the Raman signal in the deep layer of the alveolar bone;

[0101] The output light beams of all Raman signal collection and detection units are introduced into the Raman signal spectrum acquisition module through an optical fiber bundle. The Raman signal spectrum acquisition module includes a grating spectrometer and a high-sensitivity CCD detector, and the CCD detector records the corresponding Raman signal intensity at multiple spatial offset distances to form an original Raman spectrum signal group.

[0102] In this embodiment, S3 includes the following steps:

[0103] S31. During the process of irradiating the oral mucosa surface of the subject with a multi-wavelength tunable excitation light beam, synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone, and set the frequency f u and ultrasonic flux I LIFU of the low-intensity focused ultrasound, where f u represents the central frequency of the focused ultrasound, and I LIFU represents the power per unit area of the low-intensity focused ultrasound;

[0104] S32. Set an acousto-optic modulator in the incident path of the multi-wavelength tunable excitation light beam to perform frequency shift and phase modulation on the excitation lights of the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2, and set the modulation frequency to f m and the modulation amplitude to φ m , where f m represents the frequency offset applied by the acousto-optic modulator, and φ m represents the phase offset angle after modulation;

[0105] S33. Make the modulated excitation light output by the acousto-optic modulator completely coincide with the low-intensity focused ultrasound irradiation area in space and time, so that the modulated excitation light acts on the tissue interface modulated by the low-intensity focused ultrasound, inducing a change in the microscopic elastic modulus between the alveolar bone and the soft tissue covering it, and forming a Raman scattering response enhancement area;

[0106] S34. Synchronously collect the Raman scattering signals under the condition of the modulated excitation light on each spatial offset distance d i channel in the Raman scattering response enhancement area, and construct a modulated Raman spectrum signal set where represents the Raman spectrum signal intensity obtained under the action of acousto-optic modulation at the spatial coordinates (x, y) under the conditions of the excitation wavelength λ k and the spatial offset distance d i , and λ k ∈λ1,λ2;

[0107] S35. Frequency-lock extraction is performed on all Raman spectral signals in the modulated Raman spectral signal set R2 to extract the coherent Raman components synchronized with the modulation frequency f m After removing the non-modulated background signals, the frequency-locked Raman components are uniformly recombined into the second set of original Raman spectral signal sets wherein represents the original data set of Raman spectral signals collected based on the acousto-optic composite modulation mechanism and not yet subjected to bone-soft tissue separation and noise reduction.

[0108] In this embodiment, S4 includes the following steps:

[0109] S41. A short-pulse broadband fluorescence excitation light source with a central excitation wavelength of 405 - 440 nm and a pulse width of τ p is used to irradiate the gingival soft tissue on the surface of the oral mucosa of the subject to be examined with a light power per unit area of I F and the time-integrated fluorescence original image F raw (x, y) is collected and recorded;

[0110] S42. Dark current subtraction and flat-field correction are performed on the collected fluorescence original image F raw (x, y) to obtain the corrected fluorescence image F cal (x, y), and the region of interest of the gingival soft tissue is selected in the corrected fluorescence image;

[0111] S43. Calculate the average fluorescence intensity within the region of interest of the gingival soft tissue and combine the average fluorescence intensity with the light power per unit area I F and the equivalent optical path d of the oral mucosa ref to estimate the absorption coefficient μ of the gingival soft tissue a,soft ;

[0112] S44. Calculate the effective attenuation coefficient of the gingival soft tissue by combining the absorption coefficient of the gingival soft tissue and the reference scattering coefficient;

[0113] S45. Construct a set of soft tissue optical parameters Θ soft , and the set of soft tissue optical parameters includes the absorption coefficient of the gingival soft tissue and the effective attenuation coefficient of the gingival soft tissue.

[0114] In this embodiment, S5 includes the following steps:

[0115] S51. Perform spatial coordinate registration on the first set of original Raman spectral signal sets and the second set of original Raman spectral signal sets to obtain the spatially registered combined data set of Raman spectral signals

[0116] S52. Combine the spatially registered Raman spectral signal combined dataset D match and the set of soft tissue optical parameters and input them into the bone-soft tissue joint inversion equation for bone-soft tissue separation calculation;

[0117] S53. Based on the solution results of the bone-soft tissue joint inversion equation, extract and reconstruct the Raman spectral components representing the characteristics of the deep alveolar bone to form a deep alveolar bone Raman characteristic spectral dataset where, represents the Raman characteristic spectral signal intensity of the alveolar bone tissue obtained at the spatial coordinates (x, y) on the oral mucosa surface under the conditions of the excitation wavelength λ k and the spatial offset distance d i after bone-soft tissue separation calculation.

[0118] In this embodiment, S52 includes the following steps:

[0119] S521. Take the spatially registered Raman spectral signal combined dataset and the set of soft tissue optical parameters as input variables and jointly introduce them into the bone-soft tissue joint inversion equation to establish the following optimization model:

[0120]

[0121] where, represents the set of Raman characteristic signals of the alveolar bone to be solved, is the loss function, which is used to measure the residual error between the combined signal and the theoretically inverted signal under the soft tissue model;

[0122] S522. The loss function contains two parts of residual terms, namely the basic Raman channel residual term and the modulated Raman channel residual term:

[0123]

[0124] where, and are the theoretical Raman signals respectively derived through the optical propagation model based on the currently estimated Raman characteristic signals of the alveolar bone and the soft tissue optical parameters, and w1 and w2 are the residual weighting coefficients of the basic channel and the modulated channel;

[0125] S523. In each iteration process, apply the set of soft tissue optical parameters Θ soft to the estimated Raman characteristic signal R of the alveolar bone bone , and calculate the simulated spectral value;

[0126] S524. Repeat the optical propagation simulation and loss function evaluation process in step S523, and continuously adjust the Raman characteristic signal of the alveolar bone through gradient descent to make the loss function Converge to the global minimum or meet the convergence threshold condition, and finally complete the bone-soft tissue separation solution process.

[0127] In this embodiment, S6 includes the following steps:

[0128] S61. Organize each alveolar bone Raman characteristic spectral signal in the deep alveolar bone Raman characteristic spectral dataset in three dimensions according to the excitation wavelength dimension, spatial offset distance dimension, and spatial coordinate dimension to construct a spectral-depth joint distribution map stack T bone (x, y, λ k , d i ), and the constructed spectral-depth joint distribution map stack is used to describe the alveolar bone Raman characteristic signal intensity distribution under two-dimensional spatial coordinate positions, different excitation wavelengths, and different spatial offset distances. The spectral-depth joint distribution map stack represents the alveolar bone Raman characteristic signal under the spatial position (x, y), excitation wavelength λ k and spatial offset distance d i conditions;

[0129] S62. Use the spectral-depth joint distribution map stack as the input and input it into the convolutional neural network model for processing. In the convolutional neural network model, perform end-to-end auto-encoding processing on the input spectral-depth joint distribution map stack through continuous convolutional calculations, activation function applications, and downsampling operations. The encoding result is a multi-scale spatial-spectral joint feature F bone ;

[0130] S63. During the feature extraction process, introduce a supervised training strategy. Evaluate the difference between the output result of the convolutional neural network model and the corresponding alveolar bone health status label by constructing an objective loss function. The objective loss function is used to guide parameter updates, and the goal is to minimize this loss value. The input of the convolutional neural network model is the spectral-depth joint distribution map stack, and the output is the prediction result. The prediction result is compared with the pre-defined set of health status labels to measure the classification error;

[0131] S64. Update the parameter set in the convolutional neural network model through the backpropagation algorithm. Recalculate the difference between the predicted output and the health status label in each iteration, and update the network parameters according to the gradient direction of the objective loss function; repeat the process of feature extraction, prediction, and parameter update until the objective loss function converges or meets the preset early stopping condition; when the objective loss function reaches the minimum or meets the training termination condition, obtain the trained alveolar bone tissue health recognition model

[0132] In this embodiment, S7 includes the following steps:

[0133] S71. Input the deep alveolar bone Raman characteristic spectrum dataset into the alveolar bone tissue health recognition model for classification and inference to obtain the health discrimination label at the corresponding coordinate region (x, y). Form a set of alveolar bone health discrimination labels including all detection points.

[0134] S72. Match and analyze the set of alveolar bone health discrimination labels with the pre-set alveolar bone health status evaluation rules, and output the alveolar bone health assessment result. The alveolar bone health status evaluation rules are classified and judged according to the average Raman characteristic intensity, Raman peak structure integrity, and offset channel consistency index:

[0135] The alveolar bone health status is Grade I (normal): If the average intensity of the alveolar bone Raman signal and the signal-to-noise ratio SNR of the main peak peak ≥S1, and at the same time the standard deviation σ between the spatial offset channels d ≤δ1;

[0136] The alveolar bone health status is Grade II (mild absorption): If the average intensity of the alveolar bone Raman signal or SNR peak <S1, but still satisfies σ d ≤δ2;

[0137] The alveolar bone health status is Grade III (moderate to severe absorption): If the average intensity of the alveolar bone Raman signal or σ d >δ2;

[0138] Among them, is the average intensity of the alveolar bone Raman signal, SNR peak is the signal-to-noise ratio of the main Raman peak, σ d is the signal standard deviation between the spatial offset channels, and T1, T2, S1, δ1, δ2 are the evaluation thresholds built into the model;

[0139] S74. Visualize and output the alveolar bone health assessment result determined according to the evaluation rules in real time at the corresponding spatial coordinates (x, y) to the display terminal, construct a two-dimensional visualization map, and the output content includes the health status classification label, comprehensive risk rating identifier, and key feature parameter value range corresponding to each detection point.

[0140] Example 1:

[0141] In a certain alveolar bone health screening activity, a patient with a history of chronic periodontitis received non-invasive detection of alveolar bone using spatial offset Raman spectroscopy based on the present invention. The implementer first wore a standard oral Raman probe for the patient. The offset distances of the 6 offset channels in the probe were set to 2 mm, 3 mm, 4 mm, 5 mm, 6 mm, and 7 mm respectively, and the dual-laser mode was adopted. The excitation wavelengths were set to 785 nanometers and 840 nanometers respectively, and the laser incident power was automatically adjusted to 4.5 milliwatts. The simultaneously started fluorescence imaging module recorded the original image F raw (x,y). After flat-field correction and region selection, the average corrected fluorescence intensity of the patient's gingival soft tissue was obtained Combined with the optical power per unit area I F =7.6(mW / cm 2 ) and the equivalent optical path d of the oral mucosa ref =0.14 cm, the absorption coefficient μ of the gingival soft tissue was deduced a,soft =4.17(cm -1 ).

[0142] During the detection process, the original Raman signal collected in real time At the position (x0,y0), the main peak intensity under the excitation wavelength of 785 nanometers and the offset 4 mm channel was 131.5 (a.u.), the average value between the spatial offset channels was 128.2 (a.u.), and the standard deviation was 3.9 (a.u.). Through acousto-optic modulation synchronous excitation, the modulated Raman signal was obtained The main peak was enhanced to 149.7 (a.u.), and the signal-to-noise ratio was increased to 23.1 dB. The set of soft tissue optical parameters Θ soft Combined into the bone-soft tissue joint inversion equation, and inversely calculated together with the original and modulated Raman signals to obtain the Raman characteristic signal of the deep alveolar bone The average value at all detection points was 138.8 (a.u.), and the standard deviation of the spatial offset channel decreased to 3.2 (a.u.).

[0143] After three-dimensional recombination of all detection data, a characteristic map stack T bone (x,y,λ k ,d i ) was constructed and input into the convolutional neural network model For health status recognition. The network output the label of this detection area as "Grade Ⅰ (normal)", and the confidence level was 0.97. According to the evaluation rules of the present invention, the average Raman signal of the patient Was higher than the health threshold T1 = 120 (a.u.), the main peak signal-to-noise ratio SNR peak =21.5 (dB) was higher than S1 = 16 (dB), and the standard deviation σ of the channel consistency d=3.2 (a.u.) is lower than δ1 = 6 (a.u.), and it is determined as Grade I of the alveolar bone health status.

[0144] To compare with the traditional CBCT method, the system synchronously retrieves CBCT image data. The bone density value of the corresponding detection area is 986 (HU), and the doctor's manual evaluation is "no obvious bone resorption". However, CBCT cannot provide information on the bone metabolic state at the molecular level, and this patient has abnormal liver function, so CBCT cannot perform continuous dynamic tracking. Two weeks later, the present invention was applied for detection again. All indicators were basically stable, without abnormal fluctuations, and the patient was not affected by any radiation during the entire follow-up period.

[0145] There is another detected patient who was detected due to early alveolar bone resorption manifestations. During multi-wavelength Raman acquisition, the average intensity of the main peak was only 95.7 (a.u.), the signal-to-noise ratio was 14.2 (dB), and the standard deviation of the spatial offset channel was 8.8 (a.u.). After joint inversion, the health assessment was Grade II (mild absorption), and the system automatically output a heat map showing that the absorption was concentrated in the right mandibular area. The confidence level of the output label of the convolutional neural network model was 0.91. After three months of follow-up, the detection was performed again. The main peak intensity dropped to 88.2 (a.u.), and the standard deviation of the spatial offset increased to 11.5 (a.u.), and the system automatically prompted the risk progression. In contrast, the bone density value in the CBCT image decreased from 835 (HU) to 810 (HU) within three months, and it was difficult to quantitatively grade the difference between the two manual interpretations by doctors.

[0146] For the statistical analysis of 50 detected samples, the sensitivity of the method of the present invention reached 92%, the specificity was 96%, and the average time for automatic interpretation and output of each detection area was 46 seconds, and doctors did not need to perform subjective interpretation. The interpretation sensitivity of traditional CBCT was 75%, the specificity was 92%, the average time for single-area detection was 17 minutes, and two doctors were required to cooperate.

[0147] In the model training samples, the average intensity of the main peak in the healthy group was 141.1 (a.u.), in the mild absorption group was 105.9 (a.u.), and in the severe absorption group was 68.3 (a.u.). The characteristic tensors showed a hierarchical distribution law in the dimensions of multi-wavelength, depth, and multi-spatial coordinates. The accuracy rate of the test set after network training was 94.3%, and the accuracy rate of the validation set was 92.8%, both of which were better than 84.6% of the traditional statistical analysis method.

[0148] Among all high-risk patients, based on the continuous dynamic non-invasive tracking of the present invention, 8 cases of suspected absorption progression were first discovered, and all were confirmed in the subsequent surgical pathology. Compared with the traditional CBCT control group, the present invention can reflect the changes in bone health earlier, non-destructively, dynamically, and structurally, greatly improving the ability of early screening and dynamic monitoring of alveolar bone diseases, and significantly optimizing the clinical work efficiency of doctors and the medical experience of patients.

[0149] The present invention adopts a multi-wavelength spatial offset Raman spectroscopy detection device and an acousto-optic modulation synchronous enhancement mechanism, and uses a MEMS controllable offset probe array to synchronously collect Raman signals at different depths. By constructing a joint inversion equation for bone and soft tissue, the multi-channel Raman signals and the soft tissue optical parameter set are coupled and inverted. A joint optimization loss function is introduced in the signal separation link, which significantly improves the separation accuracy of the deep alveolar bone Raman characteristic signal and effectively suppresses the interference of soft tissue and background noise.

[0150] The present invention reorganizes the deep alveolar bone Raman characteristic spectrum dataset into a high-order tensor input convolutional neural network model according to multiple wavelengths and multiple offset depth dimensions, and realizes end-to-end automatic extraction of spatial-spectral joint features and health status identification. Compared with the existing traditional methods based on simple statistical features or one-dimensional spectral analysis, the proposed convolutional neural network can explore the distribution patterns of alveolar bone molecular structure at multiple scales and multiple spatial levels, and realize automatic discrimination of multiple categories such as healthy, mild absorption, moderate and severe absorption.

[0151] The hierarchical output mechanism proposed in the present invention, based on a set of discriminant labels and health status evaluation rules, can output the health status results of each detection point to the terminal in real-time visualization in the form of a two-dimensional distribution heat map. The health assessment process is combined with multi-parameter threshold adaptive rules to achieve intelligent fusion judgment of key indicators such as the average intensity of the alveolar bone Raman signal, the main peak signal-to-noise ratio, and the consistency between channels, significantly improving the accuracy and transparency of risk grading.

[0152] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy, characterized in that, It includes the following steps: S1. Select a first near-infrared excitation wavelength and a second near-infrared excitation wavelength to form a multi-wavelength tunable excitation beam, and couple it into an improved spatially offset Raman detection device; S2. Set an annular lateral collection area in the improved spatially offset Raman detection device, perform partition control on the annular lateral collection area, and synchronously collect a first set of original Raman spectral signals at multiple spatial offset distances; S3. Synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone, and perform frequency shift and phase modulation on the multi-wavelength tunable excitation beam through an acousto-optic modulator, and collect a second set of original Raman spectral signals by using the acousto-optic composite modulation Raman enhancement mechanism; S4. Obtain the fluorescence image of the gingival soft tissue, estimate the set of soft tissue optical parameters based on the fluorescence image of the gingival soft tissue, and input the set of soft tissue optical parameters into the joint inversion equation; S5. Input the first set of original Raman spectral signals and the second set of original Raman spectral signals and the set of soft tissue optical parameters into the joint inversion equation together, perform bone-soft separation calculation, and obtain the Raman characteristic spectral data set of the deep alveolar bone; S6. Reorganize the Raman characteristic spectral data set of the deep alveolar bone into a spectral depth map stack according to the multi-wavelength dimension and the multi-depth dimension, and train the alveolar bone tissue health recognition model; S7. Based on the alveolar bone tissue health recognition model, perform health status discrimination on the real-time collected Raman characteristic spectral data set of the deep alveolar bone, and generate the alveolar bone health assessment result.

2. The non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy according to claim 1, characterized in that The S1 includes the following steps: S11. Select the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2 in the excitation light source; S12. Through the electronically controlled wavelength switching driver, respectively set the excitation time window of the first near-infrared excitation wavelength and the excitation time window of the second near-infrared excitation wavelength, so that the first near-infrared excitation wavelength and the second near-infrared excitation wavelength are output in sequence within a single detection cycle; S13. Couple the multi-wavelength tunable excitation beam into the coaxial-ring dual-channel composite optical path structure in sequence. The coaxial-ring dual-channel composite optical path structure includes a central optical axis incident path and an annular concentric collection channel. The central optical axis incident path is used for the primary incidence of the multi-wavelength tunable excitation beam; S14. Vertically irradiate the multi-wavelength tunable excitation beam modulated by the coaxial-ring dual-channel compound optical path onto the oral mucosa surface of the subject to be examined, and set the incident angle θ in = 90°, to form a directional light field irradiation area. The size of the directional light field irradiation area is (L x , L y ), where L x = 5 - 10 mm, L y = 5 - 10 mm. Align the center of the irradiation area with the center coordinates (x0, y0) of the alveolar bone target area, so as to satisfy the following spatial matching constraint: |(x c - x0, y c - y0)| ≤ ∈, where L x represents the width of the laser irradiation area in the horizontal direction, L y represents the length of the laser irradiation area in the vertical direction, (x c , y c ) is the center coordinate of the laser irradiation area, and ∈ is the positioning error threshold, with ∈ ≤ 0.5 mm.

3. The non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy according to claim 2, wherein, The S2 includes the following steps: S21. An annular lateral collection area is arranged around the periphery of the improved spatially offset Raman detection device. The annular lateral collection area is arranged symmetrically around the central optical axis and is composed of a plurality of Raman signal collection and detection units. The number of Raman signal collection and detection units is set to N d , and the spatial offset distance of each Raman signal collection and detection unit relative to the central excitation optical axis is set to where d i represents the spatial offset distance of the i-th Raman signal collection and detection unit, and i = 1, 2,..., N d ; S22. Embedding a microelectromechanical system mirror assembly between the Raman signal collection and detection units in the annular lateral collection area, and adjusting the working state of the mirror assembly through a control voltage signal for adjustment; S23. Synchronously regulate the control voltage signal with the excitation period of the multi-wavelength tunable excitation beam for exciting the laser, and perform laser irradiation at each spatial offset distance d i and collect the corresponding Raman scattering signals to construct a group of original Raman spectral signals where represents the Raman spectral signal intensity obtained at the spatial coordinates (x, y) on the oral mucosa surface under the conditions of excitation wavelength λ k and spatial offset distance d i and λ k ∈ λ1, λ2; S24. Perform sampling uniformity verification on each Raman spectrum signal in the original Raman spectrum signal group R1 ; S25. Reorganize all valid Raman spectral signals that have passed the sampling uniformity verification according to the spatial offset distance d i into the first set of original Raman spectral signal sets 4. The non-invasive alveolar bone detection method based on spatially offset Raman spectroscopy according to claim 3, wherein, The improved spatially offset Raman detection device includes the following structure: Set an annular detection structure as the main collection area of Raman signals. A central laser incident channel is set in the center of the annular detection structure, and the optical axis of the central laser incident channel is defined as the central optical axis; Set N d equidistant Raman signal collection and detection units inside the annular detection structure, and the distance between the center of each Raman signal collection and detection unit and the central optical axis is set as the spatial offset distance d i ; Each Raman signal collection and detection unit is connected to a group of band-pass filter modules, and the central transmission wavelength of the band-pass filter module is set to λ R , and the spectral bandwidth of the band-pass filter module is Δλ; A microelectromechanical system mirror assembly is arranged at the front end of each Raman signal collection and detection unit to control the exit angle of the microelectromechanical system mirror assembly to form an intersection focal area with the incident angle θ of the central laser incident path in therebetween; The output light beams of all Raman signal collection and detection units are imported into the Raman signal spectrum acquisition module through an optical fiber bundle. The Raman signal spectrum acquisition module includes a grating spectrometer and a highly sensitive CCD detector. The CCD detector records the corresponding Raman signal intensities at multiple spatial offset distances respectively. A set of original Raman spectrum signals is formed.

5. The non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy according to claim 3, wherein The S3 includes the following steps: S31. During the process of irradiating the oral mucosa surface of the subject with the multi-wavelength tunable excitation beam, synchronously apply low-intensity focused ultrasound to the corresponding area of the alveolar bone; S32. Set an acousto-optic modulator in the incident path of the multi-wavelength tunable excitation beam, and perform frequency shift and phase modulation on the excitation lights of the first near-infrared excitation wavelength λ1 and the second near-infrared excitation wavelength λ2; S33. Make the modulated excitation light output by the acousto-optic modulator completely coincide with the low-intensity focused ultrasound irradiation area in space and time, induce microscopic elastic modulus changes between the alveolar bone and the soft tissue covering it, and form a Raman scattering response enhancement area; S34. Each spatial offset distance d within the Raman scattering response enhancement region i Synchronously collect Raman scattering signals under the condition of modulated excitation light on the channel, and construct a set of modulated Raman spectral signals Wherein represents the Raman spectral signal intensity obtained under the acousto-optic modulation at the spatial coordinates (x, y) under the conditions of the excitation wavelength λ k and the spatial offset distance d i λ ∈ λ1, λ2 k ∈ λ1, λ2; S35. Frequency-lock extraction is performed on all Raman spectral signals in the modulated Raman spectral signal set R2 to extract coherent Raman components synchronized with the modulation frequency f m After removing the non-modulated background signals, the frequency-locked Raman components are uniformly recombined into a second set of original Raman spectral signal sets 6. The non-invasive alveolar bone detection method based on spatially offset Raman spectroscopy according to claim 5, characterized in that, The S4 includes the following steps: S41. Irradiate the gingival soft tissue on the surface of the oral mucosa of the subject to be examined with a short-pulse broadband fluorescence excitation light source, and collect and record the time-integrated fluorescence original image F raw (x, y); S42. Perform dark current subtraction and flat field correction on the acquired raw fluorescence image F raw (x, y) to obtain the corrected fluorescence image F cal (x, y), and select the region of interest of gingival soft tissue in the corrected fluorescence image; S43. Calculate the average fluorescence intensity within the region of interest of the gingival soft tissue And combine the average fluorescence intensity With the optical power per unit area I F And the equivalent optical path d of the oral mucosa ref To estimate the absorption coefficient μ of the gingival soft tissue a,soft ; S44. Combine the absorption coefficient of the gingival soft tissue and the reference scattering coefficient to calculate the effective attenuation coefficient of the gingival soft tissue; S45. Construct a set Θ of soft tissue optical parameters soft , where the set of soft tissue optical parameters includes the absorption coefficient of gingival soft tissue and the effective attenuation coefficient of gingival soft tissue.

7. The non-invasive alveolar bone detection method based on spatially offset Raman spectroscopy according to claim 6, characterized in that, The said S5 includes the following steps: S51. Perform spatial coordinate registration on the first set of original Raman spectrum signal sets and the second set of original Raman spectrum signal sets to obtain a combined data set of Raman spectrum signals after spatial registration S52. Combine the spatially registered Raman spectral signal combined dataset D match and the set of soft tissue optical parameters into the bone-soft tissue joint inversion equation for bone-soft separation calculation; Based on the solution results of the bone-soft tissue combined inversion equation, extract and reconstruct the Raman spectral components representing the characteristics of the deep alveolar bone to form a Raman characteristic spectral dataset of the deep alveolar bone. Among them, represents the Raman characteristic spectral signal intensity of the alveolar bone tissue obtained at the spatial coordinates (x, y) on the oral mucosa surface under the conditions of the excitation wavelength λ k and the spatial offset distance d i .

8. The non-invasive alveolar bone detection method based on spatially offset Raman spectroscopy according to claim 7, characterized in that, The said S52 includes the following steps: S521. Jointly introduce the spatially registered combined Raman spectroscopy signal dataset and the soft tissue optical parameter set as input variables into the bone and soft tissue joint inversion equation to establish an optimization model: Among them, represents the set of Raman characteristic signals of the alveolar bone to be solved, ) is the loss function, Θ soft is the set of soft tissue optical parameters; S522. Loss function It includes two parts of residual terms, namely the basic Raman channel residual term and the modulated Raman channel residual term: Among them, and are respectively the theoretical Raman signals derived based on the currently estimated Raman characteristic signals of the alveolar bone and the optical parameters of soft tissues through an optical propagation model, and w1 and w2 are the residual weighting coefficients of the basic channel and the modulation channel; S523. During each iteration, for the set of soft tissue optical parameters Θ soft act on the estimated Raman characteristic signal R of the alveolar bone bone , and calculate the simulated spectral value; Repeat the optical propagation simulation and loss function evaluation process in step S523, and continuously adjust the alveolar bone Raman characteristic signal through gradient descent to make the loss function converge to the global minimum or meet the convergence threshold condition, and finally complete the bone-soft tissue separation solution process.

9. A non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy according to claim 8, characterized in that, The said S6 includes the following steps: S61. Organize each alveolar bone Raman characteristic spectral signal in the deep alveolar bone Raman characteristic spectral dataset in three dimensions according to the excitation wavelength dimension, spatial offset distance dimension, and spatial coordinate dimension to construct a spectral-depth joint distribution map stack T bone (x,y,λ k ,d i ); S62. Taking the spectral-depth joint distribution map stack as input, it is input into a convolutional neural network model for processing. In the convolutional neural network model, the input spectral-depth joint distribution map stack is subjected to end-to-end automatic encoding processing through continuous convolutional calculations, activation function applications, and downsampling operations, and the encoding result is the multi-scale spatial-spectral joint feature F bone ; S63. During the feature extraction process, a supervised training strategy is introduced to train the convolutional neural network model, and a trained alveolar bone tissue health recognition model is obtained.

10. A non-invasive detection method for alveolar bone based on spatially offset Raman spectroscopy according to claim 9, characterized in that, The said S7 includes the following steps: S71. Input the Raman characteristic spectral dataset of the deep alveolar bone into the alveolar bone tissue health recognition model for classification and inference to obtain the health discrimination label at the corresponding coordinate region (x, y). Form a set of alveolar bone health discrimination labels that includes all the detection points. S72. Match the alveolar bone health discrimination label set with the pre-set alveolar bone health status evaluation rules for matching analysis, and output the alveolar bone health assessment results. The alveolar bone health status evaluation rules are classified and judged according to the following indicators: The alveolar bone health status is Grade I (normal): If the average intensity of the Raman signal of the alveolar bone and the signal-to-noise ratio SNR of the main peak peak ≥ S1, and at the same time the standard deviation σ between the spatially offset channels d ≤ δ1; The alveolar bone health status is grade II (mild resorption): If the average intensity of the Raman signal of the alveolar bone or SNR peak < S1, but still satisfies σ d ≤δ2; The alveolar bone health status is Grade III (moderate to severe resorption): If the average intensity of the Raman signal of the alveolar bone or σ d > δ2; Among them, is the average intensity of the Raman signal of the alveolar bone, and SNR peak is the signal-to-noise ratio of the main Raman peak, and σ d is the standard deviation of the signals between the spatial offset channels. T1, T2, S1, δ1, and δ2 are the evaluation thresholds built into the model; S74. Real-time visually output the alveolar bone health assessment result determined according to the evaluation rule at the corresponding spatial coordinates (x, y) to a display terminal to construct a two-dimensional visualization map.

Images