Peri-implantitis non-invasive detection method based on spatial offset Raman spectrum technology

Abstract

The invention discloses a peri-implantitis non-invasive detection method based on a spatial offset Raman spectrum technology, which comprises the following steps: S1, constructing adjustable spatial offset Raman detection equipment to obtain a calibration parameter set; s2, selecting an offset window with an optimal signal-to-noise ratio as a target acquisition window, and outputting optimized spatial offset Raman spectrum data; s3, performing wavelet multi-scale decomposition on the optimized spatial offset Raman spectrum data by adopting a preset mother wavelet to obtain a wavelet multi-scale coefficient set; s4, suppressing a background noise coefficient by adopting a self-adaptive threshold set, and reconstructing to obtain reconstructed de-noised spatial offset Raman spectrum data; s5, generating a three-dimensional space offset wavelet spectrum tensor, performing concentrated enhancement, and outputting enhanced space offset Raman spectrum data; and S6, obtaining a periimplantitis health state classification result and an inflammation index value. According to the invention, the detection sensitivity and accuracy of early weak signals of dental periimplantitis are obviously improved.

Description

Technical Field

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

[0002] With the advancement of digital medicine and non-invasive diagnostics, early and accurate detection of peri-implantitis has become an important research area in clinical dentistry. Peri-implantitis is a chronic disease characterized by bone matrix resorption and inflammatory responses, and its early diagnosis is crucial for preventing implant failure. However, currently used clinical detection techniques, including X-ray imaging, probe depth measurement, and ultrasound testing, all have significant shortcomings in practical application.

[0003] First, X-ray imaging can only observe significant changes in the implant boundary when bone absorption reaches more than 2 mm, and it is impossible to sensitively capture early subtle bone loss. In addition, X-rays are ionizing radiation, which carries the risk of repeated detection and is not suitable for high-frequency follow-up. Secondly, the method of measuring the depth of the periodontal pocket with a probe is somewhat invasive. During the operation, the data may fluctuate due to differences in the doctor's technique, and it is difficult to obtain objective and continuous quantitative diagnostic information. For early lesions covered by some soft tissues, traditional detection methods are even more difficult to accurately reflect the true pathological state. Although existing ultrasound imaging methods are non-invasive to a certain extent, they lack specificity and resolution for changes in weak molecular characteristics at the implant-bone interface or deep soft tissue, making it difficult to meet the needs of accurate early identification of peri-implantitis.

[0004] Therefore, there is an urgent need for a new detection method that can adapt to the complex tissue structure of dental implants, improve the ability to identify deep signals, and have automated, non-invasive, and rapid quantitative analysis capabilities to solve the above technical bottlenecks. Summary of the Invention

[0005] One purpose of the present invention is to propose a non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy technology. The present invention significantly improves the detection sensitivity and accuracy of early weak signals of dental peri-implantitis.

[0006] According to an embodiment of the present invention, a non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy technology includes the following steps:

[0007] S1. Construct an adjustable spatial offset Raman detection device and align it with the mucosal surface surrounding the dental implant. Perform laser power calibration and initial offset zero calibration to obtain a calibration parameter set.

[0008] S2. Initiate dynamic spatial offset scanning control based on the calibration parameters, monitor the Raman signal-to-noise ratio changes in real time, record the original spatial offset Raman spectral data corresponding to each offset distance, select the offset window with the optimal signal-to-noise ratio as the target acquisition window, and output the optimized spatial offset Raman spectral data;

[0009] S3. Based on the optimized spatial offset Raman spectral data, the spatial offset - wavelet decomposition linkage mapping relationship is called, and the preset mother wavelet is used to perform wavelet multiscale decomposition on the optimized spatial offset Raman spectral data to obtain a set of wavelet multiscale coefficients;

[0010] S4. Calculate the coefficient variance and information entropy of the wavelet multi-scale coefficient set at each scale, and generate an adaptive threshold set, use the adaptive threshold set to suppress the background noise coefficient and reconstruct the reconstructed denoised spatially offset Raman spectral data;

[0011] S5. Stacking the reconstructed denoised spatially offset Raman spectral data and the corresponding offset distance and wavelet scale in a three-dimensional order of offset-wavenumber-scale to generate a three-dimensional spatially offset wavelet spectral tensor and performing centralized enhancement, and outputting the enhanced spatially offset Raman spectral data;

[0012] S6. Extract biochemical Raman eigenvectors from the enhanced spatially offset Raman spectroscopy data and input the biochemical Raman eigenvectors into a pre-trained machine learning model to obtain the health status classification results and inflammation index values ​​of dental peri-implantitis.

[0013] Optionally, the S1 includes the following steps:

[0014] S11. Construct an adjustable spatially offset Raman detection device with continuous spatial offset capability. The adjustable spatially offset Raman detection device includes a central excitation fiber, an annular collection fiber array, and a laser power calibration module. The central excitation fiber is used to output an excitation laser with a wavelength of 785-800 nanometers. The annular collection fiber array is evenly distributed around the excitation fiber and fixed to a piezoelectric micromotion platform. By controlling the piezoelectric micromotion platform, the annular collection fiber array is spatially offset relative to the excitation beam in the direction of the axial outer edge of the dental implant, obtaining an offset distance Δx.

[0015] S12. Install the adjustable spatial offset Raman detection device on the positioning bracket and align it with the mucosal surface around the dental implant, so that the laser emission axis of the central excitation fiber is perpendicular to the implant area. The vertical alignment distance between the adjustable spatial offset Raman detection device and the mucosal surface is d align , vertical alignment distance d align The average thickness parameter h of individual gingival tissue gum Added with the spot focus offset correction coefficient ε;

[0016] S13. Start the laser power calibration module to calibrate the laser intensity before the excitation light is emitted so that the excitation light power parameter P0 is controlled within the safe range of non-invasive detection. The laser power parameter P0 represents the laser power at the output end of the central excitation fiber;

[0017] S14. Perform zero point calibration on the initial offset position of the piezoelectric micro-motion platform, record the current offset position as the offset starting position coordinate x0, and set the offset starting position coordinate x0 as the reference point for spatial offset control. Calculate the current offset position coordinate x0 based on the offset starting position coordinate x0 and the offset distance Δx. actual , output calibration parameter set C cal ={P0,x actual ,d align}.

[0018] Optionally, S2 includes the following steps:

[0019] S21. By sequentially adjusting the spatial offset increment parameter so that each spatial offset distance is within the range of 0 to 4 mm, the spatial offset increment parameter is the difference between each two adjacent spatial offset distances;

[0020] S22. At each spatial offset, based on the offset starting position coordinates and the current spatial offset distance, the current spatial offset distance is added to the offset starting position coordinates to obtain the actual spatial position coordinates;

[0021] S23. Collecting raw spatially offset Raman spectral data at each actual spatial position coordinate;

[0022] S24. Calculate the peak intensity of the target Raman peak and the root mean square value of the corresponding background noise for each acquisition of the raw spatially offset Raman spectrum data, and calculate the signal-to-noise ratio parameter of the spatial position by dividing the peak intensity of the target Raman peak by the root mean square value of the corresponding background noise;

[0023] S25. Based on all the SNR parameters in the spatial offset sequence, use the central difference method to calculate the SNR gradient parameter by subtracting the previous SNR parameter from the next adjacent SNR parameter at the current spatial position, and then dividing the result by twice the spatial offset increment parameter. When the absolute value of the SNR gradient parameter is less than a threshold, the spatial offset increment parameter is amplified to accelerate the scan. When the absolute value of the SNR gradient parameter is greater than the threshold, the spatial offset increment parameter is reduced to refine the search.

[0024] S26. Determine the platform position where the signal-to-noise ratio parameter reaches its maximum value and the signal-to-noise ratio gradient parameter turns from positive to negative or approaches zero based on all signal-to-noise ratio parameters and their corresponding signal-to-noise ratio gradient parameters. Record the index corresponding to the platform position as the optimal signal-to-noise ratio index. Centered on this spatial offset distance, extend the target acquisition window W by one spatial offset increment parameter each forward and backward. opt ;

[0025] S27. In the target acquisition window W opt The original spatially offset Raman spectral data are repeatedly collected several times, and all spectral intensity values ​​at the same wave number point are averaged to obtain the optimized spatially offset Raman spectral data O(ν j ).

[0026] Optionally, S3 includes the following steps:

[0027] S31. Spatial offset distance Δx based on target acquisition window k The spatial offset-wavelet decomposition linkage mapping relationship between the number of wavelet decomposition layers is used to calculate the optimal number of wavelet decomposition layers L. opt = The optimal number of wavelet decomposition layers L opt represents the optimal decomposition level number of discrete wavelet decomposition for optimized spatially offset Raman spectroscopy data, γ is the mapping scale coefficient, and δ is the mapping bias coefficient;

[0028] S32. In the preset mother wavelet library For each mother wavelet ψ r (t) Perform initial decomposition to optimize the spatially offset Raman spectral data O(ν j ) is input, and the detail coefficient set W of each mother wavelet in the first decomposition layer is obtained. 1,r (ν j ), using the detail coefficient set W of each mother wavelet 1,r (ν j ) and optimized spatial offset Raman spectroscopy data O(ν j ) Calculate the energy compression rate parameter η r , in all energy compressibility parameters η r In the example, the mother wavelet with the largest value is selected as the optimal mother wavelet

[0029] S33. Using the optimal mother wavelet Optimize the spatial offset Raman spectroscopy data O(ν j ) Perform the optimal wavelet decomposition layer L opt Discrete wavelet decomposition of the layer to obtain the approximate coefficient vector corresponding to the optimal decomposition layer number The set of detail coefficient vectors corresponding to each decomposition layer Form a wavelet multi-scale coefficient set W.

[0030] Optionally, the S4 includes the following steps:

[0031] S41. For the wavelet multi-scale coefficient set W, for each decomposition scale l detail coefficient vector D l (ν j ) calculate the coefficient variance parameters respectively

[0032] S42. Detail coefficient vector D for each decomposition scale l l (ν j ) Calculate the information entropy parameter H l The information entropy parameter is obtained by multiplying the probability of all detail coefficient amplitudes appearing in different amplitude intervals under the statistical decomposition scale by their logarithms with the base 2 and taking the negative value;

[0033] S43. Binding coefficient variance parameter The arithmetic square root of and information entropy parameter H l Generate the information entropy-variance joint threshold parameter T through weighted summation l ;

[0034] S44. For each scale l, the detail coefficient vector D l (ν j ) performs coefficient threshold processing, and retains the original value when the absolute value of the detail coefficient vector is greater than or equal to the information entropy-variance joint threshold parameter, otherwise it is set to zero to obtain the detail coefficient vector after threshold processing

[0035] S45. Using the optimal mother wavelet Based on the detail coefficient vector after threshold processing Approximate coefficient vector corresponding to the optimal number of decomposition levels Perform inverse discrete wavelet transform to obtain reconstructed denoised spatially offset Raman spectral data O denoise (ν j ).

[0036] Optionally, the S5 includes the following steps:

[0037] S51. The reconstructed and denoised spatially offset Raman spectral data at each spatial offset distance are stacked in the order of spatial offset-wavenumber-wavelet scale to form a three-dimensional spatially offset wavelet spectral tensor T(i, j, l), which represents the distribution characteristics of Raman spectral intensity at different spatial offset distances and wavelet scales;

[0038] S52. Define the characteristic peak wave number ν in the detection of dental peri-implantitis bio The characteristic window wave number set ν centered j∈[ν bio -Δν,ν bio +Δν], calculate the characteristic response tensor F(i,l);

[0039] S53. The spatial offset distance Δx relative to the characteristic response tensor F(i,l) i The weighted variance calculation of the spatial offset scale adaptive focusing factor Ω is obtained. i,l :

[0040]

[0041] Among them, Ω i,l is the spatial offset scale adaptive focusing factor, which represents the spatial offset distance Δx i , characteristic peak response and spatial offset distance center at scale l the relative contribution of the degree of deviation; Represents the average distance of spatial offset after weighting of feature response tensor F(i,l), N x Indicates the total number of spatial offset positions involved in the analysis, L opt represents the optimal number of wavelet decomposition layers;

[0042] S54. Adaptive focusing factor Ω using spatial offset scale i,l The three-dimensional spatial offset wavelet spectral tensor T(i,j,l) is subjected to adaptive nonlinear focusing enhancement of the spatial offset dimension and the wavelet scale dimension to obtain enhanced spatial offset Raman spectral data O enh (ν j ):

[0043]

[0044] The nonlinear amplification parameter λ of the exponential function is used to enhance the contrast of the deep weak biological Raman signal relative to the surface interference background, and the value of λ is a positive real number.

[0045] Optionally, the S6 includes the following steps:

[0046] S61. Based on enhanced spatially offset Raman spectroscopy data O enh (ν j ) Determine the central wave number ν of the bone matrix phosphate peak P , titanium oxide layer peak center wave number ν Ti 、Inflammatory mediator peak center wave number ν I and the bacterial metabolite peak center wave number ν B , set a characteristic window with a half-width of Δν around each central wavenumber and calculate the peak area:

[0047]

[0048] Among them, Ak represents the integrated area of ​​the kth pathological characteristic peak, ν k is the corresponding central wave number;

[0049] S62. Select internal reference peak wave number ν ref Calculate the reference peak area A ref , and construct the normalized biochemical Raman feature vector:

[0050]

[0051] Wherein, v is a four-dimensional dimensionless eigenvector, and each component represents the area ratio of the corresponding pathological peak to the reference peak;

[0052] S63. Perform a normalization transformation operation on the normalized biochemical Raman eigenvector v to obtain a normalized eigenvector

[0053] S64. Normalize the feature vector Input the pre-trained machine learning model to obtain the classification result C of the health status of dental peri-implantitis state and inflammation index value I inflam .

[0054] Optionally, the dental implant peri-inflammation health status classification result C state It is a multi-category label used to indicate one of the four states of the sample: healthy, mucositis, early periarthritis or late periarthritis. The inflammation index value I inflam It is a continuous indicator with a value range of 0 to 1, and is used to quantitatively reflect the degree of inflammation in the current test sample.

[0055] The beneficial effects of the present invention are:

[0056] (1) The present invention proposes a spatial offset-wavelet decomposition linkage mapping algorithm, which adaptively determines the number of wavelet decomposition layers based on the spatial offset distance, and automatically selects the optimal mother wavelet through the energy compression rate to achieve multi-scale structured modeling of Raman signals at different detection depths. It can dynamically adjust the decomposition parameters according to the depth of the target area, effectively separate soft tissue scattering and saliva fluorescence broadband noise, enhance the detectability of deep Raman characteristic peaks of bone matrix and inflammatory mediators, and significantly improve the detection sensitivity and accuracy of weak signals in the early stage of dental peri-implantitis.

[0057] (2) The present invention constructs a characteristic response tensor based on the typical pathological peaks of peri-implantitis, performs nonlinear weighting on the three-dimensional wavelet spectrum tensor through a spatial offset scale adaptive focusing factor, adopts an exponential amplification strategy to preferentially enhance deep and weak lesion signals, and automatically suppresses surface interference. It can better adapt to different tissue depths, tissue structures and inflammatory heterogeneity, achieve focused enhancement and high-confidence output of the molecular characteristics of the lesions, and improve the molecular-level quantitative capability of non-invasive detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0059] Figure 1 This is a flow chart of a non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy technology proposed by the present invention. DETAILED DESCRIPTION

[0060] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0061] refer to Figure 1 A non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy technology comprises the following steps:

[0062] S1. Construct an adjustable spatial offset Raman detection device. Fix the central excitation fiber and an annular collection fiber array on a piezoelectric micromotion platform. This allows the annular collection fiber array to achieve a continuous spatial offset of 0–4 mm relative to the excitation beam at the axial outer edge of the dental implant. Align the adjustable spatial offset Raman detection device with the mucosal surface surrounding the dental implant via a positioning bracket. Perform laser power calibration and initial offset zero calibration to obtain a calibration parameter set.

[0063] S2. Initiate dynamic spatial offset scanning control based on calibration parameters, monitor Raman signal-to-noise ratio changes in real time, and drive the piezoelectric micro-motion stage to iteratively adjust within a preset offset range. Record the raw spatial offset Raman spectral data corresponding to each offset distance, calculate the signal-to-noise ratio gradient based on the raw spatial offset Raman spectral data, select the offset window with the optimal signal-to-noise ratio as the target acquisition window, and output the optimized spatial offset Raman spectral data.

[0064] S3. Based on the optimized spatial offset Raman spectral data, call the spatial offset - wavelet decomposition linkage mapping relationship to determine the corresponding wavelet decomposition layer number, and use the preset mother wavelet to perform wavelet multiscale decomposition on the optimized spatial offset Raman spectral data to obtain a wavelet multiscale coefficient set;

[0065] S4. Calculate the coefficient variance and information entropy of the wavelet multi-scale coefficient set at each scale, generate an adaptive threshold set based on the information entropy-variance joint threshold strategy, use the adaptive threshold set to suppress the background noise coefficient and reconstruct the reconstructed denoised spatially offset Raman spectral data;

[0066] S5. Stacking the reconstructed denoised spatially offset Raman spectral data and the corresponding offset distance and wavelet scale in a three-dimensional order of offset-wavenumber-scale to generate a three-dimensional spatially offset wavelet spectral tensor, performing a tensor projection focusing algorithm on the three-dimensional spatially offset wavelet spectral tensor to focus on the multi-offset and multi-scale joint high response regions, and outputting enhanced spatially offset Raman spectral data;

[0067] S6. Extract biochemical Raman feature vectors containing bone matrix phosphate peaks, titanium oxide layer peaks, inflammatory mediator peaks, and bacterial metabolite peaks from the enhanced spatially offset Raman spectroscopy data. Input the biochemical Raman feature vectors into the pre-trained machine learning model to obtain the health status classification results and inflammatory index values ​​of dental peri-implantitis.

[0068] In this embodiment, S1 includes the following steps:

[0069] S11. Construct an adjustable spatially offset Raman detection device with continuous spatial offset capability. The adjustable spatially offset Raman detection device includes a central excitation fiber, an annular collection fiber array, and a laser power calibration module. The central excitation fiber is used to output an excitation laser with a wavelength of 785-800 nanometers. The annular collection fiber array is evenly distributed around the excitation fiber and fixed to a piezoelectric micromotion platform. By controlling the piezoelectric micromotion platform, the annular collection fiber array is spatially offset relative to the excitation beam in the direction of the axial outer edge of the dental implant, resulting in an offset distance Δx. The offset distance ranges from 0 to 4 mm, covering the common thickness of the oral mucosa and the diameter range of the dental implant neck, ensuring that Raman scattering signals can be collected from the deep area surrounding the implant.

[0070] S12. Install the adjustable spatial offset Raman detection device on the positioning bracket and align it with the mucosal surface around the dental implant, so that the laser emission axis of the central excitation fiber is perpendicular to the implant area. The alignment operation ensures that the excitation light spot is precisely focused below the mucosal surface. The vertical alignment distance between the adjustable spatial offset Raman detection device and the mucosal surface is d align , vertical alignment distance d align The average thickness parameter h of individual gingival tissue gum Added to the spot focus offset correction coefficient ε, it is used to ensure that the excitation light energy accurately acts on the mucosa-implant interface area;

[0071] S13. Activate the laser power calibration module to calibrate the laser intensity before emitting the excitation light, ensuring that the excitation light power parameter P0 is within the safe range for non-invasive testing. The laser power parameter P0 represents the laser power at the output end of the central excitation fiber and is not to exceed 100 mW. This power control ensures that oral soft tissue is not thermally damaged by irradiation while ensuring sufficient Raman signal excitation efficiency.

[0072] S14. Perform zero point calibration on the initial offset position of the piezoelectric micro-motion platform, record the current offset position as the offset starting position coordinate x0, and set the offset starting position coordinate x0 as the reference point for spatial offset control. Calculate the current offset position coordinate x0 based on the offset starting position coordinate x0 and the offset distance Δx. actual , current offset position coordinate x actual Indicates the current spatial position of the collected optical fiber array in actual operation, which is used as the position positioning basis in the dynamic spatial offset control process and outputs the calibration parameter set C cal ={P0,x actual ,d align}.

[0073] In this embodiment, S2 includes the following steps:

[0074] S21. Adjust the spatial offset increment parameter sequentially so that each spatial offset distance is within the range of 0 to 4 mm. The spatial offset increment parameter is the difference between two adjacent spatial offset distances. All spatial offset distances are arranged sequentially to form a spatial offset sequence.

[0075] S22. At each spatial offset, based on the offset starting position coordinates and the current spatial offset distance, the current spatial offset distance is added to the offset starting position coordinates to obtain the actual spatial position coordinates. The actual spatial position coordinates are used to control the precise spatial position of the collection fiber array to achieve accurate positioning of each spectral acquisition;

[0076] S23. Collecting raw spatially offset Raman spectral data at each actual spatial position coordinate. Raw spatially offset Raman spectral data refers to the spectral intensity values ​​recorded at all Raman wavenumber points at that position. The wavenumber point is a physical quantity expressed in units of reciprocal centimeters.

[0077] S24. Calculate the peak intensity of the target Raman peak and the RMS value of the corresponding background noise for each acquisition of the raw spatially offset Raman spectrum data. Divide the peak intensity of the target Raman peak by the RMS value of the corresponding background noise to obtain a signal-to-noise ratio parameter for the spatial location. The signal-to-noise ratio parameter is used to measure the signal quality at that spatial location.

[0078] S25. Based on all the signal-to-noise ratio parameters in the spatial offset sequence, a central difference method is used to calculate the signal-to-noise ratio gradient parameter by subtracting the previous signal-to-noise ratio parameter from the next adjacent signal-to-noise ratio parameter at the current spatial position, and dividing the result by twice the spatial offset increment parameter. The signal-to-noise ratio gradient parameter reflects the rate at which the signal-to-noise ratio changes with spatial offset distance and provides a basis for dynamically adjusting the spatial offset increment parameter. When the absolute value of the signal-to-noise ratio gradient parameter is less than a threshold, the spatial offset increment parameter is amplified to accelerate the scan. When the absolute value of the signal-to-noise ratio gradient parameter is greater than the threshold, the spatial offset increment parameter is reduced to refine the search.

[0079] S26. Determine the platform position where the signal-to-noise ratio parameter reaches its maximum value and the signal-to-noise ratio gradient parameter turns from positive to negative or approaches zero based on all signal-to-noise ratio parameters and their corresponding signal-to-noise ratio gradient parameters. Record the index corresponding to the platform position as the optimal signal-to-noise ratio index. Centered on this spatial offset distance, extend the target acquisition window W by one spatial offset increment parameter each forward and backward. opt ;

[0080] S27. In the target acquisition window W opt The original spatially offset Raman spectral data are repeatedly collected several times, and all spectral intensity values ​​at the same wave number point are averaged to obtain the optimized spatially offset Raman spectral data O(ν j ).

[0081] In this embodiment, S3 includes the following steps:

[0082] S31. Spatial offset distance Δx based on target acquisition window k The spatial offset-wavelet decomposition linkage mapping relationship between the number of wavelet decomposition layers is used to calculate the optimal number of wavelet decomposition layers. The optimal number of wavelet decomposition layers L opt represents the optimal number of decomposition layers for discrete wavelet decomposition of optimized spatially offset Raman spectroscopy data, γ is the mapping scale coefficient, and δ is the mapping bias coefficient. The mapping scale coefficient and mapping bias coefficient are used to achieve adaptive selection of the number of decomposition layers under different offset distances.

[0083] S32. In the preset mother wavelet library For each mother wavelet ψ r (t) Perform initial decomposition to optimize the spatially offset Raman spectral data O(ν j ) is input, and the detail coefficient set W of each mother wavelet in the first decomposition layer is obtained. 1,r (ν j ), using the detail coefficient set W of each mother wavelet 1,r (ν j ) and optimized spatial offset Raman spectroscopy data O(ν j) Calculate the energy compression rate parameter η r , used to evaluate the ability of different mother wavelets to concentrate the energy of effective signals when decomposing optimized spatially offset Raman spectroscopy data:

[0084]

[0085] Among them, N ν represents the total number of wave number points, which is used to optimize the spatial offset Raman spectral data O(ν j ) contains the number of Raman spectral wavenumber points;

[0086] In all energy compressibility parameters η r In the example, the mother wavelet with the largest value is selected as the optimal mother wavelet

[0087] S33. Using the optimal mother wavelet Optimize the spatial offset Raman spectroscopy data O(ν j ) Perform the optimal wavelet decomposition layer L opt Discrete wavelet decomposition of the layer to obtain the approximate coefficient vector corresponding to the optimal decomposition layer number The set of detail coefficient vectors corresponding to each decomposition layer Form the wavelet multi-scale coefficient set W:

[0088]

[0089] Among them, the set of approximate coefficient vectors and detail coefficient vectors represents the multi-scale decomposition of signal energy at different scales, and l is the decomposition level index, which is used to distinguish the detail coefficient vectors at each scale.

[0090] In this embodiment, S4 includes the following steps:

[0091] S41. For the wavelet multi-scale coefficient set W, for each decomposition scale l detail coefficient vector D l (ν j ) calculate the coefficient variance parameters respectively Used to evaluate the energy dispersion of coefficients under decomposition scale:

[0092]

[0093] Among them, μ l is the l-th scale detail coefficient vector D l (ν j ) coefficient mean, N ν is the total number of Raman spectral wavenumber points, which is used to measure the background noise energy at the decomposition scale;

[0094] S42. Detail coefficient vector D for each decomposition scale l l(ν j ) Calculate the information entropy parameter H l The information entropy parameter is used to evaluate the uncertainty of the distribution of detail coefficients under the decomposition scale. The information entropy parameter is obtained by multiplying the probability of all detail coefficient amplitudes appearing in different amplitude intervals under the statistical decomposition scale by their logarithms with the base 2 and taking the negative value;

[0095] S43. Binding coefficient variance parameter and information entropy parameter H l Generate information entropy-variance joint threshold parameter T l , which is used to adaptively separate the effective signal from the background noise:

[0096] T l =α·σ l +β·H l ;

[0097] Among them, σ l represents the variance parameter of the lth scale coefficient The arithmetic square root of is used to balance the amplitude scale of the coefficient. α and β represent the weighting coefficients of variance and information entropy respectively.

[0098] S44. For each scale l, the detail coefficient vector D l (ν j ) performs coefficient threshold processing, and retains the original value when the absolute value of the detail coefficient vector is greater than or equal to the information entropy-variance joint threshold parameter, otherwise it is set to zero to obtain the detail coefficient vector after threshold processing

[0099] S45. Using the optimal mother wavelet Based on the detail coefficient vector after threshold processing Approximate coefficient vector corresponding to the optimal number of decomposition levels Perform inverse discrete wavelet transform to obtain reconstructed denoised spatially offset Raman spectral data O denoise (ν j ).

[0100] In this embodiment, S5 includes the following steps:

[0101] S51. The reconstructed and denoised spatially offset Raman spectral data at each spatial offset distance are stacked in the order of spatial offset-wavenumber-wavelet scale to form a three-dimensional spatially offset wavelet spectral tensor T(i, j, l), which represents the distribution characteristics of Raman spectral intensity at different spatial offset distances and wavelet scales;

[0102] Spatial offset-wavenumber-wavelet scale sequential stacking is a technical process in which, in the context of Raman spectroscopy non-invasive detection technology, the reconstructed and denoised Raman spectral data obtained by multiple spatial offset acquisitions are arranged, integrated, and constructed into a three-dimensional tensor data structure according to the three-dimensional coordinate relationship of spatial offset distance-Raman spectral wavenumber-wavelet decomposition scale in order to comprehensively reflect the signal characteristics of oral tissue at different spatial depths (spatial offset), different molecular vibration frequencies (wavenumber), and different signal decomposition detail levels (wavelet scale).

[0103] Spatial offset refers to obtaining Raman signals from different depths or regions of the implant by adjusting the distance of the collection fiber array relative to the excitation beam, thereby characterizing the change of the signal with the detection depth;

[0104] Wavenumber refers to the horizontal axis variable of the Raman spectrum, reflecting the difference in molecular vibration energy levels and corresponding to the molecular specific information of biochemical components;

[0105] Wavelet scale refers to the different levels of details and approximate components obtained by decomposing the Raman signal using the wavelet multi-scale decomposition algorithm, which can respectively reflect broadband noise, slowly varying background and weak characteristic signals.

[0106] The result of sequential stacking is that the denoised spectral intensity values ​​at each spatial offset distance, each wavenumber point, and each wavelet scale are integrated into the same three-dimensional data tensor according to the nested relationship of the three dimensions of spatial offset-wavenumber-wavelet scale, thus realizing the full-domain stereoscopic expression of signal information. The three-dimensional tensor preserves the comprehensive data of different detection depths, different molecular vibration sites, and different multi-scale signal decomposition levels.

[0107] In this embodiment, the spatial offset-wavenumber-wavelet scale sequential stacking breaks through the limitation that traditional two-dimensional spectral data can only process single depth or single scale signals. Through three-dimensional sequential stacking, the information fusion of the three domains of space-wavenumber-scale is realized, which has obvious technical progress in the extraction of early, deep and weak lesion characteristics.

[0108] S52. Define the characteristic peak wave number ν in the detection of dental peri-implantitis bio The characteristic window wave number set ν centered j ∈[ν bio -Δν,ν bio +Δν], calculate the characteristic response tensor F(i,l):

[0109]

[0110] Where Δν is the spectral line integration step, F(i,l) represents the cumulative response intensity in the characteristic peak area related to dental implantitis at the i-th spatial offset and the l-th wavelet scale, and νbio Phosphate bone mineralization (approximately 959 cm -1 ) and inflammatory mediators (approximately 1355cm -1 ) The Raman peak position of the pathological characteristic is used to focus on the biological characteristics of the lesion, and Δν represents the half-width of the characteristic peak wavenumber window;

[0111] S53. The spatial offset distance Δx relative to the characteristic response tensor F(i,l) i The weighted variance calculation of the spatial offset scale adaptive focusing factor Ω is obtained. i,l :

[0112]

[0113] Among them, Ω i,l is the spatial offset scale adaptive focusing factor, which represents the spatial offset distance Δx i , characteristic peak response and spatial offset distance center at scale l the relative contribution of the degree of deviation; represents the weighted spatial offset average distance of the feature response tensor F(i,l), which is used to highlight the biological signals in deep lesion areas. x Indicates the total number of spatial offset positions involved in the analysis, L opt represents the optimal number of wavelet decomposition layers;

[0114] Adaptive focusing factor Ω for spatial offset scale in S53 i,l , not only depends on the Raman signal intensity response of a certain characteristic peak area, but also dynamically integrates the response at the spatial offset distance (Δx i ), specifically, Ω i,l The distribution characteristics of the focusing signal at different detection depths are expressed by the spatial offset variance weighted by the characteristic peak intensity.

[0115] Instead of weighting all spatial offsets and scales equally, the contribution of each offset-scale combination in the target feature area and its dispersion from the center are combined; through the Δx weighted mean parameter, the weak peak response in the deep layer (dental implant-bone interface) is identified and highlighted.

[0116] Compared with existing conventional formulas, traditional tensor weighting methods mostly use energy normalization or direct weight projection, focusing only on how strong the signal is, without distinguishing the signal contribution due to the physical location of the detection (spatial offset depth). S53 incorporates spatial structure information into the signal enhancement logic and adaptively amplifies weak signals in deep areas of the lesion. In this embodiment, the spectral signal processing algorithm is organically combined with the actual physiological structure depth requirements of dental peri-implantitis, enabling the system to automatically enhance signal focus in the areas with the greatest biological value, rather than enhancing all high signal points on the surface, greatly improving the accuracy and early diagnosis capabilities of actual clinical applications.

[0117] S54. Adaptive focusing factor Ω using spatial offset scale i,l The three-dimensional spatial offset wavelet spectral tensor T(i,j,l) is adaptively nonlinearly focused enhanced in terms of spatial offset dimension and wavelet scale dimension to obtain enhanced spatial offset Raman spectral data O enh (ν j ):

[0118]

[0119] The nonlinear amplification parameter λ of the exponential function is used to enhance the contrast of the deep weak biological Raman signal relative to the surface interference background. The value of λ is a positive real number. The dental peri-implantitis lesion signal is enhanced in the deep layer through adaptive nonlinear focusing enhancement.

[0120] The S54 formula adopts the exponential function nonlinear amplification focusing mechanism, and the adaptive focusing factor Ω obtained by S53 is i,l It is embedded as an exponential factor in the weight distribution. Its function is to nonlinearly enhance the spatial-scale combination with large focusing factors (deep layers, important areas, weak characteristic responses), so that its weight in the overall synthesis is much greater than that of surface noise points or irrelevant areas. The parameter λ is adjustable, which facilitates the flexible optimization of focusing intensity for different individuals and different inflammatory phenotypes.

[0121] In this embodiment, by deeply combining the physiological structure (spatial offset distribution) with the algorithmic nonlinear focusing mechanism, the system's sensitivity to deep weak Raman signals is significantly improved while effectively suppressing surface interference, which represents a substantial technological advancement in the early diagnosis of non-invasive peri-implantitis.

[0122] In this embodiment, S6 includes the following steps:

[0123] S61. Based on enhanced spatially offset Raman spectroscopy data O enh (ν j ) Determine the central wave number ν of the bone matrix phosphate peak P , titanium oxide layer peak center wave number ν Ti 、Inflammatory mediator peak center wave number ν Iand the bacterial metabolite peak center wave number ν B , set a characteristic window with a half-width of Δν around each central wavenumber and calculate the peak area:

[0124]

[0125] Among them, A k represents the integrated area of ​​the kth pathological characteristic peak, ν k is the corresponding central wave number;

[0126] The calculation of the peak area in S61 is not only based on the enhanced Raman spectrum O enh (ν j ), and around the various biochemical characteristic peaks related to dental peri-implantitis (bone matrix phosphate, titanium oxide layer, inflammatory mediators, bacterial metabolites, etc.), a characteristic window was set at each central wave number point and the signal was integrated to obtain the peak area A k , and further normalized with reference peaks (phenylalanine, etc.) to form a feature vector.

[0127] This formula not only performs multi-point extraction and normalization, but also combines front-end enhanced focused output to accurately extract multivariate disease characteristics, adapting to the diagnostic needs of multi-molecule + multi-level + individual differences. The peak area window integration mechanism and normalization standard design achieve high robustness and universality, greatly improving the algorithm's adaptability to real and complex physiological conditions. The feature construction method in this implementation makes non-invasive Raman spectroscopy a truly quantitative molecular diagnostic tool, rather than just a qualitative screening tool.

[0128] S62. Select internal reference peak wave number ν ref Calculate the reference peak area A ref , and construct the normalized biochemical Raman feature vector:

[0129]

[0130] Wherein, v is a four-dimensional dimensionless eigenvector, and each component represents the area ratio of the corresponding pathological peak to the reference peak;

[0131] S63. Perform a normalization transformation operation on the normalized biochemical Raman feature vector v. The normalization transformation operation is used to adjust the values ​​of each dimension of the feature vector to a uniform scale range to match the input requirements of the pre-trained machine learning model. The normalization transformation operation is achieved by subtracting the value in the corresponding feature mean vector from the value of each dimension of the normalized biochemical Raman feature vector, and then dividing it by the value in the corresponding feature standard deviation vector to obtain a normalized feature vector.

[0132] S64. Normalize the feature vector Input the pre-trained machine learning model to obtain the classification result C of the health status of dental peri-implantitis state and inflammation index value I inflam .

[0133] The pre-trained machine learning model in this embodiment is used to extract molecular features and perform clinical quantitative identification of peri-implantitis lesions using Raman spectral data processed with wavelet multiscale denoising and spatial offset enhancement. The construction method mainly includes the following steps:

[0134] Dataset collection and annotation: A large number of original spatially offset Raman spectral data from actual clinical oral implant cases are collected, covering various states including health, mucositis, early periarthritis and late periarthritis. Each sample is manually annotated with pathological grades by professional physicians according to current diagnostic standards.

[0135] Preprocessing and feature engineering: The collected raw spectral data are sequentially subjected to dynamic spatial offset optimization, wavelet multi-scale denoising, and characteristic peak response enhancement processes to extract multidimensional Raman feature vectors including bone matrix phosphate peaks, titanium oxide layer peaks, inflammatory mediator peaks, and bacterial metabolite peaks. All feature vectors are normalized and standardized to eliminate amplitude differences between samples.

[0136] Feature selection and label construction, through statistical analysis and correlation screening, retain the most discriminatory Raman feature variables for the grading of dental peri-implantitis, and correspond one-to-one with the medically annotated grading labels and inflammation index labels to form a structured training set.

[0137] The machine learning model is set up and trained using support vector machines to design corresponding loss functions for classification tasks (outputting health status grades) and regression tasks (outputting inflammation index values). During model training, cross-validation and hyperparameter tuning are used to prevent overfitting, and stratified sampling is used to ensure the balance of samples of various labels.

[0138] For model evaluation and validation, part of the data is reserved as an independent validation set to evaluate the model's classification accuracy and regression mean square error indicators respectively to ensure that the model can maintain high robustness and clinical applicability for unseen samples. For models with insufficient performance, feature engineering and structural optimization are repeatedly adjusted until the set performance threshold is met.

[0139] Model storage and deployment: Store the trained model parameters in a portable format on the detection device terminal or cloud system, and combine them with the standardized input feature vector in actual detection to achieve health status classification and inflammation index output for a single Raman spectrum acquisition.

[0140] In this embodiment, the dental peri-implantitis health status classification result C stateIt is a multi-category label used to indicate one of the four states of the sample: healthy, mucositis, early periarthritis or late periarthritis. The inflammation index value I inflam It is a continuous indicator with a value range of 0 to 1, and is used to quantitatively reflect the degree of inflammation in the current test sample.

[0141] Example 1: The implant department of Oral Hospital A and the periodontology team conducted a clinical non-invasive detection comparative study on multiple patients suspected of peri-implantitis after implant surgery. The study used the method of the present invention and compared it with conventional X-ray imaging and probe measurement methods to evaluate the feasibility and detection effect of the present invention in actual clinical scenarios.

[0142] At the study site, participants were all patients who had dental implants for 12-36 months and presented with gingival redness, swelling, and suspected perigingival inflammation (bleeding easily on probing). Thirty patients (nearly 50:1 male-to-female ratio, aged 32-67 years) were selected. Each patient signed an informed consent form and underwent oral disinfection according to hospital infection control requirements. The on-site testing was conducted next to the dental chair between 9:00 AM and 12:00 PM. Participants were required to fast for two hours and maintain a clean mouth to minimize interference from saliva and food debris.

[0143] First, the basic physical signs of the patients were recorded and numbered, and the three-dimensional structural CT and periodontal pocket depth of the area around the dental implant of each patient were collected as baseline data. Subsequently, the non-invasive detection device of the present invention was used to carry out spectral acquisition. The device includes an adjustable spatial offset Raman probe, using a 785nm laser, the laser power is calibrated to 85mW, and the collection fiber array achieves a 0-4mm spatial offset. The device is fixed to the buccal mucosal surface of the implant area, and the piezoelectric micro-motion platform controls the offset increment to 0.5mm, completing data acquisition at 8 spatial offset positions. A complete Raman spectrum is collected at each offset position, and the original spectrum is transmitted to the supporting analysis system.

[0144] During the spectral processing stage, the device adaptively selects the optimal acquisition window based on the signal-to-noise ratio of each offset point, and automatically calls the spatial offset-wavelet decomposition linkage algorithm. According to the mucosal thickness and implant depth of each patient, the system dynamically determines the number of wavelet decomposition layers. The typical number of decomposition layers is 4-6 layers, and the mother wavelet type is automatically screened. The multi-scale coefficients are subjected to noise suppression and signal enhancement through the information entropy-variance joint threshold method to reconstruct the deep denoised Raman signal. The multi-channel, multi-scale signals are then input into the adaptive tensor focusing algorithm in the form of a three-dimensional tensor, and the 959cm -1 (phosphate), 1005cm -1 (phenylalanine), 1355cm -1(inflammatory mediators) to automatically enhance the characteristic signals of deep lesions. Finally, through characteristic peak area normalization, the normalized feature vector is extracted and input into a pre-trained support vector machine model to obtain a four-level classification result of healthy, mucositis, early periarthritis, and late periarthritis, as well as an inflammatory index score. The entire testing process, from positioning, collection, analysis, to report generation, takes an average of 82 seconds, with the fastest time being only 74 seconds, and no patient discomfort caused by the operation.

[0145] In order to compare the effects of the present invention with those of conventional methods, all subjects underwent manual measurement with a periodontal probe and digital X-ray examination on the same day. The probe measurement results were independently performed by two experienced attending physicians to measure the periodontal pocket depth (mm) and bleeding index (0-3 points). X-ray examination recorded the changes in bone height around the implant through standard anteroposterior films and quantitatively analyzed the bone resorption. In the comparative experiment, the above three methods output healthy and pathological grades respectively, and were retrospectively calibrated with the actual surgical pathology diagnosis results two weeks after surgery.

[0146] The actual test results show that, in all 30 samples, the Raman enhanced spatial offset denoised signal is 959cm -1 All phosphate peaks were detected within a single scan, with an average signal-to-noise ratio of 22.8±5.6, exceeding that of conventional methods (conventional back-scattered Raman signals have a signal-to-noise ratio of only 9.2±4.1). The method of the present invention can distinguish between healthy, mucositis, early-stage, and late-stage periarthritis, with an automatic classification accuracy rate of 93.3% (28 / 30). The correlation coefficient between the inflammation index and postoperative pathology score was r=0.91, and the sensitivity for early-stage cases with a bone resorption threshold of 0.5-2mm was increased to 91.7%. In contrast, the classification accuracy of conventional probe measurement was only 80.0% (24 / 30), and the detection rate for microscopic submucosal bone resorption (<2mm) was less than 60%. X-ray detection sensitivity for early bone resorption <1.5mm was only 37.5%, and the accuracy for distinguishing healthy from mucositis was less than 70%. For soft tissue thicknesses >2mm, the signal-to-noise ratio of the target peak with conventional back-scattered Raman is generally less than 10. In some cases, the bone peak is completely submerged in background noise, resulting in a high misdiagnosis rate.

[0147] Among 30 patients, 4 cases of early bone resorption (0.7-1.1mm) were all accurately detected by the method of the present invention, while only 1 case was found by probe measurement, and X-rays were unable to distinguish. After the normalized feature vector was input, the machine learning model only took 0.09 seconds to output the inflammation grade and inflammation index, greatly improving the efficiency of clinical instant diagnosis. The classifications at each level are as follows: healthy (11 cases, all correctly judged), mucositis (8 cases, 1 case was wrongly judged as healthy), early periarthritis (6 cases, all accurate), and late periarthritis (5 cases, 1 case was wrongly judged as early).

[0148] Furthermore, using clinical sample library data collected from June to August 2024 to compare the iterative training of the algorithm model, the traditional method achieved a healthy / lesion classification accuracy of 81.2% / 76.7% on the training set (N=180) and the test set (N=60), respectively. The tensor focusing enhancement + machine learning discrimination method of the present invention achieved a classification accuracy of 95.6% / 93.3% on the same samples, and the mean square error of the inflammation index was 0.047 (the traditional method was 0.188). The multi-scale adaptive denoising and tensor enhancement mechanism of the present invention significantly improved the detectability of the Raman characteristic peaks of deep weak signals, enabling non-invasive and accurate assessment at the molecular level in complex oral tissue environments.

[0149] In terms of user experience, patients reported that the testing process was comfortable, painless, bleeding-free, and short. Doctors subjectively commented that the device was convenient and easy to use, and that the data output format could automatically connect to the hospital's information system, enabling structured report storage. Combined with specific data and clinical scenarios, the method not only overcomes the technical bottleneck of traditional X-ray and probe methods, which cannot non-invasively, rapidly, and quantitatively identify early-stage peri-implantitis, but also enhances the clinical usability and diagnostic value of molecular-level testing.

[0150] The present invention proposes a spatial offset-wavelet decomposition linkage mapping algorithm, which adaptively determines the number of wavelet decomposition layers based on the spatial offset distance, and automatically screens the optimal mother wavelet through the energy compression rate to achieve multi-scale structured modeling of Raman signals at different detection depths. It can dynamically adjust the decomposition parameters according to the depth of the target area, effectively separate soft tissue scattering and saliva fluorescence broadband noise, enhance the detectability of deep Raman characteristic peaks of bone matrix and inflammatory mediators, and significantly improve the detection sensitivity and accuracy of weak signals in the early stage of dental peri-implantitis.

[0151] The present invention constructs a characteristic response tensor based on the typical pathological peaks of peri-implantitis, performs nonlinear weighting on the three-dimensional wavelet spectrum tensor through a spatial offset scale adaptive focusing factor, adopts an exponential amplification strategy to preferentially enhance deep and weak lesion signals, and automatically suppresses surface interference. It can better adapt to different tissue depths, tissue structures and inflammatory heterogeneity, achieve focused enhancement and high-confidence output of the molecular characteristics of the lesions, and improve the molecular-level quantitative capability of non-invasive detection.

[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 peri-implantitis based on spatially offset Raman spectroscopy, characterized in that: The steps include: S1. Construct an adjustable spatial offset Raman detection device and align it with the mucosal surface surrounding the dental implant. Perform laser power calibration and initial offset zero calibration to obtain a calibration parameter set. S2. Initiate dynamic spatial offset scanning control based on the calibration parameters, monitor the Raman signal-to-noise ratio changes in real time, record the original spatial offset Raman spectral data corresponding to each offset distance, select the offset window with the optimal signal-to-noise ratio as the target acquisition window, and output the optimized spatial offset Raman spectral data; S3. Based on the optimized spatial offset Raman spectral data, the spatial offset - wavelet decomposition linkage mapping relationship is called, and the preset mother wavelet is used to perform wavelet multiscale decomposition on the optimized spatial offset Raman spectral data to obtain a set of wavelet multiscale coefficients; S4. Calculate the coefficient variance and information entropy of the wavelet multi-scale coefficient set at each scale, and generate an adaptive threshold set, use the adaptive threshold set to suppress the background noise coefficient and reconstruct the reconstructed denoised spatially offset Raman spectral data; S5. Stacking the reconstructed denoised spatially offset Raman spectral data and the corresponding offset distance and wavelet scale in a three-dimensional order of offset-wavenumber-scale to generate a three-dimensional spatially offset wavelet spectral tensor and performing centralized enhancement, and outputting the enhanced spatially offset Raman spectral data; S6. Extract biochemical Raman eigenvectors from the enhanced spatially offset Raman spectroscopy data and input the biochemical Raman eigenvectors into a pre-trained machine learning model to obtain the health status classification results and inflammation index values ​​of dental peri-implantitis.

2. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Construct an adjustable spatially offset Raman detection device with continuous spatial offset capability. The adjustable spatially offset Raman detection device includes a central excitation fiber, an annular collection fiber array, and a laser power calibration module. The central excitation fiber is used to output an excitation laser with a wavelength of 785-800 nanometers. The annular collection fiber array is evenly distributed around the excitation fiber and fixed to a piezoelectric micromotion platform. By controlling the piezoelectric micromotion platform, the annular collection fiber array is spatially offset relative to the excitation beam in the direction of the axial outer edge of the dental implant, obtaining an offset distance Δx. S12. Install the adjustable spatial offset Raman detection device on the positioning bracket and align it with the mucosal surface around the dental implant, so that the laser emission axis of the central excitation fiber is perpendicular to the implant area. The vertical alignment distance between the adjustable spatial offset Raman detection device and the mucosal surface is d align , vertical alignment distance d align The average thickness parameter h of individual gingival tissue gum Added with the spot focus offset correction coefficient ε; S13. Start the laser power calibration module to calibrate the laser intensity before the excitation light is emitted so that the excitation light power parameter P0 is controlled within the safe range of non-invasive detection. The laser power parameter P0 represents the laser power at the output end of the central excitation fiber; S14. Perform zero point calibration on the initial offset position of the piezoelectric micro-motion platform, record the current offset position as the offset starting position coordinate x0, and set the offset starting position coordinate x0 as the reference point for spatial offset control. Calculate the current offset position coordinate x0 based on the offset starting position coordinate x0 and the offset distance Δx. actual , output calibration parameter set C cal ={P0,x actual ,d align }.

3. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 2, characterized in that: The S2 comprises the following steps: S21. By sequentially adjusting the spatial offset increment parameter so that each spatial offset distance is within the range of 0 to 4 mm, the spatial offset increment parameter is the difference between each two adjacent spatial offset distances; S22. At each spatial offset, based on the offset starting position coordinates and the current spatial offset distance, the current spatial offset distance is added to the offset starting position coordinates to obtain the actual spatial position coordinates; S23. Collecting raw spatially offset Raman spectral data at each actual spatial position coordinate; S24. Calculate the peak intensity of the target Raman peak and the root mean square value of the corresponding background noise for each acquisition of the raw spatially offset Raman spectrum data, and calculate the signal-to-noise ratio parameter of the spatial position by dividing the peak intensity of the target Raman peak by the root mean square value of the corresponding background noise; S25. Based on all the SNR parameters in the spatial offset sequence, use the central difference method to calculate the SNR gradient parameter by subtracting the previous SNR parameter from the next adjacent SNR parameter at the current spatial position, and then dividing the result by twice the spatial offset increment parameter. When the absolute value of the SNR gradient parameter is less than a threshold, the spatial offset increment parameter is amplified to accelerate the scan. When the absolute value of the SNR gradient parameter is greater than the threshold, the spatial offset increment parameter is reduced to refine the search. S26. Determine the platform position where the signal-to-noise ratio parameter reaches its maximum value and the signal-to-noise ratio gradient parameter turns from positive to negative or approaches zero based on all signal-to-noise ratio parameters and their corresponding signal-to-noise ratio gradient parameters. Record the index corresponding to the platform position as the optimal signal-to-noise ratio index. Centered on this spatial offset distance, extend the target acquisition window W by one spatial offset increment parameter each forward and backward. opt ; S27. In the target acquisition window W opt The original spatially offset Raman spectral data are repeatedly collected several times, and all spectral intensity values ​​at the same wave number point are averaged to obtain the optimized spatially offset Raman spectral data O(ν j ).

4. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 3, characterized in that: The S3 includes the following steps: S31. Spatial offset distance Δx based on target acquisition window k The spatial offset-wavelet decomposition linkage mapping relationship between the number of wavelet decomposition layers is used to calculate the optimal number of wavelet decomposition layers. The optimal number of wavelet decomposition layers L opt represents the optimal decomposition level number of discrete wavelet decomposition for optimized spatially offset Raman spectroscopy data, γ is the mapping scale coefficient, and δ is the mapping bias coefficient; S32. In the preset mother wavelet library For each mother wavelet ψ r (t) Perform initial decomposition to optimize the spatially offset Raman spectral data O(ν j ) is input, and the detail coefficient set W of each mother wavelet in the first decomposition layer is obtained. 1,r (ν j ), using the detail coefficient set W of each mother wavelet 1,r (ν j ) and optimized spatial offset Raman spectroscopy data O(ν j ) Calculate the energy compression rate parameter η r , in all energy compressibility parameters η r In the example, the mother wavelet with the largest value is selected as the optimal mother wavelet S33. Using the optimal mother wavelet Optimize the spatial offset Raman spectroscopy data O(ν j ) Perform the optimal wavelet decomposition layer L opt Discrete wavelet decomposition of the layer to obtain the approximate coefficient vector corresponding to the optimal decomposition layer number The set of detail coefficient vectors corresponding to each decomposition layer Form a wavelet multi-scale coefficient set W.

5. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 4, characterized in that: The S4 comprises the following steps: S41. For the wavelet multi-scale coefficient set W, for each decomposition scale l detail coefficient vector D l (ν j ) calculate the coefficient variance parameters respectively S42. Detail coefficient vector D for each decomposition scale l l (ν j ) Calculate the information entropy parameter H l The information entropy parameter is obtained by multiplying the probability of all detail coefficient amplitudes appearing in different amplitude intervals under the statistical decomposition scale by their logarithms with the base 2 and taking the negative value; S43. Binding coefficient variance parameter The arithmetic square root of and information entropy parameter H l Generate the information entropy-variance joint threshold parameter T through weighted summation l ; S44. For each scale l, the detail coefficient vector D l (ν j ) performs coefficient threshold processing, and retains the original value when the absolute value of the detail coefficient vector is greater than or equal to the information entropy-variance joint threshold parameter, otherwise it is set to zero to obtain the detail coefficient vector after threshold processing S45. Using the optimal mother wavelet Based on the detail coefficient vector after threshold processing Approximate coefficient vector corresponding to the optimal number of decomposition levels Perform inverse discrete wavelet transform to obtain reconstructed denoised spatially offset Raman spectral data O denoise (ν j ).

6. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 5, characterized in that: The S5 comprises the following steps: S51. The reconstructed and denoised spatially offset Raman spectral data at each spatial offset distance are stacked in the order of spatial offset-wavenumber-wavelet scale to form a three-dimensional spatially offset wavelet spectral tensor T(i, j, l), which represents the distribution characteristics of Raman spectral intensity at different spatial offset distances and wavelet scales; S52. Define the characteristic peak wave number ν in the detection of dental peri-implantitis bio The characteristic window wave number set ν centered j ∈[ν bio -Δν,ν bio +Δν], calculate the characteristic response tensor F(i,l); S53. The spatial offset distance Δx relative to the characteristic response tensor F(i,l) i The weighted variance calculation of the spatial offset scale adaptive focusing factor Ω is obtained. i,l : Among them, Ω i,l is the spatial offset scale adaptive focusing factor, which represents the spatial offset distance Δx i , characteristic peak response and spatial offset distance center at scale l the relative contribution of the degree of deviation; Represents the average distance of spatial offset after weighting of feature response tensor F(i,l), N x Indicates the total number of spatial offset positions involved in the analysis, L opt represents the optimal number of wavelet decomposition layers; S54. Adaptive focusing factor Ω using spatial offset scale i,l The three-dimensional spatial offset wavelet spectral tensor T(i,j,l) is adaptively nonlinearly focused enhanced in terms of spatial offset dimension and wavelet scale dimension to obtain enhanced spatial offset Raman spectral data O enh (ν j ): The nonlinear amplification parameter λ of the exponential function is used to enhance the contrast of the deep weak biological Raman signal relative to the surface interference background, and the value of λ is a positive real number.

7. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 6, characterized in that: The S6 comprises the following steps: S61. Based on enhanced spatially offset Raman spectroscopy data O enh (ν j ) Determine the central wave number ν of the bone matrix phosphate peak P , titanium oxide layer peak center wave number ν Ti 、Inflammatory mediator peak center wave number ν I and the bacterial metabolite peak center wave number ν B , set a characteristic window with a half-width of Δν around each central wavenumber and calculate the peak area: Among them, A k represents the integrated area of ​​the kth pathological characteristic peak, ν k is the corresponding central wave number; S62. Select internal reference peak wave number ν ref Calculate the reference peak area A ref , and construct the normalized biochemical Raman feature vector: Wherein, v is a four-dimensional dimensionless eigenvector, and each component represents the area ratio of the corresponding pathological peak to the reference peak; S63. Perform a normalization transformation operation on the normalized biochemical Raman eigenvector v to obtain a normalized eigenvector S64. Normalize the feature vector Input the pre-trained machine learning model to obtain the classification result C of the health status of dental peri-implantitis state and inflammation index value I inflam .

8. The non-invasive detection method for peri-implantitis based on spatially offset Raman spectroscopy according to claim 7, characterized in that: The dental peri-implantitis health status classification result C state It is a multi-category label used to indicate one of the four states of the sample: healthy, mucositis, early periarthritis or late periarthritis. The inflammation index value I inflam It is a continuous indicator with a value range of 0 to 1, and is used to quantitatively reflect the degree of inflammation in the current test sample.

Images