Method for monitoring the quality of cleaning of ophthalmic precision instruments

Abstract

The application relates to the technical field of medical equipment detection, and discloses an ophthalmic precision instrument cleaning quality monitoring method, which comprises the following steps: collecting a Raman spectrum signal of an instrument surface; sequentially performing dark current removal, cosmic ray removal and fluorescent background removal time sequence pretreatment on the signal; performing scattering correction on the pretreated signal to eliminate signal fluctuation caused by a high-reflectivity curved surface; extracting characteristic peaks of proteins, lipids and polysaccharides; and calculating residual signal-to-noise ratio according to the characteristic peaks to monitor the cleaning quality. Without introducing exogenous substances or modifying optical hardware, the application realizes nondestructive and multi-type detection of trace mixed residues on a high-reflectivity metal surface.

Description

Technical Field

[0001] This invention relates to the field of medical equipment testing technology, and in particular to a method for monitoring the cleaning quality of precision ophthalmic instruments. Background Technology

[0002] Most ophthalmic instruments are made of highly reflective metals (such as stainless steel and titanium alloys), and their surfaces exhibit complex curvature variations. This presents two inherent challenges for nondestructive testing that are difficult to overcome simultaneously. First, residues adhere to the instrument surface in the form of a solid film, generating extremely weak Raman signals. Furthermore, the metal substrate, under laser excitation, produces a broad-spectrum fluorescence background with intensity several orders of magnitude higher than the Raman signal. Simultaneously, the dark current noise of the detector itself and random spikes from cosmic rays further obscure the true signal. Second, the highly reflective curved surface of the instrument causes drastic fluctuations in signal intensity at different measurement points due to differences in surface curvature and reflectivity, rendering threshold determination methods based on absolute signal intensity inapplicable.

[0003] In existing technologies, surface-enhanced Raman scattering can improve signal strength, but the introduced nanoparticles may cause secondary contamination of the equipment; while complex optical hardware modifications can suppress reflection interference, they are costly and difficult to promote. Summary of the Invention

[0004] Therefore, the purpose of this invention is to overcome the problems of incomparable signals caused by the highly reflective curved surfaces of ophthalmic instruments and the signal of trace mixed residues being submerged by metal fluorescence and noise, and to provide a method for monitoring the cleaning quality of precision ophthalmic instruments. Without introducing external substances or modifying optical hardware, this method achieves non-destructive, multi-type detection of trace mixed residues on highly reflective metal surfaces through time-sequential preprocessing and scattering correction.

[0005] To address the aforementioned technical problems, this invention provides a method for monitoring the cleaning quality of ophthalmic precision instruments, comprising: The surface of the cleaned instrument is scanned at multiple points using a Raman spectroscopy detector, and the Raman spectral signal of each measurement point is acquired. The Raman spectral signal is preprocessed to obtain a preprocessed spectral signal. The preprocessing includes: performing a first processing on the Raman spectral signal to remove the dark current background noise of the detector to obtain a first signal; performing a second processing on the first signal to remove random spikes generated by high-energy particle impacts to obtain a second signal; and performing a third processing on the second signal to remove the broadband fluorescence background generated by the metal substrate to obtain the spectral signal. The spectral signal is subjected to scattering correction to eliminate signal intensity fluctuations caused by the highly reflective curved surface of the instrument, thereby obtaining a corrected spectral signal; Raman characteristic peaks are extracted from the corrected spectral signal; the Raman characteristic peaks include protein characteristic peaks, lipid characteristic peaks and polysaccharide characteristic peaks. The residual signal-to-noise ratio at each measurement point is calculated based on the Raman characteristic peaks to monitor the cleaning quality.

[0006] Preferably, the scattering correction of the spectral signal includes: denoting the preprocessed spectral signal as matrix X, where rows of matrix X correspond to each measurement point and columns correspond to each wavenumber point; calculating the mean spectrum of the spectra of each measurement point in matrix X, denoted as the reference spectrum; for each measurement point's spectrum xj, performing linear regression fitting with the reference spectrum as the independent variable and the spectrum xj of that measurement point as the dependent variable to obtain the offset coefficient bj and slope coefficient mj of that measurement point; correcting the spectrum of the measurement point to obtain the corrected spectrum xj*; traversing all measurement points to obtain the corrected spectral signal; wherein, the corrected spectrum xj* is calculated as: xj*=(xj-bj) / mj; the offset coefficient bj is used to correct the baseline offset caused by the change in the curvature of the instrument surface, and the slope coefficient mj is used to correct the overall intensity scaling of the spectrum caused by the difference in the reflectivity of the instrument surface.

[0007] Preferably, after obtaining the offset coefficient bj and slope coefficient mj of the measurement point, the method further includes: determining whether the measurement point is located in a first region based on the spatial coordinates of each measurement point on the instrument surface; the first region includes the cutting edge region, joint region, or tip region of the instrument; if so, then adjusting the offset coefficient bj of the measurement point. j and slope coefficient m j The local correction includes: extracting several correction measurement points in the spatial domain of the current measurement point that are not in the first region; calculating the offset coefficient of each correction measurement point, and obtaining a correction offset coefficient by weighted averaging of the offset coefficients of each correction measurement point; calculating the slope coefficient of each correction measurement point, and obtaining a correction slope coefficient by weighted averaging of the slope coefficients of each correction measurement point; and performing scattering correction on the spectrum of the current measurement point based on the correction offset coefficient and the correction slope coefficient.

[0008] Preferably, the weight of the weighted average is inversely proportional to the spatial distance from the corrected measurement point to the current measurement point.

[0009] Preferably, the first processing includes: acquiring a dark current spectrum with the same integration time as the detection while the laser source is off, and storing it as a background reference signal; for each measurement point, subtracting the intensity value of the corresponding wavenumber point in the background reference signal from the wavenumber signal point by point in the Raman spectral signal to obtain the first signal.

[0010] Preferably, the second processing includes: scanning the spectral data of the first signal point by point, calculating the mean and standard deviation of each data point and its two adjacent data points; if the intensity value of the current data point exceeds the mean plus five times the standard deviation, it is marked as a candidate peak; for each candidate peak, its peak width characteristic value is calculated; the peak width characteristic value is the number of data points crossed when the intensity value of the data point drops to 50% of its peak height; if the peak width characteristic value is less than or equal to 2, the candidate peak is determined to be a cosmic ray peak; otherwise, it is determined to be a true Raman characteristic peak; the intensity value of the data point determined to be the cosmic ray peak is replaced with the median value of its two adjacent data points to obtain the second signal.

[0011] Preferably, the third processing method for removing the broadband fluorescence background generated by the metal substrate based on the adaptive baseline correction method includes: denoting the spectral data of the second signal as the original vector y, the length of which is n; n corresponds to the number of wavenumber points in the spectrum; setting the iteration number, smoothing parameter λ, and convergence threshold ε; calculating the weight vector w(t) in the t-th iteration; the initial weight of each data point is 1, and in the iteration, for data points with intensity values ​​lower than the current fitted baseline, assigning weights higher than the current fitted baseline; obtaining the fitted baseline z(t) for the t-th iteration by solving the following weighted penalized least squares problem: ; i is the index of the data point, with a value from 1 to n; y i z is the intensity value of the i-th data point in the original vector; i w represents the intensity value of the i-th data point in the fitted baseline. i (t) represents the weight of the i-th data point in the t-th iteration; Δ 2 It is a second-order difference operator; Calculate the difference between the current fitted baseline and the original vector. If the absolute value of the difference at the i-th data point is less than the threshold, reduce the weight of that data point in the next iteration so that the fitted baseline gradually approaches the low-value region of the spectrum in subsequent iterations. Subtract the fitted baseline obtained in the final iteration from the original vector to obtain the spectral data.

[0012] Preferably, the corrected spectral signal covers continuous spectral data in the wavenumber range of 400~3200 cm⁻¹; extracting Raman characteristic peaks from the corrected spectral signal includes: reading the original intensity value at each wavenumber position from the corrected spectral signal according to a preset plurality of characteristic peak wavenumber positions, and using this as the Raman characteristic peak intensity value at that measurement point; the preset plurality of characteristic peak wavenumber positions include: 1650~1660 cm⁻¹ corresponding to protein characteristic peaks. -1 and 1230~1270cm -1The lipid characteristic peak corresponds to 2850 cm⁻¹ -1 2920cm -1 and 1745cm -1 The characteristic peak of polysaccharides corresponds to 480 cm⁻¹ -1 850~870cm -1 and 1120cm -1 .

[0013] Preferably, extracting Raman characteristic peaks from the corrected spectral signal further includes performing local baseline recalibration on the corrected spectral signal, which includes: selecting 5 to 10 wavenumber points to the left and right of each preset wavenumber position of the characteristic peak as a local window; obtaining the local baseline of the local window based on linear fitting; removing the local baseline from the original intensity value to obtain the locally corrected peak height intensity value; and extracting the intensity value of the Raman characteristic peak from the locally corrected peak height intensity value.

[0014] Preferably, calculating the residual signal-to-noise ratio (SNR) of each measurement point based on the Raman characteristic peaks includes: for each measurement point, calculating the ratio of the protein characteristic peak, lipid characteristic peak, and polysaccharide characteristic peak to their respective local noise standard deviations to obtain the residual SNR of each residual type corresponding to the measurement point; comparing the protein residual SNR, lipid residual SNR, and polysaccharide residual SNR with their respective cleanliness thresholds to obtain the cleaning quality of each measurement point.

[0015] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: First, based on a time-series preprocessing step, dark current background noise is removed to establish a zero baseline. Then, random spikes from cosmic rays are removed to avoid severely distorting the fitted curve of the fluorescence background. Finally, the broadband fluorescence background of the metal substrate is removed. Based on this processing order set according to the physical differences of the interfering signals, the submerged signal is separated from strong noise while preserving the true Raman characteristic peaks to the greatest extent possible, thus solving the fundamental problem of trace residual signals being submerged by strong noise. Specifically, it can stably extract weak Raman characteristic peaks with a signal-to-noise ratio ≥3 from the original signal submerged by dark current, cosmic rays, and metallic fluorescence backgrounds.

[0016] Furthermore, scattering correction is performed on the preprocessed spectral signal. Through mathematical transformation, the spectrum of each measurement point is corrected to a uniform quantization scale to compensate for the signal intensity fluctuations caused by changes in surface curvature and reflectivity. This makes the residual signal-to-noise ratios at different locations comparable, thereby overcoming the problem of threshold determination failure caused by highly reflective surfaces.

[0017] Finally, characteristic peaks of proteins, lipids, and polysaccharides are extracted from the calibrated spectral signals. The ternary characteristic peak system can distinguish between protein, lipid, and polysaccharide residues in parallel. It can not only determine whether there are residues, but also distinguish the specific types of residues, providing a basis for subsequent classification and cleaning, and avoiding the efficiency loss caused by blind rewashing.

[0018] In summary, the ophthalmic precision instrument cleaning quality monitoring method of the present invention achieves non-destructive, multi-type detection of trace mixed residues on highly reflective metal surfaces through time-sequential preprocessing and scattering correction without introducing exogenous substances or modifying optical hardware. Attached Figure Description

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of a method for monitoring the cleaning quality of ophthalmic precision instruments in a preferred embodiment of the present invention; Figure 2 This is a flowchart of scattering correction of spectral signals in a preferred embodiment of the present invention; Figure 3 This is a flowchart illustrating the correction of the offset coefficient and slope coefficient in a preferred embodiment of the present invention; Figure 4 This is a flowchart of the third process in a preferred embodiment of the present invention; Figure 5 This is a flowchart of the process for extracting Raman characteristic peak intensity values ​​in a preferred embodiment of the present invention; Figure 6 This is a flowchart illustrating the extraction of Raman characteristic peak intensity values ​​in another embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0021] This invention provides a method for monitoring the cleaning quality of precision ophthalmic instruments. Targeting the unique characteristics of highly reflective metallic curved surfaces and trace residues on solid surfaces of precision ophthalmic instruments, Raman spectroscopy is used to achieve non-destructive testing of cleaning quality. The following detailed description, in conjunction with the accompanying drawings and specific implementation steps, illustrates this method.

[0022] Reference Figure 1 As shown, the specific steps include: S100: Multi-point scanning of the cleaned instrument surface is performed based on a Raman spectroscopy detector, and the Raman spectral signal of each measurement point is acquired; S200. The Raman spectral signal is preprocessed to obtain a preprocessed spectral signal; the preprocessing includes: performing a first processing on the Raman spectral signal to remove the dark current background noise of the detector to obtain a first signal; performing a second processing on the first signal to remove random spikes generated by high-energy particle impacts to obtain a second signal; and performing a third processing on the second signal to remove the broadband fluorescence background generated by the metal substrate to obtain the spectral signal. S300. Perform scattering correction on the spectral signal to eliminate signal intensity fluctuations caused by the highly reflective curved surface of the instrument, and obtain the corrected spectral signal. S400. Extract Raman characteristic peaks from the corrected spectral signal; the Raman characteristic peaks include protein characteristic peaks, lipid characteristic peaks, and polysaccharide characteristic peaks. S500: Calculate the residual signal-to-noise ratio of each measurement point based on the Raman characteristic peaks to monitor the cleaning quality.

[0023] In specific implementation step S100, the Raman spectroscopy detector is first calibrated for wavelength. Using the characteristic Raman peaks of acetaminophen or polystyrene standard substances as a reference, the wavenumber axis of the detector is corrected until the wavenumber error does not exceed ±1 cm. -1 Within the specified accuracy range, ensure the accuracy of subsequent feature peak identification.

[0024] After calibration, place the precision ophthalmic instruments to be tested (such as microsurgical scissors, microforceps, needle holders, lens implanters, or corneal knives) on the detection platform of the Raman spectroscopy detector. Start the Raman spectroscopy detector using an excitation source with a wavelength of 785 nm, which effectively avoids autofluorescence interference from metallic materials such as stainless steel and titanium alloys. Set the laser power to 50–100 mW, the integration time to 0.5–2 seconds / point, and the spectral acquisition range to 400–3200 cm⁻¹. -1 .

[0025] After cleaning, the instrument surface was scanned at multiple points to acquire the raw Raman spectral signal at each measurement point. The multi-point scanning strategy was adapted according to the instrument type: for microscissors and microforceps, the focus was on scanning the blade area and tip area; for needle holders, the focus was on scanning the joint area and clamping surface; for lens implanters and corneal knives, the focus was on scanning the end face area in contact with intraocular tissues.

[0026] In specific implementation step S200, after acquiring the raw Raman spectral signal of each measurement point, a first processing step is performed to remove the dark current background noise of the detector itself. Specifically, with the excitation source turned off, the dark current spectrum is acquired at the same integration time as the formal detection, and this dark current spectrum is stored as background reference data. The acquisition frequency of this dark current spectrum is once after the detection of each instrument is completed, or once after acquiring 10 measurement points, to eliminate the influence of detector temperature drift on the dark current.

[0027] When processing the raw Raman spectral signal at each measurement point, the intensity value of the corresponding wavenumber point in the background reference data is subtracted point by point from the raw spectral signal at that measurement point to obtain the first signal after removing dark current. This operation can effectively remove the thermionic noise and readout noise of the detector CCD, avoiding misjudging the device's own noise as a residual signal.

[0028] After obtaining the first signal, a second processing step is performed to remove random spike interference generated by high-energy particle impacts. Specifically, the spectral data of the first signal after removing dark current is scanned point-by-point, and the mean and standard deviation of each data point and its two adjacent data points (a total of 5 data points) are calculated. If the intensity value of the current data point exceeds the mean plus five times the standard deviation, the data point is marked as a candidate spike.

[0029] For each candidate peak, its peak width characteristic value is further calculated. This characteristic value is the number of data points that the intensity value of the data point must cross to reach 50% of its peak height. If the peak width characteristic value is less than or equal to 2, the candidate peak is determined to be a cosmic ray peak; if the peak width characteristic value is greater than 2, it is determined to be a true Raman characteristic peak and is retained.

[0030] The intensity value of the data point identified as a cosmic ray spike is replaced with the median value of the two adjacent data points to its left and right. This scanning and replacement process is repeated until no new cosmic ray spikes are detected in a single scan, thus obtaining a second signal. This method effectively distinguishes extremely narrow cosmic ray spikes (width of 1-2 data points) from true Raman characteristic peaks (half-width at half maximum usually corresponds to 3-10 data points) by using peak width characteristic values, avoiding the erroneous rejection of narrow Raman peaks.

[0031] After obtaining the second signal, a third processing step is performed to remove the broad-spectrum fluorescence background generated by the metal substrate. This embodiment employs the Adaptive Iterative Reweighted Penalized Least Squares (airPLS) algorithm. Specifically, the spectral data of the second signal after cosmic ray removal is denoted as the original vector, and the number of iterations, smoothing parameters, and convergence threshold are set. A fitting baseline is obtained by iteratively solving a weighted penalized least squares problem. This fitting process assigns higher weights to data points in the spectrum below the current fitting baseline, gradually bringing the fitting baseline closer to the low-value region of the spectrum. Iteration stops when the convergence condition or the maximum number of iterations is reached. The original vector is subtracted from the fitting baseline obtained in the final iteration to obtain the spectral data after removing the fluorescence background. This algorithm can automatically fit the smooth broad-spectrum fluorescence background generated by the metal substrate without requiring manual setting of baseline parameters, making it particularly suitable for online detection scenarios.

[0032] When implementing specific step S300, refer to... Figure 2 As shown, after obtaining the spectral data after removing the fluorescence background, scattering correction is performed on the spectral signal to eliminate signal intensity fluctuations caused by the highly reflective curved surface of the instrument. Specifically, the preprocessed spectral signal is denoted as matrix X, where rows correspond to each measurement point and columns correspond to each wavenumber point. The mean spectrum of each measurement point in matrix X is calculated and denoted as the reference spectrum. For each measurement point's spectrum xj, a linear regression is performed with the reference spectrum as the independent variable and the spectrum xj of that measurement point as the dependent variable to obtain the offset coefficient bj and slope coefficient mj of that measurement point. The measurement point spectrum is corrected to obtain the corrected spectrum xj*. All measurement points are traversed to obtain the corrected spectral signal. Here, j is the measurement point index. The corrected spectrum xj* is calculated as: xj* = (xj - bj) / mj. The offset coefficient bj is used to correct the baseline offset caused by changes in the curvature of the instrument surface, and the slope coefficient mj is used to correct the overall spectral intensity scaling caused by differences in the reflectivity of the instrument surface. This correction significantly reduces the difference in spectral intensity at different curvature positions on the same instrument, making it possible to determine residues based on a uniform threshold on highly reflective metal instruments.

[0033] In specific implementation step S400, after obtaining the corrected spectral signal, Raman characteristic peaks are extracted from the corrected spectral signal. The corrected spectral signal covers the range of 400~3200 cm⁻¹. -1 Continuous spectral data within the wavenumber range. The three types of Raman characteristic peaks extracted in this example and their corresponding wavenumber positions are as follows: protein characteristic peaks correspond to 1650~1660 cm⁻¹. -1 and 1230~1270cm -1 The characteristic lipid peak corresponds to 2850 cm⁻¹. -1 2920cm -1 and 1745cm -1The characteristic peak of polysaccharides corresponds to 480 cm⁻¹. -1 850~870cm -1 and 1120cm -1 During extraction, the intensity values ​​at each wavenumber position are read from the calibrated spectral signal according to the preset wavenumber positions of multiple characteristic peaks, and these values ​​are used as the Raman characteristic peak intensity values ​​for that measurement point.

[0034] In specific implementation step S500, the residual signal-to-noise ratio (SNR) of each measurement point is calculated based on the Raman characteristic peak intensity values ​​to monitor cleaning quality. Specifically, for each measurement point, the ratio of the peak height of the protein characteristic peak, lipid characteristic peak, and polysaccharide characteristic peak to their respective local baseline noise is calculated to obtain the residual SNR for each residue type at that measurement point. The local baseline noise refers to the standard deviation of the spectral intensity within the wavenumber range to the left and right of the characteristic peak. The residual SNR of protein, lipid, and polysaccharide is compared with their respective cleanliness thresholds. If any SNR is lower than the corresponding cleanliness threshold, it is determined that the measurement point has an excess of the corresponding type of residue. The cleanliness thresholds are set as follows: protein SNR ≥ 3, lipid SNR ≥ 3, and polysaccharide SNR ≥ 5. After comparing the residual SNR of all measurement points with their respective cleanliness thresholds, the cleaning quality monitoring results of the device can be obtained.

[0035] The above-described embodiment first removes dark current background noise and establishes a zero baseline based on a time-series preprocessing step; then it removes random cosmic ray spikes to avoid severely distorting the fitted curve of the fluorescence background; finally, it removes the broadband fluorescence background of the metal substrate. Based on this processing order set according to the physical differences of the interfering signals, the submerged signal is separated from strong noise while preserving the true Raman characteristic peaks to the greatest extent possible, thus solving the fundamental problem of trace residual signals being submerged by strong noise. Specifically, it can stably extract weak Raman characteristic peaks with a signal-to-noise ratio ≥3 from the original signal submerged by dark current, cosmic rays, and metallic fluorescence background.

[0036] Furthermore, scattering correction is performed on the preprocessed spectral signal. Through mathematical transformation, the spectrum of each measurement point is corrected to a uniform quantization scale to compensate for the signal intensity fluctuations caused by changes in surface curvature and reflectivity. This makes the residual signal-to-noise ratios at different locations comparable, thereby overcoming the problem of threshold determination failure caused by highly reflective surfaces.

[0037] Finally, characteristic peaks of proteins, lipids, and polysaccharides are extracted from the calibrated spectral signals. The ternary characteristic peak system can distinguish between protein, lipid, and polysaccharide residues in parallel. It can not only determine whether there are residues, but also distinguish the specific types of residues, providing a basis for subsequent classification and cleaning, and avoiding the efficiency loss caused by blind rewashing.

[0038] In summary, the ophthalmic precision instrument cleaning quality monitoring method of the present invention achieves non-destructive, multi-type detection of trace mixed residues on highly reflective metal surfaces through time-sequential preprocessing and scattering correction without introducing exogenous substances or modifying optical hardware.

[0039] In the above embodiments, the offset coefficient and slope coefficient for each measurement point are obtained through linear regression fitting. This method can obtain reliable correction parameters in flat areas of the instrument surface. However, ophthalmic precision instruments have extreme curvature variations and complex reflection characteristics in their cutting edge areas (such as the blade thickness of microscissors, which can be as thin as tens of micrometers), joint areas (such as the hinge of a needle holder), and tip areas (such as the tip of tweezers). In these first regions, the signal-to-noise ratio of the Raman spectral signal is extremely low, causing linear regression fitting to fail to obtain reliable offset coefficients and slope coefficients. The fitting results may contain outliers (such as negative slopes or excessively large offsets) due to poor signal quality. If these outlier coefficients are directly used for scattering correction, it will introduce even greater errors, causing distortion of the corrected spectrum. This is an overlooked problem because conventional scattering correction methods do not exhibit localized signal quality differences when applied to powder or liquid samples.

[0040] To address the problem of unreliable correction parameters in the first region (edge, joint, tip, etc.) of ophthalmic precision instruments due to poor signal quality in scattering correction methods, this invention further provides a preferred method for monitoring the cleaning quality of ophthalmic precision instruments. Through spatial neighborhood weighted correction, reliable parameters from flat regions are used to estimate the correction parameters for the first region, achieving high-precision scattering correction across the entire instrument surface. This ensures the accuracy and reliability of cleaning quality detection. (Refer to...) Figure 2 and 3 As shown, the specific plan is as follows.

[0041] The goal of the linear regression fitting described above is to find the optimal offset coefficient bj and slope coefficient mj, such that xj≈mj×reference spectrum+bj; the offset coefficient bj is used to correct the wavelength-independent baseline offset caused by the curvature change of the instrument surface, and the slope coefficient mj is used to correct the wavelength-proportional overall intensity scaling of the spectrum caused by the difference in reflectivity of the instrument surface.

[0042] After obtaining the offset coefficient bj and slope coefficient mj for each measurement point, it is determined whether the measurement point is located in the first region based on its spatial coordinates on the instrument surface. In this embodiment, the first region includes the cutting edge region, joint region, or tip region of the instrument. The specific determination method is as follows: A three-dimensional spatial model of the instrument is pre-established, and the coordinate ranges of the cutting edge region, joint region, and tip region are marked in the model. During measurement, the spatial coordinates of each measurement point are obtained through the instrument positioning device, and these coordinates are compared with the preset coordinate range of the first region. If the coordinates fall within the range, the measurement point is determined to be located in the first region; otherwise, the measurement point is determined to be located in a non-first region (i.e., a flat surface region).

[0043] If the current measurement point is determined to be located in the first region, then the offset coefficient bj and slope coefficient mj of that measurement point are locally corrected. The specific operation of the local correction is as follows: Centered on the current measurement point, extract several correction measurement points that are not in the first region within its spatial neighborhood. The extent of the spatial neighborhood can be set according to the instrument's geometry, such as a circular area with a radius of 1-2 mm centered on the current measurement point, or a rectangular area with a side length of 2-3 mm. Correction measurement points are measurement points located within this spatial neighborhood that are determined to be outside the first region. To ensure the reliability of the correction parameters, the number of correction measurement points should be at least 3. If there are fewer than 3 in the neighborhood, the neighborhood should be appropriately expanded until a sufficient number of correction measurement points are obtained.

[0044] Calculate the offset coefficient for each corrected measurement point, and then perform a weighted average based on these offset coefficients to obtain the corrected offset coefficient. The weight of the weighted average is inversely proportional to the spatial distance between the corrected measurement point and the current measurement point; that is, the closer the corrected measurement point, the higher its weight, and the farther away it is, the lower its weight. Specifically, let the coordinates of the current measurement point P be (x0, y0), and the coordinates of its k-th corrected measurement point Pk be (x0, y ... k y k If ), then the distance d from Pk to P is... k : ; weight w k = 1 / d k (or w) k = 1 / d k ²); and perform a weighted average of the offset coefficients of each corrected measurement point.

[0045] Similarly, the slope coefficient of each corrected measurement point is calculated separately, and a weighted average is performed based on the slope coefficients of each corrected measurement point to obtain the corrected slope coefficient.

[0046] Based on the correction offset coefficient and correction slope coefficient obtained above, scattering correction is performed on the spectrum of the current measurement point.

[0047] For measurement points that are determined to be outside the first region (i.e., measurement points located in flat surface regions), the offset coefficient bj and slope coefficient mj obtained by linear regression fitting in step two are directly used for scattering correction.

[0048] The present invention employs spatial neighborhood weighted correction, using reliable parameters from adjacent flat regions to estimate the parameters of the first region, thus avoiding contamination of the correction results by outlier parameters. Secondly, a weighted average is performed using weights inversely proportional to spatial distance; closer correction measurement points have a greater impact on the current measurement point. This aligns with the physical reality of continuously changing instrument surface morphology, ensuring a smooth transition in the corrected offset and slope coefficients, and preventing discontinuous jumps in the corrected spectrum at region boundaries.

[0049] In the above embodiments, the surface of the ophthalmic instrument is made of a highly reflective metal material. The intensity and shape of its fluorescent background are affected by the surface condition of the instrument (such as the degree of polishing and the thickness of the oxide layer) and the laser incident angle, resulting in significant differences between measurement points. The above-mentioned third processing of the second signal based on the adaptive baseline correction method to remove the broadband fluorescent background generated by the metal substrate, although solving the problem of automatic parameter setting to a certain extent, still has the following problems for the specific application scenario of ophthalmic instruments: First, the fluorescent background of the metal substrate overlaps with the Raman characteristic peaks of the residue in the spectrum. Especially when the residue signal is weak, the baseline correction algorithm is prone to misjudging the weak Raman peak as part of the fluorescent background and removing it incorrectly; Second, the drastic fluctuations in signal intensity caused by the highly reflective curved surface will interfere with the convergence process of baseline fitting, making it impossible for the fitted baseline to accurately approach the low-value region of the spectrum.

[0050] To address the issues that baseline correction methods cannot adaptively remove fluorescent backgrounds of varying shapes in the detection of highly reflective metallic curved surfaces of ophthalmic precision instruments, and that they easily misinterpret weak Raman peaks as background, the preferred embodiment of this invention further provides a method for monitoring the cleaning quality of ophthalmic precision instruments. Through a weighted penalized least squares and iterative weight update strategy, this method accurately removes the fluorescent background from the metal substrate while retaining weak Raman peaks without manual intervention, thereby achieving highly sensitive detection of trace residues. The following is a combination of... Figure 4 The specific implementation process will be explained in detail.

[0051] The spectral data of the second signal after removing cosmic rays is denoted as the original vector y, with a length of n, where n corresponds to the number of wavenumber points in the spectrum. Iteration parameters are set as follows: the maximum number of iterations is 50, and the smoothing parameter λ is set to 10. 5 The convergence threshold ε is set to 10.-6 The smoothing parameter λ controls the smoothness of the fitted baseline; the larger the λ value, the smoother the fitted baseline. The convergence threshold ε is used to determine whether the iteration has reached a stable state. The number of iterations is initialized to t=0.

[0052] In the t-th iteration, the weight vector w(t) is calculated; the initial weight of each data point is 1, i.e., w i (0)=1, i=1,2,…,n. In subsequent iterations, the weight vector is updated according to the following rule: for data points whose intensity values ​​in the spectrum are lower than the current fitted baseline, data points whose intensity values ​​in the spectrum are higher than the current fitted baseline are assigned weights. Specifically, when y i <z i At time (t), the weight w i (t+1) is set to a value close to 1; when y i ≥z i At time (t), the weight w i (t+1) is set to a value close to 0. The physical meaning of this weighting strategy is that points in the spectrum below the fitted baseline are likely to be the real background region (without Raman characteristic peak interference), and should be given high weight to guide the fitted baseline to be close to these points; while points in the spectrum above the fitted baseline are likely to be the region where Raman characteristic peaks are located, and should be given low weight to eliminate their interference with the baseline fitting.

[0053] The fitted baseline z(t) for the t-th iteration is obtained by solving the following weighted penalized least squares problem: ; Where i is the data point index, with a value from 1 to n; y i z is the intensity value of the i-th data point in the original vector; i w represents the intensity value of the i-th data point in the fitted baseline. i (t) represents the weight of the i-th data point in the t-th iteration; w i (t+1) represents the weight of the i-th data point in the (t+1)-th iteration; λ is the smoothing parameter; Δ² is the second-order difference operator.

[0054] The objective function consists of two terms: the first term is a weighted fitting residual term, which is used to make the fitting baseline close to the original spectrum; the second term is a smoothing penalty term, which is used to ensure the smoothness of the fitting baseline.

[0055] Calculate the difference between the current fitted baseline z(t) and the original vector y; for each data point, calculate the absolute value of each difference |y|. i -z iIf the absolute value of the difference between the i-th data points is less than a threshold, then the weight of that data point in the next iteration is reduced, so that the fitted baseline gradually approaches the lower value region of the spectrum in subsequent iterations. Specifically, when the absolute value of the difference is less than the threshold, the weight of the data point in the next iteration is set to a smaller value, so that the contribution of that data point in the next fitting is reduced. The purpose of this operation is that when the fitted baseline has already approached a certain data point, the guiding significance of that data point for subsequent fitting is reduced, and its weight should be reduced, so that the fitted baseline can further sink to a lower spectral region.

[0056] Iteration stops when the relative change between the fitted baselines of two consecutive iterations is less than the convergence threshold ε, or when the maximum number of iterations (50) is reached. The original vector y is subtracted from the fitted baseline obtained in the final iteration to obtain the spectral data after removing the fluorescence background. This algorithm can automatically fit a smooth, broad-spectrum fluorescence background generated by a metal substrate without requiring manual setting of baseline parameters, making it particularly suitable for online detection scenarios.

[0057] The present invention, in its embodiments, firstly employs a weighting strategy that assigns high weights to points below the baseline and low weights to points above the baseline, effectively avoiding the erroneous removal of weak Raman characteristic peaks. Regardless of how weak the Raman signal of the residue is, as long as it is positively convex relative to the fluorescence background, it will be automatically identified as a characteristic peak region and excluded from baseline fitting, ensuring the sensitivity of trace residue detection. Secondly, through a second-order difference smoothing penalty term and optimized smoothing parameters, it ensures that the fitted baseline conforms to the physical characteristics of a broad, gradually changing fluorescence background, avoiding overfitting or underfitting.

[0058] In the above embodiments, after completing dark current subtraction, cosmic ray removal, fluorescence background subtraction, and scattering correction, theoretically the baseline of the corrected spectral signal is zero, and the original intensity values ​​at each wavenumber position can be directly read as the characteristic peak intensities. However, in actual detection, due to the highly reflective metallic curved surface characteristics of ophthalmic instruments, even after global preprocessing and scattering correction, there are still slight local baseline drifts in the corrected spectral signal. The reasons for this local baseline drift include: First, although scattering correction eliminates the overall signal fluctuations caused by changes in surface curvature, the microscopic morphology of the highly reflective metallic surface (such as polishing marks and micro-scratches) will produce wavelength-related local distortions in different wavenumber regions, and these distortions cannot be completely eliminated by global scattering correction; Second, the essence of the adaptive baseline correction algorithm is to fit a smooth curve passing through the bottom of the spectrum. In regions with dense characteristic peaks (such as near the amide I and amide III bands of proteins), the presence of multiple adjacent characteristic peaks will form peak clusters, making it difficult for the baseline correction algorithm to accurately determine the position of the real background, resulting in residual baseline shifts in local areas. The above problems are not present or significant in routine Raman spectroscopy of powders, liquids or biological tissues because those samples have higher signal intensity and relatively uniform background.

[0059] To address the issue of peak height measurement error caused by residual local baseline drift after global preprocessing in the detection of highly reflective metallic curved surfaces of ophthalmic precision instruments using Raman spectroscopy feature extraction methods, a further method for monitoring the cleaning quality of ophthalmic precision instruments is provided. This method selects a local window near the wavenumber position of each characteristic peak and independently fits and subtracts the local baseline to extract the peak height intensity value after local correction, thereby controlling the peak height measurement error to within 5% and improving the accuracy of trace residue detection. The following is a combination of... Figure 5 and Figure 6 The specific implementation process will be explained in detail.

[0060] In the above embodiment, after preprocessing and scattering correction of the Raman spectral signal, a corrected spectral signal is obtained. This corrected spectral signal covers a range of 400–3200 cm⁻¹. -1 Continuous spectral data across the wavenumber range, with each measurement point corresponding to a complete spectral curve.

[0061] Based on multiple preset wavenumber positions of characteristic peaks, the original intensity values ​​at each wavenumber position are read from the calibrated spectral signal and used as the Raman characteristic peak intensity values ​​for that measurement point. The preset wavenumber positions of characteristic peaks specifically include: 1650–1660 cm⁻¹ corresponding to the protein characteristic peaks. -1 and 1230~1270cm -1 The lipid characteristic peak corresponds to 2850 cm⁻¹ -1 2920cm -1 and 1745cm-1 The characteristic peak of polysaccharides corresponds to 480 cm⁻¹ -1 850~870cm -1 and 1120cm -1 For characteristic peaks over a wavenumber range rather than at a single point (e.g., 1650–1660 cm⁻¹) -1 Take the wavenumber position corresponding to the maximum intensity within this range as the peak position, and record the intensity value at the peak position.

[0062] After reading the raw intensity values, the measurement reliability of each characteristic peak is evaluated. The evaluation method is as follows: calculate the standard deviation of the spectral intensity within a range of 10 wavenumber points on both sides of the characteristic peak. If the standard deviation exceeds a preset local fluctuation threshold (such as 5% of the average spectral intensity), it is determined that there is a local baseline drift at the position of the characteristic peak, and local baseline recalibration needs to be performed; otherwise, the raw intensity value read is directly used as the intensity value of the characteristic peak.

[0063] For characteristic peaks requiring local baseline recalibration, select 5-10 wavenumber points to the left and right of each preset wavenumber position as a local window. The specific size of the local window can be selected based on the density of the characteristic peaks: for isolated characteristic peaks (such as lipid peaks at 2850 cm⁻¹), ... -1 Polysaccharide 480cm -1 Select 5 wavenumber points on each side; for dense characteristic peak regions (such as protein 1650~1660 cm⁻¹), select 5 wavenumber points on each side. -1 (Nearby), select 8-10 wavenumber points on each side to ensure that the window contains a sufficient number of background points for baseline fitting. The principle for selecting the local window is: the window range should include the shoulder regions on both sides of the characteristic peak, but not include other adjacent characteristic peaks, to avoid interference from adjacent peaks.

[0064] Within a selected local window, the local baseline is obtained based on linear fitting. The specific method for linear fitting is as follows: using wavenumber points as the independent variable and spectral intensity as the dependent variable, a least-squares linear regression is performed on all data points within the window to obtain a straight line, which is the local baseline estimate for that local window. For cases where the baselines on both sides of the characteristic peak exhibit slight curvature, quadratic polynomial fitting can be used instead of linear fitting to obtain a more accurate local baseline estimate.

[0065] The obtained local baseline is removed from the original intensity value I0 to obtain the locally corrected peak height intensity value. Specifically, at the peak wavenumber position of the characteristic peak, the value of the local baseline I1 at that position is calculated. Then, I1 is subtracted from the original intensity value I0 read in step two to obtain the locally corrected peak height intensity value I = I0 - I1.

[0066] Finally, the intensity value of the Raman characteristic peak is extracted from the locally corrected peak height intensity value, and this intensity value is taken as the final measurement result of the characteristic peak at that measurement point. For the case where the same characteristic peak has multiple wavenumber positions (such as protein characteristic peaks corresponding to two wavenumber ranges), the peak height intensity value at each wavenumber position is calculated separately, and then the average or maximum value is taken as the representative intensity value of the characteristic peak.

[0067] The present invention, by selecting a local window near the wavenumber position of each characteristic peak and independently fitting a local baseline, can effectively eliminate the measurement error caused by the inconsistency in the baseline height on both sides of the characteristic peak, reducing the relative error of peak height measurement from 20-50% to below 5%. This improvement in accuracy is crucial for the detection of trace residues: when the residual signal is only 3 times higher than the noise signal-to-noise ratio, a 5% measurement error is still within an acceptable range, while a 20% error may lead to misjudgment of the residue type.

[0068] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for monitoring the quality of cleaning ophthalmic precision instruments, characterized in that, include: The surface of the cleaned instrument is scanned at multiple points using a Raman spectroscopy detector, and the Raman spectral signal of each measurement point is acquired. The Raman spectral signal is preprocessed to obtain a preprocessed spectral signal; The preprocessing includes: performing a first processing on the Raman spectral signal to remove the dark current background noise of the detector and obtain a first signal; performing a second processing on the first signal to remove random spikes generated by high-energy particle impacts and obtain a second signal; and performing a third processing on the second signal to remove the broadband fluorescence background generated by the metal substrate and obtain the spectral signal. The spectral signal is subjected to scattering correction to eliminate signal intensity fluctuations caused by the highly reflective curved surface of the instrument, thereby obtaining a corrected spectral signal; Raman characteristic peaks are extracted from the corrected spectral signal; the Raman characteristic peaks include protein characteristic peaks, lipid characteristic peaks and polysaccharide characteristic peaks. The residual signal-to-noise ratio at each measurement point is calculated based on the Raman characteristic peaks to monitor the cleaning quality.

2. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 1, characterized in that, Scattering correction of the spectral signal includes: The preprocessed spectral signal is denoted as matrix X, where the rows of matrix X correspond to each measurement point and the columns correspond to each wavenumber point. Calculate the mean spectrum of the spectra at each measurement point in the matrix X, and denote it as the reference spectrum; For each measurement point's spectrum xj, the reference spectrum is used as the independent variable, and the spectrum xj of that measurement point is used as the independent variable. j Perform linear regression fitting on the dependent variable to obtain the offset coefficient bj and slope coefficient mj for the measurement point; j is the index of the measurement point. The spectrum of the measurement point is corrected according to the following formula to obtain the corrected spectrum xj* of the measurement point: xj*=(xj-bj) / mj; The offset coefficient bj is used to correct the baseline offset caused by the change in the curvature of the instrument surface, and the slope coefficient mj is used to correct the overall intensity scaling of the spectrum caused by the difference in reflectivity of the instrument surface. By traversing all measurement points, the corrected spectral signal is obtained.

3. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 2, characterized in that, After obtaining the offset coefficient bj and slope coefficient mj of the measurement point, the following is also included: Based on the spatial coordinates of each measurement point on the instrument surface, it is determined whether the measurement point is located in the first region; the first region includes the cutting edge region, joint region, or tip region of the instrument. If so, then the offset coefficient bj and slope coefficient mj of the measurement point are locally corrected, including: Using the current measurement point as the center, extract several correction measurement points in its spatial domain that are not in the first region; Calculate the offset coefficient for each of the correction measurement points, and obtain the correction offset coefficient by weighted averaging of the offset coefficients of each correction measurement point. Calculate the slope coefficient for each of the corrected measurement points, and obtain the corrected slope coefficient by weighted averaging of the slope coefficients for each of the corrected measurement points. The scattering correction is performed on the spectrum of the current measurement point based on the correction offset coefficient and the correction slope coefficient.

4. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 3, characterized in that, The weight of the weighted average is inversely proportional to the spatial distance from the corrected measurement point to the current measurement point.

5. The method for monitoring the cleaning quality of ophthalmic precision instruments according to any one of claims 1-4, characterized in that, The first process includes: With the laser source off, the dark current spectrum was acquired with the same integration time as the detection and stored as a background reference signal; For each measurement point, the intensity value of the corresponding wavenumber point in the background reference signal is subtracted from the wavenumber point of the Raman spectral signal to obtain the first signal.

6. The method for monitoring the cleaning quality of ophthalmic precision instruments according to any one of claims 1-4, characterized in that, The second process includes: The spectral data of the first signal is scanned point by point, and the mean and standard deviation of each data point and its two adjacent data points are calculated. If the intensity value of the current data point exceeds the mean plus five times the standard deviation, it is marked as a candidate spike; For each candidate peak, its peak width feature value is calculated; the peak width feature value is the number of data points that the data point's intensity value spans when it drops to 50% of its peak height. If the peak width characteristic value is less than or equal to 2, the candidate peak is determined to be a cosmic ray peak; otherwise, it is determined to be a true Raman characteristic peak. The intensity value of the data point identified as the cosmic ray spike is replaced with the median value of its two adjacent data points to obtain the second signal.

7. The method for monitoring the cleaning quality of ophthalmic precision instruments according to any one of claims 1-4, characterized in that, The third processing step, which removes the broad-spectrum fluorescence background generated by the metal substrate using an adaptive baseline correction method, includes: The spectral data of the second signal is denoted as the original vector y, and the length of the original vector is n; n corresponds to the number of wavenumber points in the spectrum. Set the number of iterations, the smoothing parameter λ, and the convergence threshold ε; In the t-th iteration, the weight vector w(t) is calculated; the initial weight of each data point is 1. In the iteration, for data points whose intensity value in the spectrum is lower than the current fitting baseline, the data points whose intensity value in the spectrum is higher than the current fitting baseline are assigned weights. The fitted baseline z(t) for the t-th iteration is obtained by solving the following weighted penalized least squares problem: ； i is the index of the data point, with a value from 1 to n; y i z is the intensity value of the i-th data point in the original vector; i w represents the intensity value of the i-th data point in the fitted baseline. i (t) represents the weight of the i-th data point in the t-th iteration; Δ 2 It is a second-order difference operator; Calculate the difference between the current fitted baseline and the original vector. If the absolute value of the difference of the i-th data point is less than the threshold, reduce the weight of that data point in the next iteration so that the fitted baseline gradually approaches the low value region of the spectrum in subsequent iterations. The spectral data is obtained by subtracting the fitted baseline obtained from the final iteration from the original vector.

8. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 1, characterized in that, The corrected spectral signal covers continuous spectral data in the wavenumber range of 400~3200 cm⁻¹; extracting Raman characteristic peaks from the corrected spectral signal includes: reading the original intensity value at each wavenumber position from the corrected spectral signal according to a preset plurality of characteristic peak wavenumber positions, and using this as the Raman characteristic peak intensity value at that measurement point; the preset plurality of characteristic peak wavenumber positions include: The characteristic peak of the protein corresponds to the 1650~1660cm range. -1 and 1230~1270cm -1 ; The lipid characteristic peak corresponds to 2850 cm⁻¹ -1 2920cm -1 and 1745cm -1 ; The characteristic peak of polysaccharides corresponds to 480 cm⁻¹ -1 850~870cm -1 and 1120cm -1 .

9. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 8, characterized in that, Extracting Raman characteristic peaks from the corrected spectral signal further includes performing local baseline recalibration on the corrected spectral signal, which includes: For each preset characteristic peak wavenumber position, select 5 to 10 wavenumber points to the left and right as local windows; The local baseline of the local window is obtained based on linear fitting; The local baseline is removed from the original intensity value to obtain the locally corrected peak height intensity value; The intensity values ​​of Raman characteristic peaks are extracted from the locally corrected peak height intensity values.

10. The method for monitoring the cleaning quality of ophthalmic precision instruments according to claim 1, characterized in that, Calculating the residual signal-to-noise ratio at each measurement point based on the Raman characteristic peaks includes: For each measurement point, the ratios of the protein characteristic peak, lipid characteristic peak, and polysaccharide characteristic peak to their respective local noise standard deviations are calculated to obtain the residual signal-to-noise ratio for each residual type at the measurement point. The cleaning quality of each measurement point is obtained by comparing the signal-to-noise ratio of protein residue, lipid residue, and polysaccharide residue with their respective cleanliness thresholds.

Images