Biological microstructure multi-channel near-field data processing method and system for terahertz scanning near-field optical microscope

Abstract

The application relates to the technical field of biomedical micro-imaging and multi-dimensional data processing, in particular to a biological microstructure multi-channel near-field data processing method and system for a terahertz scanning near-field optical microscope, which comprises the following steps: acquiring multi-channel original data obtained by scanning biological microstructures by using the terahertz scanning near-field optical microscope, and independently and in parallel pre-processing the original data of each channel; stripping the original packaging structure of the surface topography data and the near-field optical data after pre-processing, performing analysis and conversion, obtaining a surface topography matrix and a near-field optical matrix with a unified format, and registering to the same global coordinate system; constructing the two-dimensional surface topography matrix into a three-dimensional elevation surface, performing feature mapping on the near-field optical matrix by using a color mapping algorithm, generating a two-dimensional feature map of the near-field optical matrix, and rendering the two-dimensional feature map to the three-dimensional elevation surface according to the same coordinates to obtain a three-dimensional superposition model, so that the surface topography features and the near-field optical features are synchronously and stereoscopically presented.

Description

Technical Field

[0001] This application relates to the field of biomedical microscopic imaging and multidimensional data processing technology, and in particular to a method and system for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy. Background Technology

[0002] Terahertz scattering scanning near-field optical microscopy (THz s-SNOM) combines the nanoscale spatial resolution of atomic force microscopy (AFM) with the high sensitivity of terahertz waves to the chemical components of biological tissues (such as hydration state and organic / inorganic ratio), showing great potential for non-destructive detection of biomacromolecules, cell structures, and biomineralized tissues.

[0003] In practical research, the THz s-SNOM system simultaneously generates multi-channel raw data, including surface morphology height (H), near-field multi-order amplitudes (A1, A2), and phases (P1, P2), when scanning biological microstructures. Processing this raw data faces the following challenges: 1) Data format closure and difficulty in multi-channel fusion: The raw data generated by commercial probe microscopes is usually in a proprietary closed format, which makes it inconvenient to directly extract the underlying physical numerical matrix, thus limiting the cross-channel joint analysis of morphology and optical dielectric signals.

[0004] 2) Spatial misalignment caused by scanning delay: Due to the unavoidable time difference between the probe physical scanning and the terahertz near-field signal integration and acquisition, the morphology channel and the near-field optical channel will have physical translation error in spatial coordinates.

[0005] 3) Complexity of biological sample surfaces leads to signal distortion: The surfaces of biological microstructures (such as extracellular matrix, mineralized tissue sections, etc.) generally have macroscopic tilt and microscopic roughness. These morphological features can seriously interfere with the extraction of near-field optical signals and produce morphological artifacts.

[0006] Currently, there is a lack of a standardized and systematic full-process data processing paradigm in the field, which makes it impossible to efficiently realize the transformation from raw closed data to the final high-precision multi-dimensional physical and chemical feature mapping. Summary of the Invention

[0007] The purpose of this application is to overcome the above-mentioned deficiencies of the prior art and thus provide a method and system for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy.

[0008] To address the aforementioned technical problems, the technical solution provided in this application offers a multi-channel near-field data processing method for biological microstructures using terahertz scanning near-field optical microscopy, comprising: Step 1: Acquire multi-channel raw data of biological microstructures obtained by terahertz scanning near-field optical microscopy, and perform independent parallel preprocessing on the raw data of each channel; the multi-channel raw data includes: surface morphology data and near-field optical data; Step 2: Remove the original encapsulation structure of the preprocessed surface topography data and near-field optical data, and perform analysis and transformation to obtain a surface topography matrix and near-field optical matrix with a unified format; Step 3: Register the surface topography matrix and the near-field light matrix to the same global coordinate system; Step 4: Construct a three-dimensional elevation surface from the two-dimensional surface topography matrix, use a color mapping algorithm to perform feature mapping on the near-field optical matrix to generate a two-dimensional feature map of the near-field optical matrix, and render the two-dimensional feature map onto the three-dimensional elevation surface according to the same coordinates to obtain a three-dimensional superimposed model, so as to simultaneously and stereoscopically present the surface topography features and near-field optical features.

[0009] As an improvement to the above method, the preprocessing in step 1 includes: performing surface fitting and baseline flattening preprocessing on the surface topography data in the multi-channel raw data to eliminate macroscopic tilt errors and obtain corrected surface topography data; and performing adaptive hierarchical spatial filtering preprocessing on the near-field optical data to suppress high-frequency scanning noise interference and obtain denoised near-field optical data.

[0010] As an improvement to the above method, the near-field optical data includes: first-order near-field amplitude data, second-order near-field amplitude data, first-order near-field phase data, and second-order near-field phase data; the adaptive hierarchical spatial filtering preprocessing of the near-field optical data specifically includes: for the second-order near-field amplitude data, a nonlinear median filter is used, with the filter window being an odd-numbered matrix; for the first-order near-field amplitude data, first-order near-field phase data, and second-order near-field phase data, a linear low-pass filter is used.

[0011] As an improvement to the above method, in step 2, the original packaging structure is automatically identified and stripped using Gwyddion software. The surface topography data and near-field optical data are analyzed pixel by pixel and converted into standardized ASCII data files to obtain a discretized surface topography matrix and near-field optical matrix with complete physical dimensions.

[0012] As an improvement to the above method, step 3 specifically includes: interactively extracting biological topological boundaries with significant features from the surface morphology matrix and the near-field optical matrix as spatial registration reference points for each matrix; establishing a global coordinate system based on the grid coordinates of the surface morphology matrix; calculating the two-dimensional spatial translation vector of the near-field optical matrix relative to the global coordinate system based on the spatial registration reference points; and performing coordinate translation transformation and interpolation processing on the near-field optical matrix using the two-dimensional spatial translation vector to enable pixel-level registration between the near-field optical matrix and the surface morphology matrix in the global coordinate system.

[0013] As an improvement to the above method, step 4 specifically includes: retaining the original physical dimensions of the surface topography matrix, performing extreme value normalization on the near-field optical matrix, and mapping the near-field optical matrix to the dimensionless [0, 1] interval; in the three-dimensional coordinate system, using the two-dimensional surface topography matrix as a spatial topological skeleton, constructing a continuous three-dimensional elevation surface corresponding to the surface topography matrix; using a pseudo-color transfer function and / or a custom periodic stripe grayscale function to perform feature mapping on the extreme value normalized near-field optical matrix, generating a two-dimensional feature map of the near-field optical matrix aligned with the physical scale of the extreme value normalized near-field optical matrix; rendering the two-dimensional feature map onto the three-dimensional elevation surface according to the same coordinates, and superimposing a reflection model to obtain a three-dimensional superimposed model, and synchronously and stereoscopically presenting the surface topography features and near-field optical features through the three-dimensional superimposed model.

[0014] As an improvement to the above method, the method further includes: performing feature profile analysis and probability distribution statistics based on a surface morphology matrix and a near-field light matrix with a unified format, so as to perform multidimensional feature quantitative analysis of biological microstructures.

[0015] To achieve another objective of the present invention, the present invention also provides a multi-channel near-field data processing system for biological microstructures for terahertz scanning near-field optical microscopy, comprising: The preprocessing module is used to acquire multi-channel raw data obtained by scanning biological microstructures using a terahertz scanning near-field optical microscope, and to perform independent and parallel preprocessing on the raw data of each channel; the multi-channel raw data includes: surface morphology data and near-field optical data; The format unification module is used to strip the original encapsulation structure of the preprocessed surface topography data and near-field optical data, and to parse and transform them to obtain a surface topography matrix and near-field optical matrix with a unified format. The registration module is used to register the surface topography matrix and the near-field light matrix to the same global coordinate system; and The model generation module is used to construct a three-dimensional elevation surface from a two-dimensional surface topography matrix. It uses a color mapping algorithm to perform feature mapping on the near-field optical matrix to generate a two-dimensional feature map of the near-field optical matrix. The two-dimensional feature map is then rendered onto the three-dimensional elevation surface according to the same coordinates to obtain a three-dimensional superimposed model, so as to simultaneously and stereoscopically present the surface topography features and near-field optical features.

[0016] The advantage of this application is that the method and system for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy provided in this application provide a standardized workflow framework for the analysis of multi-channel near-field data of biological microstructures, and give a complete processing paradigm from raw data to qualitative and quantitative visualization analysis, providing an objective and reusable standardized technical solution for the nanoscale pathological characterization of biological tissues. Attached Figure Description

[0017] Figure 1 The flowchart of the multi-channel near-field data processing method for biological microstructures for terahertz scanning near-field optical microscopy provided by the present invention; Figure 2 This is a schematic diagram illustrating the principle of multi-channel matrix registration based on feature anchor points. Figure 3 This is a sub-process for multi-dimensional parameter space reconstruction and quantitative visualization. Figure 4(a) is a two-dimensional mapping diagram of the original surface topography height matrix before baseline flattening in Embodiment 1 of the present invention; Figure 4(b) is a two-dimensional mapping diagram of the surface topography height matrix after the baseline is flattened in Embodiment 1 of the present invention; Figure 5(a) shows the fidelity of the two-dimensional near-field amplitude image before filtering and noise reduction in Embodiment 1 of the present invention; Figure 5(b) shows the fidelity of the two-dimensional near-field amplitude image after filtering and noise reduction in Embodiment 1 of the present invention; Figure 5(c) shows the effect of the three-dimensional morphology-near-field multi-channel model corresponding to Figure 5(a) before filtering; Figure 5(d) shows the effect of the three-dimensional topography-near field multi-channel model corresponding to Figure 5(b) after filtering; Figure 6(a) is a two-dimensional visual effect diagram of the near-field optical matrix feature mapping using a high-contrast pseudo-color transfer function in Embodiment 1 of the present invention. Figure 6(b) is a two-dimensional visual effect diagram of the feature mapping of the near-field optical matrix by the custom periodic stripe grayscale function in Embodiment 1 of the present invention; Figure 7 This is a diagram illustrating the morphology-near-field optical multi-channel three-dimensional spatial layering modeling effect in Embodiment 1 of the present invention. Figure 8(a) is a two-dimensional mapping diagram of surface topography height with feature extraction profile trajectory in Embodiment 1 of the present invention; Figure 8(b) is a two-dimensional mapping diagram of near-field amplitude with feature extraction profile trajectory in Embodiment 1 of the present invention; Figure 8(c) is a quantitative analysis diagram of the biaxial feature profile (Line Profile) spanning the target microstructure in Embodiment 1 of the present invention; Figure 9 This is a schematic diagram of a statistical visualization (half-violin plot) model based on a global probability distribution. Detailed Implementation

[0018] The technical solutions provided in this application are further illustrated below with reference to the embodiments.

[0019] The method provided in this application for the full-process analysis, registration, and visualization reconstruction of multi-channel near-field data of biological microstructures for terahertz scattering scanning near-field optical microscopy (THz s-SNOM) is as follows: Figure 1 As shown, the entire process includes the following steps: Step 1: Baseline calibration and preprocessing of raw data.

[0020] The raw multi-channel data files generated from THz s-SNOM scanning of biological microstructures were acquired. For the surface morphology height channel, surface fitting and baseline flattening were performed to eliminate macroscopic tilting and substrate undulation errors introduced during biological sample preparation. For the near-field optical channel (containing multiple amplitude and phase values), low-pass and band-pass spatial filtering operations were performed to remove high-frequency scanning noise. High-fidelity morphology and denoised near-field feature data were obtained.

[0021] Step 2: Numerical matrix extraction and format standardization.

[0022] Import the preprocessed multi-channel closed-format data into the cross-platform parsing module; remove the proprietary header and encapsulation shell of the probe microscope to obtain the morphology height ( ), near-field amplitude ( ) and near-field phase ( Heterogeneous channels such as ASCII and ASCII are parsed pixel by pixel and converted into independent standardized plain text ASCII data files, breaking through the limitations of the single software generally used for analysis in commercial equipment, and obtaining a complete discretized original numerical matrix.

[0023] Step 3: Registration based on feature anchor points.

[0024] The standardized multidimensional numerical matrix is ​​imported into the numerical computation platform; given the inherent time delay between the probe physical scan and the integration of the terahertz photoelectric signal, such as Figure 2As shown, biological topological boundaries (such as the edges of tubular structures) with significant features are interactively extracted from the topographic numerical matrix and the near-field optical numerical matrix as spatial anchor points; the two-dimensional spatial translation vector of the near-field optical matrix is ​​calculated based on the grid coordinates of the topographic matrix; by applying the coordinate translation transformation matrix, the spatial misalignment caused by scanning delay is eliminated in the underlying data dimension, and pixel-level registration of physical topography and local dielectric response is achieved.

[0025] Step 4: Multidimensional parametric space reconstruction and quantitative visualization.

[0026] Based on the registered unified spatial coordinate system, the topography height matrix and the near-field optical parameter matrix are fused in multiple dimensions; such as Figure 3 As shown, near-field dielectric parameters (amplitude, phase change, etc.) are mapped to a specific pseudo-color transfer function to generate a pseudo-color two-dimensional feature map aligned with a physical scale. Furthermore, for specific biological microstructure regions, features such as line profiles are extracted or three-dimensional morphology-dielectric superposition distribution models are constructed to quantitatively analyze and visualize the component heterogeneity and structural degradation state of biological tissues at the nanoscale.

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and are not intended to limit this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] Example 1 This embodiment provides a multi-channel image processing method for biological microstructures using terahertz scanning near-field optical microscopy. Due to the complex hierarchical structure and heterogeneous hydration state of biomineralized tissues (such as dentin), their nanoscale characterization is extremely challenging. Therefore, this embodiment specifically uses excised human dentin sections (including cross-sections of normal dentin and diseased carious dentin) as the target biological microstructure to verify and illustrate the specific execution process and beneficial effects of the multi-channel image processing method described in this invention.

[0029] Before implementing the data processing method of this invention, it is first necessary to acquire the raw multichannel scan data to be processed. Specifically, this embodiment employs a terahertz scattering scanning near-field optical microscope (THzs-SNOM) system operating at a frequency of 100 GHz. This system is equipped with a custom-designed platinum-coated probe with a tip curvature radius of approximately 100 nm. During the data acquisition scanning process, the probe periodically taps and oscillates at a set mechanical frequency (approximately 16.82 kHz), and a radio frequency source transmits terahertz waves to the probe tip via an antenna. The near-field interaction between the probe and the sample generates a scattered signal, which the system demodulates and extracts at the probe's higher harmonic frequencies using a lock-in amplifier.

[0030] Through the above scanning process, the system finally outputs raw multi-channel closed-format files containing spatial coordinate information. These files contain at least: a surface topography height matrix for characterizing the micro-undulations of the sample, and a multi-dimensional near-field optical matrix (specifically including first-order and second-order near-field amplitude matrices and near-field phase matrices) for characterizing the local dielectric response of the sample (related to the underlying chemical composition and hydration state).

[0031] Based on the acquired raw multi-channel closed data, combined with Figure 1 and Figure 2 As shown, perform the following complete workflow steps for image parsing, registration, and reconstruction: Step 1: Baseline calibration and preprocessing of raw data In conjunction with the embodiments, this step specifically includes independent parallel processing of the morphology channel and the near-field optical channel, aiming to eliminate systematic errors introduced by the environment and physical scanning equipment, and extract data that more closely resembles the true morphology and local dielectric properties. The specific execution logic and core parameter configuration of this step are as follows: (1) Flattening of topography baseline based on first-order polynomial fitting The original surface morphology height matrix inevitably incorporates the macroscopic substrate tilt caused by the biological slicing process (such as cutting and polishing). Without processing, the absolute height difference of this macroscopic tilt would far exceed the nanoscale dimensions of real biological microstructures, leading to extreme distortion during subsequent multidimensional feature fusion. In this step, a reference plane fitting and subtraction algorithm is performed on the extracted original surface height channels. Regarding the selection of core parameters in this process, to avoid artificial topological distortion of the real microstructure of biological tissues (such as the edge structure of dentinal tubules) caused by high-order mathematical fitting, the fitting order is chosen to be first-order, calculating only one optimal two-dimensional tilt plane as the substrate. Simultaneously, a global thresholdless strategy should be adopted for the fitting sampling parameters, without discarding any extreme highs or lows, ensuring that all pixel coordinates within the field of view participate in the calculation of the first-order reference plane with equal weight. Subsequently, the original height matrix is ​​subtracted from the reference plane pixel by pixel to accurately eliminate the macroscopic tilt of the sample, while preserving the authenticity of the nanoscale dentin surface morphology without loss. The effect comparison before and after the surface morphology height matrix baseline is flattened is shown in Figure 4(a) and Figure 4(b). Figure 4(a) is the two-dimensional mapping of the original height matrix, and Figure 4(b) is the two-dimensional mapping of the height matrix after the first-order curved surface fitting and flattening.

[0032] (2) Adaptive hierarchical spatial filtering based on near-field features of signal dimension The true terahertz near-field scattering signal generated by the probe tip is extremely weak and easily affected by mechanical vibration and background scattering. The acquired multidimensional near-field optical matrix (including first / second-order near-field amplitude and near-field phase) requires rigorous filtering and noise reduction. The innovation of this invention lies in abandoning a single global filtering strategy and proposing a hierarchical processing model that adaptively configures the filter type and core parameters based on the order and physical properties of the near-field signal. The specific rules are as follows: For second-order near-field amplitude images: the absolute intensity of the second-order near-field scattered signal drops sharply, and the image is highly susceptible to randomly distributed extreme outliers (i.e., "salt-and-pepper noise"). Simultaneously, amplitude data is more sensitive to scattering clutter caused by minute surface roughness. Therefore, linear filtering, which leads to edge blurring, must be excluded from such data, and nonlinear median filtering should be preferred instead. In terms of core parameter settings, the window order for the nonlinear filter should be set to [value missing]. or larger odd matrices (such as This parameter configuration directly replaces the contaminated center pixel with the median of the local neighborhood pixels, which can strip away discrete "salt and pepper noise" and preserve the "step-like" physical abrupt change edges of the near-field signal at the junction of different components (such as peritubular and intertubular dentin) within the biological tissue.

[0033] For first-order near-field amplitude and first- and second-order near-field phase images: the signal intensity of the first-order harmonic signal is relatively high, and the phase signal is highly sensitive to boundary abrupt changes in the intrinsic dielectric constant of the material (compositional differences). To filter out conventional high-frequency mechanical scanning noise while preventing spatial dispersion at physical boundaries, linear low-pass filtering (such as Gaussian filtering) is preferred for this type of data, with a small neighborhood window set on the core parameter (convolution kernel size). (pixels). This lightweight filtering parameter can smooth high-frequency noise while maintaining the clarity of the macroscopic component distribution contours to the maximum extent.

[0034] The hierarchical filtering mechanism follows the principle that "the denoising parameter intensity of higher-order signals is greater than that of lower-order signals, and the nonlinear fidelity requirement of amplitude signals is stricter than that of phase signals." This step provides reliable input for the subsequent generation of artifact-free, high-fidelity underlying numerical matrices. The comparison of the effects before and after near-field optical feature space filtering is shown in Figures 5(a)-5(d). Figures 5(a) and 5(b) show the fidelity comparison of the two-dimensional near-field amplitude image before and after filtering and denoising (Figure 5(a) is before filtering, and Figure 5(b) is after filtering). Figures 5(c) and 5(d) show the comparison of the presentation effects of the three-dimensional morphology-near-field multi-channel models corresponding to Figures 5(a) and 5(b) before and after filtering, respectively.

[0035] Step 2: Numerical Matrix Extraction and Format Standardization Raw data generated by commercial scanning probe microscopy (SPM) software typically uses proprietary, closed encapsulation formats (such as .flt or vendor-specific binary formats), storing topographic, amplitude, and phase data in proprietary binary files. This data format hinders numerical computation and spatial registration across software platforms. Therefore, this step utilizes the open-source scanning probe microscopy data processing framework Gwyddion (an open-source software for SPM data visualization) to perform the stripping and standardization of the underlying data. The specific implementation logic is as follows: (1) Low-level parsing of closed formats and file header stripping Import the multichannel proprietary format file, preprocessed in Step 1, into Gwyddion. Gwyddion automatically identifies and removes proprietary headers, redundant hardware calibration information, and color mapping shells added by commercial devices. This operation restores the previously black-box data file to reflect the probe-sample interaction. Two-dimensional discrete numerical matrix.

[0036] (2) Independent mapping and lossless transcoding of multidimensional physical channels After the packaging shell is removed, the system can independently determine the surface height matrix in a unified physical coordinate system. Near-field amplitude matrix With near-field phase matrix Subsequently, cross-platform standardized transcoding is performed on each of the separated independent matrices. At this point, directly exporting to common image formats (such as JPEG, PNG, TIFF) may introduce interpolation errors. Therefore, it is necessary to specify the output format, losslessly converting each pixel coordinate of each dimension matrix and its corresponding physical dimensions (such as nanometer values ​​for height and millivolt values ​​for amplitude) into a standardized ASCII data file (saved in .txt format). The final output ASCII matrix is ​​no longer a visual image with precision limitations, but contains floating-point data of physical dimensions. This provides a precise and standardized underlying data foundation for subsequent (step three) subpixel-level mathematical calculations and coordinate translation registration of dielectric feature anchor points in the numerical computing platform MATLAB.

[0037] (3) Automated batch processing based on Gwyddion's built-in Python environment (Pygwy) When dealing with large-scale multi-channel datasets (such as massive amounts of data generated from continuous multi-region scanning of multiple biological sample slices), relying on traditional graphical user interfaces (GUIs) to perform the aforementioned file extraction and transcoding operations one by one can lead to severe efficiency bottlenecks. To address this, automated batch processing can be built using the Python scripting interface (Pygwy) provided by the Gwyddion framework. Batch processing eliminates the time-consuming bottleneck of human-computer interaction, prevents data omissions and channel confusion that may result from manual operation, and ensures the scalability and high repeatability of the underlying numerical matrix extraction process. By writing and deploying Python scripts that call the core objects and methods of the underlying gwy module in the system, the algorithm can silently traverse the specified dataset directory. Without manual intervention, the script automatically extracts the corresponding channel identifiers (Data IDs) from each encapsulated file using a loop structure and calls the corresponding underlying output functions (export_ascii and its variants) to batch perform file header stripping, channel separation, and lossless export of massive amounts of multi-channel data into plain text format.

[0038] Step 3: Registration based on feature anchor points After data standardization and transcoding, an unavoidable millisecond-level time difference exists between the physical mechanical scanning of the probe cantilever (acquiring morphology) and the integration and acquisition of the terahertz photoelectric signal (acquiring near-field optical information). This results in a two-dimensional relative translational misalignment of tens to hundreds of nanometers between the morphology height matrix and the near-field optical matrix in physical space coordinates. If this misalignment is not eliminated, subsequent component mapping will be ambiguous. Combined with... Figure 2The diagram shown illustrates the registration principle. This step imports the standardized plain text ASCII matrices for each channel into the numerical computing platform MATLAB, where a matrix-level coordinate transformation is performed. The specific registration algorithm and implementation logic are as follows: (1) Interactive extraction of multi-channel topological feature anchor points The height matrix in an unregistered state With near-field amplitude matrix In this process, boundaries with significant physical or chemical abrupt changes are identified as spatial registration anchor points. In this embodiment, for dentin samples, the luminal margins of dentinal tubules are preferably extracted as feature objects. First, in the height matrix... In this study, the physical topological boundary was located by utilizing the abrupt drop in surface height extreme values ​​(indicating the entry of pericanal dentin into the hollow cavity), and the spatial two-dimensional grid coordinates of the boundary feature points were recorded and set as reference anchor points. Subsequently, in the near-field amplitude matrix Within the corresponding neighborhood, the near-field optical boundary is identified by the step-like attenuation of the dielectric signal intensity (due to the lack of a highly polarizable hydrated collagen network and minerals within the lumen), and the coordinates of this feature point are recorded and designated as the misalignment feature point. .

[0039] (2) Spatial translation vector calculation and sub-pixel level interpolation resampling Based on the extracted feature coordinate pairs, the two-dimensional coordinate translation vector of the near-field optical matrix relative to the height reference matrix is ​​calculated. ,in and (Offset under normal circumstances) Since the actual physical scan delay, when converted to a discrete pixel grid, is often not an integer number of pixels—a phenomenon particularly noticeable in scans with large strides—this invention does not employ a simple overall matrix translation. Instead, it introduces a sub-pixel-level resampling algorithm to achieve high-resolution matrix resampling. A global coordinate system is established using the spatial grid as the absolute reference, and the calculated translation vectors are used. A translational affine transformation is performed on the near-field amplitude and phase matrices. During the transformation, bilinear interpolation or bicubic spline interpolation algorithms are used to recalculate and assign near-field dielectric values ​​at non-integer coordinate points.

[0040] (3) Generation of pixel-level absolute registration matrix After translation transformation and sub-pixel interpolation resampling, a new corrected near-field matrix is ​​generated. and At this point, the height matrix any coordinate point on It not only accurately corresponds to its physical, true nanometer-level elevation, but also corresponds to the same coordinates in the correction matrix. The local terahertz dielectric response at the location was observed. Multi-channel pixel-level registration of the physical morphology and near-field optical signal was achieved.

[0041] Step 4: Multidimensional Parametric Space Reconstruction and Quantitative Visualization After registering the morphological height matrix and the near-field optical matrix in step three, the underlying data matrix in a unified coordinate system is imported into the numerical computing platform. This step aims to transform abstract terahertz dielectric data into atlases that intuitively reveal the microstructure and pathological evolution characteristics of biological tissues through color mapping, 3D modeling, and multi-dimensional quantitative analysis algorithms. The specific implementation logic is as follows: (1) Feature mapping and high-contrast reconstruction For matrices with different physical properties, this step employs differentiated dimensionality and mapping strategies. The physical variations in biological tissues are a core objective indicator for assessing their structural integrity. For the surface height matrix, its original physical dimensions (micrometers or nanometers) are preserved. For the near-field optical matrix (containing amplitudes and phases of various orders), baseline drift in absolute signal intensity (millivolts) occurs due to probe wear, RF power perturbations, and environmental factors in different batches of scanning experiments. Therefore, extreme value normalization must be performed only on the near-field optical matrix, uniformly mapping it to the dimensionless [0, 1] interval.

[0042] After completing the differential dimension processing, specific color mapping algorithms are configured for each matrix. For the normalized near-field matrix, a high-contrast pseudo-color transfer function (generally a thermodynamic color scale, Hot colormap) and a custom periodic fringe grayscale function are applied. These mapping algorithms can transform terahertz dielectric differences into significant visual abrupt changes in two-dimensional space. Figure 6(a) shows the two-dimensional visual effect of feature mapping the near-field optical matrix using the high-contrast pseudo-color transfer function, and Figure 6(b) shows the two-dimensional visual effect of feature mapping the near-field optical matrix using the custom periodic fringe grayscale function.

[0043] (2) Morphology-Near-field optical multi-channel three-dimensional spatial overlay drawing To provide a three-dimensional representation of the complex heterogeneous component distribution on the surface of biological microstructures, this invention introduces a morphology-dielectric multi-channel 3D spatial rendering algorithm. Specifically, the algorithm first extracts the two-dimensional surface height matrix. This is used as a spatial topological skeleton to construct a continuous physical elevation surface in a three-dimensional coordinate system. Subsequently, the near-field amplitude matrix under the same spatial coordinate system is extracted. The numerical values ​​are converted into pseudo-color texture maps, which are then directly rendered and applied to the aforementioned 3D elevation surface. By overlaying lighting and material reflection models such as Glaude shading, this 3D layered modeling process successfully achieves the simultaneous three-dimensional presentation of surface physical morphology undulations and internal deep dielectric properties / chemical composition within the same microscopic region. Figure 7 The rendering effect of multi-channel 3D spatial overlay modeling of morphology and near-field optics is shown. The surface height matrix is ​​used as the 3D topological skeleton, and multi-order near-field optical features (such as first-order and second-order amplitudes) are rendered in a stereoscopic synchronous manner as independent floating layers and texture mapping.

[0044] (3) Quantitative analysis of biomedical multidimensional characteristics In the field of biomedical microscopic imaging, to eliminate the subjectivity of purely visual observation, dimensionality reduction profile analysis and global nonparametric statistics are performed based on the aforementioned reconstruction matrix to output objective quantitative indicators: Line Profile Analysis: On the reconstructed two-dimensional map, a one-dimensional discrete data series is extracted along a specific spatial vector direction spanning the target microstructure (such as dentinal tubules). Surface height curves and near-field amplitude curves are simultaneously generated under the same physical distance abscissa. This dimensionality reduction characterization can quantitatively confirm the strong correlation between the physical state and dielectric response of a specific physiological structure with sub-micron precision. Figure 8(a) shows the profile extraction trajectory across the selected target microstructure (dentinal tubule) on the surface topography height map. Figure 8(b) shows the profile extraction trajectory synchronized to a near-field amplitude map strictly aligned with the height map. Figure 8(c) shows the quantitative analysis diagram of the biaxial line profile across the target microstructure, i.e., the comparison of the synchronous extraction and strong spatial correlation between the surface height curve (left axis) and the near-field amplitude curve (right axis) under a unified physical distance abscissa.

[0045] Global probability distribution statistical analysis: For the massive matrix of pixels within the field of view, a kernel density estimation algorithm (KDE) is introduced to calculate the continuous probability density distribution of the dielectric signal, and a half-violin diagram statistical model is generated by fusing quartile box plots. This statistical analysis can quantitatively reveal changes in the median signal and changes in local dielectric heterogeneity (data dispersion) caused by pathological processes from a global macroscopic perspective. Figure 9 A schematic diagram of a statistical visualization (half-violin plot) model based on a global probability distribution is shown.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy, comprising: Step 1: Acquire multi-channel raw data of biological microstructures obtained by terahertz scanning near-field optical microscopy, and perform independent parallel preprocessing on the raw data of each channel. The multi-channel raw data includes: surface topography data and near-field optical data; Step 2: Remove the original encapsulation structure of the preprocessed surface topography data and near-field optical data, and perform analysis and transformation to obtain a surface topography matrix and near-field optical matrix with a unified format; Step 3: Register the surface topography matrix and near-field light matrix with a uniform format to the same global coordinate system; Step 4: Construct a three-dimensional elevation surface from the two-dimensional surface topography matrix, use a color mapping algorithm to perform feature mapping on the near-field optical matrix to generate a two-dimensional feature map of the near-field optical matrix, and render the two-dimensional feature map onto the three-dimensional elevation surface according to the same coordinates to obtain a three-dimensional superimposed model, so as to simultaneously and stereoscopically present the surface topography features and near-field optical features.

2. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 1, characterized in that, The preprocessing in step 1 includes: performing surface fitting and baseline flattening preprocessing on the surface topography data in the multi-channel raw data to eliminate macroscopic tilt errors and obtain corrected surface topography data; and performing adaptive hierarchical spatial filtering preprocessing on the near-field optical data to suppress high-frequency scanning noise interference and obtain denoised near-field optical data.

3. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 2, characterized in that, The near-field optical data includes: first-order near-field amplitude data, second-order near-field amplitude data, first-order near-field phase data, and second-order near-field phase data; the adaptive hierarchical spatial filtering preprocessing of the near-field optical data specifically includes: for the second-order near-field amplitude data, a nonlinear median filter is used, with the filter window being an odd-numbered matrix; for the first-order near-field amplitude data, first-order near-field phase data, and second-order near-field phase data, a linear low-pass filter is used.

4. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 1, characterized in that, In step 2, the original packaging structure is automatically identified and stripped using Gwyddion software. The surface topography data and near-field optical data are analyzed pixel by pixel and converted into standardized ASCII data files to obtain a discretized surface topography matrix and near-field optical matrix with complete physical dimensions.

5. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 1, characterized in that, Step 3 specifically includes: In the surface morphology matrix and near-field light matrix, biological topological boundaries with significant features are interactively extracted as spatial registration reference points for each matrix; A global coordinate system is established based on the grid coordinates of the surface topography matrix; Based on the spatial registration reference point, the two-dimensional spatial translation vector of the near-field optical matrix relative to the global coordinate system is calculated. The two-dimensional spatial translation vector is used to perform coordinate translation transformation and interpolation on the near-field optical matrix, so that the near-field optical matrix is ​​registered with the surface topography matrix in the global coordinate system at the pixel level.

6. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 1, characterized in that, Step 4 specifically includes: Preserving the original physical dimensions of the surface topography matrix, extreme value normalization is performed on the near-field optical matrix, mapping the near-field optical matrix to the dimensionless interval [0, 1]. In a three-dimensional coordinate system, the two-dimensional surface topology matrix is ​​used as a spatial topological skeleton to construct a continuous three-dimensional elevation surface corresponding to the surface topology matrix. The near-field optical matrix after extreme value normalization is characterized by using a pseudo-color transfer function and / or a custom periodic stripe grayscale function to generate a two-dimensional feature map of the near-field optical matrix that is aligned with the physical scale of the near-field optical matrix after extreme value normalization. The two-dimensional feature map is rendered onto a three-dimensional elevation surface with the same coordinates, and a reflection model is superimposed to obtain a three-dimensional superimposed model. The surface morphology features and near-field optical features are presented synchronously and three-dimensionally through the three-dimensional superimposed model.

7. The method for processing multi-channel near-field data of biological microstructures for terahertz scanning near-field optical microscopy according to claim 1, characterized in that, Also includes: Feature profile analysis and probability distribution statistics are performed based on surface morphology matrices and near-field light matrices with a unified format to conduct multidimensional quantitative analysis of biological microstructures.

8. A multi-channel near-field data processing system for biological microstructures used in terahertz scanning near-field optical microscopy, comprising: The preprocessing module is used to acquire multi-channel raw data obtained from scanning biological microstructures using a terahertz scanning near-field optical microscope, and to perform independent and parallel preprocessing on the raw data of each channel. The multi-channel raw data includes: surface topography data and near-field optical data; The format unification module is used to strip the original encapsulation structure of the preprocessed surface topography data and near-field optical data, and to parse and transform them to obtain a surface topography matrix and near-field optical matrix with a unified format. The registration module is used to register surface topography matrices and near-field light matrices with a uniform format to the same global coordinate system; and The model generation module is used to construct a three-dimensional elevation surface from a two-dimensional surface topography matrix. It uses a color mapping algorithm to perform feature mapping on the near-field optical matrix to generate a two-dimensional feature map of the near-field optical matrix. The two-dimensional feature map is then rendered onto the three-dimensional elevation surface according to the same coordinates to obtain a three-dimensional superimposed model, so as to simultaneously and stereoscopically present the surface topography features and near-field optical features.

Images