A method for in vivo detection of malignant tumors based on laser raman

Abstract

This invention belongs to the field of biomedical optics technology, specifically relating to an in vivo detection method for malignant tumors based on laser Raman spectroscopy. The method involves real-time acquisition of laser Raman spectral data of the target detection area, in-situ OCT tissue depth and scattering coefficient data, and real-time intraoperative imaging data during surgery. Tumor characteristic Raman peak datasets are extracted, and the relative intensity ratios of these peaks are calculated to establish a precise point-to-point mapping between the Raman characteristic data and the spatial location of the tissue. The corrected Raman spectral characteristic data are compared with standard Raman characteristic spectra of malignant tumors. A quantitative grading model is constructed based on the relative intensity ratios of the characteristic peaks, and the microscopic invasion boundary of the tumor tissue is delineated according to spatial coordinates. Real-time visual detection results and diagnostic prompts are generated and synchronously transmitted to an intraoperative navigation system to assist in surgical resection. This solves the problem of insufficient accuracy in spectral feature extraction in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of biomedical optics technology, specifically relating to an in vivo detection method for malignant tumors based on laser Raman spectroscopy. Background Technology

[0002] Early and accurate detection of malignant tumors and real-time intraoperative monitoring are crucial for improving patient prognosis and treatment outcomes. Laser Raman spectroscopy, a non-invasive detection method based on molecular vibrational properties, plays an indispensable role in in vivo detection of malignant tumors. This technology utilizes near-infrared laser irradiation of living tissue, capturing the Raman shifts of different biomolecules to form a unique molecular fingerprint. This allows for precise differentiation of the biochemical composition differences between malignant tumor tissue and normal or benign lesion tissue, such as the content and structural changes of molecules like proteins, lipids, and amino acids. It eliminates the need for tissue sectioning, staining, or injection of exogenous contrast agents, enabling real-time, in-situ detection of tumor lesions. Combining the advantages of non-ionizing radiation, high chemical specificity, and sub-millimeter-level detection accuracy, it not only assists in the screening and diagnosis of early malignant tumors but can also be applied to intraoperative tumor margin assessment, helping doctors accurately identify residual cancer cells, reducing the risk of secondary surgery. Furthermore, it enables dynamic monitoring of tumor development, providing reliable molecular-level evidence for the formulation and optimization of clinical treatment plans.

[0003] Current in vivo malignant tumor detection technologies have many shortcomings, making it difficult to meet the needs of precise intraoperative diagnosis and treatment: First, intraoperative detection often acquires single-type data independently, failing to achieve simultaneous acquisition of laser Raman spectroscopy data, in-situ OCT tissue depth and scattering coefficient data, and real-time intraoperative imaging data, thus hindering the collaborative application of multi-source data; Second, the processing of laser Raman spectroscopy data is limited to simple feature extraction and qualitative differentiation of benign and malignant tumors, without combining it with real-time intraoperative imaging data to complete spatial coordinate registration, making it impossible to establish a precise point-to-point mapping relationship between Raman feature data and tissue spatial location; Third, it does not combine in-situ OCT tissue depth and scattering coefficient data for depth attenuation adaptive compensation correction of Raman spectra, resulting in insufficient accuracy of spectral feature extraction and affecting the accuracy of benign and malignant tumor differentiation; Fourth, it cannot construct a quantitative grading model based on the relative intensity ratio of tumor characteristic Raman peaks, making it difficult to grade and quantify the degree of tumor malignancy, and it cannot delineate the microscopic invasion boundary of tumor tissue based on spatial coordinates, easily leading to incomplete surgical resection; Fifth, the detection results lack real-time visualization and cannot be synchronously transmitted to the intraoperative navigation system, making it difficult to effectively assist surgical resection operations and failing to meet the actual needs of precise intraoperative diagnosis and treatment. Summary of the Invention

[0004] Therefore, it is necessary to provide an in vivo detection method for malignant tumors based on laser Raman spectroscopy to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for in vivo detection of malignant tumors based on laser Raman spectroscopy, including:

[0006] Intraoperative real-time acquisition of laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data;

[0007] The laser Raman spectroscopy data is processed to extract the tumor characteristic Raman peak dataset, calculate the relative intensity ratio of the tumor characteristic Raman peaks, and simultaneously combine the intraoperative real-time image data to complete spatial coordinate registration, establishing a point-to-point precise mapping relationship between Raman characteristic data and tissue spatial location;

[0008] Based on the spatial mapping relationship and combined with the in-situ OCT tissue depth and scattering coefficient data, the Raman spectrum is subjected to depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign or malignant attributes of the target area tissue and obtain the tissue benign or malignant discrimination result. A quantitative grading model is constructed based on the relative intensity ratio of the feature peaks. The intensity ratio of the feature peaks of benign and normal tissues is used as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy. The microscopic invasion boundary of tumor tissue is delineated according to the spatial coordinates.

[0009] Based on the results of tissue benign or malignant differentiation, tumor malignancy degree and microscopic invasion boundary, real-time intraoperative visual detection results and diagnostic prompts are generated and synchronously transmitted to the intraoperative navigation system to assist in surgical resection.

[0010] In one embodiment, the intraoperative real-time acquisition of laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data includes:

[0011] Construct a multimodal data synchronous acquisition and timing calibration architecture;

[0012] Based on the aforementioned multimodal data synchronous acquisition and timing calibration architecture, a unified hardware clock source is used to trigger laser Raman spectroscopy acquisition, OCT tomography and intraoperative image capture, eliminating the time delay deviation of multi-device acquisition and obtaining timing-aligned multimodal raw data.

[0013] Motion artifact filtering is applied to the original multimodal data to remove jitter interference caused by breathing, heartbeat, and instrument operation, thereby completing data noise reduction and format standardization, and outputting a spatiotemporally synchronized multimodal data volume of the target detection area.

[0014] In one embodiment, processing the laser Raman spectroscopy data to extract a dataset of tumor characteristic Raman peaks includes:

[0015] Construct a fluorescence background suppression and denoising model;

[0016] Based on the fluorescence background suppression and denoising model, and constrained by the characteristic shift range in the standard Raman feature map, wavelet decomposition and polynomial baseline fitting are performed on the original spectral data to remove the autofluorescence background in in vivo tissues and obtain the effective signal components.

[0017] Threshold filtering is applied to the effective signal components for noise reduction, and curve fitting is performed on the Raman peaks within the wavenumber range to screen and extract the tumor characteristic Raman peak dataset.

[0018] In one embodiment, the step of combining the intraoperative real-time image data to complete spatial coordinate registration and establish a point-to-point precise mapping relationship between Raman feature data and tissue spatial location includes:

[0019] Constructing an intraoperative image-Raman spectrum spatial registration model;

[0020] Based on the intraoperative image-Raman spectroscopy spatial registration model, the spatial coordinates of the tissue contour feature points and the laser Raman detection spot in the real-time intraoperative image are extracted. The soft tissue deformation is compensated by combining the tissue depth parameters calculated by OCT, and the registration transformation between image space and physical space is completed.

[0021] Based on the coordinate mapping parameters after registration transformation, a mapping relationship is established between spectral detection sites, image pixel coordinates, and the three-dimensional position of in vivo tissue.

[0022] In one embodiment, the depth attenuation adaptive compensation correction of the Raman spectrum involves comparing the corrected Raman spectral feature data with the standard Raman feature map of malignant tumors to determine the benign or malignant nature of the target region tissue, thereby obtaining the tissue benign or malignant determination result, including:

[0023] By combining in situ OCT tissue depth and scattering coefficient data, the relative intensity ratio of tumor characteristic Raman peaks is compensated by photon transmission attenuation inverse operation to eliminate the influence of tissue depth differences on Raman signal intensity, and the biochemical component feature matrix after depth normalization correction is obtained.

[0024] The feature data of each detection site in the biochemical component feature matrix are compared point by point with the pre-established standard Raman feature map of malignant tumors, and the feature matching degree of each detection site is calculated.

[0025] Based on the feature matching degree, a benign or malignant probability distribution map of the target region tissue is generated, and a binary tissue benign or malignant discrimination result is output by segmentation through a preset probability threshold.

[0026] In one embodiment, the step of constructing a quantitative grading model based on the relative intensity ratio of the characteristic peaks, using the intensity ratio of characteristic peaks of benign and normal tissues as a quantitative benchmark threshold, grading and quantifying the degree of tumor malignancy, and delineating the microscopic invasion boundary of tumor tissue according to the spatial coordinates, includes:

[0027] A quantitative grading model was established based on the characteristic peak intensity ratio of benign and normal tissue samples, using at least one combination of the characteristic peak intensity ratio of nucleic acid to lipid, the characteristic peak intensity ratio of nucleic acid to protein, or the characteristic peak intensity ratio of lipid to protein as input parameters.

[0028] According to the quantitative grading model, the deviation of the characteristic peak intensity ratio of each spatial coordinate point in the target detection area from the quantitative benchmark threshold is calculated. The deviation is divided into multiple numerical intervals, which correspond to the grading standards of normal tissue, precancerous lesions, low-grade malignancy, intermediate-grade malignancy and high-grade malignancy, respectively, to obtain the tumor malignancy grading results of each coordinate point.

[0029] Based on the tumor malignancy grading results at each coordinate point, and combined with the point-to-point precise mapping relationship between the Raman feature data and the spatial location of the tissue, the critical position where the malignancy grading changes abruptly in space is identified. A three-dimensional closed contour is generated by the isosurface extraction algorithm to delineate the microscopic invasion boundary between tumor tissue and normal tissue.

[0030] In one embodiment, the generation of real-time intraoperative visual detection results and diagnostic prompts, and their simultaneous transmission to the intraoperative navigation system to assist surgical resection, includes:

[0031] The results of tissue benign and malignant discrimination, tumor malignancy grade, and microscopic invasion boundary contour data are fused and superimposed with intraoperative real-time images to generate a distribution map of tumor benign and malignant attributes, a heat map of malignancy grade, and boundary contour lines displayed in pseudo-color image form as a visual detection result. At the same time, combined with the preset surgical safety margin distance parameter, diagnostic and treatment prompts are generated, including the suggested boundary of the surgical resection range, the location of residual lesions that need to be focused on, and the safety margin distance prompt.

[0032] The visualized detection results and the diagnostic prompts are transmitted to the intraoperative navigation system via a standard medical digital imaging and communication protocol interface and displayed in real time on the navigation system's monitor to assist the surgeon in making decisions and executing the surgical resection.

[0033] Secondly, this application also provides an in vivo detection system for malignant tumors based on laser Raman spectroscopy, comprising:

[0034] The data acquisition module is used to acquire laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time image data in real time during the operation;

[0035] The spectral denoising and spatial registration module is used to process the laser Raman spectral data, extract the tumor characteristic Raman peak dataset, calculate the relative intensity ratio of the tumor characteristic Raman peaks, and combine the intraoperative real-time image data to complete spatial coordinate registration, establishing a point-to-point accurate mapping relationship between Raman characteristic data and tissue spatial location.

[0036] The benign / malignant differentiation and boundary delineation module is used to perform depth attenuation adaptive compensation correction on the Raman spectrum based on the spatial mapping relationship and the in-situ OCT tissue depth and scattering coefficient data. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign / malignant attributes of the target area tissue and obtain the tissue benign / malignant differentiation result. A quantitative grading model is constructed based on the relative intensity ratio of the feature peaks. The intensity ratio of the feature peaks of benign and normal tissues is used as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy. The microscopic invasion boundary of the tumor tissue is delineated according to the spatial coordinates.

[0037] The intraoperative visualization module generates real-time intraoperative visualization detection results and diagnostic prompts based on the tissue benignity / malignancy determination results, tumor malignancy degree, and microscopic invasion boundary, and transmits them synchronously to the intraoperative navigation system to assist in surgical resection.

[0038] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a laser Raman-based in vivo detection method for malignant tumors as described in the above embodiments.

[0039] A fourth aspect of this application provides a computer program product, including a computer program or instructions, for implementing a laser Raman-based in vivo detection method for malignant tumors as described in the above embodiments.

[0040] Therefore, this application includes the following beneficial effects: By simultaneously fusing intraoperative in vivo laser Raman spectroscopy, in-situ OCT tissue depth scattering parameters, and real-time image multimodal data, and relying on point-to-point precise spatial coordinate registration to establish an accurate mapping between Raman features and tissue location, and combining tissue depth information to complete Raman spectral depth attenuation adaptive compensation correction, the interference of spectral signal distortion caused by tissue depth is effectively eliminated, significantly improving the accuracy of tumor characteristic peak detection; it can perform rapid identification of benign and malignant tissues, grading and quantifying the degree of tumor malignancy, and high-precision delineation of microscopic invasion boundaries non-invasively in vivo during surgery, solving the technical pain points of traditional detection in vitro lag, coarse definition of invasion range, and inability to locate in real time. Finally, it generates visualized detection results in real time and connects to the intraoperative navigation system, which can accurately guide the surgeon to completely remove the tumor lesion and preserve normal tissue to the maximum extent, significantly improving the accuracy of tumor surgical resection, reducing residual lesions and the risk of postoperative recurrence, and demonstrating outstanding clinical real-time diagnostic and treatment application value.

[0041] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0042] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0043] Figure 1 This is a flowchart of a method for in vivo detection of malignant tumors based on laser Raman spectroscopy, according to an embodiment of this application.

[0044] Figure 2 This is a schematic diagram of an in vivo malignant tumor detection method based on laser Raman spectroscopy according to an embodiment of this application;

[0045] Figure 3 This is a schematic diagram of the structure of an in vivo malignant tumor detection system based on laser Raman spectroscopy, according to an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] It should be noted that the term "comprising" and any variations thereof, as used in this application, are intended to cover non-exclusive inclusion. The term "multiple" as used in this application refers to two or more. The term "and / or" as used in this application refers to one of the solutions, or any combination of multiple solutions.

[0048] The following describes an embodiment of a laser Raman-based in vivo detection method for malignant tumors according to the present application, with reference to the accompanying drawings. Addressing the issue of insufficient accuracy in spectral feature extraction mentioned in the background section, this application provides a laser Raman-based in vivo detection method for malignant tumors. This method simultaneously fuses intraoperative in vivo laser Raman spectra, in-situ OCT tissue depth scattering parameters, and real-time multimodal image data. It establishes a precise mapping between Raman features and tissue location based on precise point-to-point spatial coordinate registration. Combined with tissue depth information, it performs adaptive compensation correction for Raman spectral depth attenuation, effectively eliminating spectral signal distortion interference caused by tissue depth and significantly improving the accuracy of tumor feature peak detection. It enables rapid intraoperative non-invasive in vivo differentiation of benign and malignant tissues, quantification of tumor malignancy, and high-precision delineation of microscopic invasion boundaries. This solves the technical pain points of traditional in vitro detection methods, such as lag, coarse definition of invasion range, and inability to locate tumors in real time. Finally, it generates real-time visualized detection results and integrates them into an intraoperative navigation system, accurately guiding the surgeon to completely remove the tumor lesion while preserving normal tissue to the maximum extent. This significantly improves the accuracy of tumor surgical resection, reduces residual lesions and the risk of postoperative recurrence, demonstrating outstanding clinical real-time diagnostic and therapeutic value.

[0049] In one exemplary embodiment, such as Figure 1 As shown, an in vivo detection method for malignant tumors based on laser Raman spectroscopy is provided, which may include the following steps:

[0050] In step S101, laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data are acquired in real time during the operation.

[0051] For example, this provides accurate, comprehensive, and synchronous raw data support for subsequent tumor detection processes. Laser Raman spectroscopy data can provide evidence of tumor biochemical characteristics, in-situ OCT data can supplement information on tissue depth and scattering characteristics, and intraoperative real-time imaging can help establish the correlation between Raman features and tissue spatial location. The combination of these three can improve the accuracy and timeliness of intraoperative detection, providing a reliable data foundation for tissue benignity / malignancy differentiation, tumor malignancy grading, microscopic invasion boundary delineation, and intraoperative diagnostic and treatment prompts. This, in turn, assists surgeons in making accurate decisions on the extent of surgical resection and reduces the risk of residual lesions.

[0052] In an exemplary embodiment, intraoperative real-time acquisition of laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data includes:

[0053] Construct a multimodal data synchronous acquisition and timing calibration architecture;

[0054] Based on a multimodal data synchronous acquisition and timing calibration architecture, a unified hardware clock source is used to trigger laser Raman spectroscopy acquisition, OCT tomography and intraoperative imaging, eliminating the time delay deviation of acquisition from multiple devices and obtaining timing-aligned multimodal raw data.

[0055] Motion artifact filtering is applied to the raw multimodal data to remove jitter interference caused by breathing, heartbeat, and instrument operation, thereby completing data denoising and format standardization, and outputting a spatiotemporally synchronized multimodal data volume of the target detection area.

[0056] Motion artifact filtering is a signal processing technique used to suppress or eliminate false structures or distortions in signals or images caused by relative motion between the observed object and the acquisition system.

[0057] For example, it can effectively eliminate the vibration interference caused by intraoperative breathing, heartbeat, and instrument operation, avoiding the distortion and signal mixing of multimodal raw data caused by such interference. This ensures the integrity and accuracy of laser Raman spectroscopy data, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data. It provides high-quality, interference-free raw data support for subsequent core steps such as fluorescence background suppression, characteristic peak extraction, spatial coordinate registration, and adaptive compensation correction of Raman spectroscopy depth attenuation. It reduces the impact of interference factors on tissue benign and malignant differentiation, tumor malignancy grading, and microscopic invasion boundary delineation, ensuring the accuracy and stability of intraoperative detection results and providing reliable data support for subsequent diagnosis and treatment prompts and surgical navigation assistance.

[0058] Motion artifact filtering is applied to the multimodal raw data to remove jitter interference caused by respiration, heartbeat, and instrument operation. Data denoising and format standardization are completed, outputting a spatiotemporally synchronized multimodal data volume of the target detection area. First, through a multimodal joint motion feature extraction process, the rigid and non-rigid deformation displacement vectors of the tissue contour are calculated frame-by-frame using optical flow in the intraoperative real-time image sequence. Axial and lateral tissue displacement signals are extracted from the raw OCT interferometric signal through phase demodulation. Defocusing and spot shift interference are identified in the raw laser Raman spectroscopy data by analyzing baseline fluctuations and abrupt changes in the signal-to-noise ratio of characteristic peaks at adjacent acquisition points. Simultaneously, precise localization and interval marking of periodic physiological motion jitter caused by respiration and heartbeat, as well as non-periodic sudden motion artifacts caused by instrument operation and lens shift, are completed. Then, an adaptive phase-locked loop filtering algorithm based on motion trajectory fitting is used to complete frame-by-frame tissue deformation compensation for the marked periodic motion artifacts. To compensate for jitter interference and suppress artifacts, abnormal threshold discrimination and invalid data removal are employed for non-periodic sudden motion artifacts. Data completion is achieved by combining linear interpolation of adjacent valid acquisition frames. Simultaneously, wavelet threshold denoising is applied to laser Raman spectroscopy data, coherence noise and speckle noise suppression is applied to OCT data, and edge-preserving denoising filtering is applied to intraoperative image data to complete the denoising optimization of all modal data. Then, the multimodal data that has undergone artifact removal and denoising processing is subjected to full-process format standardization processing, unifying the timestamp encoding rules, three-dimensional spatial coordinate reference and data storage format of each modal data, and completing the secondary alignment verification of each modal data in the time and spatial dimensions. Finally, a multimodal data volume covering the complete target detection area and accurately synchronized spatiotemporally point by point is output, providing a standardized basic data source for subsequent spectral feature extraction, spatial coordinate registration and tissue benign and malignant discrimination.

[0059] In step S102, the laser Raman spectroscopy data is processed to extract the tumor characteristic Raman peak dataset, calculate the relative intensity ratio of the tumor characteristic Raman peaks, and simultaneously combine the intraoperative real-time image data to complete spatial coordinate registration, establishing a precise point-to-point mapping relationship between the Raman characteristic data and the spatial location of the tissue.

[0060] The relative intensity ratio of tumor characteristic Raman peaks refers to the ratio of the intensity of characteristic peaks representing tumor-related biomolecules in the Raman spectrum, which is used to quantitatively assess the biochemical differences between tumor tissues and normal tissues.

[0061] For example, it can effectively avoid the shortcomings of a single Raman peak intensity being easily affected by tissue depth, detection environment and instrument error, realize the normalization calibration of Raman characteristic signals, and improve the stability and reliability of characteristic data. As a core characteristic parameter, it can not only provide a key basis for the accurate comparison of the corrected Raman spectrum with the standard Raman characteristic spectrum of malignant tumors, helping to efficiently distinguish the benign and malignant nature of target tissues, but also serve as an input parameter for a quantitative grading model. Using the ratio of the characteristic peak intensity of benign and normal tissues as the benchmark threshold, it can realize the grading and quantification of the degree of tumor malignancy. At the same time, combined with the mapping relationship between Raman features and tissue spatial location, it can help identify the spatial mutation critical position of tumor malignancy, provide data support for accurately delineating the microscopic invasion boundary of tumor tissue, and ensure the accuracy of intraoperative detection and surgical resection decisions.

[0062] To calculate the relative intensity ratios of tumor characteristic Raman peaks, precise peak position calibration and quantitative calculation of peak intensity are first performed for all tumor characteristic Raman peaks at each spectral acquisition site within the target detection area. The peak intensity is calculated using the integral peak area after baseline correction to improve quantitative accuracy. Simultaneously, conservative characteristic peaks that are stably expressed in normal physiological tissues and unaffected by biochemical component fluctuations caused by malignant transformation of tumors are selected as normalized internal reference peaks. The intensity of all tumor characteristic Raman peaks at the same site is normalized using these internal reference peaks to eliminate systematic errors caused by laser power fluctuations, acquisition optical path loss, and site focusing differences. Based on the normalized intensity of each tumor characteristic peak, the pairwise relative intensity ratios of nucleic acid characteristic peaks and lipid characteristic peaks, nucleic acid characteristic peaks and protein characteristic peaks, and lipid characteristic peaks and protein characteristic peaks are calculated. Furthermore, based on the specific biochemical markers corresponding to the target malignant tumor type, the relative intensity ratios of tumor-specific characteristic peaks and internal reference peaks, as well as tumor-related characteristic peaks, can be calculated. This generates a dataset of relative intensity ratios of tumor characteristic Raman peaks corresponding to each spectral acquisition site, providing core quantitative parameters for subsequent quantitative grading of tumor malignancy and benign / malignant differentiation.

[0063] In one exemplary embodiment, laser Raman spectroscopy data is processed to extract a dataset of tumor characteristic Raman peaks, including:

[0064] Construct a fluorescence background suppression and denoising model;

[0065] Based on the fluorescence background suppression and denoising model, and constrained by the characteristic shift range in the standard Raman feature map, wavelet decomposition and polynomial baseline fitting are performed on the original spectral data to remove the autofluorescence background in in vivo tissues and obtain the effective signal components.

[0066] Threshold filtering is applied to the effective signal components for noise reduction, and curve fitting is performed on the Raman peaks within the wavenumber range to screen and extract the Raman peak dataset of tumor features.

[0067] In vivo tissue autofluorescence background refers to the non-specific fluorescence signal generated by endogenous fluorescent substances in living tissue after being irradiated with excitation light, which can interfere with the detection of exogenous fluorescent probes.

[0068] For example, it can effectively eliminate the interference of autofluorescence on the laser Raman spectral signal, avoid the fluorescence signal from masking the tumor characteristic Raman peaks, and ensure the purity of the extracted effective signal components. This provides a reliable foundation for subsequent Raman peak threshold filtering and noise reduction, curve fitting, and accurate screening and extraction of tumor characteristic Raman peak datasets. At the same time, it can reduce problems such as characteristic peak shift and intensity distortion caused by fluorescence interference, ensuring the authenticity and accuracy of tumor characteristic Raman peak data. This provides high-quality characteristic data support for core steps such as subsequent calculation of the relative intensity ratio of characteristic peaks, Raman spectral depth correction, and comparison with standard Raman characteristic maps of malignant tumors. In this way, it ensures the accuracy of tissue benign and malignant differentiation, tumor malignancy grading, and microscopic invasion boundary delineation, providing a reliable basis for intraoperative diagnosis and treatment prompts and surgical navigation.

[0069] Wavelet decomposition and polynomial baseline fitting were performed on the raw spectral data to remove the in vivo tissue autofluorescence background and obtain the effective signal components. Based on a pre-constructed fluorescence background suppression and denoising model, and using the clearly defined characteristic shift wavenumber intervals in the standard Raman characteristic spectra of malignant tumors and normal tissues as the core constraints, for the raw laser Raman spectral data of the in vivo target detection area acquired in real time during surgery, wavelet basis functions adapted to the Raman spectral signals and autofluorescence background characteristics of biological tissues were selected. Multi-level wavelet decomposition was performed on the raw spectral data, decomposing the raw spectral signal into low-frequency approximate components corresponding to the broadband slowly varying autofluorescence background, and multi-scale high-frequency detail components containing narrow-band abrupt signals of Raman characteristic peaks and residual random noise. Then, the silence of the standard Raman characteristic spectrum without characteristic Raman peaks was used as the basis for further analysis. Using the wavenumber interval as the fitting anchor point, a high-order polynomial baseline fitting is performed on the low-frequency approximation components obtained by wavelet decomposition using an iterative weighted least squares method. During the iteration process, Raman peak signal points exceeding the preset deviation threshold are automatically identified and removed to avoid the bias effect of sharp characteristic peaks on the baseline fitting accuracy. A continuous baseline profile that perfectly matches the autofluorescence background of in vivo tissue is accurately fitted. Then, the fitted autofluorescence baseline is subtracted from the original spectral data to complete the main removal of the broadband autofluorescence background. The background-removed spectral signal is then reconstructed by inverse wavelet transform in combination with the effective high-frequency detail components obtained by wavelet decomposition, while simultaneously suppressing residual baseline drift and random noise. Finally, an effective spectral signal component without autofluorescence background interference and with complete preservation of effective information of tumor characteristic Raman peaks is obtained.

[0070] Curve fitting was performed on Raman peaks within the wavenumber range to screen and extract a dataset of tumor-specific Raman peaks. For the effective signal components of the Raman spectrum after autofluorescence background removal, all candidate peak sites conforming to the Raman peak line shape characteristics were identified within the target wavenumber range corresponding to malignant tumor biochemical markers using a second-derivative peak-finding algorithm. The center wavenumber, peak width boundary, and initial peak height of each candidate peak were calibrated. Then, using the calibrated parameters as initial values ​​for iteration, nonlinear least-squares curve fitting was performed on each candidate Raman peak using a Voigt line shape function adapted to Raman intrinsics and instrument broadening characteristics. The precise center wavenumber, integral peak area, and other core characteristic parameters of each candidate peak were obtained iteratively. Subsequently, the center wavenumber of each candidate peak was matched and verified with the standard wavenumber of tumor-specific characteristic peaks in the standard Raman characteristic spectrum of malignant tumors. A reasonable wavenumber deviation threshold was set, and invalid peaks, impurity peaks, and interference peaks were removed, retaining only the characteristic peaks of tumor-related biochemical markers and the internal reference conservative characteristic peaks. Finally, a dataset of tumor-specific Raman peaks corresponding to each spectral acquisition site was constructed.

[0071] In one exemplary embodiment, spatial coordinate registration is performed by combining intraoperative real-time image data to establish a precise point-to-point mapping relationship between Raman feature data and tissue spatial location, including:

[0072] Constructing an intraoperative image-Raman spectrum spatial registration model;

[0073] Based on the intraoperative image-Raman spectroscopy spatial registration model, the spatial coordinates of tissue contour feature points and laser Raman detection spot in the real-time intraoperative image are extracted. The soft tissue deformation is compensated by combining the tissue depth parameters calculated by OCT, and the registration transformation between image space and physical space is completed.

[0074] Based on the coordinate mapping parameters after registration transformation, a mapping relationship is established between spectral detection sites, image pixel coordinates, and the three-dimensional position of in vivo tissue.

[0075] Among them, the intraoperative image-Raman spectroscopy spatial registration model is a computational model used to spatially align the Raman spectroscopy acquisition points with the pixel coordinates of the intraoperative images during surgery, so as to achieve the fusion and localization of molecular information and anatomical structures.

[0076] For example, it can combine intraoperative real-time image tissue contour features, Raman detection spot spatial coordinates, and OCT tissue depth parameters to compensate for soft tissue deformation, complete the unified registration transformation between image space and physical space, establish a point-to-point precise mapping relationship between spectral detection sites, image pixel coordinates, and in vivo tissue three-dimensional position, solve the defects of Raman biochemical features being unable to spatially locate and intraoperative images lacking biochemical diagnostic information, achieve precise binding of tumor biochemical features with actual tissue position, provide reliable spatial positioning basis for subsequent Raman spectral depth attenuation correction, point-by-point benign and malignant discrimination, spatial distribution analysis of tumor malignancy degree, and three-dimensional delineation of microscopic invasion boundaries, and support the fusion and visualization of tumor detection results and intraoperative images, ensure accurate and reliable intraoperative navigation and positioning, and significantly improve the spatial accuracy of tumor microscopic invasion boundary identification and surgical resection.

[0077] By combining tissue depth parameters calculated by OCT, soft tissue deformation is compensated, and registration transformation between image space and physical space is completed. Key feature points of tissue contours in intraoperative real-time images and the physical space coordinates of laser Raman detection spots are extracted as registration reference control points. Simultaneously, the three-dimensional tissue depth parameters, soft tissue layer structure information, and axial and lateral non-rigid deformations of soft tissue caused by intraoperative traction and physiological movement are calculated from the spatiotemporally aligned in-situ OCT data. First, a rigid transformation matrix is ​​used to complete the global registration of the two-dimensional pixel coordinate system of intraoperative images and the three-dimensional physical coordinate system of Raman detection. The model aims to eliminate global spatial offset caused by overall translation, rotation, and scaling. Then, using the point-by-point tissue depth parameters and local deformation variables calculated by OCT as constraints, a thin-plate spline non-rigid deformation compensation model adapted to the deformation characteristics of soft tissue is constructed. This model compensates for the local tissue deformation corresponding to each pixel in the image point by point, correcting the axial depth deformation and lateral local deformation deviation that cannot be represented by two-dimensional images. At the same time, by iteratively optimizing the registration transformation parameters, the model minimizes the reprojection error of feature points in image space and physical space, and finally completes the accurate registration transformation from the two-dimensional image space of intraoperative images to the real three-dimensional physical space of in vivo tissue.

[0078] A mapping relationship was established between spectral detection sites, image pixel coordinates, and the three-dimensional position of in vivo tissue. Based on the coordinate mapping parameters that have been registered and transformed, and using the three-dimensional world coordinate system of the intraoperative navigation system as a unified reference, the coordinate systems of the laser Raman, intraoperative images, and OCT systems were globally calibrated. For each Raman spectral detection site within the target detection area, the three-dimensional physical coordinates and unique identifier of its spot center were accurately calibrated. Simultaneously, through the intrinsic and extrinsic parameter matrices calibrated by the imaging equipment, combined with the point-by-point tissue depth parameters calculated by OCT, the two-dimensional pixel coordinates of the intraoperative images were transformed pixel by pixel to a unified coordinate system and mapped to the real three-dimensional physical position of the in vivo tissue. Then, using the registration reference feature points as anchor points, a continuous spatial coordinate mapping function was constructed to complete the point-to-point precise binding of spectral sites and corresponding image pixel coordinates, correcting depth-related mapping deviations, and optimizing mapping accuracy through reprojection error verification. Finally, a precise end-to-end mapping relationship with temporal synchronization identifiers, corresponding spectral detection sites, image pixel coordinates, and the three-dimensional position of in vivo tissue was established for each spatial sampling unit within the target detection area.

[0079] In step S103, based on the spatial mapping relationship and combined with the in-situ OCT tissue depth and scattering coefficient data, the Raman spectrum is subjected to depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign or malignant attributes of the target area tissue and obtain the tissue benign or malignant discrimination result. A quantitative grading model is constructed based on the relative intensity ratio of feature peaks. The intensity ratio of feature peaks of benign and normal tissues is used as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy. The microscopic invasion boundary of tumor tissue is delineated according to the spatial coordinates.

[0080] Among them, the quantitative grading model refers to an analytical tool that uses mathematical methods to transform multiple indicators of the assessment object into comparable values ​​and classifies different risk or capability levels based on preset thresholds.

[0081] For example, using the ratio of characteristic peak intensities of benign and normal tissues as a unified quantitative benchmark, and employing the ratio of characteristic peak intensities among nucleic acids, lipids, and proteins as quantitative evaluation parameters, this approach overcomes the limitations of traditional methods that can only qualitatively determine tissue benignity or malignancy. It can quantitatively calculate the deviation of characteristic ratios at each spatial coordinate point from the benchmark threshold, enabling refined grading and assessment of malignancy levels for normal tissues, precancerous lesions, low-grade malignancy, intermediate-grade malignancy, and high-grade malignancy. Simultaneously, by combining spatial point-to-point mapping relationships, it can accurately identify the critical location of spatial mutations in malignancy, providing a core basis for the objective quantitative delineation of tumor microscopic invasion boundaries. This enhances the refinement, objectivity, and repeatability of tumor diagnosis, and provides reliable quantitative diagnostic and treatment support for accurately determining the surgical resection range and safe resection margin planning during surgery.

[0082] The quantitative grading model uses the ratio of characteristic peak intensities of benign to normal tissue samples as the quantification benchmark threshold. in These can correspond to the intensity ratios of characteristic peaks for nucleic acids and lipids, respectively. The ratio of characteristic peak intensity of nucleic acids to proteins Lipid and protein characteristic peak intensity ratio A quantitative grading model is constructed, using at least one combination of the aforementioned characteristic peak intensity ratios as model input parameters. Corresponding input , , Through the formula: Calculate the deviation between the characteristic peak intensity ratio of each spatial coordinate point within the target detection area and the quantization reference threshold. Then the degree of deviation Divided into multiple numerical intervals, among which Corresponding to normal tissue, Corresponding to precancerous lesions, Corresponding to low-grade malignancy, Corresponding to moderate malignancy, Corresponding to high malignancy, the tumor malignancy grade is obtained for each coordinate point. Finally, based on this grade, and combining the precise point-to-point mapping relationship between Raman feature data and tissue spatial location, the critical locations where the malignancy grade undergoes abrupt changes in space are identified. A three-dimensional closed contour is generated using an isosurface extraction algorithm, the core formula of which includes the voxel vertex isosurface intersection interpolation formula: in These are the interpolation coefficients. This represents the severity rating of the voxel apex. The isosurface threshold, , These are the maximum and minimum grading values ​​at the voxel vertices, and the formula for calculating the intersection coordinates: in Let be the coordinates of the intersection point of the isosurface and the voxel edge. , Let be the coordinates of the two vertices of the voxel edge. The intersection points of all voxel edges are calculated using the above formula. These intersection points are then connected to form triangular facets, which are then stitched together to generate a three-dimensional closed contour, defining the microscopic invasive boundary between tumor tissue and normal tissue.

[0083] In an exemplary embodiment, the Raman spectrum is subjected to depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is then compared with the standard Raman feature map of malignant tumors to determine the benign or malignant nature of the target region tissue, resulting in a tissue benign or malignant determination result, including:

[0084] By combining in situ OCT tissue depth and scattering coefficient data, the relative intensity ratio of tumor characteristic Raman peaks is compensated by photon transmission attenuation inverse operation to eliminate the influence of tissue depth differences on Raman signal intensity, and the biochemical component feature matrix after depth normalization correction is obtained.

[0085] The feature data of each detection site in the biochemical component feature matrix are compared point by point with the pre-established standard Raman feature map of malignant tumors, and the feature matching degree of each detection site is calculated.

[0086] Based on the feature matching degree, a benign or malignant probability distribution map of the target region tissue is generated, and the tissue benign or malignant discrimination result is output by segmentation through a preset probability threshold.

[0087] Among them, the standard Raman characteristic spectrum of malignant tumors refers to a standardized spectral library of specific biomolecular vibrational peaks measured by Raman spectroscopy and used to identify malignant tumors.

[0088] For example, this provides a unified, standardized, and objective authoritative reference benchmark for the comparison of corrected in vivo Raman spectral features, eliminating discrimination bias caused by individual tissue differences and fluctuations in the detection environment. By calculating the matching degree of each detection site through point-by-point feature matching, a probability distribution map of tissue benign and malignant can be generated and threshold segmentation can be completed to distinguish benign and malignant attributes, replacing subjective experience judgment. This ensures that the results of tumor benign and malignant identification have consistency, stability, and repeatability. At the same time, it provides standardized biochemical discrimination basis for the quantitative grading of characteristic peak intensity ratio, determination of malignancy degree, and identification of microscopic invasion boundaries, further improving the clinical credibility and accuracy of intraoperative in vivo tumor detection results.

[0089] The feature matching degree of each detection site is calculated. For the biochemical component feature matrix of each detection site after depth attenuation adaptive compensation correction and depth normalization, the Raman feature data of each detection site is first precisely aligned with the pre-constructed standard Raman feature maps of malignant tumors and normal tissues to unify the feature dimensions and data normalization scale, ensuring that the feature wavenumber range and sampling interval of the data to be detected are completely matched with those of the standard maps. Then, the core parameters of the corresponding feature peaks in the detection site and the standard maps are extracted. Based on the correlation between each feature peak and the malignant transformation of the tumor, a differential weight is preset, and the feature peaks of specific biomarkers related to tumor proliferation are given higher weights. Then, the weighted cosine similarity between the feature vector of the detection site and the feature vectors of the two types of standard maps is calculated. The similarity is corrected by combining the verification results of feature peak position matching deviation and peak shape fitting goodness, eliminating the matching error caused by wavenumber drift and peak shape distortion. Finally, the corrected weighted similarity value is used as the feature matching degree of the standard feature map of the malignant tumor corresponding to the detection site.

[0090] In an exemplary embodiment, a quantitative grading model is constructed based on the relative intensity ratio of characteristic peaks. The intensity ratio of characteristic peaks in benign normal tissue is used as the quantification benchmark threshold to grade and quantify the malignancy of the tumor. Furthermore, the microscopic invasion boundary of the tumor tissue is delineated based on spatial coordinates, including:

[0091] A quantitative grading model was established based on the characteristic peak intensity ratio of benign and normal tissue samples, using at least one combination of the characteristic peak intensity ratio of nucleic acid to lipid, the characteristic peak intensity ratio of nucleic acid to protein, or the characteristic peak intensity ratio of lipid to protein as input parameters.

[0092] Based on the quantitative grading model, the deviation of the characteristic peak intensity ratio of each spatial coordinate point in the target detection area from the quantitative benchmark threshold is calculated. The deviation is divided into multiple numerical intervals, which correspond to the grading standards of normal tissue, precancerous lesions, low-grade malignancy, intermediate-grade malignancy and high-grade malignancy, respectively, to obtain the tumor malignancy grading results of each coordinate point.

[0093] Based on the tumor malignancy grading results at each coordinate point, and combined with the point-to-point precise mapping relationship between Raman feature data and tissue spatial location, the critical location where the malignancy grading changes spatially is identified. A three-dimensional closed contour is generated through the isosurface extraction algorithm to delineate the microscopic invasion boundary between tumor tissue and normal tissue.

[0094] Among them, a three-dimensional closed contour refers to a continuous curve in space that is connected end to end without interruption, forming a closed loop boundary.

[0095] For example, it can completely and continuously characterize the true invasive morphology and boundary range of tumors in three-dimensional space, solving the problem of fragmented two-dimensional tomographic boundaries that cannot reflect three-dimensional infiltration. It can accurately locate the spatial critical area of ​​microscopic invasion between tumors and normal tissues, clearly present the range of lesions with micro-infiltration and irregular spread at the tumor edge, and provide intuitive and reliable spatial basis for objectively defining the actual resection boundary of the tumor and planning reasonable and safe resection margins during surgery. At the same time, it can be directly integrated with intraoperative imaging and navigation systems for visualization display, assisting surgeons in accurately identifying lesion edges and completely removing invasive tissue, effectively reducing micro-lesion residues, reducing the risk of postoperative recurrence, and significantly improving the accuracy and radical cure of tumor resection.

[0096] The deviation of the characteristic peak intensity ratio of each spatial coordinate point within the target detection area from the quantification benchmark threshold is calculated. Using the quantification benchmark threshold, pre-calibrated with a large sample of benign and normal tissue and corresponding to the characteristic peak intensity ratio input to the quantification grading model, as a reference, the validity of the characteristic peak intensity ratio of each spatial coordinate point within the target detection area is first verified, invalid data is removed, and it is ensured that its calculation method, normalization rules, and characteristic peak selection are completely consistent with the quantification benchmark threshold. For each valid coordinate point, the relative deviation of the characteristic peak intensity ratio is calculated based on the normal reference mean of the benchmark threshold. This is then converted into a standardized deviation factor by combining the standard deviation of the benign sample ratio distribution to eliminate dimensional differences. Next, the standardized deviation factors of multiple characteristic peak ratios are weighted and fused according to the model's preset weights to obtain a comprehensive deviation quantification value. Simultaneously, the excess range relative to the benchmark threshold confidence interval is verified, ultimately yielding the deviation result of each spatial coordinate point corresponding to the quantification benchmark threshold.

[0097] In step S104, based on the results of tissue benign or malignant differentiation, the degree of tumor malignancy, and the microscopic invasion boundary, real-time intraoperative visual detection results and diagnostic prompts are generated and simultaneously transmitted to the intraoperative navigation system to assist in surgical resection.

[0098] Among them, the microscopic invasion boundary refers to the indistinct boundary between tumor tissue and surrounding normal tissue observed under a microscope, exhibiting invasive growth.

[0099] For example, it can accurately identify the critical boundary of tumor cells infiltrating and spreading into surrounding normal tissues in a hidden, small, and irregular manner. This overcomes the shortcomings of traditional detection methods, which can only identify macroscopic tumor lesions and cannot detect microscopic infiltrating areas that are difficult to detect with the naked eye and conventional imaging. It can clearly distinguish the true boundary between tumor-infiltrating tissue and normal healthy tissue, providing core clinical evidence for accurately planning the surgical resection range and determining reasonable and safe resection margins during surgery. It can guide surgeons to completely remove microscopic infiltrating lesions to reduce the risk of postoperative recurrence, while avoiding excessive resection of normal tissue, maximizing the achievement of precise minimally invasive resection. At the same time, it supports the visualization of detection results and the precise guidance of intraoperative navigation systems.

[0100] In one exemplary embodiment, generating real-time intraoperative visual detection results and diagnostic prompts, and synchronously transmitting them to the intraoperative navigation system to assist in surgical resection, includes:

[0101] The results of tissue benign and malignant discrimination, tumor malignancy grade, and microscopic invasion boundary contour data are fused and overlaid with intraoperative real-time images to generate a distribution map of tumor benign and malignant attributes, a heat map of malignancy grade, and boundary contour lines displayed in pseudo-color image form as a visual detection result. At the same time, combined with the preset surgical safety margin distance parameter, diagnostic and treatment prompts are generated, including the suggested boundary of the surgical resection range, the location of residual lesions that need to be focused on, and the safety margin distance prompt.

[0102] The visualized detection results and diagnostic prompts are transmitted to the intraoperative navigation system through a standard medical digital imaging and communication protocol interface, and displayed in real time on the navigation system's monitor to assist the surgeon in making decisions and executing the surgical resection.

[0103] The location of residual lesions refers to the specific anatomical location of the diseased tissue that still exists in the body after treatment.

[0104] For example, it can accurately identify high-risk sites of tiny, occult tumor infiltration residues that cannot be distinguished by visual observation or conventional imaging during surgery, clarify the extent of microscopic invasive lesions that are easily missed during surgery, promptly issue key attention reminders to the surgeon, guide the surgeon to treat high-risk residual areas in a targeted manner, effectively avoid leaving behind tiny tumor lesions, significantly reduce the risk of local recurrence after surgery, and optimize surgical resection decisions by combining safe resection margin distance, improve the complete tumor resection rate and surgical radical cure effect, and also provide precise targeted positioning basis for targeted follow-up after surgery.

[0105] The following specific embodiment will illustrate a method for in vivo detection of malignant tumors based on laser Raman spectroscopy. Figure 2 As shown, it includes:

[0106] During intraoperative real-time detection of malignant tumors in the target tissue region, a multimodal data synchronous acquisition and timing calibration architecture is first activated. A unified 100MHz hardware clock source is used to synchronously trigger the laser Raman spectroscopy acquisition module, the optical coherence tomography (OCT) module, and the intraoperative real-time image acquisition module. The laser Raman spectroscopy acquisition module uses a 785nm narrow-linewidth continuous laser as the excitation source, with the excitation light output power stably controlled at 40mW. The spectral acquisition range covers 700cm⁻¹ to 3000cm⁻¹, the spectral resolution is set to 2cm⁻¹, and the spectral integration time for a single detection site is set to 25ms. The OCT module... Using a broadband light source with a center wavelength of 1310nm, the system has an axial resolution of 10μm, a lateral resolution of 15μm, a maximum imaging depth of 2mm, and a scanning rate of 50kHz. The intraoperative real-time image acquisition module uses a 4K resolution medical endoscope imaging unit, and the imaging frame rate is stably maintained at 30fps. Through a unified clock source synchronization triggering mechanism, the inherent acquisition delay deviations of 12ms for the laser Raman spectroscopy acquisition module, 8ms for the optical coherence tomography scanning module, and 16ms for the intraoperative real-time image acquisition module are eliminated. The timing alignment deviation of the data acquired by multiple devices is strictly controlled within 3ms, resulting in multimodal raw data with complete timing alignment.

[0107] Subsequently, motion artifact filtering was performed on the multimodal raw data. An adaptive temporal filtering algorithm was used to remove interference from in vivo tissues caused by respiratory movements (0.2Hz to 0.7Hz), periodic deformation interference from heartbeats (0.8Hz to 1.2Hz), and random jitter interference from surgical instrument manipulation (5Hz to 20Hz). The signal-to-noise ratio of the processed data was improved by 28dB. Simultaneously, data denoising and format standardization were performed, ultimately outputting a spatiotemporally synchronized multimodal data volume of the target detection area. This completed the multimodal raw data acquisition and processing workflow. After the multimodal data acquisition and processing, the acquired laser Raman spectral data was analyzed. The fluorescence background suppression and denoising process was initiated. First, a fluorescence background suppression and denoising model was constructed, using the characteristic shift intervals in the pre-established standard Raman characteristic spectra of malignant tumors and normal tissues as constraints. The original spectral data underwent db4 wavelet 5-level decomposition, and baseline fitting was performed simultaneously using a 5th-order polynomial to effectively remove the strong autofluorescence background generated in in vivo tissues, separating the effective spectral signal components. Subsequently, a 3σ threshold filtering algorithm was used to remove random noise from the effective signal components. Gaussian curve fitting was performed on the Raman peaks in the full wavenumber range of 700 cm⁻¹ to 3000 cm⁻¹, accurately screening and extracting a dataset of tumor characteristic Raman peaks highly correlated with the biochemical components of malignant tumors. The dataset contains seven core Raman peaks: 785 cm⁻¹, 1004 cm⁻¹, 1090 cm⁻¹, 1265 cm⁻¹, 1302 cm⁻¹, 1445 cm⁻¹, and 1658 cm⁻¹. Specifically, 785 cm⁻¹ corresponds to the pyrimidine ring breathing vibration of nucleic acids; 1004 cm⁻¹ corresponds to the benzene ring stretching vibration of phenylalanine in proteins; 1090 cm⁻¹ corresponds to the stretching vibration of the phosphate backbone of nucleic acids; 1265 cm⁻¹ corresponds to the vibration of the amide III bond in proteins; 1302 cm⁻¹ corresponds to the CH₂ bending vibration of lipid molecules; and 1445 cm⁻¹ corresponds to the CH₂ deformation vibration of both lipid and protein molecules. The 58 cm⁻¹ corresponds to the vibrational characteristics of the amide I bond in proteins. After extracting the characteristic peaks, the relative intensity ratios of the tumor characteristic Raman peaks are calculated simultaneously. Specifically, this includes three core relative intensity ratios: I1090 / I1445 (nucleic acid characteristic peak to lipid characteristic peak), I1090 / I1658 (nucleic acid characteristic peak to protein characteristic peak), and I1302 / I1658 (lipid characteristic peak to protein characteristic peak). These provide core quantitative parameters for subsequent benign / malignant differentiation and quantitative grading of malignancy. While completing the Raman characteristic peak extraction and relative intensity ratio calculation, the spatial coordinate registration process between intraoperative images and Raman spectra is initiated to construct an intraoperative image-Raman spectrum spatial registration model.

[0108] First, tissue contour feature points were extracted from the intraoperative real-time images, resulting in 128 highly stable tissue contour feature points. Simultaneously, the physical spatial coordinates of the laser Raman detection spot were acquired, with a sampling step size of 50 μm and a spatial coordinate sampling density of 400 detection sites per square millimeter. Combined with tissue depth parameters calculated by the optical coherence tomography (OCT) module, precise compensation was performed for the nonlinear deformation of soft tissue caused by traction and pulsation during surgery, achieving a deformation compensation accuracy of 8 μm. Rigid and non-rigid registration transformations were completed between the image pixel space and the tissue physical space, with the reprojection error strictly controlled within 10 μm. Finally, based on the coordinate mapping parameters after registration transformation, the coordinate mapping between each spectral detection site and the intraoperative image pixels was established. A precise point-to-point mapping relationship between coordinates and the three-dimensional physical location of in vivo tissue is established, realizing a one-to-one correspondence between Raman feature data and tissue spatial location. This provides a spatial coordinate foundation for subsequent depth correction, boundary delineation, and visualization fusion. After establishing the spatial coordinate mapping relationship, tissue depth and scattering coefficient data obtained from in-situ optical coherence tomography (OCT) are combined to perform depth attenuation adaptive compensation correction on the Raman spectral data. The tissue scattering coefficient distribution range within the target detection area measured by the OCT module is 8 mm⁻¹ to 22 mm⁻¹, and the effective tissue detection depth range is 0 mm to 1.5 mm. Based on the modified Beer-Lambert photon transport model, photon transport attenuation inverse operation compensation is performed on the relative intensity ratio of tumor characteristic Raman peaks. This method effectively eliminates the influence of tissue depth differences and scattering coefficient heterogeneity on Raman signal intensity. After compensation processing, the relative deviation of characteristic peak intensity of the same type of normal tissue at different depths is reduced from 35% to less than 4%, resulting in a depth-normalized corrected biochemical component feature matrix. Subsequently, the feature data of each detection site in the biochemical component feature matrix are compared point-by-point with a pre-established standard Raman feature map of malignant tumors. This standard Raman feature map is constructed based on 1200 benign and malignant tissue samples verified by pathological gold standards, including 620 malignant tumor tissue samples and 580 benign normal tissue samples. The feature comparison uses the cosine similarity algorithm to calculate the feature matching degree of each detection site, and the feature matching degree discrimination threshold is set to 0.85. A feature matching degree of 0.85 or higher at a detection site is considered malignant, while a feature matching degree of less than 0.85 is considered benign. Simultaneously, a benign / malignant probability distribution map of the target area tissue is generated based on the feature matching degree of each point within the entire detection area. The probability values ​​range from 0 to 1, and 0.5 is used as the probability segmentation threshold. After binarization, the benign / malignant tissue discrimination result for the entire detection area is output. After completing the tissue benign / malignant discrimination, a tumor malignancy grading model is constructed based on the extracted feature peak relative intensity ratio. The feature peak intensity ratio of pathologically verified benign / normal tissue samples is used as the quantification benchmark threshold, where the I1090 / I1445 benchmark threshold for benign / normal tissue is 0.62, with a standard deviation of 0.The baseline thresholds for I1090 / I1658 were 0.58 with a standard deviation of 0.07, and for I1302 / I1658, they were 0.91 with a standard deviation of 0.09. The weighted combination of the three sets of feature peak intensity ratios was used as the input parameters for the quantization grading model. The weights of the three parameters were set to 0.4, 0.35, and 0.25, respectively. Based on the quantization grading model, the deviation of the feature peak intensity ratio of each spatial coordinate point within the target detection area from the quantization baseline threshold was calculated. The deviation was quantified using standardized Z-score values. Malignancy was graded according to the Z-score range: -1 to 1 corresponded to normal tissue, 1 to 2 to precancerous lesions, 2 to 3 to low-grade malignant tumors, 3 to 4 to moderate-grade malignant tumors, and greater than 4 to high-grade malignant tumors. Finally, the tumor malignancy grading results for each spatial coordinate point within the entire detection area were obtained.

[0109] Subsequently, based on the tumor malignancy grading results at each coordinate point, and combining the precise point-to-point mapping relationship between Raman feature data and tissue spatial location, the critical locations where tumor malignancy grading abruptly changes in spatial distribution are identified. The MarchingCubes isosurface extraction algorithm is used to generate a three-dimensional closed contour, accurately delineating the microscopic invasion boundary between tumor tissue and normal tissue. The boundary positioning accuracy reaches 25μm, achieving precise characterization of the tumor's microscopic invasion range. After completing tissue benign / malignant differentiation, tumor malignancy grading, and microscopic invasion boundary delineation, all detection results are visualized intraoperatively in real time. The processing and diagnosis prompt generation process first involves pixel-level fusion and overlay of tissue benign / malignant differentiation results, tumor malignancy grading results, and microscopic invasive boundary contour data with intraoperative real-time images. A pseudo-color coding method is then used to generate visualized detection results. Normal tissue areas are displayed in green pseudo-color, precancerous lesions in yellow, low-grade malignant tumors in orange, moderate-grade malignant tumors in light red, and high-grade malignant tumors in dark red. Simultaneously, a tumor malignancy grading heatmap is generated, using color gradients. The system visually presents the spatial distribution differences in tumor malignancy, clearly marking the microscopic invasion boundaries of tumor tissue with solid white lines. Combined with a preset safe surgical margin distance parameter of 2mm, it generates diagnostic and treatment prompts including suggested surgical resection boundaries, locations of residual lesions requiring close monitoring, and real-time alerts on safe margin distances. The suggested surgical resection boundaries are marked with dashed white lines, and the real-time alert accuracy for safe margin distances reaches 0.1mm. The generated visual detection results and diagnostic and treatment prompts are synchronously transmitted to the intraoperative navigation system via the standard medical digital imaging and communication DICOM 3.0 protocol interface. The end-to-end data transmission latency is strictly controlled within 100ms. The visualization results displayed on the navigation system monitor are completely synchronized with the real-time intraoperative images, maintaining a stable 30fps without significant stuttering or delay. This assists the surgeon in real-time monitoring of the tumor's location, extent, malignancy, and invasion boundaries during surgery, enabling precise tumor resection. Simultaneously, it verifies the benign or malignant properties of the surgical margins, ensuring complete tumor removal while preserving normal functional tissue to the greatest extent possible, improving surgical precision and treatment effectiveness, and reducing the risk of postoperative tumor recurrence.

[0110] In summary, this invention enables real-time in-situ detection of malignant tumors in target tissue areas during surgery. Through multimodal device synchronous collaboration, tumor-specific biochemical feature identification and analysis, high-precision spatial registration of images and spectra, and adaptive correction for tissue depth signal attenuation, combined with a pathological gold standard feature library, it achieves intelligent discrimination of tissue benignity and malignancy. This enables quantitative grading of tumor malignancy and precise delineation of microscopic invasion boundaries. The detection results are visualized through real-time pixel-by-pixel fusion of intraoperative images, simultaneously generating surgical resection range and safe resection margin diagnostic prompts, and are integrated into the intraoperative navigation system in real time. This allows for clear real-time presentation of tumor location, malignancy grade, and invasion range during surgery, guiding surgeons to precisely resect lesions. While completely removing tumor tissue, it maximizes the preservation of normal functional tissue, effectively improving the accuracy of intraoperative tumor diagnosis and treatment and reducing the risk of postoperative tumor recurrence.

[0111] Figure 3 This is a schematic diagram of the structure of an in vivo malignant tumor detection system based on laser Raman spectroscopy, according to an embodiment of this application.

[0112] like Figure 3 As shown, the in vivo malignant tumor detection system 10 based on laser Raman includes: a data acquisition module 100, a spectral denoising and spatial registration module 200, a benign / malignant tumor discrimination and boundary delineation module 300, and an intraoperative visualization module 400.

[0113] The system includes the following modules: Data acquisition module 100, used for real-time acquisition of laser Raman spectral data, in-situ OCT tissue depth and scattering coefficient data, and real-time intraoperative image data of the target detection area; spectral denoising and spatial registration module 200, used for processing the laser Raman spectral data, extracting tumor characteristic Raman peak datasets, calculating the relative intensity ratios of tumor characteristic Raman peaks, and simultaneously performing spatial coordinate registration in conjunction with real-time intraoperative image data to establish a precise point-to-point mapping relationship between Raman characteristic data and tissue spatial location; and benign / malignant discrimination and boundary delineation module 300, used for processing the laser Raman spectral data based on the spatial mapping relationship and in-situ OCT tissue depth and scattering coefficient data, and for determining the Raman spectral characteristics of the target area. The spectrum undergoes depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign or malignant attributes of the target area tissue, obtaining the tissue benign or malignant discrimination result. A quantitative grading model is constructed based on the relative intensity ratio of feature peaks, using the intensity ratio of characteristic peaks of benign and normal tissues as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy, and delineate the microscopic invasion boundary of tumor tissue according to spatial coordinates. The intraoperative visualization module 400 generates real-time intraoperative visualization detection results and diagnostic prompts based on the tissue benign or malignant discrimination result, the degree of tumor malignancy, and the microscopic invasion boundary, and transmits them synchronously to the intraoperative navigation system to assist in surgical resection.

[0114] It should be noted that the foregoing explanation of an embodiment of a method for in vivo detection of malignant tumors based on laser Raman spectroscopy also applies to the in vivo detection system for malignant tumors based on laser Raman spectroscopy in this embodiment, and will not be repeated here.

[0115] According to the embodiments of this application, a laser Raman-based in vivo malignant tumor detection system is proposed. By simultaneously fusing intraoperative in vivo laser Raman spectra, in-situ OCT tissue depth scattering parameters, and real-time image multimodal data, it establishes a precise mapping between Raman features and tissue location based on point-to-point spatial coordinate registration. Combined with tissue depth information, it completes adaptive compensation correction for Raman spectral depth attenuation, effectively eliminating spectral signal distortion interference caused by tissue depth and significantly improving the accuracy of tumor characteristic peak detection. It can perform rapid intraoperative non-invasive in vivo identification of benign and malignant tissues, grading and quantifying the degree of tumor malignancy, and high-precision delineation of microscopic invasion boundaries, solving the technical pain points of traditional in vitro detection such as lag, coarse definition of invasion range, and inability to locate in real time. Finally, it generates visualized detection results in real time and connects to the intraoperative navigation system, which can accurately guide the surgeon to completely remove the tumor lesion and preserve normal tissue to the maximum extent, significantly improving the accuracy of tumor surgical resection, reducing residual lesions and the risk of postoperative recurrence, and demonstrating outstanding clinical real-time diagnostic and therapeutic value.

[0116] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for in vivo detection of malignant tumors based on laser Raman spectroscopy.

[0117] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the aforementioned method for in vivo detection of malignant tumors based on laser Raman spectroscopy.

[0118] In the description of this specification, the references to the terms "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0119] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0120] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0121] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by suitable instructions. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0122] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0123] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for in vivo detection of malignant tumors based on laser Raman spectroscopy, characterized in that, include: Intraoperative real-time acquisition of laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data; The laser Raman spectroscopy data is processed to extract the tumor characteristic Raman peak dataset, calculate the relative intensity ratio of the tumor characteristic Raman peaks, and simultaneously combine the intraoperative real-time image data to complete spatial coordinate registration, establishing a point-to-point precise mapping relationship between Raman characteristic data and tissue spatial location; Based on the spatial mapping relationship and combined with the in-situ OCT tissue depth and scattering coefficient data, the Raman spectrum is subjected to depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign or malignant attributes of the target area tissue and obtain the tissue benign or malignant discrimination result. A quantitative grading model is constructed based on the relative intensity ratio of the feature peaks. The intensity ratio of the feature peaks of benign and normal tissues is used as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy. The microscopic invasion boundary of tumor tissue is delineated according to the spatial coordinates. Based on the results of tissue benign or malignant differentiation, tumor malignancy degree and microscopic invasion boundary, real-time intraoperative visual detection results and diagnostic prompts are generated and synchronously transmitted to the intraoperative navigation system to assist in surgical resection.

2. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The intraoperative real-time acquisition of laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time imaging data includes: Construct a multimodal data synchronous acquisition and timing calibration architecture; Based on the aforementioned multimodal data synchronous acquisition and timing calibration architecture, a unified hardware clock source is used to trigger laser Raman spectroscopy acquisition, OCT tomography and intraoperative image capture, eliminating the time delay deviation of multi-device acquisition and obtaining timing-aligned multimodal raw data. Motion artifact filtering is applied to the original multimodal data to remove jitter interference caused by breathing, heartbeat, and instrument operation, thereby completing data noise reduction and format standardization, and outputting a spatiotemporally synchronized multimodal data volume of the target detection area.

3. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The process of processing the laser Raman spectroscopy data to extract a dataset of tumor characteristic Raman peaks includes: Construct a fluorescence background suppression and denoising model; Based on the fluorescence background suppression and denoising model, and constrained by the characteristic shift range in the standard Raman feature map, wavelet decomposition and polynomial baseline fitting are performed on the original spectral data to remove the autofluorescence background in in vivo tissues and obtain the effective signal components. Threshold filtering is applied to the effective signal components for noise reduction, and curve fitting is performed on the Raman peaks within the wavenumber range to screen and extract the tumor characteristic Raman peak dataset.

4. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The step of combining the intraoperative real-time image data to complete spatial coordinate registration and establish a point-to-point precise mapping relationship between Raman feature data and tissue spatial location includes: Constructing an intraoperative image-Raman spectrum spatial registration model; Based on the intraoperative image-Raman spectroscopy spatial registration model, the spatial coordinates of the tissue contour feature points and the laser Raman detection spot in the real-time intraoperative image are extracted. The soft tissue deformation is compensated by combining the tissue depth parameters calculated by OCT, and the registration transformation between image space and physical space is completed. Based on the coordinate mapping parameters after registration transformation, a mapping relationship is established between spectral detection sites, image pixel coordinates, and the three-dimensional position of in vivo tissue.

5. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The Raman spectrum undergoes depth attenuation adaptive compensation correction. The corrected Raman spectral feature data is then compared with the standard Raman feature map of malignant tumors to determine the benign or malignant nature of the target tissue, yielding a tissue benign / malignant discrimination result, including: By combining in situ OCT tissue depth and scattering coefficient data, the relative intensity ratio of tumor characteristic Raman peaks is compensated by photon transmission attenuation inverse operation to eliminate the influence of tissue depth differences on Raman signal intensity, and the biochemical component feature matrix after depth normalization correction is obtained. The feature data of each detection site in the biochemical component feature matrix are compared point by point with the pre-established standard Raman feature map of malignant tumors, and the feature matching degree of each detection site is calculated. Based on the feature matching degree, a benign or malignant probability distribution map of the target region tissue is generated, and a binary tissue benign or malignant discrimination result is output by segmentation through a preset probability threshold.

6. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The quantitative grading model constructed based on the relative intensity ratio of the characteristic peaks, using the intensity ratio of characteristic peaks of benign and normal tissues as the quantitative benchmark threshold, quantifies the degree of tumor malignancy and delineates the microscopic invasion boundary of tumor tissue according to the spatial coordinates, including: A quantitative grading model was established based on the characteristic peak intensity ratio of benign and normal tissue samples, using at least one combination of the characteristic peak intensity ratio of nucleic acid to lipid, the characteristic peak intensity ratio of nucleic acid to protein, or the characteristic peak intensity ratio of lipid to protein as input parameters. According to the quantitative grading model, the deviation of the characteristic peak intensity ratio of each spatial coordinate point in the target detection area from the quantitative benchmark threshold is calculated. The deviation is divided into multiple numerical intervals, which correspond to the grading standards of normal tissue, precancerous lesions, low-grade malignancy, intermediate-grade malignancy and high-grade malignancy, respectively, to obtain the tumor malignancy grading results of each coordinate point. Based on the tumor malignancy grading results at each coordinate point, and combined with the point-to-point precise mapping relationship between the Raman feature data and the spatial location of the tissue, the critical position where the malignancy grading changes abruptly in space is identified. A three-dimensional closed contour is generated by the isosurface extraction algorithm to delineate the microscopic invasion boundary between tumor tissue and normal tissue.

7. The in vivo detection method for malignant tumors based on laser Raman spectroscopy according to claim 1, characterized in that, The generated real-time visual detection results and diagnostic prompts during surgery are synchronously transmitted to the intraoperative navigation system to assist in surgical resection, including: The results of tissue benign and malignant discrimination, tumor malignancy grade, and microscopic invasion boundary contour data are fused and superimposed with intraoperative real-time images to generate a distribution map of tumor benign and malignant attributes, a heat map of malignancy grade, and boundary contour lines displayed in pseudo-color image form as a visual detection result. At the same time, combined with the preset surgical safety margin distance parameter, diagnostic and treatment prompts are generated, including the suggested boundary of the surgical resection range, the location of residual lesions that need to be focused on, and the safety margin distance prompt. The visualized detection results and the diagnostic prompts are transmitted to the intraoperative navigation system via a standard medical digital imaging and communication protocol interface and displayed in real time on the navigation system's monitor to assist the surgeon in making decisions and executing the surgical resection.

8. A system for in vivo detection of malignant tumors based on laser Raman spectroscopy, characterized in that, include: The data acquisition module is used to acquire laser Raman spectral data of the in vivo target detection area, in-situ OCT tissue depth and scattering coefficient data, and intraoperative real-time image data in real time during the operation; The spectral denoising and spatial registration module is used to process the laser Raman spectral data, extract the tumor characteristic Raman peak dataset, calculate the relative intensity ratio of the tumor characteristic Raman peaks, and combine the intraoperative real-time image data to complete spatial coordinate registration, establishing a point-to-point accurate mapping relationship between Raman characteristic data and tissue spatial location. The benign / malignant differentiation and boundary delineation module is used to perform depth attenuation adaptive compensation correction on the Raman spectrum based on the spatial mapping relationship and the in-situ OCT tissue depth and scattering coefficient data. The corrected Raman spectral feature data is compared with the standard Raman feature spectrum of malignant tumors to determine the benign / malignant attributes of the target area tissue and obtain the tissue benign / malignant differentiation result. A quantitative grading model is constructed based on the relative intensity ratio of the feature peaks. The intensity ratio of the feature peaks of benign and normal tissues is used as the quantitative benchmark threshold to grade and quantify the degree of tumor malignancy. The microscopic invasion boundary of the tumor tissue is delineated according to the spatial coordinates. The intraoperative visualization module generates real-time intraoperative visualization detection results and diagnostic prompts based on the tissue benignity / malignancy determination results, tumor malignancy degree, and microscopic invasion boundary, and transmits them synchronously to the intraoperative navigation system to assist in surgical resection.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the in vivo detection method for malignant tumors based on laser Raman as described in any one of claims 1-6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed, it implements the in vivo detection method for malignant tumors based on laser Raman as described in any one of claims 1-6.

Images