Dynamic tissue segmentation method based on transmission difference

Through the dynamic tissue segmentation method based on transmission differences, using broadband light source and fluorescent dye labeling combined with convolutional neural network analysis, the clarity problem of intraoperative imaging technology when distinguishing fascial, ligament, interdisc tissues and vascular nerves is solved, and the precise segmentation and real-time visualization of tissues are achieved, improving surgical efficiency and accuracy.

CN120259269AInactive Publication Date: 2025-07-04FUXIN MINING GENERAL HOSPITAL OF LIAONING HEALTH IND GRP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510417759.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-03
Publication Date
2025-07-04
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing intraoperative imaging technology has low clarity when distinguishing structures such as fascia, ligaments, disc tissues and vascular nerves, especially in the case of tissue adhesions, which leads to a decrease in intraoperative identification accuracy and lacks real-time dynamic analysis capabilities, which affects surgical efficiency.

Method used

The dynamic segmentation method based on transmission difference is adopted to obtain multi-spectral images through broadband light sources and fluorescent dye labels, and the tissue characteristics are analyzed using convolutional neural networks, combined with anatomical space depth data, dynamically adjust the light source band and angle, and combined with real-time visual rendering and three-dimensional reconstruction technology to achieve accurate segmentation and visualization of the tissue.

Benefits of technology

It improves the accuracy and real-time nature of intraoperative tissue recognition, enhances image support in complex surgical environments, and ensures clear visualization and precise positioning of key structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120259269A_ABST
    Figure CN120259269A_ABST
Patent Text Reader

Abstract

The invention discloses a dynamic tissue segmentation method based on transmission difference, which is characterized in that a multispectral image is acquired by combining irradiation of a broadband light source with specific fluorescence labeling based on the transmission difference of different tissues to light, and tissue characteristics are analyzed by using an image processing algorithm. A convolutional neural network is adopted to extract spectral features of fascia, ligaments, vascular nerves and intervertebral disc tissues, and precise segmentation of the tissues is realized in combination with anatomical space depth data. For a complex tissue adhesion scene, a dynamic light source wave band adjusting strategy is introduced, the contrast ratio between different tissues is optimized, and the resolution ratio is improved. Besides, through real-time visual rendering and a three-dimensional reconstruction technology, the spatial presentation effect of an intraoperative tissue structure is enhanced, and an optical flow algorithm and a time sequence filtering technology are utilized to stably track a tissue boundary. According to the invention, the accuracy of tissue image recognition can be improved in a complex imaging environment, and dynamic segmentation and visualization processing of different tissue structures can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of image dynamic segmentation, and particularly relates to a method and device for tissue dynamic segmentation based on transmission difference. Background Art

[0002] In the fields of joint and spinal surgery, intraoperative imaging technology is playing an increasingly prominent role in minimally invasive surgery, and its development directly affects the accuracy of intraoperative operations and surgical efficiency. Currently, spinal minimally invasive surgeries (such as endoscopic discectomy and UBE surgery) usually require high-precision operations in a narrow anatomical space, posing high requirements for the imaging resolution of intraoperative tissues. However, existing intraoperative imaging methods still have certain limitations, affecting the ability to identify key tissue structures during the surgical process.

[0003] Currently, intraoperative imaging mainly relies on white light imaging or single fluorescence staining technology. When resolving structures such as fascia, ligaments, intervertebral disc tissues, and blood vessels and nerves, the clarity is relatively low. Especially in the case of tissue adhesions, traditional imaging methods are difficult to accurately distinguish the boundaries of different tissues, resulting in a decrease in intraoperative recognition accuracy and affecting the doctor's operation judgment. In addition, due to the relatively single light source band of existing intraoperative imaging technology, the transmission characteristics of different tissues to light have not been fully utilized, resulting in poor visualization effects of some key tissues.

[0004] In addition, most existing intraoperative imaging methods rely on static images and lack the ability to perform real-time analysis based on different tissue characteristics, making it difficult to meet the dynamic recognition requirements of tissue structures during the surgical process. In complex surgical scenarios, insufficient tissue imaging resolution may prolong the surgical time and increase the operation difficulty. Therefore, how to optimize imaging technology and improve intraoperative tissue resolution by utilizing the transmission characteristics of different tissues to different wavelengths of light has become an urgent problem to be solved in the current field of intraoperative imaging. Therefore, there is an urgent need for a technology to overcome the deficiencies of existing intraoperative imaging methods and improve the accuracy and real-time performance of tissue recognition. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a method and device for tissue dynamic segmentation based on transmission difference to solve the problems existing in the above-mentioned prior art.

[0006] In the first aspect, to achieve the above object, the present invention provides a method for tissue dynamic segmentation based on transmission difference, including the following steps:

[0007] Obtain multispectral image data of the surgical area, irradiate fascia, ligaments, blood vessels and nerves, and intervertebral disc tissues with a broadband light source, and generate a first resolution image by using the transmission difference of each tissue to different wavelengths of light;

[0008] For the first resolution image, a fluorescent dye labeling scheme is adopted to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves, generating a second enhanced image;

[0009] Extract the superimposed features of the fluorescent signal and the broadband light source signal from the second enhanced image, analyze the signal intensity and spatial distribution through a convolutional neural network, and determine the boundary between the fascia and the ligament to generate a third segmentation image;

[0010] Combine the third segmentation image and the real-time anatomical space depth data, and perform secondary feature extraction on the blood vessels, nerves, and intervertebral disc tissues using the transmission difference to generate a fourth localization image;

[0011] After generating the fourth localization image, when overlapping signals in the adhesion scenario are detected, optimize the irradiation angle and wavelength combination through a dynamic light source band adjustment algorithm to generate a fifth optimized image;

[0012] Based on the fifth optimized image, fuse the spatial coordinates and depth data of the blood vessels, nerves, and intervertebral disc tissues, and generate a sixth dynamic image through real-time visual rendering;

[0013] Classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image;

[0014] Analyze the tissue boundary change trend of the seventh classification image using the optical flow algorithm, and smooth the jitter through time series filtering to generate an eighth stable image;

[0015] Extract the key structure contours from the eighth stable image and map them to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate a ninth visualization image, completing the tissue segmentation in the image.

[0016] Optionally, the process of obtaining the multispectral image data of the surgical area, irradiating the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues with a broadband light source, and generating the first resolution image using the transmission difference of each tissue for different wavelengths of light includes:

[0017] Perform denoising and enhancement processing on the multispectral image data to obtain a preliminary image;

[0018] Extract the transmission characteristics of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues from the preliminary image, and perform boundary segmentation on the fascia tissue using the wavelength difference threshold;

[0019] Extract the texture features of the fascia and the ligament based on the fascia segmentation image, and judge the junction area between the two through a support vector machine algorithm;

[0020] Analyze the change trend of the transmission characteristics of the blood vessels and nerves and the spectral characteristics of the intervertebral disc tissue, and use the random forest algorithm to judge the contact area between the intervertebral disc and the blood vessels and nerves, generating a multi-tissue distribution map as the first resolution image.

[0021] Optionally, for the first resolution image, the process of generating the second enhanced image by using a fluorescent dye labeling scheme to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves includes:

[0022] Mark the excitation signals of the fascia signal, ligament signal, and blood vessels and nerves in the first resolution image to generate a preliminary fluorescence image;

[0023] Adjust the wavelength to enhance the intensity distribution of the fluorescent signal and separate the fascia signal from the ligament signal;

[0024] Judge the boundary region between the fascia and the ligament through a convolutional neural network to generate a blood vessel distribution feature map;

[0025] Extract the excitation difference between the blood vessels and nerves and the ligament signal, analyze the spatial distribution characteristics of the fluorescent signal, and generate the second enhanced image.

[0026] Optionally, the process of extracting the superimposed features of the fluorescent signal and the broadband light source signal from the second enhanced image, analyzing the signal intensity and spatial distribution through a convolutional neural network, and determining the demarcation line between the fascia and the ligament to generate the third segmentation image includes:

[0027] Use a convolutional neural network to extract the fluorescent signal intensity and spatial distribution information and judge the boundary continuity between the fascia and the ligament;

[0028] Separate the fascia feature map based on the fluorescent distribution characteristics at the fascia demarcation, and adjust the spatial distribution weight of the broadband light source signal at the ligament demarcation;

[0029] Extract the overlapping region of the fluorescent signal and the broadband light source, determine the independent regions of the fascia and the ligament through the difference characteristics, and generate the third segmentation image.

[0030] Optionally, after generating the fourth positioning image, when overlapping signals in the adhesion scenario are detected, the process of generating the fifth optimized image by optimizing the irradiation angle and wavelength combination through a dynamic light source band adjustment algorithm includes:

[0031] After generating the fourth positioning image, detect the overlapping signals in the adhesion scenario, adjust the light source wavelength to the red light band and optimize the irradiation angle;

[0032] Adopt an edge detection algorithm to extract the boundary information between the blood vessels and nerves and the intervertebral disc tissue, and repair the broken boundary through morphological operations to generate the fifth optimized image.

[0033] Optionally, the process of fusing the spatial coordinates and depth data of the blood vessels and nerves and the intervertebral disc tissue based on the fifth optimized image and generating the sixth dynamic image through real-time visual rendering includes:

[0034] Extract the spatial coordinates of the blood vessels and nerves and the intervertebral disc tissue based on the fifth optimized image and map them to a three-dimensional coordinate system through depth data;

[0035] Fuse the boundary information with the depth distribution of the anatomical space, and use the interpolation method to smooth the mutation area of the spatial coordinates;

[0036] Enhance the spatial hierarchy of blood vessels, nerves and intervertebral disc tissues by adjusting the light and shadow distribution, and generate the sixth dynamic image.

[0037] Optionally, the process of analyzing the tissue boundary change trend of the seventh classification image using the optical flow algorithm and generating the eighth stable image by smoothing the jitter through time series filtering includes:

[0038] Extract the displacement change data of the tissue boundary in the seventh classification image, and calculate the pixel motion vector between consecutive frames;

[0039] Eliminate the random jitter of the displacement data through time series filtering to generate a smoothed boundary change curve;

[0040] Based on the smoothed data, correct the spatial position of the tissue boundary to generate the eighth stable image.

[0041] Optionally, the process of extracting the key structure contours from the eighth stable image and mapping them to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate the ninth visualization image and complete the tissue segmentation in the image includes:

[0042] Extract the contour data of the fascia, ligament, blood vessel, nerve and intervertebral disc tissue from the eighth stable image, and analyze their spatial positioning through stereology methods;

[0043] Fuse the anatomical space depth data with the contour data, and use interpolation technology to adjust the boundary details of the three-dimensional model;

[0044] Generate a three-dimensional visualization image based on volume rendering technology to generate the ninth visualization image.

[0045] In a second aspect, the present invention also provides a tissue dynamic segmentation device based on transmission difference for implementing a tissue dynamic segmentation method based on transmission difference. The device includes:

[0046] A multispectral imaging module for acquiring multispectral image data of the surgical area, irradiating the fascia, ligament, blood vessel, nerve and intervertebral disc tissue with a broadband light source, and generating a first resolution image by using the transmission difference of each tissue to different wavelengths of light;

[0047] A fluorescence labeling module for exciting the fluorescence signals of the fascia, ligament and blood vessel nerves in the first resolution image to generate a second enhanced image;

[0048] A feature analysis module, configured to extract the superimposed features of the fluorescence signal and the broadband light source signal from the second enhanced image, analyze the signal intensity and spatial distribution through a convolutional neural network, and determine the boundary line between the fascia and the ligament to generate a third segmentation image;

[0049] A secondary positioning module, configured to combine real-time anatomical space depth data, and perform secondary feature extraction on blood vessels, nerves, and intervertebral disc tissues by using transmission differences to generate a fourth positioning image;

[0050] A dynamic light source optimization module, configured to generate a fifth optimized image by adjusting the light source wavelength and irradiation angle when overlapping signals in an adhesion scenario are detected;

[0051] A visualization rendering module, configured to fuse the spatial coordinates and depth data of blood vessels, nerves, and intervertebral disc tissues to generate a sixth dynamic image;

[0052] A pixel classification module, configured to classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image;

[0053] An optical flow tracking module, configured to analyze the change trend of the tissue boundary and smooth the jitter through time series filtering to generate an eighth stable image;

[0054] A three-dimensional reconstruction module, configured to extract the key structure contours and map them to a three-dimensional anatomical model to generate a ninth visualization image.

[0055] Optionally, the multispectral imaging module includes:

[0056] An image denoising unit, configured to perform denoising and enhancement processing on the multispectral image data to obtain a preliminary image;

[0057] A transmission characteristic extraction unit, configured to extract the transmission characteristics of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues from the preliminary image;

[0058] A fascia segmentation unit, configured to perform boundary segmentation on the fascia tissue by using a wavelength difference threshold;

[0059] A texture analysis unit, configured to extract the texture features of the fascia and the ligament based on the fascia segmentation image, and judge the junction area between the two through a support vector machine algorithm;

[0060] A contact analysis unit, configured to analyze the change trend of the transmission characteristics of blood vessels and nerves and the spectral characteristics of intervertebral disc tissues, and judge the contact area through a random forest algorithm.

[0061] In a third aspect, the present invention further provides a computer terminal device, including:

[0062] One or more processors;

[0063] A memory, coupled to the processor, for storing one or more programs;

[0064] When the one or more programs are executed by the one or more processors, the one or more processors implement a tissue dynamic segmentation method based on transmission difference.

[0065] In a fourth aspect, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, a tissue dynamic segmentation method based on transmission difference is implemented.

[0066] Compared with the prior art, the present invention has the following advantages and technical effects:

[0067] A tissue dynamic segmentation method and device based on transmission difference provided by the present invention. The method is based on the transmission difference of light by different tissues, obtains a multi-spectral image by irradiating with a broadband light source in combination with specific fluorescence labeling, and analyzes tissue characteristics using an image processing algorithm. A convolutional neural network is used to extract the spectral characteristics of fascia, ligaments, blood vessels, nerves, and intervertebral disc tissues, and combined with anatomical space depth data to achieve precise segmentation of tissues. For complex tissue adhesion scenarios, the present invention introduces a dynamic light source band adjustment strategy to optimize the contrast between different tissues and improve the resolution. In addition, through real-time visualization rendering and three-dimensional reconstruction techniques, the spatial presentation effect of intraoperative tissue structures is enhanced, and an optical flow algorithm and time series filtering technique are used to stably track tissue boundaries.

[0068] The present invention can improve the accuracy of tissue image recognition in a complex imaging environment, achieve dynamic segmentation and visualization processing of different tissue structures, and provide clearer image support for intraoperative image analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0070] Figure 1 is a flowchart of the tissue dynamic segmentation method based on transmission difference according to an embodiment of the present invention;

[0071] Figure 2 is a partial flow schematic diagram of the tissue dynamic segmentation method based on transmission difference according to an embodiment of the present invention;

[0072] Figure 3 is a partial flow schematic diagram of the tissue dynamic segmentation method based on transmission difference according to an embodiment of the present invention;

[0073] Figure 4Schematic diagram of the structure of the tissue dynamic segmentation device based on transmission difference according to the embodiment of the present invention. Detailed implementation manners

[0074] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0075] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0076] Embodiment 1

[0077] As Figure 1 shown, in this embodiment, a tissue dynamic segmentation method based on transmission difference is provided, including:

[0078] Obtain multi-spectral image data of the surgical area, irradiate fascia, ligaments, blood vessels, nerves and disc tissues with a broadband light source, and generate a first resolution image by using the transmission difference of each tissue for light of different wavelengths;

[0079] For the first resolution image, adopt a fluorescent dye labeling scheme to excite the fluorescent signals of fascia, ligaments and blood vessels and nerves, and generate a second enhanced image;

[0080] Extract the superimposed features of the fluorescent signal and the broadband light source signal from the second enhanced image, analyze the signal intensity and spatial distribution through a convolutional neural network, and determine the boundary line between the fascia and the ligament to generate a third segmentation image;

[0081] Combine real-time anatomical space depth data, and use the transmission difference to perform secondary feature extraction on blood vessels, nerves and disc tissues to generate a fourth localization image;

[0082] When overlapping signals in the adhesion scenario are detected, optimize the irradiation angle and wavelength combination through the dynamic light source band adjustment algorithm to generate a fifth optimized image;

[0083] Fuse the spatial coordinates and depth data of blood vessels, nerves and disc tissues, and generate a sixth dynamic image through real-time visual rendering;

[0084] Classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image;

[0085] Use the optical flow algorithm to analyze the change trend of the tissue boundary, smooth the jitter through time series filtering to generate an eighth stable image; extract the key structure contours and map them to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate a ninth visualization image, and complete the tissue segmentation in the image.

[0086] Step S101: Obtain multispectral image data of the surgical area. Irradiate the fascia, ligament, blood vessels, nerves, and intervertebral disc tissue with a broadband light source, and generate a first resolution image by utilizing the transmission differences of different wavelengths of light by each tissue.

[0087] As an implementation manner in this embodiment, the process of obtaining multispectral image data of the surgical area, irradiating the fascia, ligament, blood vessels, nerves, and intervertebral disc tissue with a broadband light source, and generating a first resolution image by utilizing the transmission differences of different wavelengths of light by each tissue includes:

[0088] Perform denoising and enhancement processing on the multispectral image data to obtain a preliminary image;

[0089] First, extract the transmission characteristics of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissue from the preliminary image, and then perform boundary segmentation on the fascia tissue by using a wavelength difference threshold;

[0090] Extract the texture features of the fascia and ligament based on the fascia segmentation image, and judge the junction area between the two by using a support vector machine algorithm;

[0091] Analyze the change trend of the transmission characteristics of the blood vessels and nerves and the spectral characteristics of the intervertebral disc tissue, and judge the contact area between the intervertebral disc and the blood vessels and nerves by using a random forest algorithm to generate a multi-tissue distribution map as the first resolution image.

[0092] As an additional implementation manner in this embodiment, obtain multispectral image data of the surgical area. The multispectral image data includes image information of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissue. Perform denoising and enhancement processing on the multispectral image data to obtain a preliminary image. Extract the transmission characteristics of each tissue from the preliminary image, perform boundary segmentation on the fascia tissue by using a preset wavelength difference threshold to obtain a fascia segmentation image. For the fascia segmentation image, extract the texture features of the fascia tissue and the ligament tissue, and judge the junction area between the fascia tissue and the ligament tissue by using a support vector machine algorithm to obtain a junction distribution image. According to the junction distribution image, obtain the change trend of the transmission characteristics of the blood vessels and nerves in the fascia tissue to generate a blood vessel and nerve distribution map. Analyze the spectral characteristics of the intervertebral disc tissue by using a preset wavelength difference threshold to obtain an intervertebral disc tissue feature map. Extract the transmission characteristics from the intervertebral disc tissue feature map, and judge the contact area between the intervertebral disc tissue and the blood vessels and nerves by using a random forest algorithm to obtain a contact distribution image. According to the contact distribution image, obtain the spatial relationship of the fascia tissue, ligament tissue, and intervertebral disc tissue in the surgical area to generate a multi-tissue distribution map, and use the tissue distribution map as the first resolution image.

[0093] Exemplarily, obtaining multispectral image data of the surgical area is an important part of modern medical image processing. The multispectral technology captures the characteristics of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissue by light of different wavelengths.

[0094] As an optional implementation in this embodiment, near-infrared light can be used to distinguish the high transmittance of blood vessels, or mid-infrared light can be used to highlight the absorption characteristics of intervertebral disc tissue.

[0095] For example, assume that in spinal surgery, image data with a wavelength range of 400-1000 nanometers is collected, in which the fascia appears light gray and the blood vessels and nerves appear dark due to hemoglobin absorption.

[0096] As an optional implementation in this embodiment, the denoising and enhancement processing can use wavelet transform technology to remove high-frequency noise, and at the same time improve the visibility of tissues through contrast stretching.

[0097] As an optional implementation in this embodiment, discrete wavelet transform is applied to the noisy image, and the image is reconstructed after retaining the low-frequency signal to obtain a preliminary image with clearer fascia edges and about 30% fewer noise points. This processing can improve the accuracy of subsequent segmentation.

[0098] Specifically, when extracting the transmission characteristics from the preliminary image, the intensity changes of each tissue at different wavelengths can be analyzed. The fascia usually has a higher transmittance at 700 nanometers, while the ligament has a lower transmittance. A preset wavelength difference threshold, such as setting the transmittance difference to 0.2, is used to perform boundary segmentation on the fascia.

[0099] As an optional implementation in this embodiment, in the lumbar region, after the transmission difference between the fascia and the surrounding tissue exceeds the threshold, the segmented image shows that the fascia boundary error is only 1-2 mm, which improves the surgical positioning accuracy. To extract texture features from the fascia segmentation image, the gray level co-occurrence matrix can be used to calculate the roughness of the fascia and the fiber directionality of the ligament.

[0100] As an optional implementation in this embodiment, the fascia has a uniform texture and a correlation value of 0.8, while the ligament has a correlation value of 0.6 due to the fiber arrangement. The support vector machine algorithm determines the boundary area based on these features.

[0101] As an optional implementation in this embodiment, the junction distribution image shows that the width of the junction between the lumbar fascia and the ligament is about 3 mm, and the boundary transition is smooth, which is helpful for separation during surgery. According to the change trend of the transmission characteristics of blood vessels and nerves analyzed based on the junction distribution image, it can be observed that the transmittance of blood vessels at 850 nanometers decreases by 10%, and a distribution map is generated to show its winding path in the fascia.

[0102] As an optional implementation in this embodiment, in cervical spine surgery, the distribution map reveals that the blood vessels avoid the fascia center, which facilitates avoiding key areas.

[0103] As an alternative implementation in this embodiment, the spectral feature analysis of the intervertebral disc tissue can focus on its characteristic of high water content, with an obvious absorption peak at 950 nanometers. By setting the wavelength difference threshold to 0.15, the obtained feature map can clearly distinguish the intervertebral disc from the surrounding tissues.

[0104] As an alternative implementation in this embodiment, during arthroscopic knee surgery, the popliteal cyst (Baker's cyst) is anatomically adjacent to the popliteal vascular nerve bundle (popliteal artery, popliteal vein, and tibial nerve). The cyst often presents as a translucent cystic structure, while the vascular nerve bundle is a dark tubular tissue. Under traditional white light imaging, the contrast between the two is low, which easily leads to accidental injury to blood vessels and nerves during the operation. This embodiment uses the method disclosed in the present invention to achieve dynamic segmentation.

[0105] As an alternative implementation in this embodiment, the process of multispectral imaging and first-resolution image generation includes:

[0106] Broadband light source selection: A 400 - 1000 nm broadband light source is used, and the following bands are analyzed in detail:

[0107] 1. 520 - 550 nm (green light): The popliteal cyst has a transmittance > 70% due to its high proportion of mucoprotein; the blood vessels have a transmittance < 30% due to the absorption of hemoglobin.

[0108] 2. 800 - 850 nm (near-infrared): The transmittance of the vascular nerve bundle increases to 50%, and the transmittance of the cyst decreases to 40%.

[0109] It also includes denoising processing, including: performing wavelet threshold denoising on the original image to retain the characteristics of the fascia (transmittance 55%), cyst (70%), and vascular nerve (30%).

[0110] The example results include: In the first-resolution image, the cyst appears as a bright area (green channel), and the vascular nerve bundle is a dark red strip-shaped structure (near-infrared channel), with a preliminary separation error < 1 mm.

[0111] The implementation related to fluorescence labeling and the second enhanced image, and the fluorescence dye scheme includes:

[0112] Cyst labeling: Inject indocyanine green (ICG), which emits 835 nm fluorescence when excited at 805 nm, and specifically binds to the glycoprotein in the cyst;

[0113] Vascular nerve labeling: Intravenous injection of the patented fluorescent probe VF-8 (targeting vascular endothelial growth factor receptor), which emits 710 nm fluorescence when excited at 680 nm.

[0114] Based on the above scheme, the process of dynamic excitation includes:

[0115] Alternately use 680nm and 805nm excitation lights to capture vascular nerve (red) and cyst (near-infrared) signals respectively;

[0116] Separate overlapping signals through a convolutional neural network (U-Net architecture) to eliminate cross-interference.

[0117] Through the above operations, the enhancement effects obtained are as follows: the fluorescence signal intensity ratio of vascular nerves to cysts is increased from 1:1 to 5:1, and the boundary clarity is increased by 80%.

[0118] As an optional implementation method in this embodiment, during the dynamic light source optimization (adhesion scenario processing), the situation where the posterior wall of the cyst adheres to the popliteal vein and cannot be distinguished by traditional imaging may be encountered. Here, three optional optimization strategies are proposed, including:

[0119] 1. Red light band switching: Adjust to 620 - 650nm red light, with the peak vascular absorption rate (hemoglobin absorption coefficient μ_a = 0.3mm-1), and the cyst absorption rate μ_a = 0.1mm-1.

[0120] 2. Oblique illumination: Adjust the light source angle to 45°, and utilize the specular reflection characteristics of blood vessels to enhance the edge contrast.

[0121] 3. Morphological repair: Use a 3×3 circular kernel to perform closing operation on the broken boundary, and the connection rate is increased by 90%.

[0122] Through the optimization strategy, the integrity of the blood vessel contour in the adhesion area reaches 95%, and the measurement error of the cyst-vessel distance is <0.5mm.

[0123] As an optional implementation method in this embodiment, the three-dimensional reconstruction and intraoperative navigation specifically include:

[0124] 1. First, perform depth data fusion:

[0125] Obtain the depth map of the popliteal fossa area by laser scanning (accuracy 0.2mm), with the depth of the cyst surface being 3 - 5mm and the depth of vascular nerves being 6 - 8mm.

[0126] Register the fluorescence boundary (2D) with the depth data to generate three-dimensional coordinates.

[0127] 2. Next, perform real-time rendering:

[0128] Volume rendering parameters: The cyst is set to semi-transparent blue (Opacity = 0.3), and the vascular nerves are solid red (Opacity = 0.8).

[0129] Dynamic tracking: Use the optical flow algorithm (Lucas-Kanade method) to track the movement of the arthroscope, and the boundary jitter is reduced from ±3 pixels to ±0.5 pixels.

[0130] 3. Finally, intraoperative calibration was completed: During the operation, the distance between the cyst and the popliteal artery was shown to be 2.7 mm (the error was verified by postoperative MRI to be 0.3 mm).

[0131] As an alternative implementation in this embodiment, the L4-L5 intervertebral disc feature map shows sharp edges. After the transmission characteristics are extracted, the random forest algorithm judges the contact area with blood vessels and nerves, with an error of less than 1 mm, and the contact distribution image intuitively presents the fitting degree between the two. It can be understood and explained that a multi-tissue distribution map is generated according to the contact distribution image, and the spatial relationship of the fascia, ligament and intervertebral disc is clear at a glance.

[0132] As an alternative implementation in this embodiment, during scoliosis surgery, the first resolution image shows that the fascia covers the ligament above and the intervertebral disc is below, with clear spatial levels, providing a basis for preoperative planning. This method significantly improves the tissue recognition accuracy and reduces the risk of misoperation.

[0133] Step S102: For the first resolution image, adopt a preset fluorescent dye labeling scheme to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves at different wavelengths to obtain a second enhanced image.

[0134] As an implementation in this embodiment, the process of generating a second enhanced image by using a fluorescent dye labeling scheme to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves for the first resolution image includes:

[0135] Mark the fascia signal, ligament signal, and excitation signal of blood vessels and nerves in the first resolution image to generate a preliminary fluorescent image;

[0136] Adjust the wavelength to enhance the intensity distribution of the fluorescent signal and separate the fascia signal from the ligament signal;

[0137] Judge the boundary area between the fascia and the ligament through a convolutional neural network to generate a blood vessel distribution feature map;

[0138] Extract the excitation difference between blood vessels and nerves and ligament signals, analyze the spatial distribution characteristics of the fluorescent signal, and generate a second enhanced image.

[0139] As an additional implementation method in this embodiment, obtain a first-resolution image; for the first-resolution image, perform fluorescent dye labeling using a preset scheme to obtain fascia signals, ligament signals, and excitation signals of blood vessels and nerves; generate a preliminary fluorescence image based on the fascia signals, ligament signals, and excitation signals of blood vessels and nerves; for the preliminary fluorescence image, adjust the excitation signal using different wavelengths, extract the intensity distribution of the fluorescence signal, and obtain an enhanced fluorescence image; separate the fascia signal and the ligament signal from the enhanced fluorescence image; use a convolutional neural network to determine the boundary region between the fascia signal and the ligament signal, and obtain a boundary distribution image; based on the boundary distribution image, obtain the change trend of the fluorescence signal of blood vessels and nerves in the fascia signal, and generate a blood vessel distribution feature map; through the blood vessel distribution feature map, extract the difference in excitation signals at different wavelengths to obtain an overlapping region image of blood vessels and nerves and ligaments; for the overlapping region image, analyze the spatial distribution characteristics of the fluorescence signal using a preset scheme to generate a second enhanced image, as Figure 2 shown.

[0140] As an alternative implementation method in this embodiment, after obtaining the first-resolution image, the tissue characteristics can be further highlighted through fluorescent dye labeling.

[0141] As an alternative implementation method in this embodiment, green fluorescent dyes such as FITC can be selected for the fascia, which is excited at a wavelength of 488 nm and emits fluorescence at 515 nm, while red dyes such as TRITC can be used for the ligaments, which emit red signals when excited at 550 nm. Near-infrared dyes such as Cy5 are used for blood vessels and nerves, with an excitation wavelength of 650 nm. This labeling scheme uses the specific binding ability of the dyes to distinguish tissues. When generating a preliminary fluorescence image, the fascia presents a green fluorescent band, the ligaments are red texture areas, and the blood vessels and nerves show dark red paths.

[0142] As an alternative implementation method in this embodiment, adjusting the wavelength for the preliminary fluorescence image can optimize the signal. For example, the fascia signal is enhanced in the range of 500 - 520 nm. After filtering out stray light, the fluorescence intensity distribution is more concentrated, and the brightness of the fascia area in the image is increased by about 20%. The ligament signal is adjusted at 570 nm, the background noise is reduced, and the red area is more prominent. The enhanced fluorescence image can thus clearly reflect the boundaries of each tissue, facilitating subsequent separation.

[0143] Specifically, when separating the fascia signal and the ligament signal, it can be based on the differences in fluorescence color and intensity. The peak of the green signal of the fascia is at 515 nm, with an intensity of about 1000 units, while the peak of the red signal of the ligament is at 580 nm, with an intensity of 800 units. By setting the intensity threshold and wavelength range, the boundary of the separated fascia area is smooth, and the texture of the ligament area is retained intact. This separation lays the foundation for boundary judgment.

[0144] When a convolutional neural network determines the boundary region, an enhanced fluorescence image can be input, and the network extracts edge features through the convolutional layer. For example, in the scenario of lumbar spine surgery, the width of the color transition region at the junction of the fascia and the ligament is about 2 millimeters. The boundary distribution image generated after network recognition shows that the edge error of the transition region is only 0.5 millimeters.

[0145] During network training, diverse samples can be added to ensure the adaptability of boundary judgment to the tissue morphologies of different patients. By analyzing the fluorescence signal change trend of blood vessels and nerves based on the boundary distribution image, their distribution in the fascia can be observed. For example, in the range of 650 - 670 nanometers, the blood vessel signal intensity drops from 800 units to 600 units, reflecting its traversing path. In the generated blood vessel distribution feature map, the blood vessel path is serpentinely distributed in the fascia, and the feature map intuitively presents its spatial position, which helps to avoid key areas during the operation.

[0146] When extracting the signal difference at a specific wavelength from the blood vessel distribution feature map, 670 nanometers can be focused on. At this time, the difference in signal intensity between the blood vessel and the ligament reaches 300 units, and the overlapping region image shows that the width of their contact area is about 1 millimeter. This difference extraction helps to accurately distinguish the tissue overlapping parts and avoid confusion.

[0147] As an alternative implementation in this embodiment, for analyzing the spatial distribution characteristics of fluorescence signals in the overlapping region image, the grid division method can be used. Exemplarily, the image is divided into 10x10 grids, and the mean signal intensity in each grid is calculated. The intensity fluctuation range in the overlapping region of the blood vessel and the ligament is 200 - 500 units. In the generated second enhanced image, the boundary of the overlapping region is sharper, and the spatial sense of hierarchy is enhanced, which is convenient for intraoperative identification and operation planning. This method improves the practicality of the image by refining the signal distribution.

[0148] Step S103, extract the superimposed features of the fluorescence signal and the broadband light source signal from the second enhanced image, analyze the signal intensity and spatial distribution through the convolutional neural network algorithm, determine the demarcation line between the fascia and the ligament, and generate the third segmentation image.

[0149] As an implementation in this embodiment, the process of extracting the superimposed features of the fluorescence signal and the broadband light source signal from the second enhanced image, analyzing the signal intensity and spatial distribution through the convolutional neural network, and determining the demarcation line between the fascia and the ligament to generate the third segmentation image includes:

[0150] Use the convolutional neural network to extract the fluorescence signal intensity and spatial distribution information, and judge the boundary continuity between the fascia and the ligament;

[0151] Based on the fluorescence distribution characteristics at the fascia demarcation, separate the fascia feature map and adjust the spatial distribution weight of the broadband light source signal at the ligament demarcation;

[0152] The overlapping area of ​​the fluorescence signal and the broadband light source is extracted, and the independent areas of the fascia and ligament are determined by the difference characteristics to generate the third segmentation image.

[0153] As an additional implementation method in this embodiment, a second enhanced image is obtained, which includes the superposition characteristics of the fluorescence signal and the broadband light source; the second enhanced image is analyzed by using a convolutional neural network to obtain an analysis image, which includes the fluorescence signal intensity and spatial distribution information; based on the analysis image, the spatial distribution characteristics of the fascia boundary and the ligament boundary are obtained; the continuity of the boundary area between the fascia boundary and the ligament boundary is determined by the convolutional neural network to obtain a boundary enhanced image; based on the boundary enhanced image, the distribution characteristics of the fluorescence signal at the fascia boundary are extracted; the distribution characteristics are separated by using a preset threshold to obtain a fascia feature map; based on the fascia feature map, the broadband light source is obtained. The superposition characteristics of the signal at the ligament boundary; if the superposition characteristics exceed the preset range, the spatial distribution weight of the broadband light source signal is adjusted to obtain a ligament feature map; the overlapping area of ​​the fluorescence signal and the superposition characteristics is extracted from the ligament feature map; a convolutional neural network is used to analyze the spatial distribution of the overlapping area to obtain an overlapping distribution image; based on the overlapping distribution image, the difference characteristics of the fluorescence signal intensity and spatial distribution are obtained; by comparing the difference characteristics, the independent areas of fascia and ligament are determined to obtain a regional segmentation image; based on the regional segmentation image, the distribution characteristics of the fluorescence signal in the independent areas of fascia and ligament are extracted; the superposition characteristics of the fluorescence signal and the broadband light source are adjusted using a preset scheme to obtain a third segmentation image.

[0154] As an optional implementation in this embodiment, when acquiring the second enhanced image, the superposition characteristic of the fluorescence signal and the broadband light source can be achieved through multi-spectral imaging technology.

[0155] As an optional implementation in this embodiment, the fluorescent signal is emitted by a specific dye, such as green fluorescence to mark fascia, and the broadband light source provides natural light reflection from the background tissue. After the two are superimposed, the image retains the fluorescence specificity and contains the anatomical details of the broadband light source.

[0156] Specifically, the wavelength range of the broadband light source can be set to 400-700 nanometers, which works together with the fluorescence signal to form rich visual information. This superposition helps to distinguish tissue boundaries in subsequent analysis. When a convolutional neural network is used to analyze the second enhanced image, features can be extracted through multiple layers of convolution. The first layer of the network focuses on the spatial distribution of the fluorescence signal, and the second layer combines the signal intensity of the broadband light source to generate an analysis image. In the analysis image, the fluorescence signal intensity of the fascia may be displayed as 1200 units, with a spatial distribution in a band shape, while the ligament signal intensity is 900 units, distributed in blocks. This combination of intensity and distribution information lays the foundation for subsequent boundary extraction.

[0157] When obtaining the spatial distribution characteristics of the fascial boundary and the ligament boundary, the color and morphological differences in the analysis image can be used. The fascial boundary usually extends along the green fluorescent band with a width of about 3 mm, while the ligament boundary appears as a red block area with a width of about 5 mm. This characteristic reflects the natural differences in their anatomical structures and facilitates further processing.

[0158] As an alternative implementation in this embodiment, when judging the continuity of the boundary region between the fascial boundary and the ligament boundary, the convolutional neural network can run through the edge detection module.

[0159] As an alternative implementation in this embodiment, the network identifies the continuity at the junction of the fascia and the ligament to generate a boundary-enhanced image. In the boundary-enhanced image, the width of the transition zone in the junction area is reduced to 1 mm, and the boundary lines are smoother and clearer. This continuity judgment is crucial for subsequent feature extraction.

[0160] It should be noted that when extracting the distribution characteristics of the fluorescence signal at the fascial boundary, the signal changes within a specific wavelength range can be focused on. In the range of 510 - 520 nm, the fascial fluorescence signal shows a uniform distribution, and the intensity is stable at 1100 units. This distribution characteristic reflects the uniformity of the fascial tissue and provides a basis for separation.

[0161] As an alternative implementation in this embodiment, when separating the fascial feature map using a preset threshold, the intensity threshold can be set to 1000 units. The areas above this threshold are classified as fascia to generate a fascial feature map. In the map, the fascial area shows a clear green banded structure, and the background noise is significantly reduced. This separation method is simple and efficient.

[0162] Specifically, when obtaining the superimposition characteristics of the broadband light source signal at the ligament boundary, the light reflection intensity in the range of 400 - 600 nm can be analyzed. If the superimposition characteristics exceed the preset range, for example, the intensity exceeds 1500 units, then the spatial distribution weight of the broadband light source is adjusted to reduce the background interference and generate a ligament feature map. The ligament area is more prominent in the adjusted image. When extracting the overlapping area from the ligament feature map, the overlapping part of the fluorescence signal and the broadband light source can be focused on. At the wavelength of 550 nm, the overlapping width of the fluorescence signal and the broadband light source signal is about 2 mm. After the spatial distribution of the overlapping area is analyzed by the network, an overlapping distribution image is generated, and the boundary of the overlapping part is clearly visible in the image.

[0163] As an alternative implementation in this embodiment, when obtaining the difference characteristics of the fluorescence signal intensity and spatial distribution, the signal values in the fascia and ligament regions can be compared. The fascia signal intensity is 1200 units with a uniform distribution, while the ligament is 900 units with a concentrated distribution. This difference characteristic supports regional segmentation.

[0164] As an optional implementation in this embodiment, when determining the independent areas of fascia and ligament, a regional segmentation image is generated by comparing the difference characteristics. In the image, the fascia and ligament are independent of each other, and the boundary error is controlled within 0.8 mm. This segmentation result is intuitive and practical. When extracting the distribution characteristics of the fluorescence signal in the independent area, the fascia area signal shows a band-like extension, while the ligament area signal is block-like aggregation. This feature extraction helps to further optimize the image.

[0165] As an optional implementation in this embodiment, when adjusting the superposition characteristics of the fluorescent signal and the broadband light source, the contrast can be enhanced by a preset scheme. For example, the weight of the fluorescent signal is increased to 60%, and the weight of the broadband light source is reduced to 40%, and a third segmented image is generated. The image has a stronger sense of tissue hierarchy, which is convenient for subsequent applications.

[0166] Step S104, based on the third segmented image, combined with the anatomical spatial depth data collected in real time, the transmission difference is used to perform secondary feature extraction on the blood vessels, nerves and intervertebral disc tissues to obtain a fourth positioning image.

[0167] As an implementation method of this embodiment, when an overlapping signal in a sticky scene is detected, the process of optimizing the illumination angle and wavelength combination by a dynamic light source band adjustment algorithm to generate a fifth optimized image includes:

[0168] Detect overlapping signals in the adhesion scene, adjust the wavelength of the light source to the red light band and optimize the irradiation angle;

[0169] The edge detection algorithm is used to extract the boundary information of blood vessels, nerves and intervertebral disc tissues, and the broken boundaries are repaired through morphological operations to generate the fifth optimized image.

[0170] As an additional implementation method in this embodiment, a third segmented image is obtained, and the third segmented image includes anatomical spatial depth data collected in real time; for the third segmented image, a transmission difference analysis method is used to extract the initial feature distribution to obtain a preliminary distribution image; the superposition characteristics of the depth data and the transmission difference in the preliminary distribution image are analyzed by a convolutional neural network to determine the distribution area of ​​the blood vessels and nerves to obtain a blood vessel positioning map; based on the blood vessel positioning map, the feature distribution of the disc tissue under the transmission difference is extracted, and the boundary characteristics of the blood vessels, nerves and disc tissues are separated by a preset threshold to obtain a boundary distribution image; the overlapping area of ​​the spatial information and the feature distribution is obtained from the boundary distribution image, and the continuity of the overlapping area is judged to obtain an enhanced distribution image; for the enhanced distribution image, the spatial distribution weights of the blood vessels, nerves and disc tissues are adjusted by using the transmission difference to obtain a weight-adjusted image; by analyzing the weight-adjusted image, the independent regional distribution of the blood vessels, nerves and disc tissues in the anatomical space is obtained to obtain a regional division image; the difference characteristics of the feature distribution and the spatial information are extracted from the regional division image, and a fourth positioning image is generated by a clustering algorithm.

[0171] As an alternative implementation in this embodiment, when acquiring the third segmented image, anatomical space depth data can be collected by a real-time imaging device.

[0172] Using laser scanning technology, the depth data records the three-dimensional contour of the tissue surface in millimeters, with an accuracy of up to 0.5 mm. This method can provide rich spatial information for subsequent analysis. For the third segmented image, when using the transmission difference analysis method, the penetration depth difference of light in the tissue can be focused on. Due to the hemoglobin absorption characteristics, the transmission rate in the vascular region is relatively low, while due to the fibrous structure, the transmission rate in the intervertebral disc tissue is relatively high. A preliminary distribution image is thus generated, showing the initial characteristics of different tissues. When analyzing the preliminary distribution image through a convolutional neural network, the superimposed characteristics of depth data and transmission difference can be extracted.

[0173] As an alternative implementation in this embodiment, the first layer of the network identifies the depth contour, and the second layer analyzes the change in transmission rate to output the distribution area of blood vessels and nerves. For example, the depth change in the vascular region is gentle, and the transmission rate is below 20%, while the depth fluctuation in the intervertebral disc tissue is large, and the transmission rate is about 50%. This superimposed characteristic helps to generate a blood vessel localization map. It should be noted that when extracting the characteristics of the intervertebral disc tissue from the blood vessel localization map, the boundary region of the transmission difference can be focused on. Set the transmission rate threshold to 30%. Those below this value are classified as blood vessels and nerves, and those above this value are classified as intervertebral disc tissue to generate a boundary distribution image. When obtaining the overlapping area from the boundary distribution image, the overlapping part of the spatial information and the feature distribution can be analyzed.

[0174] Specifically, if the width of the overlapping area is about 2 mm, then through continuity judgment, it is determined whether it is a natural transition zone. An enhanced distribution image is thus generated, and the boundary of the overlapping area is clearer. When adjusting the weight for the enhanced distribution image, the spatial distribution ratio can be reallocated according to the transmission difference. For example, increase the weight of blood vessels and nerves to 70% and decrease the weight of intervertebral disc tissue to 30% to generate a weight-adjusted image, highlighting their respective characteristics. When obtaining the independent region distribution by analyzing the weight-adjusted image, it can be observed that the blood vessels and nerves show a linear extension, while the intervertebral disc tissue shows a sheet-like distribution. A region division image is thus formed, and the boundary error is controlled within 1 mm.

[0175] When extracting the difference characteristics from the region division image, the clustering algorithm can group based on depth and transmission rate. The depth change of blood vessels and nerves is less than 2 mm, and the transmission rate is low, while the depth change of intervertebral disc tissue reaches 5 mm, and the transmission rate is high. A fourth localization image is thus generated, and the tissue partition is more intuitive. This method improves the localization accuracy and provides a reliable basis for subsequent anatomical analysis.

[0176] As an alternative implementation, the combination of transmission difference analysis and depth data can effectively distinguish complex tissue structures. Real-time depth data reveals blood vessels hidden beneath the intervertebral disc, while transmission difference highlights their light absorption characteristics. The complementarity of the two enhances the accuracy of feature extraction. The application of the clustering algorithm makes the region division more automated, reduces human intervention, and improves efficiency. The algorithm automatically identifies the transmission rate gradient and quickly generates a positioning image. This combination of technologies has important value in anatomical research.

[0177] Step S105, for the fourth positioning image, when overlapping signals in the adhesion scenario are detected, optimize the irradiation angle and wavelength combination through the dynamic light source band adjustment algorithm, determine the boundary of the key structure, and generate the fifth optimized image.

[0178] As an implementation in this embodiment, obtain the fourth positioning image. For the fourth positioning image, use an image processing tool to detect overlapping signals in the adhesion scenario; according to the detected overlapping signals, adjust the irradiation angle and wavelength combination of the light source to obtain optimized lighting conditions; under the optimized lighting conditions, use an edge detection algorithm to determine the boundary of the key structure in the fourth positioning image to obtain the boundary determination result of the key structure; according to the boundary determination result, generate the fifth optimized image, as Figure 3 shown.

[0179] As an alternative implementation, when using an image processing tool to detect overlapping signals in the adhesion scenario for the fourth positioning image, the gray-scale analysis method can be used to identify the signal overlapping area.

[0180] Exemplarily, in the scenario of adhesion between blood vessels and intervertebral disc tissue, the gray value may have an abnormal transition band due to overlap. By setting a gray threshold, such as in the range of 50 - 100, the overlapping signal area can be preliminarily marked.

[0181] As an alternative implementation of this embodiment, after detecting overlapping signals in the adhesion scenario, the system automatically selects the optimal wavelength combination by analyzing the reflectance differences of different tissues for a specific wavelength in real time. The specific steps include:

[0182] Reflectance analysis: Perform multi-band reflectance sampling on the current image (at intervals of 50 nm in the range of 400 - 1000 nm) to establish a reflectance curve database for blood vessels, nerves, and intervertebral disc tissue.

[0183] Dynamic matching: When an adhesion area is detected, compare the reflectance differences of each band in the database (for example, at 650 nm, the blood vessel reflectance < 20%, and the intervertebral disc reflectance > 60%), and select the band with the largest reflectance difference (such as the combination of 650 nm and 950 nm).

[0184] Angle optimization: Adjust the incident angle of the light source according to the surface curvature of the tissue (calculated from depth data), for example, adjust from vertical incidence to 30° oblique incidence, to reduce the interference of specular reflection.

[0185] Experimental data shows that after adopting this strategy, the contrast of the adhesion area is increased by 40% (from the original 0.3 to 0.42).

[0186] In spinal surgery simulation, the recognition error of the boundary between blood vessels and intervertebral discs is reduced from 2.1 mm to 0.8 mm.

[0187] As an alternative implementation, use a Gaussian filter to smooth the image noise, and then enhance the local contrast to highlight the signal characteristics of the adhesion site.

[0188] If the gray value of normal blood vessels is about 30 and that of intervertebral disc tissue is 80, the overlapping area may present an intermediate value of about 50. This method helps to accurately locate the adhesion points. When adjusting the irradiation angle and wavelength combination of the light source according to the detected overlapping signal, the imaging effect can be optimized from the characteristics of the light source. For example, aiming at the high absorption characteristics of blood vessels, the light source wavelength can be adjusted to the red light range, such as adjusting the wavelength to the red light band of 650 - 700 nm, such as 650 nm, and at the same time changing the irradiation angle from vertical to 45 degrees to reduce the interference of surface reflection.

[0189] As an alternative implementation, if the original light source is white light and the angle is vertical, the overlapping signal may be blurred due to diffuse reflection; after adjustment, the red light penetrates deeper into the blood vessels, and the reflection of the intervertebral disc tissue is enhanced, and the contrast between the two is increased.

[0190] This optimization can effectively separate the signal characteristics of the adhesion area. Under the optimized lighting conditions, when using an edge detection algorithm to judge the boundary of key structures, the Sobel operator can be selected to process the image gradient.

[0191] Aiming at the linear distribution of blood vessels, the operator can detect sudden changes in the horizontal gradient, such as areas where the depth change is less than 2 mm; while the sheet-like characteristics of intervertebral disc tissue are manifested as obvious vertical gradients. The significant boundaries can also be screened by setting a gradient threshold, such as 20, to avoid noise interference. This method can clearly outline the edge contours of blood vessels and intervertebral disc tissue, providing an accurate basis for subsequent analysis. When generating the fifth optimized image according to the boundary judgment result, morphological operations can be combined to improve the image.

[0192] As an alternative implementation in this embodiment, use dilation and erosion to process the boundary noise. For example, after expanding the boundary width from 1 mm to a thicker area and then shrinking it back to the original range, isolated points are removed.

[0193] If the blood vessel boundary is broken due to adhesion, morphological operations can connect the breakpoints and enhance its continuity. For example, the original boundary may have a 0.5-mm gap, which is completely closed after processing. This method can improve the readability of the image and make the distribution of key structures more intuitive.

[0194] As an alternative implementation in this embodiment, a coherent process is formed from overlapping signal detection to light source adjustment, then to edge detection and image optimization.

[0195] For example, in the adhesion scenario, the boundary between the blood vessel and the intervertebral disc is initially mixed. By illuminating obliquely with red light, the difference is highlighted. Edge detection locks the contour, and finally morphological operations repair the details. The finally generated fifth optimized image can clearly show their respective regions. This progressive logic ensures the integrity of the solution and supports the reliability of the technical implementation through multi-faceted analysis. The optimized image provides a higher-quality data basis for subsequent anatomical research.

[0196] Step S106: Obtain the spatial coordinates of the blood vessel nerves and the intervertebral disc tissue from the fifth optimized image, and fuse the boundary information and depth data through real-time visualization rendering technology to obtain the sixth dynamic image.

[0197] As an implementation in this embodiment, the process of fusing the spatial coordinates of the blood vessel nerves and the intervertebral disc tissue with the depth data and generating the sixth dynamic image through real-time visualization rendering includes:

[0198] Extract the spatial coordinates of the blood vessel nerves and the intervertebral disc tissue based on the fifth optimized image and map them to a three-dimensional coordinate system through depth data;

[0199] Fuse the boundary information and the anatomical space depth distribution, and use interpolation methods to smooth the mutation regions of the spatial coordinates;

[0200] Enhance the spatial hierarchy of the blood vessel nerves and the intervertebral disc tissue through light and shadow distribution adjustment to generate the sixth dynamic image.

[0201] As an additional implementation manner in this embodiment, obtain the fifth optimized image, extract the position coordinates of blood vessels, nerves and intervertebral disc tissues from the fifth optimized image by using a segmentation algorithm to obtain a preliminary positioning result; for the preliminary positioning result, obtain the corresponding relationship between boundary data and depth information, and fuse the boundary data and depth information through real-time rendering technology to generate a dynamic intermediate image; detect the spatial trend of blood vessels and nerves from the dynamic intermediate image, and if an abnormal curvature is detected, adjust the rendering perspective to obtain a perspective-optimized image; according to the perspective-optimized image, extract the depth distribution characteristics of the intervertebral disc tissue, and use an interpolation method to smooth the depth distribution characteristics to generate a smoothed image; for the smoothed image, fuse the boundary data and the smoothed depth information, and adjust the light and shadow distribution through rendering technology to obtain a light and shadow enhanced image; extract the spatial relationship between blood vessels, nerves and intervertebral disc tissues from the light and shadow enhanced image, and use a clustering algorithm to judge the boundary between tissues to generate a sixth dynamic image.

[0202] As an alternative implementation manner in this embodiment, when extracting the position coordinates of blood vessels, nerves and intervertebral disc tissues from the fifth optimized image, an image feature can be processed by using a segmentation algorithm.

[0203] The segmentation method based on region growing can identify pixel points with similar gray values and gradually expand to the complete region. For example, if the gray value of the blood vessel region is concentrated between 20-40, while the intervertebral disc tissue is 70-90, the position coordinates of both can be extracted by setting seed points and defining a growth threshold such as 15. This method can quickly lock the target region and is especially suitable for scenarios with clear tissue boundaries.

[0204] As an alternative implementation manner in this embodiment, when obtaining the corresponding relationship between boundary data and depth information, the mapping between depth sensor data and image pixel points can be utilized.

[0205] Specifically, assume that the depth sensor records the depth of the blood vessel surface as 3 mm, while the intervertebral disc tissue is 5 mm. By matching the pixel coordinates with the depth values, a correspondence table between the boundary and the depth is formed. The real-time rendering technology then fuses these data to generate a dynamic intermediate image, dynamically displaying the spatial hierarchy of the tissues. When detecting the spatial trend of blood vessels and nerves from the dynamic intermediate image, the curvature can be analyzed through curve fitting.

[0206] If the curvature radius of a certain section of blood vessel is less than 1 mm, it may indicate an abnormal bend. At this time, adjust the rendering perspective to 45 degrees laterally to generate a perspective-optimized image. This adjustment can highlight the spatial details of the abnormal part for subsequent analysis. When extracting the depth distribution characteristics of the intervertebral disc tissue for the perspective-optimized image, the interpolation method can be used for the smooth transition region.

[0207] If the depth value of a certain area jumps from 4 mm to 6 mm, the intermediate value such as 5 mm is filled by linear interpolation to form a continuous depth distribution curve. After generating the smoothed image, the depth change of the tissue is more natural, which helps to accurately analyze.

[0208] As an alternative implementation in this embodiment, when fusing the boundary data and the smoothed depth information, the light and shadow distribution can be adjusted by rendering technology.

[0209] As an alternative implementation in this embodiment, if the light source irradiates from the top, the shadow may cover the boundary details. At this time, it is adjusted to an oblique light source such as a 30-degree angle to generate a light and shadow enhanced image. The improvement of the light and shadow contrast can better distinguish the spatial levels of blood vessels and intervertebral disc tissues. When extracting the spatial relationship from the light and shadow enhanced image, the clustering algorithm can be used to judge the boundary between tissues.

[0210] Specifically, based on K-means clustering, the pixel points are grouped according to the gray scale and depth features. For example, blood vessels are grouped into one category, and intervertebral disc tissues are grouped into another category, and the boundary is located at the junction of the two categories. This method can effectively divide complex areas. After generating the sixth dynamic image, the spatial distribution of the tissue is more intuitive.

[0211] From the coordinate extraction of segmentation to the generation of the dynamic image by clustering, a set of coherent processes are formed. For the depth difference between blood vessels and intervertebral disc tissues, interpolation smoothing and light and shadow adjustment enhance the visualization of features, while clustering further clarifies the boundary. This progressive logic ensures the integrity of the scheme and provides a reliable basis for subsequent analysis.

[0212] Step S107, according to the sixth dynamic image, classify the pixel distributions of fascia, ligament, blood vessels, nerves and intervertebral disc tissues by using a preset threshold, judge the relative positions of each tissue in the anatomical space, and generate a seventh classification image.

[0213] As an implementation in this embodiment, the process of classifying the tissue pixel distribution based on the sixth dynamic image includes:

[0214] Preliminarily classify the pixel gray scale values of fascia, ligament, blood vessels, nerves and intervertebral disc tissues by using a preset threshold;

[0215] Analyze the spatial features of the pixel distribution by the clustering algorithm to judge the relative positions of each tissue in the anatomical space;

[0216] If an overlapping area of the pixel distribution is detected, use the interpolation method to adjust the classification boundary to generate a seventh classification image.

[0217] As an additional implementation method in this embodiment, obtain the sixth dynamic image, perform a preliminary classification on the pixel distribution of the sixth dynamic image through a preset threshold to obtain a preliminary classification result; for the preliminary classification result, obtain the spatial features of the fascia distribution and ligament distribution, and use a clustering algorithm to judge the boundary of the spatial features to obtain a boundary division image; extract the spatial trend of blood vessels and nerves from the boundary division image, and fuse the spatial trend with depth information to obtain a trend enhancement image; according to the trend enhancement image, obtain the distribution area of the intervertebral disc tissue. If the distribution areas overlap, adjust the area boundary through an interpolation method to obtain an area optimization image; for the area optimization image, fuse the corresponding relationship between the fascia distribution and the anatomical space, and adjust the perspective distribution through a rendering technique to obtain a perspective adjustment image; extract the relative position information from the perspective adjustment image, and use a judgment logic to verify the relative position to obtain a position verification image; according to the position verification image, adjust the tissue boundary through a smoothing technique to obtain a seventh classification image.

[0218] As an alternative implementation method in this embodiment, when performing a preliminary classification on the pixel distribution from the sixth dynamic image through a preset threshold, the standard can be set according to the gray value range. For example, assume that the gray value of the blood vessel area is between 30 - 50, while that of the intervertebral disc tissue is between 80 - 100. Setting the threshold to 60 can initially distinguish the two types of pixels. This method is simple and intuitive, and can quickly divide the image into different regions for subsequent analysis. The selection of the threshold can be adjusted according to the actual contrast of the image to adapt to the lighting conditions of different scenarios. When obtaining the spatial features of the fascia distribution and ligament distribution for the preliminary classification result, the features can be extracted through an edge detection method.

[0219] Specifically, the fascia usually presents a relatively smooth linear distribution, while the ligament may show a relatively rough blocky structure. If the gray difference between the edge pixels of the fascia is less than 10, while that of the ligament area is greater than 20, the spatial forms of the two can be initially identified. When using a clustering algorithm such as K - means to judge the boundary, the pixels can be grouped according to gray value and position. The fascia is grouped into one category, and the ligament is grouped into another category, and the boundary naturally appears at the junction of the two categories. This method can clearly divide complex regions. When extracting the spatial trend of blood vessels and nerves from the boundary division image, the direction gradient can be used to analyze its extension trend.

[0220] As an alternative implementation method in this embodiment, if the pixel direction change of a certain section of blood vessel is less than 15 degrees, its trend is considered stable; if the change exceeds 30 degrees, it may indicate a bend. When fusing depth information, assume that the surface depth of the blood vessel is 2 mm and the deep layer is 4 mm. By matching the depth and trend, a trend enhancement image can be generated. This enhancement can highlight the spatial continuity of the blood vessel. When obtaining the distribution area of the intervertebral disc tissue according to the trend enhancement image, if area overlap is found, it can be adjusted through an interpolation method.

[0221] For example, if the boundary of a certain area jumps from coordinate x = 10 to x = 15, x = 12 and x = 13 can be inserted as transition points to form a smooth boundary. The region-optimized image thus conforms better to anatomical logic. When fusing the fascia distribution with the corresponding relationship of the anatomical space, the rendering perspective can be adjusted. The initial perspective is a frontal view. If the fascia details are blocked, the perspective can be switched to 30 degrees laterally to generate a perspective-adjusted image. This adjustment can better display the spatial hierarchy of the tissue. When extracting the relative position information from the perspective-adjusted image, the relationship can be judged by the distance between pixels.

[0222] As an optional implementation manner in this embodiment, if the closest distance between a blood vessel and the intervertebral disc tissue is less than 5 pixels, it may indicate an adjacent relationship. When verifying using a judgment logic, a rule can be set: a distance less than 3 pixels is considered overlapping, and a distance greater than 10 pixels is considered separated, to generate a position verification image. This verification ensures the accuracy of the position relationship. When adjusting the tissue boundary through a smoothing technique based on the position verification image, mean filtering can be used to process the boundary pixels.

[0223] For example, if the gray value of a certain boundary point is 50, and the adjacent points are 45 and 55, then the average value 50 is taken for smooth transition. The seventh classified image thus has a more natural boundary and a more intuitive tissue distribution. This smoothing process can reduce noise interference, improve the image quality, and provide reliable support for subsequent anatomical analysis.

[0224] Step S108, for the seventh classified image, use the optical flow algorithm to analyze the change trend of the tissue boundary in consecutive frames, and smooth the boundary jitter through time series filtering technology to obtain the eighth stable image.

[0225] As an implementation manner in this embodiment, the process of using the optical flow algorithm to analyze the change trend of the tissue boundary and generating the eighth stable image by smoothing the jitter through time series filtering includes:

[0226] Extract the displacement change data of the tissue boundary in the seventh classified image, and calculate the pixel motion vector between consecutive frames;

[0227] Eliminate the random jitter of the displacement data through time series filtering to generate a smoothed boundary change curve;

[0228] Based on the smoothed data, correct the spatial position of the tissue boundary to generate the eighth stable image.

[0229] As an additional implementation manner in this embodiment, a seventh classified image is obtained. For the seventh classified image, an optical flow algorithm is used to analyze the displacement change of the tissue boundary in consecutive frames to obtain boundary change data; the feature distribution of the displacement change is extracted from the boundary change data, and the change trend is determined through time series analysis to obtain trend distribution data; for the trend distribution data, filtering processing is performed to smooth the boundary jitter to obtain smoothed boundary data; according to the smoothed boundary data, the stable position information of the tissue boundary in consecutive frames is obtained to obtain position stability data; the dynamic features of the time series are fused from the position stability data, and the boundary display is adjusted through rendering technology to obtain a boundary-optimized image; for the boundary-optimized image, it is judged whether the stable result meets a preset threshold. If not, the boundary position is adjusted through an interpolation method to obtain an eighth stable image.

[0230] As an alternative implementation manner in this embodiment, the optical flow algorithm is a method for analyzing the pixel motion in consecutive frames. For the seventh classified image, the dynamic change of the tissue boundary can be obtained by calculating the displacement vector of pixels between adjacent frames.

[0231] In anatomical images, the fascia boundary may be displaced due to breathing or slight movement. Exemplarily, if a boundary point is located at coordinates x = 100, y = 200 in the first frame and moves to x = 102, y = 201 in the second frame, the optical flow algorithm can capture a displacement of 2 pixels horizontally and 1 pixel vertically, generating boundary change data. This method can intuitively reflect the minute movement of the tissue, facilitating subsequent dynamic analysis. When extracting the feature distribution from the boundary change data, time series analysis can reveal the regular trend of the displacement. Specifically, the displacement values of a certain boundary point within 10 frames can be statistically analyzed, such as 2, 3, 2, 4, 3 pixels, and its mean value and fluctuation range can be analyzed to determine whether it shows periodic changes.

[0232] As an alternative implementation manner in this embodiment, if the displacement value fluctuates around 3 pixels, it can be considered that the boundary change is stable, and trend distribution data is generated. This trend helps to understand the dynamic behavior of the tissue. For the trend distribution data, filtering processing can effectively smooth the boundary jitter. For example, if the displacement sequence of the boundary point is 2, 5, 3, 6, 4 pixels, after using mean filtering, it can be adjusted to 3, 3.3, 4, 4.3 pixels, removing the mutation points.

[0233] As an alternative implementation in this embodiment, key feature points are selected in the seventh-classification image (such as the corner points at the junction of fascia and ligament, detected using the Shi-Tomasi algorithm). The Lucas-Kanade algorithm is used to calculate the displacement vectors of feature points between adjacent frames (window size 15×15 pixels, maximum number of iterations 20 times). A motion field is generated based on the displacement vectors, and reverse compensation is performed on the tissue boundary (such as pushing back the boundary points with displacement > 2 pixels to the position of the previous frame). Under the interference of respiratory motion, the boundary jitter amplitude is reduced from ±3 pixels to ±0.5 pixels. The algorithm has high computational efficiency, and the single-frame processing time < 10 ms (1080p resolution).

[0234] As an alternative implementation in this embodiment, an improved RAFT (Recurrent All-Pairs Field Transforms) architecture is adopted. The input is the seventh-classification images of 5 consecutive frames, and the output is a dense optical flow field. A lightweight model is deployed (parameter quantity < 5M) to calculate the motion trend of the tissue boundary in real time. The optical flow prediction result is combined with Kalman filtering to smooth the mutation noise (such as enabling filtering correction when the single-frame displacement > 5 pixels).

[0235] As an alternative implementation in this embodiment, if the jitter amplitude is less than 1 pixel, the smoothing effect can be considered ideal, and smooth boundary data is obtained. This kind of smoothing can improve the boundary continuity and facilitate position analysis. When obtaining stable position information based on the smooth boundary data, the core position of the boundary can be determined by multi-frame averaging.

[0236] As an alternative implementation in this embodiment, if the positions of a certain boundary point in 5 frames are x = 100, 101, 100, 102, 101 respectively, the average value x = 101 is taken as the stable position to generate position-stable data. This method can reduce random interference and highlight the true distribution of tissues. When fusing the dynamic features of the time series, the rendering technology can enhance the visual performance of the boundary. If the displacement range of the fascia boundary within 5 frames is 1 - 3 pixels, its motion trajectory can be rendered by color gradient to generate an optimized boundary image. This kind of dynamic rendering can clearly show the spatial change trend of tissues and is convenient for anatomical observation. When judging whether the stable result of the optimized boundary image meets the preset threshold, a standard can be set, such as a displacement fluctuation less than 2 pixels is considered stable. Exemplarily, if the displacement of a certain boundary area is 1, 1.5, 1.2 pixels, it meets the requirements and no adjustment is needed; if it is 3, 4, 2 pixels, exceeding the threshold, interpolation adjustment is required. If the boundary jumps from x = 10 to x = 14, x = 12 can be inserted as a transition point to generate the eighth stable image. This adjustment can optimize the naturalness of the boundary and improve the reliability of anatomical analysis.

[0237] Step S109, extract the final key structural contours from the eighth stable image, and map the spatial relationships of the fascia, ligaments, blood vessels, nerves, and intervertebral disc tissues to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate a ninth visualization image.

[0238] As an implementation manner in this embodiment, the process of extracting key structural contours, mapping them to a three-dimensional anatomical model through three-dimensional reconstruction technology, and generating a ninth visualization image to complete tissue segmentation in the image includes:

[0239] Extract the contour data of the fascia, ligaments, blood vessels, nerves, and intervertebral disc tissues from the eighth stable image, and analyze their spatial positions through stereological methods;

[0240] Fuse the anatomical space depth data and contour data, and use interpolation technology to adjust the boundary details of the three-dimensional model;

[0241] Generate a three-dimensional visualization image based on volume rendering technology to generate a ninth visualization image.

[0242] As an additional implementation manner in this embodiment, obtain the eighth stable image, extract the key structural contour data from the eighth stable image to obtain contour distribution information; for the contour distribution information, use stereological methods to analyze the spatial positions of the fascia tissue and ligament distribution to obtain preliminary spatial data; according to the preliminary spatial data, fuse the anatomical features of blood vessels, nerves, and intervertebral disc tissues to generate three-dimensional spatial relationship data; for the three-dimensional spatial relationship data, use interpolation technology to adjust the boundary details of the fascia tissue and ligament distribution to obtain optimized spatial data; extract the rendering parameters of the three-dimensional anatomical model from the optimized spatial data, and generate a preliminary visualization image through volume rendering technology; for the preliminary visualization image, determine whether the pixel distribution exceeds a preset threshold. If it exceeds the preset threshold, adjust the image details through smoothing filtering processing to obtain a smoothed visualization image; according to the smoothed visualization image, fuse the time series features of the key structures to generate a ninth visualization image.

[0243] As an alternative implementation manner in this embodiment, when extracting key structural contour data from the eighth stable image, the main boundaries of the fascia and ligaments can be identified through edge detection technology.

[0244] In anatomical images, the fascia edge may appear as a line area with a relatively high gray value. If the pixel value of the fascia boundary in an image is 150 and the background area is 50, a threshold of 100 can be set to extract the contour and generate contour distribution information. This method can clearly separate key structures and facilitate subsequent analysis. When using stereological methods to analyze the spatial positions of the fascia tissue and ligament distribution for the contour distribution information, grid division technology can be used.

[0245] Specifically, the image is divided into a 10×10 grid, and the pixel proportion of fascia and ligaments in each grid is counted. In a certain grid, the fascia accounts for 60% and the ligaments account for 30%. By superimposing multiple frames of data, preliminary spatial data is obtained. This grid-based analysis can reflect the spatial distribution characteristics of tissues. When fusing the anatomical features of blood vessels, nerves, and intervertebral disc tissues based on the preliminary spatial data, it can be achieved through feature matching technology.

[0246] As an alternative implementation in this embodiment, if the fascia boundary point is located at x = 50 and y = 60, and the nearby blood vessel position is x = 52 and y = 62, the two can be associated to generate three-dimensional spatial relationship data. This fusion can fully present the relative positions between anatomical structures. When using interpolation technology to adjust the boundary details for the three-dimensional spatial relationship data, linear interpolation can be used to optimize the boundary transition. If the fascia boundary jumps from x = 10 to x = 14, x = 12 can be inserted as an intermediate point to form a smooth curve, obtaining optimized spatial data. This adjustment can enhance the natural connection of the structure. When extracting rendering parameters from the optimized spatial data, colors and transparencies can be assigned according to tissue types. The fascia is set to semi-transparent blue and the ligaments are set to solid yellow, and a preliminary visualization image is generated through volume rendering technology. This parameter setting can highlight the visual differences between different tissues. When judging whether the pixel distribution of the preliminary visualization image exceeds a preset threshold, a pixel value fluctuation range such as 50 within 0 - 255 can be set as the boundary.

[0247] As an alternative implementation in this embodiment, if the pixel value in a certain area suddenly changes from 100 to 160 and exceeds the threshold, it is adjusted to 120 through mean filtering to obtain a smooth visualization image. This processing can reduce noise interference. When fusing time series features based on the smooth visualization image, dynamic displacement data can be introduced to render the motion trajectory.

[0248] As an alternative implementation in this embodiment, if the fascia boundary displaces 1 - 2 pixels within 5 frames, the trajectory can be marked with a gradient color to generate the ninth visualization image. This dynamic fusion can intuitively display the spatio-temporal changes of tissues and improve the accuracy of anatomical analysis.

[0249] As an implementation in this embodiment, the implementation process of volume rendering technology includes:

[0250] Adjust the light projection angle according to the anatomical space coordinates of the three-dimensional model to enhance the spatial hierarchical contrast between blood vessels, nerves, and intervertebral disc tissues;

[0251] Use smoothing filtering to eliminate pixel noise in the three-dimensional visualization image and optimize the contour clarity of key structures.

[0252] Based on this, an organization dynamic segmentation method based on transmission difference provided by an embodiment of the present invention discloses a multi-spectral imaging method for identifying key structures in a surgical area. This method obtains multi-spectral images through broadband light source irradiation and fluorescent dye labeling, analyzes tissue features using a convolutional neural network, and combines anatomical space depth data to achieve precise segmentation of fascia, ligaments, blood vessels, nerves, and intervertebral disc tissues. For the adhesion scenario, the present invention adopts a dynamic light source band adjustment algorithm to optimize the imaging effect. Through real-time visualization rendering and three-dimensional reconstruction technology, the spatial relationship of key structures is mapped to a three-dimensional anatomical model. Finally, an optical flow algorithm and time series filtering technology are used to achieve stable tracking of tissue boundaries. The present invention can accurately identify and locate key anatomical structures in the image in a complex surgical environment, enhancing the accuracy of the image.

[0253] Embodiment 2

[0254] In this embodiment, a computer terminal device is provided, including:

[0255] One or more processors;

[0256] A memory, coupled to the processor, for storing one or more programs;

[0257] When one or more programs are executed by one or more processors, the one or more processors implement the methods in the above embodiments.

[0258] In this embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the methods in the above embodiments are implemented.

[0259] In this embodiment, an electronic device is further provided, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the methods in the above embodiments.

[0260] The above program can run in a processor or can also be stored in a memory (or referred to as a computer-readable medium). A computer-readable medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0261] These computer programs can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps for the functions specified in one process Figure 1 one process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one box or multiple boxes can be implemented by different modules corresponding to different steps.

[0262] In this embodiment, such a device is provided. The device is called a tissue dynamic segmentation device based on transmission difference, as Figure 4 shown, and specifically includes:

[0263] A multispectral imaging module for acquiring multispectral image data of the surgical area, irradiating fascia, ligaments, blood vessels, nerves, and disc tissues with a broadband light source, and generating a first resolution image by using the transmission difference of each tissue for different wavelengths of light;

[0264] A fluorescence labeling module for exciting the fluorescence signals of fascia, ligaments, and blood vessels and nerves in the first resolution image to generate a second enhanced image;

[0265] A feature analysis module for extracting the superimposed features of the fluorescence signal and the broadband light source signal from the second enhanced image, analyzing the signal intensity and spatial distribution through a convolutional neural network, and determining the boundary line between fascia and ligaments to generate a third segmentation image;

[0266] A secondary positioning module for combining real-time anatomical space depth data and performing secondary feature extraction on blood vessels, nerves, and disc tissues by using transmission difference to generate a fourth positioning image;

[0267] A dynamic light source optimization module, which is used to generate a fifth optimized image by adjusting the light source wavelength and irradiation angle when overlapping signals in an adhesion scene are detected;

[0268] A visualization rendering module, which is used to fuse the spatial coordinates and depth data of blood vessels, nerves and intervertebral disc tissues to generate a sixth dynamic image;

[0269] A pixel classification module, which is used to classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image;

[0270] An optical flow tracking module, which is used to analyze the change trend of tissue boundaries and smooth jitters through time series filtering to generate an eighth stable image;

[0271] A three-dimensional reconstruction module, which is used to extract key structural contours and map them to a three-dimensional anatomical model to generate a ninth visualization image.

[0272] As an implementation method in this embodiment, the multispectral imaging module includes:

[0273] An image denoising unit, which is used to perform denoising and enhancement processing on multispectral image data to obtain a preliminary image;

[0274] A transmission characteristic extraction unit, which is used to extract the transmission characteristics of fascia, ligaments, blood vessels, nerves and intervertebral disc tissues from the preliminary image;

[0275] A fascia segmentation unit, which is used to segment the boundary of fascia tissue using a wavelength difference threshold;

[0276] A texture analysis unit, which is used to extract the texture features of fascia and ligaments based on the fascia segmentation image and judge the junction area between the two through a support vector machine algorithm;

[0277] A contact analysis unit, which is used to analyze the change trend of the transmission characteristics of blood vessels and nerves and the spectral characteristics of intervertebral disc tissues, and judge the contact area through a random forest algorithm.

[0278] As an implementation method in this embodiment, the fluorescence labeling module includes:

[0279] A signal generation unit, which is used to label the excitation signals of fascia signals, ligament signals and blood vessels and nerves in the first resolution image to generate a preliminary fluorescence image;

[0280] A wavelength adjustment unit, which is used to adjust a specific wavelength to enhance the intensity distribution of the fluorescence signal;

[0281] A signal separation unit, which is used to separate fascia signals and ligament signals;

[0282] A boundary determination unit, which is used to judge the boundary area between fascia and ligaments through a convolutional neural network;

[0283] A distribution analysis unit for extracting the excitation differences between vascular nerves and ligament signals and analyzing the spatial distribution characteristics.

[0284] As an implementation manner in this embodiment, the feature analysis module includes:

[0285] A superimposed feature extraction unit for extracting the superimposed features of fluorescence signals and broadband light source signals;

[0286] A boundary continuity analysis unit for judging the boundary continuity between fascia and ligaments through a convolutional neural network;

[0287] A feature separation unit for separating the fascia feature map based on the fluorescence distribution characteristics at the fascia boundary;

[0288] A weight adjustment unit for adjusting the spatial distribution weight of the broadband light source signal at the ligament boundary;

[0289] A difference determination unit for determining independent regions through the difference characteristics between fluorescence signals and broadband light sources.

[0290] As an implementation manner in this embodiment, the dynamic light source optimization module includes:

[0291] An overlapping signal detection unit for detecting overlapping signals in adhesion scenarios;

[0292] A wavelength optimization unit for adjusting the light source wavelength to the red light band and optimizing the irradiation angle;

[0293] An edge repair unit for repairing the broken boundaries between vascular nerves and intervertebral disc tissues through morphological operations.

[0294] As an implementation manner in this embodiment, the visualization rendering module includes:

[0295] A coordinate mapping unit for extracting the spatial coordinates of vascular nerves and intervertebral disc tissues and mapping them to a three-dimensional coordinate system;

[0296] A data fusion unit for fusing boundary information and anatomical spatial depth distribution;

[0297] A mutation smoothing unit for smoothing the mutation regions of spatial coordinates through an interpolation method;

[0298] A light and shadow enhancement unit for adjusting the light and shadow distribution to enhance the spatial hierarchy.

[0299] As an implementation manner in this embodiment, the pixel classification module includes:

[0300] A grayscale classification unit for preliminarily classifying the pixel grayscale values of fascia, ligaments, vascular nerves, and intervertebral disc tissues through a preset threshold;

[0301] A spatial feature analysis unit for analyzing the spatial features of pixel distribution through a clustering algorithm;

[0302] A boundary correction unit for adjusting the classification boundary by using an interpolation method when detecting an overlapping area of pixel distribution.

[0303] As an implementation manner in this embodiment, the optical flow tracking module includes:

[0304] A displacement calculation unit for extracting the displacement change data of the tissue boundary in the seventh classification image and calculating the pixel motion vector between consecutive frames;

[0305] A jitter elimination unit for eliminating the random jitter of the displacement data through time series filtering;

[0306] A position correction unit for correcting the spatial position of the tissue boundary based on the smoothed data.

[0307] As an implementation manner in this embodiment, the 3D reconstruction module includes:

[0308] A contour extraction unit for extracting the contour data of fascia, ligament, blood vessel nerve and intervertebral disc tissues from the eighth stable image;

[0309] A spatial positioning unit for analyzing its spatial positioning through stereology methods;

[0310] A model optimization unit for fusing the anatomical space depth data and adjusting the boundary details of the 3D model by using interpolation techniques;

[0311] A volume rendering unit for generating a 3D visualization image.

[0312] As an implementation manner in this embodiment, the volume rendering unit includes:

[0313] A light source projection unit for adjusting the light source projection angle according to the anatomical space coordinates of the 3D model;

[0314] A contrast enhancement unit for enhancing the spatial level contrast between blood vessel nerves and intervertebral disc tissues;

[0315] A noise elimination unit for eliminating pixel noise in the 3D visualization image through smoothing filtering.

[0316] The device is used to implement the functions of the method in the above embodiment. Each module in the device corresponds to each step in the method, and those that have been described in the method will not be repeated here.

[0317] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for dynamic segmentation of tissues based on transmission differences, characterized in that It includes the following steps: Obtain multi-spectral image data of the surgical area. Irradiate the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues with a broadband light source, and generate a first resolution image by utilizing the transmission differences of different wavelengths of light by each tissue; For the first resolution image, adopt a fluorescent dye labeling scheme to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves, and generate a second enhanced image; Extract the superimposed features of the fluorescent signal and the broadband light source signal from the second enhanced image, analyze the signal intensity and spatial distribution through a convolutional neural network, and determine the boundary line between the fascia and the ligament to generate a third segmentation image; Combine the third segmentation image and the real-time anatomical space depth data, and perform secondary feature extraction on the blood vessels, nerves, and intervertebral disc tissues by utilizing the transmission differences to generate a fourth localization image; After generating the fourth localization image, when overlapping signals in the adhesion scenario are detected, optimize the irradiation angle and wavelength combination through a dynamic light source band adjustment algorithm to generate a fifth optimized image; Based on the fifth optimized image, fuse the spatial coordinates and depth data of the blood vessels, nerves, and intervertebral disc tissues, and generate a sixth dynamic image through real-time visualization rendering; Classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image; Analyze the tissue boundary change trend of the seventh classification image by using the optical flow algorithm, and smooth the jitter through time series filtering to generate an eighth stable image; Extract the key structure contours from the eighth stable image and map them to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate a ninth visualization image, completing the tissue segmentation in the image.

2. The method according to claim 1, wherein The process of obtaining multi-spectral image data of the surgical area, irradiating the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues with a broadband light source, and generating a first resolution image by utilizing the transmission differences of different wavelengths of light by each tissue includes: Perform denoising and enhancement processing on the multi-spectral image data to obtain a preliminary image; Extract the transmission characteristics of the fascia, ligament, blood vessels, nerves, and intervertebral disc tissues from the preliminary image, and perform boundary segmentation on the fascia tissue by using a wavelength difference threshold; Extract the texture features of the fascia and the ligament based on the fascia segmentation image, and judge the junction area between the two through a support vector machine algorithm; Analyze the change trend of the transmission characteristics of the blood vessels and nerves and the spectral characteristics of the intervertebral disc tissue, and judge the contact area between the intervertebral disc and the blood vessels and nerves through a random forest algorithm to generate a multi-tissue distribution map as the first resolution image.

3. The method according to claim 1, characterized in that, The process of adopting a fluorescent dye labeling scheme to excite the fluorescent signals of the fascia, ligament, and blood vessels and nerves for the first resolution image to generate a second enhanced image includes: Mark the fascia signal, ligament signal, and excitation signal of the blood vessels and nerves in the first resolution image to generate a preliminary fluorescent image; Adjust the wavelength to enhance the intensity distribution of the fluorescent signal and separate the fascia signal and the ligament signal; Judge the boundary area between the fascia and the ligament through a convolutional neural network to generate a blood vessel distribution feature map; Extract the excitation differences between the blood vessels, nerves, and ligament signals, and analyze the spatial distribution characteristics of the fluorescent signal to generate a second enhanced image.

4. The method according to claim 1, characterized in that, The process of extracting the superimposed features of the fluorescent signal and the broadband light source signal from the second enhanced image, analyzing the signal intensity and spatial distribution through a convolutional neural network, and determining the boundary line between the fascia and the ligament to generate a third segmentation image includes: Extract the fluorescence signal intensity and spatial distribution information using a convolutional neural network to judge the boundary continuity of the fascia and ligament; Separate the fascia feature map based on the fluorescence distribution characteristics at the fascia boundary, and adjust the spatial distribution weight of the broadband light source signal at the ligament boundary; Extract the overlapping area of the fluorescence signal and the broadband light source, determine the independent areas of the fascia and ligament through the difference characteristics, and generate the third segmentation image.

5. The method according to claim 1, characterized in that, After generating the fourth positioning image, when overlapping signals in the adhesion scenario are detected, the process of generating the fifth optimized image by optimizing the irradiation angle and wavelength combination through the dynamic light source band adjustment algorithm includes: After generating the fourth positioning image, detect the overlapping signals in the adhesion scenario, adjust the light source wavelength to the red light band and optimize the irradiation angle; Adopt an edge detection algorithm to extract the boundary information of blood vessels, nerves and intervertebral disc tissue, repair the broken boundary through morphological operations, and generate the fifth optimized image.

6. The method according to claim 1, characterized in that The process of fusing the spatial coordinates and depth data of blood vessels, nerves and intervertebral disc tissue based on the fifth optimized image and generating the sixth dynamic image through real-time visualization rendering includes: Extract the spatial coordinates of blood vessels, nerves and intervertebral disc tissue based on the fifth optimized image, and map them to a three-dimensional coordinate system through depth data; Fuse the boundary information and the anatomical space depth distribution, and use the interpolation method to smooth the mutation area of the spatial coordinates; Enhance the spatial hierarchy of blood vessels, nerves and intervertebral disc tissue through light and shadow distribution adjustment, and generate the sixth dynamic image.

7. The method according to claim 1, characterized in that The process of analyzing the tissue boundary change trend of the seventh classification image using the optical flow algorithm and generating the eighth stable image by smoothing the jitter through time series filtering includes: Extract the displacement change data of the tissue boundary in the seventh classification image, and calculate the pixel motion vector between consecutive frames; Eliminate the random jitter of the displacement data through time series filtering to generate a smoothed boundary change curve; Correct the spatial position of the tissue boundary based on the smoothed data to generate the eighth stable image.

8. The method according to claim 1, wherein The process of extracting the key structure contours from the eighth stable image and mapping them to a three-dimensional anatomical model through three-dimensional reconstruction technology to generate the ninth visualization image and completing the tissue segmentation in the image includes: Extract the contour data of the fascia, ligament, blood vessels, nerves and intervertebral disc tissue from the eighth stable image, and analyze their spatial positioning through stereology methods; Fuse the anatomical space depth data and the contour data, and use the interpolation technology to adjust the boundary details of the three-dimensional model; Generate a three-dimensional visualization image based on the volume rendering technology to generate the ninth visualization image.

9. A tissue dynamic segmentation device based on transmission difference, characterized in that The device includes: A multispectral imaging module for acquiring multispectral image data of the surgical area, irradiating the fascia, ligament, blood vessels, nerves and intervertebral disc tissue with a broadband light source, and generating a first resolution image using the transmission differences of different tissues for different wavelengths of light; A fluorescence labeling module for exciting the fluorescence signals of the fascia, ligament and blood vessels and nerves for the first resolution image to generate a second enhanced image; A feature analysis module for extracting the superimposed features of the fluorescence signal and the broadband light source signal from the second enhanced image, analyzing the signal intensity and spatial distribution through a convolutional neural network, and determining the boundary line between the fascia and ligament to generate a third segmentation image; The secondary positioning module is used to combine real-time anatomical spatial depth data, utilize transmission differences to perform secondary feature extraction on blood vessels, nerves, and intervertebral disc tissues, and generate a fourth positioning image; The dynamic light source optimization module is used to generate a fifth optimized image by adjusting the light source wavelength and irradiation angle when overlapping signals in an adhesion scenario are detected; The visualization rendering module is used to fuse the spatial coordinates and depth data of blood vessels, nerves, and intervertebral disc tissues to generate a sixth dynamic image; The pixel classification module is used to classify the tissue pixel distribution based on the sixth dynamic image to generate a seventh classification image; The optical flow tracking module is used to analyze the change trend of tissue boundaries and smooth jitters through time series filtering to generate an eighth stable image; The 3D reconstruction module is used to extract key structural contours and map them to a 3D anatomical model to generate a ninth visualization image.

10. The device according to claim 9, characterized in that, The multispectral imaging module includes: The image denoising unit is used to perform denoising and enhancement processing on multispectral image data to obtain a preliminary image; The transmission characteristic extraction unit is used to extract the transmission characteristics of fascia, ligaments, blood vessels, nerves, and intervertebral disc tissues from the preliminary image; The fascia segmentation unit is used to perform boundary segmentation on fascia tissues using a wavelength difference threshold; The texture analysis unit is used to extract the texture features of fascia and ligaments based on the fascia segmentation image and judge the junction area between the two through a support vector machine algorithm; The contact analysis unit is used to analyze the change trend of the transmission characteristics of blood vessels and nerves and the spectral characteristics of intervertebral disc tissues, and judge the contact area through a random forest algorithm.