Laparoscopic uterine fibroid localization assistance method and system

Abstract

The application relates to the technical field of medical images, and particularly discloses a laparoscopic uterine fibroid positioning auxiliary method and system, the method comprising the following steps: acquiring a laparoscopic video original image frame, correcting the original image frame through homomorphic filtering to obtain a corrected image; performing multi-scale coarse-fine registration based on an image pyramid to obtain a dense displacement vector field; performing full-image curl analysis on the vector field to obtain a tissue curl distribution map; constructing texture features of the corrected image to generate a texture atlas; combining the curl value and the texture features to construct a five-dimensional fusion feature vector, dividing a feature space according to a joint judgment rule, and extracting a fibroid candidate region; superimposing and displaying the fibroid candidate region on the original image after contour tracking, so that uterine fibroid positioning and marking visualization is realized; the application combines tissue deformation and texture features to realize adaptive positioning of uterine fibroids, reduces illumination and motion interference, and improves the robustness and reliability of surgical auxiliary identification.

Description

Technical Field

[0001] This invention relates to the field of medical imaging technology, and in particular to a method and system for assisting in the localization of uterine fibroids under laparoscopy. Background Technology

[0002] Existing technologies for uterine fibroid localization based on preoperative MRI or computed tomography images lack the ability to adapt in real time to intraoperative tissue dynamic deformation. They struggle to address tissue displacement and deformation caused by the establishment of pneumoperitoneum pressure, changes in uterine position, and respiratory motion during laparoscopic surgery. This leads to a significant increase in the registration error between preoperative images and the actual intraoperative anatomical structure, failing to provide surgeons with precise real-time localization guidance. While existing intraoperative ultrasound-assisted localization methods can acquire real-time images, they lack the utilization of the differences in the mechanical response characteristics of the uterine myometrium and fibroid tissue. They struggle to effectively distinguish fibroids with blurred boundaries from surrounding normal myometrial tissue in a two-dimensional ultrasound view, resulting in a high degree of reliance on the operator's ultrasound scanning technique and hindering the achievement of standardized and repeatable intraoperative localization.

[0003] Existing fibroid identification methods based on traditional image processing lack a mechanism for fusing tissue biomechanical features with image texture features. This makes it difficult to simultaneously capture differences in tissue rotational deformation under respiratory motion and microscopic texture changes on the tissue surface within a single modal image. This leads to false positives due to texture similarity or false negatives due to blurred boundaries. Current methods often employ fixed grayscale thresholds or preset morphological rules in their judgment logic, making it difficult to adapt to differences in uterine shape, fibroid size and location, and dynamic changes in the surgical field of view. This results in insufficient generalization ability of the algorithms, failing to meet the dual clinical requirements of high sensitivity and high specificity. Therefore, there is an urgent need to develop an intraoperative localization method that can dynamically capture tissue respiratory motion response features and integrate multi-dimensional biomechanical and texture information for adaptive judgment. This would address the problems of low localization accuracy, poor adaptability, and high operational dependence in existing technologies, improving the accuracy, stability, and clinical applicability of lesion localization during laparoscopic uterine fibroid surgery. Summary of the Invention

[0004] This invention provides a laparoscopic method and system for locating uterine fibroids, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a laparoscopic method for locating uterine fibroids, comprising:

[0006] A1: Obtain the original image frames from the laparoscopic surgery video stream, and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames;

[0007] A2: Based on the hierarchical coarse and fine registration of the preset image pyramid, multi-scale registration is performed on the corrected image frame to obtain the dense displacement vector field of the original image frame;

[0008] A3: Perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame;

[0009] A4: Construct texture features from the corrected image frame to obtain the texture feature map of the original image frame;

[0010] A5: Combine the curl values ​​of the pixels in the tissue curl distribution map with the texture features of the pixels in the texture feature map to obtain the five-dimensional fusion feature vector of the original image frame. Then, according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features, divide the fusion feature vector into feature space to obtain the fibroid candidate region of the original image frame.

[0011] A6: Perform boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and overlay the contour line on the original image frame as the fibroid localization marker of the original image frame.

[0012] In a preferred embodiment, the step of acquiring the original image frames from the laparoscopic surgery video stream and performing homomorphic filtering illumination correction on the original image frames to obtain corrected image frames includes:

[0013] The laparoscopic surgery video stream is segmented into complete respiratory cycles at fixed time intervals to obtain the original image frames of the laparoscopic surgery video stream;

[0014] The original image frame is transformed from the spatial domain to the logarithmic domain to obtain the logarithmic domain image of the original image frame;

[0015] Perform a Fast Fourier Transform on the logarithmic domain image to obtain the frequency domain image of the original image frame;

[0016] Based on a preset high-pass filter, the frequency domain image is subjected to high-pass filtering to obtain the corrected image of the original image frame;

[0017] The corrected image is reconstructed using inverse Fourier transform to obtain the corrected image frame of the original image frame.

[0018] In a preferred embodiment, the hierarchical coarse-fine registration based on a preset image pyramid, which performs multi-scale registration on the corrected image frame to obtain the dense displacement vector field of the original image frame, includes:

[0019] The corrected image frame is subjected to multi-level Gaussian reduction to obtain the image pyramid of the original image;

[0020] Based on the corrected image frame, gradient optical flow displacement fitting is performed on the top layer of the image pyramid to obtain the displacement vector field of the top layer;

[0021] The displacement vector field is upsampled and transferred to obtain the initial values ​​of the middle layer iteration of the image pyramid;

[0022] Based on the middle layer image frames of the image pyramid, optical flow displacement refinement is performed on the initial values ​​of the middle layer iteration to obtain the middle layer refined displacement vector field of the image pyramid.

[0023] The intermediate-level refined displacement vector field is refined layer by layer until the bottom layer of the image pyramid is reached, thus obtaining the bottom-level displacement vector field of the image pyramid. The bottom-level displacement vector field is then used as the dense displacement vector field of the original image frame.

[0024] In a preferred embodiment, the step of performing gradient optical flow displacement fitting on the top layer of the image pyramid based on the corrected image frame to obtain the displacement vector field of the top layer includes:

[0025] Multi-dimensional gradients are constructed on the current layer of the image pyramid to obtain the gradient image of the current layer;

[0026] The current layer is divided into blocks to obtain a rectangular window of the current layer;

[0027] Based on the gradient information of the gradient image, gradient coefficients are constructed for the rectangular window to obtain the overdetermined linear equation of the current layer;

[0028] The overdetermined linear equations are solved by the least squares method to obtain the sparse displacement vector field of the current layer;

[0029] The sparse displacement vector field is densified by bilinear interpolation to obtain the displacement vector field of the current layer.

[0030] In a preferred embodiment, the step of performing full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame includes:

[0031] The direction component of the dense displacement vector field is analyzed to obtain the horizontal displacement component field and the vertical displacement component field of the original image frame.

[0032] By performing vertical partial derivatives on the pixels in the horizontal displacement component field, the vertical rate of change of the horizontal displacement of the original image frame is obtained.

[0033] The horizontal partial derivative of the pixels in the vertical displacement component field is used to obtain the horizontal rate of change field of the vertical displacement of the original image frame.

[0034] The horizontal displacement vertical rate of change field and the vertical displacement horizontal rate of change field are differentially analyzed pixel by pixel to obtain the full-pixel curl field of the original image frame;

[0035] Based on the full-pixel curl field, the pixels of the original image frame are arranged in a regular manner to obtain the organization curl distribution map of the original image frame.

[0036] In a preferred embodiment, the step of constructing texture features on the corrected image frame to obtain the texture feature map of the original image frame includes:

[0037] The corrected image frame is subjected to color space grayscale mapping to obtain the grayscale corrected image frame of the original image frame;

[0038] A local neighborhood window is extracted from the target pixel of the grayscale corrected image frame to obtain the local window of the target pixel;

[0039] The gray-level co-occurrence matrix feature analysis is performed on the pixel set within the local window to obtain the texture contrast feature value, texture correlation feature value, texture energy feature value and texture homogeneity feature value of the pixel set within the window;

[0040] The texture contrast feature value, texture correlation feature value, texture energy feature value, and texture homogeneity feature value are combined to form a four-dimensional texture feature vector of the original image frame;

[0041] The four-dimensional texture feature vector is anchored to the center pixel of the grayscale corrected image frame to obtain the pixel-by-pixel texture feature map of the original image frame;

[0042] The full feature vectors of the pixel-by-pixel texture feature map are integrated to obtain the texture feature map of the original image frame.

[0043] In a preferred embodiment, the step of combining the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the five-dimensional fused feature vector of the original image frame includes:

[0044] Pixel-level registration is performed between the tissue curl distribution map and the texture feature map to obtain the registration feature set of pixels at corresponding positions in the original image frame;

[0045] Based on the registration feature set, curl value feature is taken for the corresponding position pixel to obtain the first dimension feature component of the corresponding position pixel;

[0046] Based on the registration feature set, feature values ​​are taken from the corresponding pixel positions to obtain the four-dimensional feature components of the four-dimensional texture feature vector;

[0047] The first-dimensional feature component and the four-dimensional feature component are concatenated in an ordered manner to obtain the five-dimensional fused feature vector of the original image frame.

[0048] In a preferred embodiment, the step of dividing the fused feature vector into feature spaces based on the joint determination rule of the curl feature in the tissue curl and the texture feature in the pixel texture feature to obtain the fibroid candidate region of the original image frame includes:

[0049] Obtain the curl value, texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the fused feature pixels in the five-dimensional fused feature vector;

[0050] The fibroid feature response value of the fused feature pixel is obtained by performing feature weighting calculation on the curl value, the texture contrast feature value, the texture energy feature value and the texture homogeneity feature value;

[0051] Based on a preset response threshold, the fibroid feature response value is thresholded to obtain the fibroid candidate pixel of the fused feature pixel;

[0052] The connected region formed by the candidate pixels of the fibroid is used as the candidate region of the fibroid in the original image frame.

[0053] The formula for calculating the characteristic response value of the fibroid is as follows:

[0054] ;

[0055] in, The curl value of the pixel. , , These are the texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the pixel, respectively. For a preset minimum positive number, This represents the theoretical maximum value of the texture homogeneity eigenvalue.

[0056] In a preferred embodiment, the step of performing boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and superimposing the contour line on the original image frame as a fibroid localization marker of the original image frame, includes:

[0057] A morphological closing operation is performed on the fibroid candidate region to obtain a binary mask image of the fibroid candidate region.

[0058] Eight-neighbor contour tracing is performed on the binary mask image to obtain the original contour point set of the fibroid candidate region;

[0059] The original set of contour points is simplified to obtain the contour line of the fibroid candidate region.

[0060] The contour lines are vector-mapped to obtain an annotated image frame of the original image frame, and the annotated image frame is used as a fibroid localization marker of the original image frame.

[0061] To address the above problems, the present invention also provides a laparoscopic uterine fibroid localization assistance system, the system comprising:

[0062] The illumination correction module is used to acquire the original image frames in the laparoscopic surgery video stream and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames.

[0063] A multi-scale registration module is used to perform multi-scale registration on the corrected image frame based on a hierarchical coarse and fine registration of a preset image pyramid, so as to obtain the dense displacement vector field of the original image frame.

[0064] The curl analysis module is used to perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame;

[0065] The texture feature construction module is used to construct texture features on the corrected image frame to obtain the texture feature map of the original image frame;

[0066] The fusion feature and candidate region segmentation module is used to combine the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the fusion feature vector of the original image frame, and to perform feature space segmentation on the fusion feature vector according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features to obtain the fibroid candidate region of the original image frame.

[0067] The contour tracking and marking module is used to perform boundary contour tracking on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and to overlay the contour line on the original image frame as a fibroid localization marker of the original image frame.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. This invention accurately captures the rotational deformation characteristics of uterine tissue under respiratory motion through coarse-to-fine image registration and curl calculation. Existing technologies struggle to adapt to dynamic tissue deformation and cannot handle displacement deformation caused by pneumoperitoneum, body position, and respiration. This invention first eliminates overall displacement using rigid body transformation at the top layer of the image pyramid, then obtains non-rigid deformation through local matching at the bottom layer, generating a dense displacement vector field of all pixels, and obtaining a curl distribution map through curl analysis. Fibroids have higher rigidity and weaker rotational deformation than normal tissue, creating a clear contrast between low-curvature lesion areas and high-curvature normal areas in the image. This method does not rely on grayscale differences, can solve the problem of blurred boundaries, and improves the accuracy and stability of lesion localization.

[0070] 2. This invention integrates tissue curl and texture features to construct a multi-dimensional adaptive recognition mechanism. Traditional fibroid recognition relies on only a single feature and fixed rules, which cannot adapt to patient differences and changes in intraoperative field of vision. This invention uses a gray-level co-occurrence matrix to extract texture features, constructs a texture atlas, and concatenates the curl value with the corresponding pixel texture vector to form a three-dimensional fusion feature. Then, unsupervised clustering is used to classify pixel categories, automatically identifying candidate fibroid regions. This mechanism does not require manual threshold setting and can adapt to different uterine shapes, fibroid differences, and changes in intraoperative field of vision, improving detection sensitivity while reducing the false detection rate, and solving the problem of insufficient generalization ability of existing technologies. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating a laparoscopic method for locating uterine fibroids according to an embodiment of the present invention.

[0072] Figure 2 This is a functional block diagram of a laparoscopic uterine fibroid localization assistance system provided in an embodiment of the present invention;

[0073] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0074] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0075] This application provides a laparoscopic method for assisting in the localization of uterine fibroids. The executing entity of this laparoscopic method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the laparoscopic method for assisting in the localization of uterine fibroids can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0076] Reference Figure 1 The diagram shown is a schematic flowchart of a laparoscopic uterine fibroid localization assistance method according to an embodiment of the present invention. In this embodiment, the laparoscopic uterine fibroid localization assistance method includes:

[0077] A1: Obtain the original image frames from the laparoscopic surgery video stream, and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames;

[0078] In this embodiment of the invention, the step of acquiring the original image frames from the laparoscopic surgery video stream and performing homomorphic filtering illumination correction on the original image frames to obtain corrected image frames includes:

[0079] The laparoscopic surgery video stream is segmented into complete respiratory cycles at fixed time intervals to obtain the original image frames of the laparoscopic surgery video stream;

[0080] The original image frame is transformed from the spatial domain to the logarithmic domain to obtain the logarithmic domain image of the original image frame;

[0081] Perform a Fast Fourier Transform on the logarithmic domain image to obtain the frequency domain image of the original image frame;

[0082] Based on a preset high-pass filter, the frequency domain image is subjected to high-pass filtering to obtain the corrected image of the original image frame;

[0083] The corrected image is reconstructed using inverse Fourier transform to obtain the corrected image frame of the original image frame.

[0084] Homomorphic filtering illumination correction is performed on the original image frames in the laparoscopic surgery video stream to eliminate the influence of uneven illumination generated by the laparoscopic light source on the uterine surface, providing uniformly illuminated corrected image frames for subsequent precise positioning based on tissue mechanical and texture features.

[0085] The laparoscopic surgery video stream is segmented at fixed time intervals to capture complete respiratory cycles, yielding the original image frames of the laparoscopic surgery video stream. During laparoscopic surgery, the patient's respiratory movements cause periodic displacement and deformation of the uterine tissue. By segmenting video frames within a complete respiratory cycle at fixed time intervals, an image sequence reflecting the complete dynamic changes of the uterus from the inspiratory to the expiratory phase under the influence of respiratory movements can be obtained. This image sequence contains morphological information of the uterine tissue at different time phases, providing a data foundation for subsequently capturing the rotational deformation characteristics of the tissue under respiratory action.

[0086] The original image frame is transformed from the spatial domain to the logarithmic domain to obtain its logarithmic domain image. In the spatial domain, the illumination and reflection components of the original image frame are coupled together in a product form, which is not conducive to separating the uneven illumination components. By taking the logarithm of the grayscale value of each pixel in the original image frame, the product relationship between the illumination and reflection components is transformed into an additive relationship, thus obtaining the logarithmic domain image. In the logarithmic domain, the illumination component appears as a low-frequency component, while the reflection component appears as a high-frequency component, creating conditions for subsequent separation of the illumination component through frequency domain filtering.

[0087] A Fast Fourier Transform (FFT) is performed on the logarithmic domain image to obtain the frequency domain image of the original image frame. The FFT transforms the logarithmic domain image from the spatial domain to the frequency domain. In the frequency domain, each frequency component of the image corresponds to grayscale changes at different scales in the spatial domain. After the Fourier transform, the low-frequency components of the logarithmic domain image are concentrated in the central region of the frequency domain image, mainly reflecting the slow changes in uneven illumination; the high-frequency components are distributed in the surrounding areas of the frequency domain image, mainly reflecting details such as the texture of the uterine surface tissue and the course of blood vessels. The frequency domain image allows for the separation of the illumination component and the tissue texture component along the frequency dimension.

[0088] Based on a preset high-pass filter, the frequency domain image is high-pass filtered to obtain the corrected image of the original image frame. The preset high-pass filter is a filtering window in the frequency domain that attenuates in the central low-frequency region and retains in the surrounding high-frequency regions. This high-pass filter is multiplied point-by-point with the frequency domain image, causing the illumination components in the central low-frequency region of the frequency domain image to be filtered out or attenuated, while the tissue texture components in the surrounding high-frequency regions are retained. After high-pass filtering, only the high-frequency components reflecting the details of the uterine surface texture are retained in the frequency domain image, thereby eliminating the influence of uneven illumination caused by changes in the angle and distance of the laparoscopic light source.

[0089] The corrected image is reconstructed using Inverse Fourier Transform (IFT) to obtain a corrected image frame from the original image frame. IFT is the process of converting a high-pass filtered frequency domain image back to the spatial domain. In the spatial domain, the corrected image obtained through IFT recovers the tissue texture information of the uterine surface, and uneven illumination is effectively suppressed. Since a logarithmic domain transformation was performed previously, an exponential operation is then performed on the corrected image to restore the original grayscale dimensions, ultimately obtaining a corrected image frame with the same spatial dimensions and pixel positions as the original image frame. In this corrected image frame, the brightness of each region of the uterine surface is uniform, and the tissue texture is clear, providing a high-quality image foundation for subsequent extraction of the uterine tissue curl distribution map and texture feature map.

[0090] The beneficial effects are as follows: by constructing a coarse-to-fine image registration process to generate a dense displacement vector field and performing curl analysis, the degree of rotational deformation of uterine tissue under respiratory motion is quantified into a curl distribution map. The difference in mechanical properties between fibroid tissue and normal myometrium in terms of rotational deformation ability is used to achieve precise lesion localization. At the same time, texture feature maps are extracted by gray-level co-occurrence matrix and fused and clustered with the curl distribution map to construct an adaptive recognition mechanism for multi-dimensional feature collaborative judgment. This eliminates the subjectivity of manually setting fixed thresholds in traditional methods, significantly improves the accuracy and stability of lesion localization, and enhances the algorithm's adaptability to different patients, different fibroid morphologies, and dynamic changes in the surgical field. It effectively reduces the false detection rate and improves clinical applicability.

[0091] A2: Based on the hierarchical coarse and fine registration of the preset image pyramid, multi-scale registration is performed on the corrected image frame to obtain the dense displacement vector field of the original image frame;

[0092] In this embodiment of the invention, the hierarchical coarse-fine registration based on a preset image pyramid, which performs multi-scale registration on the corrected image frame to obtain the dense displacement vector field of the original image frame, includes:

[0093] The corrected image frame is subjected to multi-level Gaussian reduction to obtain the image pyramid of the original image;

[0094] Based on the corrected image frame, gradient optical flow displacement fitting is performed on the top layer of the image pyramid to obtain the displacement vector field of the top layer;

[0095] The displacement vector field is upsampled and transferred to obtain the initial values ​​of the middle layer iteration of the image pyramid;

[0096] Based on the middle layer image frames of the image pyramid, optical flow displacement refinement is performed on the initial values ​​of the middle layer iteration to obtain the middle layer refined displacement vector field of the image pyramid.

[0097] In an embodiment of the present invention, the step of performing gradient optical flow displacement fitting on the top layer of the image pyramid based on the corrected image frame to obtain the displacement vector field of the top layer includes:

[0098] Multi-dimensional gradients are constructed on the current layer of the image pyramid to obtain the gradient image of the current layer;

[0099] The current layer is divided into blocks to obtain a rectangular window of the current layer;

[0100] Based on the gradient information of the gradient image, gradient coefficients are constructed for the rectangular window to obtain the overdetermined linear equation of the current layer;

[0101] The overdetermined linear equations are solved by the least squares method to obtain the sparse displacement vector field of the current layer;

[0102] The sparse displacement vector field is densified by bilinear interpolation to obtain the displacement vector field of the current layer.

[0103] A multi-level Gaussian reduction is performed on the corrected image frames to construct an image pyramid. Gradient optical flow displacement fitting is used at the top level to obtain the overall displacement estimate. Then, sparse displacement vector fields are obtained by constructing block gradient coefficients and solving overdetermined linear equations layer by layer. Finally, bilinear interpolation is used to densify the displacement vector fields of each layer to obtain the complete displacement vector fields of each layer. Finally, multi-scale registration from coarse to fine is achieved to generate a dense displacement vector field covering the entire pixel.

[0104] Multi-level Gaussian reduction is performed on the corrected image frames to obtain the image pyramid of the original image. Multi-level Gaussian reduction involves performing Gaussian filtering and downsampling operations on the corrected image frames at each level, forming a series of image layers with progressively decreasing resolution. First, Gaussian filtering is performed on the original resolution corrected image frames. A Gaussian kernel is then convolved with the image to smooth grayscale fluctuations caused by noise from the laparoscopic light source and subtle textures on the tissue surface. Then, pixels are extracted at fixed intervals, reducing the image size by half in both width and height, resulting in the first layer image. Using the first layer image as input, the Gaussian filtering and downsampling operations are repeated to obtain the second layer image. This process continues, and after multiple reductions, an image pyramid structure is formed from high resolution at the bottom to low resolution at the top. The top image has the smallest size, preserving the overall outline of the uterine tissue and the significant displacement trends caused by respiratory movements, while eliminating local details and noise interference. The bottom image has the largest size, fully preserving all details of the uterine surface texture and minor deformations.

[0105] Based on the corrected image frames, gradient optical flow displacement fitting is performed on the top layer of the image pyramid to obtain the displacement vector field of the top layer. At the top layer of the image pyramid, due to the smallest image size, the displacement of uterine tissue during the respiratory cycle mainly exhibits overall rigid body motion. Gradient optical flow displacement fitting solves for the displacement by analyzing the pixel brightness changes between the start and end phases of the top layer image. Specifically, for each pixel in the top layer image, a neighborhood window is defined centered on it. Assuming that the brightness of all pixels within the window remains constant between the start and end phases, there is a linear relationship between the displacement of each pixel and the spatial grayscale gradient. The linear relationships of all pixels within the window are combined to form an overdetermined system of linear equations concerning the displacement component of that pixel. By solving this system of equations, the displacement vector of that pixel is obtained. Traversing all pixels in the top layer image, a displacement vector field covering the entire top layer image is finally formed. This displacement vector field reflects the large-scale overall displacement trend of uterine tissue under respiratory action.

[0106] Multi-dimensional gradient construction is performed on the current layer of the image pyramid to obtain the gradient image of the current layer. The current layer in the image pyramid can be any layer after the top layer, with a higher resolution than the previous layer. Multi-dimensional gradient construction refers to simultaneously calculating the gray-level change rate of the current layer image in both the horizontal and vertical directions. Specifically, for each pixel in the current layer image, the gray-level value of the pixel in the horizontal direction is obtained by subtracting the gray-level value of the pixel in the left direction from the gray-level value of the pixel in the right direction; the gray-level value of the pixel in the vertical direction is obtained by subtracting the gray-level value of the pixel in the upper direction from the gray-level value of the pixel in the lower direction. The horizontal and vertical gradients are stored as two gradient images, which together constitute the gradient image of the current layer, used to describe the degree of gray-level change of the uterine tissue in different directions, providing spatial change information for subsequent displacement calculation.

[0107] The current layer is divided into blocks to obtain rectangular windows for that layer. The block division operation divides the current layer image into multiple non-overlapping rectangular windows, each containing a certain number of pixels. During the division, the image is uniformly divided horizontally and vertically according to the image size and a preset window size, so that each rectangular window covers a local image region. By dividing into blocks, the global displacement field solution problem is decomposed into multiple independent solution problems within local regions. Each rectangular window corresponds to a local region of uterine tissue, and the displacement of pixels within the window is assumed to conform to the same motion model, thereby reducing the solution complexity and improving adaptability to local non-rigid deformations.

[0108] Based on the gradient information of the gradient image, gradient coefficients are constructed for the rectangular window to obtain the overdetermined linear equation of the current layer. For each rectangular window, a linear equation about the displacement component is established for each pixel within the window using the horizontal gradient values, vertical gradient values, and grayscale difference between the start and end phases of all pixels within the window. Combining the linear equations of all pixels results in an overdetermined linear equation, since the number of pixels within the window is much greater than the number of displacement components to be solved. The core of gradient coefficient construction is to extract the gradient information and grayscale difference information of all pixels within the window, and organize them into a coefficient matrix and a constant term vector, providing a data foundation for subsequent solution of the displacement vector.

[0109] The overdetermined linear equations are solved using the least squares method to obtain the sparse displacement vector field of the current layer. The least squares method involves finding a set of displacement components that minimizes the sum of squared errors of the linear equations for all pixels. Specifically, for each rectangular window, the linear equations for all pixels within the window are represented in matrix form. The displacement vector corresponding to that window is obtained by solving the normal equations. This displacement vector represents the overall displacement estimate of all pixels within the rectangular window. Since only one displacement vector is obtained for each window, the displacement field composed of the displacement vectors from all windows is sparse, with displacement values ​​only at the center of the window, rather than each pixel having an independent displacement value. This sparse displacement vector field reflects the local movement trend of uterine tissue on a larger scale, providing a foundation for subsequent densification.

[0110] The sparse displacement vector field is densed using bilinear interpolation to obtain the displacement vector field of the current layer. Bilinear interpolation densening is the process of extending the displacement vector of each window in the sparse displacement vector field to every pixel within that window. Specifically, for each pixel in the current layer image, the rectangular window to which the pixel belongs is first determined, and the displacement vector corresponding to that window is obtained. Then, using the pixel's position relative to the window center, weighted interpolation is performed by combining the displacement vectors of adjacent windows. Since each window has only one displacement vector, directly assigning it to all pixels within the window would result in blocky discontinuities in the displacement field. Therefore, bilinear interpolation is used, taking into account the relative coordinates of the pixel within the window and the displacement vectors of the window and its adjacent windows, to calculate the fine displacement vector of the pixel through linear weighting. After traversing all pixels in the current layer, a dense displacement vector field is formed where each pixel has an independent displacement vector. This displacement field can continuously and smoothly describe the displacement distribution of uterine tissue at the current layer resolution.

[0111] The beneficial effect is that by constructing a pre-defined image pyramid and performing multi-level Gaussian reduction on the corrected image frames, the displacement of uterine tissue during the respiratory cycle is decomposed into a progressive optimization problem. The top layer of the pyramid uses gradient optical flow displacement fitting to quickly lock the overall displacement trend caused by respiration, generating a top-level displacement vector field. Subsequent layers construct horizontal and vertical gradient images, divide rectangular windows, and construct overdetermined linear equations based on gradient information. The sparse displacement vectors are solved using the least squares method, and then densed using bilinear interpolation to obtain a continuous and smooth displacement vector field across all pixels. This coarse-to-fine registration mechanism avoids the matching failure problem of directly calculating the displacement field from the original resolution image, while ensuring the spatial continuity and numerical accuracy of the dense displacement vector field. It can completely describe the differences in rotational deformation of each pixel, providing high-precision and highly reliable pixel-level displacement data support for subsequent curl analysis and lesion localization.

[0112] A3: Perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame;

[0113] In this embodiment of the invention, the step of performing full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame includes:

[0114] The direction component of the dense displacement vector field is analyzed to obtain the horizontal displacement component field and the vertical displacement component field of the original image frame.

[0115] By performing vertical partial derivatives on the pixels in the horizontal displacement component field, the vertical rate of change of the horizontal displacement of the original image frame is obtained.

[0116] The horizontal partial derivative of the pixels in the vertical displacement component field is used to obtain the horizontal rate of change field of the vertical displacement of the original image frame.

[0117] The horizontal displacement vertical rate of change field and the vertical displacement horizontal rate of change field are differentially analyzed pixel by pixel to obtain the full-pixel curl field of the original image frame;

[0118] Based on the full-pixel curl field, the pixels of the original image frame are arranged in a regular manner to obtain the organization curl distribution map of the original image frame.

[0119] By performing directional component analysis, partial derivative field construction, and difference calculation on the dense displacement vector field, the dense displacement vector field describing the movement law of uterine tissue is transformed into a full-pixel curl field reflecting the degree of tissue rotation and deformation. After regular arrangement, a tissue curl distribution map is generated, providing a mechanical characteristic basis for distinguishing fibroids with different hardness from normal myometrium.

[0120] Directional component analysis is performed on the dense displacement vector field to obtain the horizontal and vertical displacement component fields of the original image frame. Each pixel in the dense displacement vector field contains two directional components: the horizontal displacement value and the vertical displacement value of the pixel. Directional component analysis decomposes the displacement vector of all pixels in the dense displacement vector field into two independent two-dimensional fields. Specifically, it iterates through each pixel in the dense displacement vector field, extracts the horizontal displacement value of the pixel, and fills it into a new two-dimensional field according to the row and column position of the pixel in the image to form the horizontal displacement component field; it also extracts the vertical displacement value of the pixel and fills it into another two-dimensional field according to the same row and column position to form the vertical displacement component field. After directional component analysis, the displacement information that was originally coupled in one field is separated into two independent scalar fields, which describe the displacement distribution of the uterine tissue in the horizontal and vertical directions, respectively.

[0121] The vertical partial derivative of the horizontal displacement component field is calculated to obtain the vertical rate of change field of the horizontal displacement of the original image frame. The vertical partial derivative is used to quantify the rate of change of the horizontal displacement component in the vertical direction. Specifically, in the horizontal displacement component field, for each pixel, the horizontal displacement value of its directly below neighboring pixel is subtracted from the horizontal displacement value of its directly above neighboring pixel. The difference is then divided by the physical distance between the neighboring pixels to obtain the vertical rate of change of the horizontal displacement at that pixel. This process is repeated for all pixels in the horizontal displacement component field, and the calculated rate of change value for each pixel is filled into a new two-dimensional field according to its row and column position, forming the vertical rate of change field of the horizontal displacement. This rate of change field reflects the gradient information of the horizontal displacement of the uterine tissue in the vertical direction, that is, the degree of drastic change in the tissue's horizontal displacement with respect to its vertical position.

[0122] The horizontal partial derivative of the vertical displacement component field is solved to obtain the horizontal rate of change field of the vertical displacement of the original image frame. The horizontal partial derivative is used to quantify the rate of change of the vertical displacement component in the horizontal direction. Specifically, in the vertical displacement component field, for each pixel, the vertical displacement value of its right-hand neighbor pixel is subtracted from the vertical displacement value of its left-hand neighbor pixel. The difference is then divided by the physical distance between the neighboring pixels to obtain the horizontal rate of change of the vertical displacement at that pixel. This process is repeated for all pixels in the vertical displacement component field, and the calculated rate of change value for each pixel is filled into a new two-dimensional field according to its row and column positions, forming the horizontal rate of change field of the vertical displacement. This rate of change field reflects the gradient information of the vertical displacement of the uterine tissue in the horizontal direction, that is, the degree of drastic change in the tissue's vertical displacement with respect to its horizontal position.

[0123] The full-pixel curl field of the original image frame is obtained by performing a pixel-by-pixel difference operation between the vertical displacement horizontal rate of change field and the horizontal displacement vertical rate of change field. The specific operation of the difference operation is as follows: for each pixel in the image, the vertical displacement horizontal rate of change value at that pixel is taken from the vertical displacement horizontal rate of change field, and the horizontal displacement vertical rate of change value at that pixel is taken from the horizontal displacement vertical rate of change field. The former is subtracted from the latter, and the difference is the curl value at that pixel. After performing this subtraction operation on all pixels, the full-pixel curl field is formed. The curl value of each pixel in this curl field characterizes the degree of rotational deformation of the tissue in the neighborhood of that pixel under the action of respiratory motion; the larger the value, the more intense the tissue rotation, and the smaller the value, the weaker the tissue rotation.

[0124] Based on the full-pixel curl field, the pixels of the original image frame are regularly arranged to obtain a tissue curl distribution map of the original image frame. Regular arrangement refers to organizing the curl value of each pixel in the full-pixel curl field according to its spatial coordinates in the original image frame, forming a grayscale image with the same size as the original image frame and corresponding pixel positions. Specifically, a blank image with the same width and height as the original image frame is created, and the curl value of each pixel in the full-pixel curl field is used as the grayscale value of the corresponding pixel in the blank image. In the tissue curl distribution map generated after regular arrangement, the grayscale value of each pixel directly reflects the degree of rotational deformation of the uterine tissue at that pixel location during the respiratory cycle. Areas with high curl values ​​appear as bright areas in the image, and areas with low curl values ​​appear as dark areas, thus visually presenting the mechanical properties of the uterine tissue in an image format.

[0125] The beneficial effect is that by analyzing the directional components of the dense displacement vector field, the composite displacement information of uterine tissue is decomposed into horizontal and vertical displacement component fields. The vertical partial derivative of the horizontal displacement component field is solved to obtain the vertical rate of change field of the horizontal displacement; the horizontal partial derivative of the vertical displacement component field is solved to obtain the horizontal rate of change field of the vertical displacement. Through pixel-by-pixel interpolation, the rotational information implicit in the two rate of change fields is extracted into a full-pixel curl field, which is then regularly arranged to generate a tissue curl distribution map, making the tissue rotation deformation characteristics intuitively presented. In this map, high-curvature regions correspond to normal muscle layers that rotate violently under respiration, while low-curvature regions correspond to fibroid tissue with high hardness and restricted rotation, forming a clear mechanical feature contrast. This provides a reliable discrimination basis for subsequent lesion identification that is unaffected by grayscale differences and adapts to situations with blurred boundaries.

[0126] A4: Construct texture features from the corrected image frame to obtain the texture feature map of the original image frame;

[0127] In this embodiment of the invention, the step of constructing texture features from the corrected image frame to obtain the texture feature map of the original image frame includes:

[0128] The corrected image frame is subjected to color space grayscale mapping to obtain the grayscale corrected image frame of the original image frame;

[0129] A local neighborhood window is extracted from the target pixel of the grayscale corrected image frame to obtain the local window of the target pixel;

[0130] The gray-level co-occurrence matrix feature analysis is performed on the pixel set within the local window to obtain the texture contrast feature value, texture correlation feature value, texture energy feature value and texture homogeneity feature value of the pixel set within the window;

[0131] The texture contrast feature value, texture correlation feature value, texture energy feature value, and texture homogeneity feature value are combined to form a four-dimensional texture feature vector of the original image frame;

[0132] The four-dimensional texture feature vector is anchored to the center pixel of the grayscale corrected image frame to obtain the pixel-by-pixel texture feature map of the original image frame;

[0133] The full feature vectors of the pixel-by-pixel texture feature map are integrated to obtain the texture feature map of the original image frame.

[0134] The process involves color space grayscale mapping, local neighborhood window extraction, grayscale co-occurrence matrix feature analysis, feature vector combination and anchoring, and full feature integration of the corrected image frame. This transforms the texture information in the corrected image frame into a pixel-by-pixel four-dimensional texture feature vector, ultimately forming a texture feature map that corresponds to the original image frame space and contains multi-dimensional texture features for each pixel.

[0135] The corrected image frame is subjected to color space grayscale mapping to obtain a grayscale corrected image frame of the original image frame. The original image frame acquired during laparoscopic surgery is typically a color image containing red, green, and blue color channels, while texture feature analysis is mainly based on the variation patterns of pixel grayscale values. Color space grayscale mapping involves weighting and fusing the red, green, and blue components of each pixel in the corrected image frame according to the sensitivity of the human eye to brightness perception, calculating the brightness value of that pixel as its grayscale value. Specifically, for each pixel in the corrected image frame, its red, green, and blue components are multiplied by preset weighting coefficients and then summed to obtain the grayscale value of that pixel. After traversing all pixels of the corrected image frame, the grayscale value of each pixel is filled into a two-dimensional image of the same size as the corrected image frame, forming a grayscale corrected image frame. This grayscale corrected image frame preserves the brightness distribution and texture details of the uterine surface tissue, eliminating the interference of color information on texture feature extraction.

[0136] A local neighborhood window is extracted from the target pixel of the grayscale-corrected image frame to obtain a local window for the target pixel. Local neighborhood window extraction uses each pixel to be analyzed in the grayscale-corrected image frame as the target pixel, delineates a fixed-size square region around the pixel, and extracts the grayscale values ​​of all pixels within this region to form a local window. Specifically, for each pixel in the grayscale-corrected image frame, a square window with a predetermined odd number of pixels on each side is determined, covering the pixel and several pixels above, below, to the left, and to the right. When the target pixel is close to the image boundary, the portion of the window exceeding the boundary is either mirrored or discarded, ensuring that each target pixel obtains a local window of a full size. This local window contains the grayscale distribution information of the target pixel and its surrounding neighboring pixels, providing local region data for subsequent texture feature analysis.

[0137] The gray-level co-occurrence matrix (GLCM) feature analysis is performed on the pixel set within the local window to obtain the texture contrast feature value, texture correlation feature value, texture energy feature value, and texture homogeneity feature value of the pixel set within the window. GLCM feature analysis constructs a GLCM by statistically analyzing the frequency of simultaneous occurrence of gray-level values ​​of pixel pairs with specific spatial relationships within the local window, and then extracts statistical quantities reflecting texture characteristics from this matrix. Specifically, the gray-level values ​​of the pixels within the local window are first quantized to a preset number of gray levels to reduce the matrix dimension. Then, a specific spatial relationship is selected, such as two horizontally adjacent pixels. All pixel pairs in the window that satisfy this positional relationship are statistically analyzed, and the frequency of each gray-level combination is recorded to form the GLCM. Next, four eigenvalues ​​were calculated based on the gray-level co-occurrence matrix: texture contrast eigenvalue, reflecting the drastic change in gray-level values ​​within the window, was obtained by weighted summation of elements far from the diagonal in the matrix; texture correlation eigenvalue, reflecting the linear dependence of gray-level values ​​within the window, was obtained by calculating the correlation coefficients between each row and column of the matrix; texture energy eigenvalue, reflecting the uniformity of texture within the window, was obtained by calculating the sum of squares of each element in the matrix; and texture homogeneity eigenvalue, reflecting the local consistency of texture within the window, was obtained by weighted summation of elements closest to the diagonal in the matrix. These four eigenvalues ​​characterize the texture properties of the uterine surface tissue within the local window from different dimensions.

[0138] The texture contrast feature value, texture correlation feature value, texture energy feature value, and texture homogeneity feature value are combined to form a four-dimensional texture feature vector of the original image frame. The four-dimensional texture feature vector is formed by arranging the four feature values ​​corresponding to each target pixel in a fixed order to create a vector containing four components. Specifically, an array of length four is created, and the texture contrast feature value is used as the first component, the texture correlation feature value as the second component, the texture energy feature value as the third component, and the texture homogeneity feature value as the fourth component, and these are stored sequentially in the array. This four-dimensional texture feature vector integrates multiple texture attributes of the local region where the target pixel is located, providing multi-dimensional feature input for subsequent pixel classification.

[0139] The four-dimensional texture feature vector is anchored to the center pixel of the grayscale-corrected image frame to obtain the pixel-by-pixel texture feature map of the original image frame. Anchoring refers to assigning the calculated four-dimensional texture feature vector to the center pixel of the local window, establishing a correspondence between the pixel and its texture feature. Specifically, for each target pixel in the grayscale-corrected image frame, after parsing the grayscale co-occurrence matrix feature of its local window and generating a four-dimensional texture feature vector, the four-dimensional texture feature vector is stored in a storage structure at the same coordinate position as the target pixel. After traversing all pixels in the grayscale-corrected image frame, each pixel has a corresponding four-dimensional texture feature vector. These vectors are arranged according to the spatial coordinates of the pixels to form a pixel-by-pixel texture feature map. This pixel-by-pixel texture feature map has the same size as the grayscale-corrected image frame, and each position stores a four-dimensional vector instead of a single grayscale value.

[0140] The texture feature map of the original image frame is obtained by integrating all feature vectors from the pixel-by-pixel texture feature map. Full feature vector integration involves organizing the four-dimensional texture feature vectors of all pixels in the pixel-by-pixel texture feature map into a unified data structure, forming a texture feature map that facilitates subsequent processing and analysis. Specifically, a multi-dimensional array of the same size as the pixel-by-pixel texture feature map is created, and the four-dimensional texture feature vector of each pixel is filled into the corresponding position in this multi-dimensional array according to its original spatial location. In this texture feature map, the data corresponding to each pixel is a four-dimensional feature vector containing four texture feature values: contrast, correlation, energy, and homogeneity of the local region where the pixel is located. The texture feature map completely preserves the spatial distribution information of the uterine surface tissue texture features, providing structured multi-dimensional feature data for subsequent feature fusion with the tissue curl distribution map.

[0141] The beneficial effect is that by mapping the color space of the corrected image frame to grayscale, the color image is converted into a grayscale corrected image, eliminating the interference of color information on texture feature extraction. Based on this, a local neighborhood window is extracted for each target pixel in the grayscale corrected image to obtain a pixel set reflecting the local grayscale distribution. Through grayscale co-occurrence matrix feature analysis, the spatial distribution law of pixel grayscale within the window is quantified from four dimensions: contrast, correlation, energy, and homogeneity. The four-dimensional feature values ​​are combined into a four-dimensional texture feature vector and anchored to the corresponding target pixel to form a pixel-by-pixel texture feature map. Finally, all feature vectors are integrated to generate a texture feature map. Each pixel in this map contains four-dimensional texture information, which complement each other, completely depicting the differences in the microstructure of uterine surface tissue. This provides structured texture feature data support for subsequent multi-dimensional feature fusion with tissue curl distribution maps.

[0142] A5: Combine the curl values ​​of the pixels in the tissue curl distribution map with the texture features of the pixels in the texture feature map to obtain the five-dimensional fusion feature vector of the original image frame. Then, according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features, divide the fusion feature vector into feature space to obtain the fibroid candidate region of the original image frame.

[0143] In this embodiment of the invention, the step of combining the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the five-dimensional fused feature vector of the original image frame includes:

[0144] Pixel-level registration is performed between the tissue curl distribution map and the texture feature map to obtain the registration feature set of pixels at corresponding positions in the original image frame;

[0145] Based on the registration feature set, curl value feature is taken for the corresponding position pixel to obtain the first dimension feature component of the corresponding position pixel;

[0146] Based on the registration feature set, feature values ​​are taken from the corresponding pixel positions to obtain the four-dimensional feature components of the four-dimensional texture feature vector;

[0147] The first-dimensional feature component and the four-dimensional feature component are concatenated in an ordered manner to obtain the five-dimensional fused feature vector of the original image frame.

[0148] In this embodiment of the invention, the step of dividing the fused feature vector into feature spaces according to the joint determination rule of the curl feature in the tissue curl and the texture feature in the pixel texture feature to obtain the fibroid candidate region of the original image frame includes:

[0149] Obtain the curl value, texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the fused feature pixels in the five-dimensional fused feature vector;

[0150] The fibroid feature response value of the fused feature pixel is obtained by performing feature weighting calculation on the curl value, the texture contrast feature value, the texture energy feature value and the texture homogeneity feature value;

[0151] Based on a preset response threshold, the fibroid feature response value is thresholded to obtain the fibroid candidate pixel of the fused feature pixel;

[0152] The connected region formed by the candidate pixels of the fibroid is used as the candidate region of the fibroid in the original image frame.

[0153] The formula for calculating the characteristic response value of the fibroid is as follows:

[0154] ;

[0155] in, The curl value of the pixel. , , These are the texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the pixel, respectively. For a preset minimum positive number, This represents the theoretical maximum value of the texture homogeneity eigenvalue.

[0156] By aligning the tissue curl distribution map and texture feature map at the pixel level, a registration feature set is formed, where each pixel possesses both a curl value and a four-dimensional texture feature vector. Based on this set, the pixel curl value is extracted as the first dimension of the five-dimensional fused feature vector, reflecting the degree of tissue rotational deformation and serving as a key mechanical feature for distinguishing fibroids from normal muscle layers. The four-dimensional texture feature vector is extracted as the second to fifth dimensions of the vector, representing tissue grayscale changes, texture regularity, uniformity, and local consistency, respectively. By concatenating these five components in a fixed order, the five-dimensional fused feature vector for each pixel is obtained, achieving the fusion of mechanical and texture features. This provides comprehensive input for subsequent feature space segmentation, and finally, candidate regions for fibroids are extracted through thresholding.

[0157] In the formula, This is the fibroid feature response value for a pixel, which is the final comprehensive score used to determine whether the pixel belongs to a fibroid region. The higher the value, the more likely the pixel belongs to a fibroid region. The curl value of this pixel is derived from the tissue curl map. , , These are the texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the pixel, respectively, all derived from the texture feature map. It is a preset, extremely small positive number used to avoid the denominator being zero; The theoretical maximum value of the texture homogeneity eigenvalue is determined by the definition of the gray-level co-occurrence matrix. For the logarithmic smoothing term, the normalized homogeneity values ​​are logarithmically transformed. The whole as a modulation factor, for The item is scaled.

[0158] The curl value is derived from the curl value of the corresponding pixel in the tissue curl distribution map. This value reflects the degree of rotational deformation of the uterine tissue at that pixel location under respiratory motion. The texture contrast feature value, texture energy feature value, and texture homogeneity feature value are derived from the four-dimensional texture feature vector of the corresponding pixel in the texture feature map. The texture contrast feature value characterizes the drastic change in pixel grayscale values ​​within a local window, the texture energy feature value characterizes the uniformity of the texture within a local window, and the texture homogeneity feature value characterizes the local consistency of the texture within a local window. The preset minimum positive number is a very small pre-defined normal number used to ensure the calculation stability of the adjustment term when the curl value is extremely small, while avoiding a denominator of zero. The theoretical maximum value of the texture homogeneity feature value is the maximum mathematically achievable value. This maximum value is determined by the normalization method of the gray-level co-occurrence matrix and the number of gray levels, and is used to normalize the texture homogeneity feature value to the range of zero to one.

[0159] A higher response value indicates a greater likelihood that the pixel belongs to a fibroid candidate region. The base term of the formula is the texture energy feature value. A larger texture energy feature value indicates a more uniform texture within the local window. Fibroid tissue typically has a uniform grayscale and texture distribution, so the base term directly contributes to the response value. The adjustment term is constructed by dividing the texture contrast feature value by the square of the curl value, multiplying by a logarithmic factor related to the texture homogeneity feature value, and then multiplying by a preset minimum positive number for amplitude control. A larger texture contrast feature value indicates a more drastic change in grayscale within the local window. The boundary between the fibroid and surrounding tissue usually has a significant grayscale difference, so the adjustment term increases. The curl value appears in the denominator and is in square form. When the curl value is small, the adjustment term increases significantly. Because fibroid tissue has high rigidity and weak rotational deformation ability, its curl value is often small, so pixels in low-curvature regions receive greater response enhancement. Texture homogeneity features appear within the logarithmic term. The larger the texture homogeneity feature, the larger the logarithmic term and the larger the modulating term. Fibroid tissue usually has high local consistency, so pixels in high homogeneity regions receive additional response enhancement.

[0160] When the curl value approaches zero, the denominator of the adjustment term approaches zero, and the value of the adjustment term increases sharply, making the response value of low-curvature regions much higher than other regions, thus highlighting fibroid candidate pixels. When the curl value increases, the adjustment term decreases rapidly, and the main response of high-curvature regions is contributed by the texture energy feature value. These regions usually correspond to normal muscle tissue, and their response values ​​are relatively low. When the texture contrast feature value increases, the adjustment term increases linearly, allowing pixels with obvious boundary contrast to obtain higher responses. When the texture homogeneity feature value increases, the logarithmic term value increases, and the adjustment term also increases, but the growth rate gradually slows down to avoid over-enhancing regions with excessive homogeneity. The texture energy feature value is directly superimposed on the response value, ensuring that even if the contribution of the adjustment term is small, regions with uniform texture can still obtain a certain response basis. In summary, pixels that simultaneously possess low curl values, high texture contrast values, and high texture homogeneity values ​​will have significantly higher fibroid feature response values ​​than other pixels, thus being accurately screened as fibroid candidate pixels in subsequent threshold determination.

[0161] Based on a preset response threshold, a threshold determination is performed on the fibroid feature response value to obtain fibroid candidate pixels for fusion feature pixels. Threshold determination involves comparing the fibroid feature response value of each pixel with the preset response threshold. When the fibroid feature response value is greater than or equal to the preset response threshold, the pixel is marked as a fibroid candidate pixel; when the fibroid feature response value is less than the preset response threshold, the pixel is excluded. Through threshold determination, pixels that meet the fibroid feature response conditions are selected from all pixels, and these pixels constitute the potential fibroid region.

[0162] Connected regions formed by candidate fibroid pixels are used as candidate fibroid regions in the original image frame. Connected region extraction involves analyzing the connectivity of all pixels marked as candidate fibroid pixels, grouping adjacent pixels belonging to the same connected region into one region. Specifically, using the eight-neighbor connectivity rule, each candidate fibroid pixel in the image is traversed, checking for other candidate fibroid pixels in its eight surrounding directions, and connecting these pixels into the same connected region. Each connected region corresponds to an independent candidate fibroid region. These regions serve as the basis for subsequent boundary tracking and contour extraction, ultimately assisting doctors in locating the uterine fibroid.

[0163] The beneficial effect is that by fusing pixel curl values ​​and texture features, a five-dimensional fused feature vector is constructed, unifying tissue mechanical properties and surface microstructure into a multi-dimensional feature space. Based on a joint judgment rule of curl value, texture contrast, energy, and homogeneity, the fibroid feature response value is calculated. Using texture energy as the basic response term, the texture contrast is divided by the curl value and adjusted by the logarithm of texture homogeneity to form an enhancement term, significantly enhancing the response of pixels satisfying low curl, high contrast, and high homogeneity. After screening candidate fibroid pixels through a preset response threshold, they are merged into continuous fibroid candidate regions through connected component analysis. This mechanism synergizes the dual differences in mechanics and texture, using low curl to lock in high-hardness fibroid regions, high contrast to lock in regions with boundary differences, and high homogeneity to ensure regional uniformity. Multiple features corroborate each other, effectively reducing false positives from single features and providing accurate and reliable candidate regions for subsequent contour extraction.

[0164] A6: Perform boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and overlay the contour line on the original image frame as the fibroid localization marker of the original image frame.

[0165] In this embodiment of the invention, the step of performing boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and superimposing the contour line on the original image frame as a fibroid localization marker of the original image frame, includes:

[0166] A morphological closing operation is performed on the fibroid candidate region to obtain a binary mask image of the fibroid candidate region.

[0167] Eight-neighbor contour tracing is performed on the binary mask image to obtain the original contour point set of the fibroid candidate region;

[0168] The original set of contour points is simplified to obtain the contour line of the fibroid candidate region.

[0169] The contour lines are vector-mapped to obtain an annotated image frame of the original image frame, and the annotated image frame is used as a fibroid localization marker of the original image frame.

[0170] Morphological closing operations, eight-neighbor contour tracing, contour point simplification, and vector mapping are performed on the candidate fibroid regions to transform the candidate regions into smooth and continuous contour lines, which are then superimposed on the original image frame to form clear and intuitive fibroid localization markers.

[0171] A morphological closing operation is performed on the fibroid candidate region to obtain a binary mask image of the fibroid candidate region. The morphological closing operation involves sequentially performing dilation and erosion operations on the binarized fibroid candidate region image. The dilation operation slides a structuring element onto the binary image; when at least one pixel in the area covered by the structuring element is a foreground pixel, the output pixel corresponding to the center point of the structuring element is set as the foreground, thereby expanding the boundary of the candidate region and filling small holes and gaps within the region. The erosion operation slides the structuring element onto the dilated image; only when all pixels in the area covered by the structuring element are foreground pixels is the output pixel corresponding to the center point of the structuring element set as the foreground, thereby shrinking the boundary and restoring the approximate shape of the candidate region. After the dilation-erosion closing operation, any previously existing gaps are connected, and edge burrs are smoothed, ultimately forming a complete and continuous binary mask image. In this binary mask image, the foreground pixel value is one, representing the fibroid candidate region, and the background pixel value is zero, representing the non-fibroid region.

[0172] Eight-neighbor contour tracing is performed on the binary mask image to obtain the original contour point set of the fibroid candidate region. Eight-neighbor contour tracing is the process of extracting the boundary pixels of the foreground region from the binary mask image. Specifically, the process involves: first, scanning the binary mask image to find the first foreground pixel as the starting point; then, using this starting point as the center, checking its eight neighboring pixels clockwise to find the next boundary point, adding it to the contour point set. Next, using the newly found boundary point as the center and the direction of the previous boundary point as the starting search direction, continuing to check its eight neighboring pixels clockwise to find the next boundary point. This process is repeated until the starting point is reached and the search direction is consistent with the initial search direction, completing the tracing of a closed contour. Through eight-neighbor contour tracing, the boundary of the fibroid candidate region is extracted as an original contour point set consisting of a series of pixels arranged in sequence, which describes the boundary shape of the candidate region.

[0173] The original contour point set is simplified to obtain the contour line of the fibroid candidate region. Contour point simplification is a process of removing redundant and noise points while retaining the main shape features of the original contour, thereby reducing the number of contour points. Specifically, it involves traversing each point in the original contour point set and calculating the angle between the vector formed by that point and its two adjacent points. When the angle is close to a straight angle, it indicates that the point is located on an approximate straight line, and the point is removed. Simultaneously, the local curvature formed by the point and its adjacent points is calculated. When the curvature is less than a preset threshold, it indicates that the point contributes little to the shape, and the point is removed. Through contour point simplification, a large number of midpoints and tiny jagged noise points on straight line segments in the original contour point set are removed. The remaining contour points can accurately describe the boundary shape of the candidate region with fewer points, forming a smooth and concise contour line.

[0174] The contour lines are vector-mapped to obtain a labeled image frame of the original image frame, which serves as the fibroid location marker for the original image frame. Vector mapping is the process of transforming the simplified contour lines from the coordinate space of the binary mask image to the coordinate space of the original image frame, and then drawing them on the original image frame. Specifically, based on the spatial correspondence between the binary mask image and the original image frame, the coordinates of each point on the contour lines are mapped to the corresponding pixel position in the original image frame. Then, on the original image frame, adjacent contour points are connected sequentially with straight line segments to form a closed contour line, which is then outlined with a preset highlight color to make it clearly visible on the original image frame. After drawing, the highlighted fibroid contour lines are superimposed on the original image frame to form a labeled image frame, which directly presents the location and boundary of the fibroid, serving as the final fibroid location marker.

[0175] The beneficial effect is the precise optimization and visualization of fibroid boundaries through morphological closing operations and eight-neighbor contour tracking. First, morphological closing operations are performed on the candidate fibroid region, and internal micro-holes and edge breaks are filled through dilation followed by erosion, generating a complete and continuous binary mask. Then, eight-neighbor contour tracking is used to extract the boundary pixel set. Redundant and noise points are removed through contour point simplification, retaining key feature points to obtain a smooth and concise contour line. Finally, through coordinate space mapping and highlight rendering, the contour is superimposed onto the original image frame to generate an annotated image. This mechanism ensures that the final contour is complete, closed, smooth, and burr-free, clearly and accurately delineating the boundary between the fibroid and surrounding tissue, providing surgeons with intuitive and precise lesion boundary indications, and effectively assisting in rapid intraoperative fibroid localization.

[0176] like Figure 2 The diagram shown is a functional block diagram of a laparoscopic uterine fibroid localization assistance system provided in an embodiment of the present invention.

[0177] The laparoscopic uterine fibroid localization assistance system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the laparoscopic uterine fibroid localization assistance system 100 may include a required illumination correction module 101, a multi-scale registration module 102, a curl resolution module 103, a texture feature construction module 104, a fusion feature and candidate region segmentation module 105, and a contour tracking and marking module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0178] In this embodiment, the functions of each module / unit are as follows:

[0179] The illumination correction module 101 is used to acquire the original image frames in the laparoscopic surgery video stream and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames.

[0180] The multi-scale registration module 102 is used to perform multi-scale registration on the corrected image frame based on the hierarchical coarse and fine registration of the preset image pyramid, so as to obtain the dense displacement vector field of the original image frame.

[0181] The curl analysis module 103 is used to perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame;

[0182] The texture feature construction module 104 is used to construct texture features on the corrected image frame to obtain the texture feature map of the original image frame;

[0183] The fusion feature and candidate region segmentation module 105 is used to combine the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the fusion feature vector of the original image frame, and to perform feature space segmentation on the fusion feature vector according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features to obtain the fibroid candidate region of the original image frame.

[0184] The contour tracking and marking module 106 is used to perform boundary contour tracking on the fibroid candidate region to obtain the contour line of the fibroid candidate region, and to overlay the contour line on the original image frame as a fibroid location marker of the original image frame.

[0185] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0186] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0187] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0188] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0189] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A laparoscopic method for locating uterine fibroids, characterized in that, The method includes: A1: Obtain the original image frames from the laparoscopic surgery video stream, and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames; A2: Based on the hierarchical coarse and fine registration of the preset image pyramid, multi-scale registration is performed on the corrected image frame to obtain the dense displacement vector field of the original image frame; A3: Perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame; A4: Construct texture features from the corrected image frame to obtain the texture feature map of the original image frame; A5: Combine the curl values ​​of the pixels in the tissue curl distribution map with the texture features of the pixels in the texture feature map to obtain the five-dimensional fusion feature vector of the original image frame. Then, according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features, divide the fusion feature vector into feature space to obtain the fibroid candidate region of the original image frame. A6: Perform boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and overlay the contour line on the original image frame as the fibroid localization marker of the original image frame.

2. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The process of acquiring the original image frames from the laparoscopic surgery video stream and performing homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames includes: The laparoscopic surgery video stream is segmented into complete respiratory cycles at fixed time intervals to obtain the original image frames of the laparoscopic surgery video stream; The original image frame is transformed from the spatial domain to the logarithmic domain to obtain the logarithmic domain image of the original image frame; Perform a Fast Fourier Transform on the logarithmic domain image to obtain the frequency domain image of the original image frame; Based on a preset high-pass filter, the frequency domain image is subjected to high-pass filtering to obtain the corrected image of the original image frame; The corrected image is reconstructed using inverse Fourier transform to obtain the corrected image frame of the original image frame.

3. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The hierarchical coarse-fine registration based on a preset image pyramid performs multi-scale registration on the corrected image frame to obtain the dense displacement vector field of the original image frame, including: The corrected image frame is subjected to multi-level Gaussian reduction to obtain the image pyramid of the original image; Based on the corrected image frame, gradient optical flow displacement fitting is performed on the top layer of the image pyramid to obtain the displacement vector field of the top layer; The displacement vector field is upsampled and transferred to obtain the initial values ​​of the middle layer iteration of the image pyramid; Based on the middle layer image frames of the image pyramid, optical flow displacement refinement is performed on the initial values ​​of the middle layer iteration to obtain the middle layer refined displacement vector field of the image pyramid. The intermediate-level refined displacement vector field is refined layer by layer until the bottom layer of the image pyramid is reached, thus obtaining the bottom-level displacement vector field of the image pyramid. The bottom-level displacement vector field is then used as the dense displacement vector field of the original image frame.

4. The laparoscopic uterine fibroid localization aid method as described in claim 3, characterized in that, The step of fitting gradient optical flow displacement to the top layer of the image pyramid based on the corrected image frame to obtain the displacement vector field of the top layer includes: Multi-dimensional gradients are constructed on the current layer of the image pyramid to obtain the gradient image of the current layer; The current layer is divided into blocks to obtain a rectangular window of the current layer; Based on the gradient information of the gradient image, gradient coefficients are constructed for the rectangular window to obtain the overdetermined linear equation of the current layer; The overdetermined linear equations are solved by the least squares method to obtain the sparse displacement vector field of the current layer; The sparse displacement vector field is densified by bilinear interpolation to obtain the displacement vector field of the current layer.

5. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The step of performing full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame includes: The direction component of the dense displacement vector field is analyzed to obtain the horizontal displacement component field and the vertical displacement component field of the original image frame. By performing vertical partial derivatives on the pixels in the horizontal displacement component field, the vertical rate of change of the horizontal displacement of the original image frame is obtained. The horizontal partial derivative of the pixels in the vertical displacement component field is used to obtain the horizontal rate of change field of the vertical displacement of the original image frame. The horizontal displacement vertical rate of change field and the vertical displacement horizontal rate of change field are differentially analyzed pixel by pixel to obtain the full-pixel curl field of the original image frame; Based on the full-pixel curl field, the pixels of the original image frame are arranged in a regular manner to obtain the organization curl distribution map of the original image frame.

6. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The process of constructing texture features from the corrected image frame to obtain the texture feature map of the original image frame includes: The corrected image frame is subjected to color space grayscale mapping to obtain the grayscale corrected image frame of the original image frame; A local neighborhood window is extracted from the target pixel of the grayscale corrected image frame to obtain the local window of the target pixel; The gray-level co-occurrence matrix feature analysis is performed on the pixel set within the local window to obtain the texture contrast feature value, texture correlation feature value, texture energy feature value and texture homogeneity feature value of the pixel set within the window; The texture contrast feature value, texture correlation feature value, texture energy feature value, and texture homogeneity feature value are combined to form a four-dimensional texture feature vector of the original image frame; The four-dimensional texture feature vector is anchored to the center pixel of the grayscale corrected image frame to obtain the pixel-by-pixel texture feature map of the original image frame; The full feature vectors of the pixel-by-pixel texture feature map are integrated to obtain the texture feature map of the original image frame.

7. The laparoscopic uterine fibroid localization aid method as described in claim 6, characterized in that, The step of combining the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the five-dimensional fused feature vector of the original image frame includes: Pixel-level registration is performed between the tissue curl distribution map and the texture feature map to obtain the registration feature set of pixels at corresponding positions in the original image frame; Based on the registration feature set, curl value feature is taken for the corresponding position pixel to obtain the first dimension feature component of the corresponding position pixel; Based on the registration feature set, feature values ​​are taken from the corresponding pixel positions to obtain the four-dimensional feature components of the four-dimensional texture feature vector; The first-dimensional feature component and the four-dimensional feature component are concatenated in an ordered manner to obtain the five-dimensional fused feature vector of the original image frame.

8. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The step of dividing the fused feature vector into feature spaces based on the joint determination rule of the curl feature in the tissue curl and the texture feature in the pixel texture feature to obtain the fibroid candidate region of the original image frame includes: Obtain the curl value, texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the fused feature pixels in the five-dimensional fused feature vector; The fibroid feature response value of the fused feature pixel is obtained by performing feature weighting calculation on the curl value, the texture contrast feature value, the texture energy feature value and the texture homogeneity feature value; Based on a preset response threshold, the fibroid feature response value is thresholded to obtain the fibroid candidate pixel of the fused feature pixel; The connected region formed by the candidate pixels of the fibroid is used as the candidate region of the fibroid in the original image frame. The formula for calculating the characteristic response value of the fibroid is as follows: ； in, The curl value of the pixel. , , These are the texture contrast feature value, texture energy feature value, and texture homogeneity feature value of the pixel, respectively. For a preset minimum positive number, This represents the theoretical maximum value of the texture homogeneity eigenvalue.

9. The laparoscopic uterine fibroid localization aid method as described in claim 1, characterized in that, The step of performing boundary contour tracing on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and superimposing the contour line on the original image frame as a fibroid localization marker of the original image frame, includes: A morphological closing operation is performed on the fibroid candidate region to obtain a binary mask image of the fibroid candidate region. Eight-neighbor contour tracing is performed on the binary mask image to obtain the original contour point set of the fibroid candidate region; The original set of contour points is simplified to obtain the contour line of the fibroid candidate region. The contour lines are vector-mapped to obtain an annotated image frame of the original image frame, and the annotated image frame is used as a fibroid localization marker of the original image frame.

10. A laparoscopic uterine fibroid localization assistance system, characterized in that, For implementing the laparoscopic uterine fibroid localization assistance method according to claim 1, the system comprises: The illumination correction module is used to acquire the original image frames in the laparoscopic surgery video stream and perform homomorphic filtering illumination correction on the original image frames to obtain the corrected image frames of the original image frames. A multi-scale registration module is used to perform multi-scale registration on the corrected image frame based on a hierarchical coarse and fine registration of a preset image pyramid, so as to obtain the dense displacement vector field of the original image frame. The curl analysis module is used to perform full-image curl analysis on the dense displacement vector field to obtain the tissue curl distribution map of the original image frame; The texture feature construction module is used to construct texture features on the corrected image frame to obtain the texture feature map of the original image frame; The fusion feature and candidate region segmentation module is used to combine the curl values ​​of pixels in the tissue curl distribution map with the texture features of pixels in the texture feature map to obtain the fusion feature vector of the original image frame, and to perform feature space segmentation on the fusion feature vector according to the joint determination rule of the curl features in the tissue curl and the texture features in the pixel texture features to obtain the fibroid candidate region of the original image frame. The contour tracking and marking module is used to perform boundary contour tracking on the candidate fibroid region to obtain the contour line of the candidate fibroid region, and to overlay the contour line on the original image frame as a fibroid localization marker of the original image frame.

Images