A recurrence intelligent early warning system for liver transplantation after liver cancer reduction treatment

Abstract

The present application relates to the technical field of medical data analysis, and particularly relates to a liver transplantation postoperative recurrence intelligent early warning system for liver cancer reduction treatment. In the data acquisition module, the postoperative pathological image and treatment duration of the patient are acquired; in the node identification and fiber feature extraction module, the color deconvolution technology is used to separate channels, the pixel position and gray scale features are combined to automatically identify microvessel and cancer nest nodes, and the fiber orientation vector and anisotropy degree are calculated based on the gray scale gradient distribution to provide fine information for recurrence risk assessment; in the transmission cost analysis module, in view of the dynamic competition nature of material transmission, the distance cost between pixel points is dynamically calculated through the fiber orientation vector and the anisotropy degree, the reference transmission path is acquired, and the effective transmission cost is obtained by combining the number of pixel points on the path; finally, the early warning module solves the effective transmission cost, calculates the space-time drug resistance score in combination with the treatment duration, quantitatively assesses the recurrence risk and performs early warning.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, specifically to an intelligent early warning system for recurrence after liver transplantation in the treatment of liver cancer downstaging. Background Technology

[0002] Liver transplantation is an important means for some patients with advanced hepatocellular carcinoma to achieve long-term survival. Before undergoing liver transplantation, patients with hepatocellular carcinoma often need to undergo downstaging therapy, which aims to reduce the tumor stage through transarterial chemoembolization or drug control. However, tumor recurrence after liver transplantation remains a key issue affecting the patient's prognosis, and the median survival time is significantly shortened after recurrence. Therefore, it is necessary to monitor and provide early warning of recurrence after liver transplantation in patients with hepatocellular carcinoma.

[0003] Current technologies for monitoring and warning of recurrence typically rely on morphological features such as changes in tumor diameter observed in imaging. However, under the pressure of long-term downstaging therapy and the hypoxic microenvironment, the stroma of residual tumor undergoes significant structural remodeling. In this case, simple morphological assessment is insufficient to distinguish whether the fibrotic tissue is a dense scar that completely blocks the tumor or a disguised structure that has only undergone remodeling but still retains stroma channels, which can easily lead to misjudgment and thus affect the accuracy of subsequent warnings. Summary of the Invention

[0004] To address the technical problem that under the drug stress and hypoxic microenvironment of long-term downstaging therapy, the stroma of residual tumor undergoes significant structural remodeling, making it difficult for simple morphological assessment to distinguish whether fibrotic tissue is a dense scar completely sealing the tumor or a disguised structure that has only undergone remodeling but still retains stroma channels, easily leading to misjudgment and affecting the accuracy of subsequent early warning, the present invention aims to provide an intelligent early warning system for recurrence after liver transplantation in the treatment of liver cancer downstaging. The specific technical solution adopted is as follows: This invention proposes an intelligent early warning system for recurrence after liver transplantation in the context of downstaging treatment for liver cancer. The system includes: The data acquisition module is used to acquire postoperative pathological images and treatment duration; The node recognition and fiber feature extraction module is used to perform color deconvolution processing on pathological images to obtain channel pathological images under different channels; based on the position distribution and gray-level features of pixels in the channel pathological images, microvascular nodes and cancer nest nodes are identified; based on the gray-level gradient distribution features of pixels in the channel pathological images, the fiber orientation vector and fiber anisotropy of each pixel are determined. The transmission cost analysis module is used to determine the distance cost between pixels based on the fiber orientation vector and fiber anisotropy of the pixels to obtain the baseline transmission path between the microvascular node and the cancer nest node; it counts the superposition of the occurrence times of pixels on all baseline transmission paths, and corrects the distance cost between pixels by combining the fiber anisotropy of the pixels, thereby calculating the effective transmission cost between the microvascular node and the cancer nest node. The early warning module is used to solve the effective transmission cost between microvascular nodes and cancer nest nodes based on the bipartite graph matching algorithm, and to calculate the spatiotemporal drug resistance score in combination with the treatment duration for postoperative early warning.

[0005] Furthermore, the pathological images under different channels include pathological images under the hematoxylin channel and pathological images under the eosin channel.

[0006] Furthermore, the determination of microvascular nodes and cancer nest nodes includes: The channel pathological image under the hematoxylin channel was segmented using a pre-trained segmentation network to obtain all tumor cell nucleus regions. In each tumor cell nucleus region, the pixel at the geometric centroid coordinates was used as a cancer nest node. For any pixel, if the gray value of the pixel in the eosin channel pathology image is greater than the preset red intensity threshold, and the gray value of the pixel in the hematoxylin channel pathology image is less than the preset blue intensity threshold, then the pixel is regarded as a blood vessel candidate pixel. Morphological closing operations are performed on all candidate pixels of blood vessels to obtain all red blood cell aggregation regions. In each red blood cell aggregation region, the pixel at the geometric centroid coordinates is taken as a microvascular node.

[0007] Further, determining the fiber orientation vector and fiber anisotropy degree for each pixel includes: In the eosin channel pathology image, the gray-level gradient of each pixel is calculated, and a structure tensor matrix is ​​constructed based on the gray-level gradient; Eigenvalue decomposition is performed on the structural tensor matrix to obtain two eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is used as the fiber orientation vector of the pixel. The sum of two eigenvalues ​​and the sum of a preset constant are used as the denominator, and the absolute value of the difference between the two eigenvalues ​​is used as the numerator. The resulting ratio is used as the fiber anisotropy degree of each pixel.

[0008] Furthermore, the method for obtaining the distance cost includes: In a pathological image, a pixel is randomly selected as the pixel to be tested, and the Euclidean distance between the pixel to be tested and each neighboring pixel in the preset neighborhood is used as the distance factor between the pixel to be tested and each neighboring pixel. The position of the pixel to be tested is taken as the starting point, and the position of each neighboring pixel is taken as the ending point to obtain the displacement vector between the pixel to be tested and each neighboring pixel. The absolute value of the sine of the angle between the displacement vector and the fiber orientation vector of the pixel to be tested is used as the cost factor. The distance cost between the pixel to be tested and each neighboring pixel is obtained based on the cost factor between the pixel to be tested and each neighboring pixel, the fiber anisotropy of the pixel to be tested, and the distance factor between the pixel to be tested and each neighboring pixel. The cost factor, fiber anisotropy, and distance factor are all positively correlated with the distance cost.

[0009] Furthermore, the method for obtaining the reference transmission path includes: In pathological images, each microvascular node is taken as the starting point and each cancer nest node as the ending point. The shortest path algorithm is used to search for the path with the minimum cumulative distance cost connecting the starting point and the ending point, which serves as the baseline transmission path between each microvascular node and each cancer nest node.

[0010] Furthermore, the method for obtaining the effective transmission cost includes: In pathological images, the superposition of the occurrence frequency of pixels on the baseline transmission path is analyzed to determine the path load weight of each pixel. Based on the path load weight and fiber anisotropy of each pixel, the matrix resistance coefficient of each pixel is calculated, and the path load weight and fiber anisotropy are positively correlated with the matrix resistance coefficient. In pathological images, within a preset neighborhood of each pixel, the mean of the matrix retardation coefficients of each pixel and each neighboring pixel is multiplied by the distance cost between each pixel and each neighboring pixel, and the resulting product is used as the diffusion distance of each pixel. Using each microvascular node as the starting point and each cancer nest node as the ending point, the shortest path algorithm is used to search for the minimum cumulative diffusion distance connecting the starting point and the ending point, which serves as the effective transmission cost between each microvascular node and each cancer nest node.

[0011] Furthermore, the method for obtaining the path load weight includes: In pathological images, the path overlay count for each pixel is initialized to 0; Traverse each baseline transmission path and increment the path superposition count of the pixels traversed by the baseline transmission path by 1. Traverse all baseline transmission paths to obtain the path superposition count of all pixels. The ratio of the path stack count of each pixel to the maximum path stack count is used as the path load weight of each pixel.

[0012] Furthermore, the method for obtaining the spatiotemporal drug resistance score includes: A cost matrix is ​​constructed based on the maximum value of the number of microvascular nodes and the number of cancer nest nodes. The element value at each position in the cost matrix is ​​the effective transmission cost between the corresponding microvascular node and cancer nest node. When there is no corresponding effective transmission cost at the position of the element, if the number of microvascular nodes is less than the number of cancer nest nodes, the element value is set to a preset penalty value, otherwise it is set to 0. The preset penalty value is greater than 0. The cost matrix is ​​solved using the Hungarian algorithm, and the minimum total cost output is taken as the global total transmission resistance. The ratio of the global total transmission resistance to the number of cancer nest nodes is taken as the unit transmission resistance. The treatment duration is numerically adjusted using a logarithmic function to obtain a time pressure factor. The time pressure factor is used as the numerator, and the sum of the unit transmission resistance and a preset constant is used as the denominator. The resulting ratio is used as the spatiotemporal drug resistance score.

[0013] Furthermore, the postoperative early warning includes: If the spatiotemporal drug resistance score is greater than the preset high-risk threshold, a high-risk warning is required. If the spatiotemporal drug resistance score is greater than or equal to the preset low-risk threshold and less than or equal to the preset high-risk threshold, a medium-risk warning is required. If the spatiotemporal drug resistance score is less than the preset low-risk threshold, a low-risk warning is required.

[0014] The present invention has the following beneficial effects: First, the data acquisition module obtained the patient's postoperative pathological images and treatment duration as the data foundation for subsequent analysis. Due to the upregulation of lysyl oxidase expression in the tumor microenvironment under long-term drug stress, linear rearrangement of collagen fibers occurs. To quantify this physical characteristic, pixel-level texture orientation and other features need to be extracted. Therefore, in the node recognition and fiber feature extraction module, color deconvolution technology is used to separate the pathological image channels. Microvascular nodes and cancer nest nodes are automatically identified by combining pixel location distribution and grayscale features to simulate resource competition and physical congestion in the microenvironment. Furthermore, based on the grayscale gradient distribution features, the fiber orientation vector and anisotropy of each pixel are calculated, transforming the spatial heterogeneity of fibrosis into a quantifiable indicator, providing more refined microenvironment information for recurrence risk assessment. Furthermore, given the dynamic competitive nature of material transport in the microenvironment—that morphologically similar fibrous septa exhibit drastically different transport efficiencies when faced with metabolic demands of varying densities—the transport cost analysis module dynamically calculates the distance cost between pixels using fiber orientation vectors (guidance) and anisotropy (resistance) to obtain the baseline transport path between microvascular nodes and cancer nest nodes. The effective transport cost is then obtained by adjusting the distance cost based on the frequency of pixel occurrences along the baseline transport path, realistically simulating the migration of tumor cells in a fibrotic microenvironment. Finally, in the early warning module, a spatiotemporal drug resistance score is calculated using a bipartite graph matching algorithm and treatment duration to quantitatively assess the postoperative recurrence risk. Based on the spatiotemporal drug resistance score, an early warning signal is automatically generated, effectively assisting clinicians in adjusting adjuvant therapy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system block diagram of an intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer, provided in one embodiment of the present invention. Figure 2 This is a flowchart of a method for obtaining effective transmission cost according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the system structure of an intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer, provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a system block diagram of an intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer, provided by an embodiment of the present invention. The system includes: a data acquisition module 101, a node identification and fiber feature extraction module 102, a transmission cost analysis module 103, and an early warning module 104.

[0021] The data acquisition module 101 is used to acquire postoperative pathological images and treatment duration.

[0022] Since postoperative pathological sections can reflect microscopic morphology, routine pathological sections are prepared from liver tissue samples of liver transplant patients. Hematoxylin-eosin (HE) staining is used, and the section thickness can be controlled at 3-5 μm. The pathological sections are digitally scanned using a high-resolution whole-section scanner to generate pathological images. At the same time, it is also necessary to obtain the patient's treatment duration. This can be done by accessing the hospital's electronic medical record system and extracting the date of the patient's first downstaging treatment (such as the first transarterial chemoembolization, TACE) and the date of liver transplantation. The ratio of the difference between the date of liver transplantation and the date of the first treatment to 7 days (unit: weeks) is used as the treatment duration.

[0023] It should be noted that the collection and acquisition of personal information data in this embodiment of the invention are all authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals. All numerical values ​​involved in the calculations in this embodiment of the invention have undergone data preprocessing to eliminate the influence of dimensions. The specific methods for eliminating the influence of dimensions are well-known to those skilled in the art and will not be limited or elaborated upon here.

[0024] The node recognition and fiber feature extraction module 102 is used to perform color deconvolution processing on the pathological image to obtain channel pathological images under different channels; based on the position distribution and gray-level features of the pixels in the channel pathological image, the microvascular nodes and cancer nest nodes are determined; based on the gray-level gradient distribution features of the pixels in the channel pathological image, the fiber orientation vector and fiber anisotropy of each pixel are determined.

[0025] Under long-term drug stress, the expression of lysyl oxidase in the tumor microenvironment is upregulated, prompting linear rearrangement of collagen fibers. This oriented fiber structure constitutes a low-resistance path for material diffusion. To quantify this physical characteristic, pixel-level texture orientation and coherence need to be extracted. In pathological images (HE staining), different tissue components (nucleus, cytoplasm, collagen fibers) exhibit different colors due to differences in staining characteristics. Therefore, in this module, the pathological image can first be processed by color deconvolution to obtain channel pathological images under different channels. Preferably, in this embodiment of the invention, the pathological image is processed using a color deconvolution algorithm (a known technique, the specific process is not described in detail) to separate the hematoxylin channel and the eosin channel. The channel pathological image under the hematoxylin channel mainly displays the cell nucleus and can be used to identify cancer nests with dense cell nuclei; the channel pathological image under the eosin channel mainly displays the cytoplasm and extracellular matrix and can be used to analyze collagen fiber texture.

[0026] Furthermore, vascular endothelial cells are loosely arranged and often accompanied by red blood cells, while tumor cells have high nuclear atypia and are densely arranged. These characteristics can be quantified by the grayscale representation and positional distribution of pixels. Therefore, in channel pathological images, microvascular nodes and cancer nest nodes can be determined based on the positional distribution and grayscale characteristics of pixels.

[0027] Preferably, in one embodiment of the present invention, the method for determining microvascular nodes and cancer nest nodes includes: Hematoxylin primarily stains cell nuclei (appearing blue-purple). The channel pathology image under the hematoxylin channel after deconvolution eliminates background interference from eosin staining. Furthermore, cancer nests are composed of densely packed tumor cells, typically distributed in clumps. Therefore, a pre-trained segmentation network (U-Net is used in this embodiment) can be directly used to segment the channel pathology image under the hematoxylin channel, automatically identifying and labeling the blue-highlighted tumor cell nuclei, thus obtaining all tumor cell nucleus regions. For each segmented tumor cell nucleus region, its geometric centroid (i.e., the average coordinates of all pixels in the region) is calculated. The pixel at the geometric centroid coordinates is used as a cancer nest node, representing the "center position" of a cancer nest.

[0028] In eosin staining, erythrocytes contain a large amount of hemoglobin, appearing as a strong red (high grayscale value). Erythrocytes absorb almost no hematoxylin, thus exhibiting a low grayscale value in the hematoxylin channel. Therefore, for any given pixel, if the grayscale value of that pixel in the pathological image under the eosin channel is greater than a preset red intensity threshold, and the grayscale value of that pixel in the pathological image under the hematoxylin channel is less than a preset blue intensity threshold, then that pixel is considered a candidate vascular pixel. This dual-channel thresholding strategy can more accurately capture erythrocyte characteristics and exclude cell nuclei or other basophilic structures that are mistakenly stained by eosin.

[0029] Since red blood cells are often unevenly distributed or have gaps in the blood vessel lumen, morphological closing operations (dilation followed by erosion) are performed on all candidate pixels of blood vessels to integrate discrete candidate pixels of blood vessels and obtain all red blood cell aggregation regions. Finally, consistent with the method for obtaining cancer nest nodes, the pixel at the geometric centroid coordinates in each red blood cell aggregation region is taken as a microvascular node.

[0030] It should be noted that since the gray value of the background matrix is ​​usually between 100 and 180, and the gray value of red blood cells is usually greater than 220, the preset red intensity threshold here is 200, and the preset blue intensity threshold is 50. The specific values ​​can be adjusted according to the implementation scenario and are not limited here. The morphological closing operation and the training process of the neural network are well-known technologies, and the specific process will not be described in detail here.

[0031] Collagen fibers in the tumor microenvironment are typically linearly arranged, forming a "highway" (low-resistance channel) for material diffusion. Diffusion along the fiber direction is easy, while diffusion perpendicular to the fiber direction is difficult. Furthermore, the more orderly and dense the fiber arrangement, the "harder" (more rigid) the matrix, and the greater the resistance to vertical material diffusion. Conversely, a disordered arrangement indicates a loose matrix. Gray-level gradients reflect the rate of change in image brightness. In fiber texture images, the direction of the largest gradient is perpendicular to the fiber edge, while the direction of the smallest gradient is parallel to the fiber axis. Therefore, based on the gray-level gradient distribution characteristics of pixels in a pathological image, the fiber orientation vector and fiber anisotropy of each pixel can be determined, which can be used to quantify the "direction" and "resistance" of material transport in the tumor microenvironment.

[0032] Preferably, in one embodiment of the present invention, the method for determining the fiber orientation vector and fiber anisotropy of each pixel includes: In the eosin channel pathology image, the gray-level gradient of each pixel is calculated, in the form of: ,in, This represents the rate of change of grayscale in the horizontal direction. It represents the rate of change of grayscale in the vertical direction. The grayscale gradient points in the direction of the fastest change in grayscale, that is, perpendicular to the fiber axis.

[0033] Then, a structure tensor matrix is ​​constructed based on the gray-level gradient. The structure tensor matrix is ​​as follows: in, This represents the structure tensor matrix at the pixel. The standard deviation is expressed as A Gaussian smoothing kernel, used to suppress local noise, is set to 2 pixels; This represents the convolution operation; This represents the rate of change of grayscale in the horizontal direction. This represents the rate of change of grayscale in the vertical direction.

[0034] At this point, the Gaussian-smoothed structural tensor matrix can represent the dominant direction of the texture in the local region. Then, eigenvalue decomposition is performed on the structural tensor matrix to obtain two eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the largest eigenvalue points to the direction of the most drastic grayscale change (perpendicular to the fiber edge), and the eigenvector corresponding to the smallest eigenvalue points to the direction of the most gradual grayscale change (parallel to the fiber axis). Therefore, the eigenvector corresponding to the smallest eigenvalue is taken as the fiber orientation vector of the pixel. Simultaneously, coherence (anisotropy) needs to be analyzed: the sum of the two eigenvalues ​​and a preset constant (used to prevent the denominator from being 0, a value of 0.001 can be taken as the denominator, and the absolute value of the difference between the two eigenvalues ​​is taken as the numerator. The resulting ratio is taken as the fiber anisotropy of each pixel. When the fiber anisotropy is larger, it indicates a denser fiber arrangement, higher uniformity, and greater matrix stiffness, resulting in extremely high resistance to material transport. Conversely, when the fiber anisotropy is smaller, it indicates a disordered fiber arrangement, random fiber orientation, relatively loose matrix, and relatively easy material transport.

[0035] The transmission cost analysis module 103 is used to determine the distance cost between pixels based on the fiber orientation vector and fiber anisotropy of the pixels to obtain the baseline transmission path between the microvascular node and the cancer nest node; to count the superposition of the occurrence times of pixels on all baseline transmission paths, and to correct the distance cost between pixels in combination with the fiber anisotropy of the pixels, thereby calculating the effective transmission cost between the microvascular node and the cancer nest node.

[0036] The dynamic competition inherent in material transport within the microenvironment—that is, morphologically similar fibrous spaces exhibit drastically different transport efficiencies when faced with metabolic demands of varying densities—is crucial. Since downphase therapy-induced collagen fiber bundles possess significant orientation, the resistance to material diffusion along the fiber's long axis is far less than the resistance perpendicularly passing through the fiber bundle. Therefore, to establish a physical connectivity benchmark in a congestion-free state, this invention transforms simple Euclidean distance into a biophysically defined directional diffusion distance. This involves calculating the distance cost between pixels based on their fiber orientation vector and fiber anisotropy, forcing the transport path to preferentially follow the "texture" direction, thus more realistically simulating directional diffusion within the body.

[0037] Preferably, in one embodiment of the present invention, the method for obtaining the distance cost includes: In a pathological image, a pixel is randomly selected as the test pixel. The Euclidean distance between the test pixel and each adjacent neighboring pixel within a preset neighborhood is used as the distance factor between the test pixel and each neighboring pixel. In this embodiment of the invention, the preset neighborhood is an 8-neighborhood. For horizontally or vertically connected neighboring pixels, the Euclidean distance is set to 1. Then, the Euclidean distance between diagonally connected neighboring pixels is... The distance factor is the lowest-level input for determining the reference transmission path, ensuring the basic connectivity of physical space. The larger the distance factor, the farther the positional distance between pixels, and the greater the resistance.

[0038] Then, taking the position of the pixel to be tested as the starting point and the position of each neighboring pixel as the ending point, the displacement vector between the pixel to be tested and each neighboring pixel is obtained. The angle between the displacement vector and the fiber orientation vector of the pixel to be tested is calculated. If the displacement vector is parallel to the fiber orientation vector of the pixel to be tested, it means that the movement of the pixel to be tested to that neighboring pixel is along the texture. In this case, the resistance is smaller and the cost is lower. Conversely, if the displacement vector is perpendicular to the fiber orientation vector of the pixel to be tested, it means that the movement of the pixel to be tested to that neighboring pixel is perpendicular to the fiber. In this case, the resistance is greater and the cost is higher. Therefore, the sine value of the angle is calculated here, and the absolute value of the sine value is used as the cost factor. The reason for taking the absolute value is that the resistance is not related to the direction of movement along the fiber, but only to the deviation angle. At this time, the larger the cost factor, the greater the movement resistance.

[0039] Based on the foregoing analysis, it is known that the cost factor between the measured pixel and its neighboring pixels, the fiber anisotropy of the measured pixel, and the distance factor between the measured pixel and each of its neighboring pixels are all positively correlated with the distance cost between the measured pixel and each of its neighboring pixels. Therefore, in this embodiment of the present invention, the following formula model can be constructed to calculate the distance cost: in, This represents the distance cost between the pixel p to be measured and its neighboring pixels q. This represents the distance factor between the pixel p to be measured and its neighboring pixels q. This represents the fiber anisotropy of the pixel p under test; This represents the cost factor between the pixel p to be tested and its neighboring pixels q. This indicates the preset anisotropy penalty weight.

[0040] In the formula model of distance cost, the distance factor represents the baseline cost of matter diffusion between pixels; the larger the value, the higher the cost. The larger the value, the more ordered the fibers are, and the higher the probability of perpendicularly traversing the fiber texture; the higher the cost. The preset anisotropy penalty weight is used to amplify the intensity of the anisotropic obstruction effect. In this embodiment of the invention, its value is set to 3, but the specific value can also be adjusted according to the implementation scenario. Since even in a completely ideal unobstructed environment, the diffusion of matter between two pixels requires crossing physical space, in order to prevent... When the distance cost is 0, the distance cost is also 0, so the formula model is set to... This is used to ensure the inherent distance or cost in physical space, thus guaranteeing the stability and rationality of the calculation.

[0041] In a biological context, tumor cells (cancer nests) must obtain oxygen and nutrients from microvessels to survive. Although the two are spatially separated, they can transfer substances to each other through a simple channel. Therefore, microvessel nodes can be regarded as suppliers and cancer nest nodes can be regarded as demanders. Based on this, combined with the distance cost between pixels obtained above, the baseline transmission path between any microvessel node and any cancer nest node can be obtained.

[0042] Preferably, in one embodiment of the present invention, the method for obtaining the reference transmission path includes: In pathological images, each microvascular node is taken as the starting point and each cancer nest node as the ending point. The shortest path algorithm (Dijkstra's algorithm) is used to search for the path with the minimum cumulative distance cost connecting the starting point and the ending point, which serves as the baseline transmission path between each microvascular node and each cancer nest node.

[0043] It should be noted that Dijkstra's algorithm is a well-known technique, and its specific process will not be elaborated here.

[0044] In the real microenvironment, many cancer nests compete for a limited number of blood vessels. Low-resistance transport channels are scarce resources, meaning that baseline transport paths often overlap in certain key low-resistance areas. The degree of overlap of these paths directly reflects the metabolic load pressure. Furthermore, in areas with excessive load pressure, such as rigid fibrotic regions, channels lack expansion space, and transport resistance increases sharply. Therefore, while statistically analyzing the overlap of pixel occurrences on the baseline transport path, we also consider the fiber anisotropy of pixels to correct for the distance cost between pixels. This yields the effective transport cost between microvascular nodes and cancer nest nodes. The effective transport cost includes dynamic congestion information, thus allowing for a more accurate quantification of the true biological permeability of the tumor microenvironment under long-term metabolic pressure. This helps distinguish between "completely blocked benign scars" and "malignant stroma with hidden transport channels."

[0045] Preferably, in one embodiment of the present invention, the method for obtaining the effective transmission cost includes: Please see Figure 2 The diagram illustrates a method flowchart for obtaining effective transmission cost in one embodiment of the present invention, which includes the following steps: Step S301: In the pathological image, analyze the superposition of the occurrence frequency of pixels on the baseline transmission path and determine the path load weight of each pixel.

[0046] In pathological images, the path stack count of each pixel is initialized to 0. Then, each baseline transport path is traversed, and the path stack count of each pixel traversed by the baseline transport path is incremented by 1. This process is repeated for all baseline transport paths to obtain the path stack count of all pixels. The larger the path stack count, the more baseline transport paths the pixel has been traversed, and the higher the metabolic flux pressure. The ratio of the path stack count of each pixel to the maximum path stack count is calculated and normalized to obtain the path load weight of each pixel. Based on the aforementioned logic, the larger the path load weight, the greater the path load pressure of the pixel.

[0047] Step S302: Calculate the matrix resistance coefficient of each pixel based on the path load weight and fiber anisotropy of each pixel.

[0048] Simple path superposition does not necessarily lead to blockage; its impact also depends on the physical compliance of the channel walls. In loose tissue, channels can expand to accommodate high throughput; however, in highly cross-linked, rigid fibrous regions, channels lack expansion space, and high throughput leads to an exponential increase in transport resistance. Therefore, in this sub-step, the path load weight is combined with fiber anisotropy to determine the matrix resistance coefficient, and both the path load weight and fiber anisotropy are positively correlated with the matrix resistance coefficient. In this embodiment of the invention, the formula model for the matrix resistance coefficient includes: in, Represents the matrix retardation coefficient of pixel i; This represents the path load weight of pixel i; This represents the fiber anisotropy of pixel i; This indicates the preset hysteresis gain coefficient.

[0049] In the formula model of the matrix resistance coefficient, the path load weight reflects how much material attempts to pass through that pixel. The larger the value, the higher the risk of congestion. The squaring term here simulates nonlinear resistance in fluid dynamics, causing the square term to increase significantly and the resistance to rise sharply when the flow rate approaches saturation (tending to 1). Greater fiber anisotropy indicates a denser fiber arrangement, higher uniformity, and greater matrix stiffness, resulting in extremely high resistance to material transport. The larger the value, the greater the flow pressure at the pixel point, and the more rigid and dense the matrix environment. The matrix resistance coefficient of the pixel point is obtained by weighting it using a preset resistance gain coefficient and finally adding it to a constant 1. When the matrix resistance coefficient is 1, it means that there is no additional resistance. When the matrix resistance coefficient is greater than 1, the larger the value, the more severe the congestion resistance is considered.

[0050] It should be noted that the preset blocking gain coefficient is 2, and the specific value can be adjusted according to the implementation scenario, without limitation here.

[0051] Step S303: In the pathological image, the effective transmission cost between microvascular nodes and cancer nest nodes is calculated by adjusting the distance cost using the matrix retardation coefficient between pixels.

[0052] In pathological images, the distance cost between the original pixels only considers the fiber direction. Therefore, within the preset neighborhood (8-neighborhood) of each pixel, the mean of the matrix retardation coefficient between each pixel and each neighboring pixel is multiplied by the distance cost between each pixel and each neighboring pixel. The resulting product is used as the diffusion distance of each pixel. The diffusion distance at this time takes into account the global congestion and local stiffness, so it can more realistically reflect the material transport in the microenvironment. The larger the value, the greater the transport difficulty.

[0053] Finally, using the same method, each microvascular node is taken as the starting point and each cancer nest node as the ending point. The shortest path algorithm (Dijkstra's algorithm) is used to search for the minimum cumulative diffusion distance connecting the starting point and the ending point. This minimum cumulative diffusion distance is used as the effective transmission cost between each microvascular node and each cancer nest node. The smaller the effective transmission cost, the lower the transmission difficulty.

[0054] The early warning module 104 is used to solve the effective transmission cost between microvascular nodes and cancer nest nodes based on the bipartite graph matching algorithm, and to calculate the spatiotemporal drug resistance score in combination with the treatment duration for postoperative early warning.

[0055] Tumor metabolism follows the biological principle of minimizing overall energy consumption. This means the entire system spontaneously seeks a globally optimal supply-demand allocation scheme, rather than engaging in localized, greedy connections. Cancer nests that fail to receive effective blood supply face "starvation," and this survival pressure constitutes the upper limit of transmission impedance. Simultaneously, based on the mutual exclusion principle of "evolutionary pressure and physical barrier," if the microenvironment of residual lesions can maintain a low material transmission impedance during long-term downstaging therapy, it indicates that the tumor clone has adapted to environmental pressure and broken through the physical blockade of fibrosis, thus possessing extremely high recurrence potential. Therefore, spatial transmission impedance was combined with temporal span parameters to calculate a spatiotemporal drug resistance score.

[0056] Preferably, in one embodiment of the present invention, the method for obtaining spatiotemporal drug resistance scores includes: To solve for the global optimal state and handle the situation of supply and demand imbalance, that is, the number of microvascular nodes and cancer nest nodes are not equal, a cost matrix (square matrix) can be constructed based on the maximum number of microvascular nodes and cancer nest nodes. The rows and columns are the microvascular node index and the cancer nest node index, respectively. The element value at each position in the cost matrix is ​​the effective transmission cost between the corresponding microvascular node and cancer nest node.

[0057] When there is no corresponding effective transmission cost at the location of an element, it indicates that the location is a virtual node. In this case, if the number of microvascular nodes is less than the number of cancer nest nodes, it indicates insufficient supply, and the element value is set to a preset penalty value to simulate the risk of necrosis caused by ischemia. Otherwise, it indicates abundant supply, and the element value is set to 0. The preset penalty value is greater than 0 and should be infinite. In this embodiment of the invention, it is set to 10. 6 , used to replace infinity, can be adjusted according to the implementation scenario.

[0058] Then, the Hungarian algorithm is used to solve the cost matrix. The minimum total cost output is used as the global transmission total resistance. The ratio of the global transmission total resistance to the number of cancer nest nodes is used as the unit transmission resistance. The unit transmission resistance eliminates the influence of tumor burden (number of cancer nests) on the index. The larger the value, the more effective the fibrosis barrier is, and the higher the possibility that the tumor is in a starved dormant state. Conversely, the smaller the value, the higher the channel patency, which means that there is a good transport channel. This is a signal of high drug resistance and high risk of recurrence.

[0059] Finally, biologically, drug stress does not increase linearly and infinitely over time. The drug's killing effect is most significant in the early stages of treatment. As time progresses, drug-resistant clones are selected, and the additional stress from prolonged treatment gradually diminishes. Therefore, a logarithmic function (sharp increase in the early stages, gradual decrease in the later stages) is used to numerically adjust the treatment duration, resulting in a time stress factor. ,in, As a time pressure factor, For treatment duration, This represents a logarithmic function with the natural constant e as the base. The larger the time pressure factor, the higher the likelihood of drug resistance. Therefore, the time pressure factor is used as the numerator, and the sum of the unit transmission resistance and a preset constant (which is used to prevent the denominator from being 0, and can be 0.001) is used as the denominator. The resulting ratio is used as the spatiotemporal drug resistance score. At this time, the higher the spatiotemporal drug resistance score, the greater the risk of relapse.

[0060] Once the spatiotemporal drug resistance score is obtained, postoperative early warning can be made based on this indicator.

[0061] Preferably, in one embodiment of the present invention, postoperative early warning includes: If the spatiotemporal resistance score is greater than the preset high-risk threshold, a high-risk warning is required, corresponding to an "escape lesion" in the pathological scenario. This means the tumor has successfully evaded fibrosis by mimicking blood vessels or using matrix channels, and is highly susceptible to metastasis and recurrence through residual microchannels after surgery. If the spatiotemporal resistance score is greater than or equal to the preset low-risk threshold and less than or equal to the preset high-risk threshold, a medium-risk warning is required, corresponding to a "partially inhibited lesion." This means the physical barrier exists but is not complete, and tumor activity is somewhat limited but not completely lost. If the spatiotemporal resistance score is less than the preset low-risk threshold, a low-risk warning is required, corresponding to a "blockade lesion." This means collagen fibers form a dense physical fence, effectively blocking the exchange of substances between the inside and outside, and tumor cells are in a dormant or necrotic state due to metabolic starvation. Therefore, these warning signals can assist doctors in formulating and adjusting treatment plans.

[0062] It should be noted that, in the embodiments of the present invention, the preset high-risk threshold can be calculated and obtained using historical patient data: data of liver cancer liver transplant patients whose recurrence status is labeled as high risk by doctors are screened, the spatiotemporal drug resistance scores corresponding to the data of historical patients from the United States and Russia are calculated through the above modules, and the lowest value is used as the preset high-risk threshold; the method for obtaining the preset low-risk threshold is the same.

[0063] In summary, the data acquisition module first obtained postoperative pathological images and treatment duration as the data foundation for subsequent analysis. Due to the upregulation of lysyl oxidase expression in the tumor microenvironment under long-term drug stress, collagen fibers undergo linear rearrangement. To quantify this physical characteristic, pixel-level texture orientation and other features need to be extracted. Therefore, in the node recognition and fiber feature extraction module, color deconvolution technology is used to separate pathological image channels. Microvascular nodes and cancer nest nodes are automatically identified by combining pixel location distribution and grayscale features to simulate resource competition and physical congestion in the microenvironment. Furthermore, based on the grayscale gradient distribution features, the fiber orientation vector and anisotropy of each pixel are calculated, transforming the spatial heterogeneity of fibrosis into a quantifiable indicator, providing more refined microenvironment information for recurrence risk assessment. Furthermore, given the dynamic competitive nature of material transport in the microenvironment—that morphologically similar fibrous septa exhibit drastically different transport efficiencies when faced with metabolic demands of varying densities—the transport cost analysis module dynamically calculates the distance cost between pixels using fiber orientation vectors (guidance) and anisotropy (resistance) to obtain the baseline transport path between microvascular nodes and cancer nest nodes. The effective transport cost is then obtained by adjusting the distance cost based on the frequency of pixel occurrences along the baseline transport path, realistically simulating the migration of tumor cells in a fibrotic microenvironment. Finally, in the early warning module, a spatiotemporal drug resistance score is calculated using a bipartite graph matching algorithm and treatment duration to quantitatively assess the postoperative recurrence risk. Based on the spatiotemporal drug resistance score, an early warning signal is automatically generated, effectively assisting clinicians in adjusting adjuvant therapy.

[0064] Please see Figure 3This document illustrates a schematic diagram of the system structure of an intelligent early warning system for recurrence after liver transplantation in the treatment of liver cancer downstaging, according to an embodiment of the present invention. The system includes a processor 400, a memory 401, a bus 402, and a communication interface 403. The processor 400, communication interface 403, and memory 401 are connected via the bus 402. The memory 401 may include a high-speed random access memory, and the bus 402 may be an ISA bus, PCI bus, or EISA bus, etc. The processor 400 may be an integrated circuit chip with signal processing capabilities. The memory 401 stores at least one instruction, at least one program, code set, or instruction set. When the processor loads and executes the at least one instruction, at least one program, code set, or instruction set, it implements the steps in the aforementioned modules.

[0065] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0066] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent early warning system for recurrence after liver transplantation in the treatment of liver cancer downstaging, characterized in that, The system includes: The data acquisition module is used to acquire postoperative pathological images and treatment duration; The node recognition and fiber feature extraction module is used to perform color deconvolution processing on pathological images to obtain channel pathological images under different channels; based on the position distribution and gray-level features of pixels in the channel pathological images, microvascular nodes and cancer nest nodes are identified; based on the gray-level gradient distribution features of pixels in the channel pathological images, the fiber orientation vector and fiber anisotropy of each pixel are determined. The transmission cost analysis module is used to determine the distance cost between pixels based on the fiber orientation vector and fiber anisotropy of the pixels to obtain the baseline transmission path between the microvascular node and the cancer nest node; it counts the superposition of the occurrence times of pixels on all baseline transmission paths, and corrects the distance cost between pixels by combining the fiber anisotropy of the pixels, thereby calculating the effective transmission cost between the microvascular node and the cancer nest node. The early warning module is used to solve the effective transmission cost between microvascular nodes and cancer nest nodes based on the bipartite graph matching algorithm, and to calculate the spatiotemporal drug resistance score in combination with the treatment duration for postoperative early warning.

2. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The pathological images of the different channels include pathological images of the hematoxylin channel and pathological images of the eosin channel.

3. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 2, characterized in that, The determination of microvascular nodes and cancer nest nodes includes: The channel pathological image under the hematoxylin channel was segmented using a pre-trained segmentation network to obtain all tumor cell nucleus regions. In each tumor cell nucleus region, the pixel at the geometric centroid coordinates was used as a cancer nest node. For any pixel, if the gray value of the pixel in the eosin channel pathology image is greater than the preset red intensity threshold, and the gray value of the pixel in the hematoxylin channel pathology image is less than the preset blue intensity threshold, then the pixel is regarded as a blood vessel candidate pixel. Morphological closing operations are performed on all candidate pixels of blood vessels to obtain all red blood cell aggregation regions. In each red blood cell aggregation region, the pixel at the geometric centroid coordinates is taken as a microvascular node.

4. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 2, characterized in that, Determining the fiber orientation vector and fiber anisotropy degree for each pixel includes: In the eosin channel pathology image, the gray-level gradient of each pixel is calculated, and a structure tensor matrix is ​​constructed based on the gray-level gradient; Eigenvalue decomposition is performed on the structural tensor matrix to obtain two eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is used as the fiber orientation vector of the pixel. The sum of two eigenvalues ​​and the sum of a preset constant are used as the denominator, and the absolute value of the difference between the two eigenvalues ​​is used as the numerator. The resulting ratio is used as the fiber anisotropy degree of each pixel.

5. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The method for obtaining the distance cost includes: In a pathological image, a pixel is randomly selected as the pixel to be tested, and the Euclidean distance between the pixel to be tested and each neighboring pixel in the preset neighborhood is used as the distance factor between the pixel to be tested and each neighboring pixel. The position of the pixel to be tested is taken as the starting point, and the position of each neighboring pixel is taken as the ending point to obtain the displacement vector between the pixel to be tested and each neighboring pixel. The absolute value of the sine of the angle between the displacement vector and the fiber orientation vector of the pixel to be tested is used as the cost factor. The distance cost between the pixel to be tested and each neighboring pixel is obtained based on the cost factor between the pixel to be tested and each neighboring pixel, the fiber anisotropy of the pixel to be tested, and the distance factor between the pixel to be tested and each neighboring pixel. The cost factor, fiber anisotropy, and distance factor are all positively correlated with the distance cost.

6. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The method for obtaining the reference transmission path includes: In pathological images, each microvascular node is taken as the starting point and each cancer nest node as the ending point. The shortest path algorithm is used to search for the path with the minimum cumulative distance cost connecting the starting point and the ending point, which serves as the baseline transmission path between each microvascular node and each cancer nest node.

7. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The method for obtaining the effective transmission cost includes: In pathological images, the superposition of the occurrence frequency of pixels on the baseline transmission path is analyzed to determine the path load weight of each pixel. Based on the path load weight and fiber anisotropy of each pixel, the matrix resistance coefficient of each pixel is calculated, and the path load weight and fiber anisotropy are positively correlated with the matrix resistance coefficient. In pathological images, within a preset neighborhood of each pixel, the mean of the matrix retardation coefficients of each pixel and each neighboring pixel is multiplied by the distance cost between each pixel and each neighboring pixel, and the resulting product is used as the diffusion distance of each pixel. Using each microvascular node as the starting point and each cancer nest node as the ending point, the shortest path algorithm is used to search for the minimum cumulative diffusion distance connecting the starting point and the ending point, which serves as the effective transmission cost between each microvascular node and each cancer nest node.

8. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 7, characterized in that, The method for obtaining the path load weight includes: In pathological images, the path overlay count for each pixel is initialized to 0; Traverse each baseline transmission path and increment the path superposition count of the pixels traversed by the baseline transmission path by 1. Traverse all baseline transmission paths to obtain the path superposition count of all pixels. The ratio of the path stack count of each pixel to the maximum path stack count is used as the path load weight of each pixel.

9. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The method for obtaining the spatiotemporal drug resistance score includes: A cost matrix is ​​constructed based on the maximum value of the number of microvascular nodes and the number of cancer nest nodes. The element value at each position in the cost matrix is ​​the effective transmission cost between the corresponding microvascular node and cancer nest node. When there is no corresponding effective transmission cost at the position of the element, if the number of microvascular nodes is less than the number of cancer nest nodes, the element value is set to a preset penalty value, otherwise it is set to 0. The preset penalty value is greater than 0. The cost matrix is ​​solved using the Hungarian algorithm, and the minimum total cost output is taken as the global total transmission resistance. The ratio of the global total transmission resistance to the number of cancer nest nodes is taken as the unit transmission resistance. The treatment duration is numerically adjusted using a logarithmic function to obtain a time pressure factor. The time pressure factor is used as the numerator, and the sum of the unit transmission resistance and a preset constant is used as the denominator. The resulting ratio is used as the spatiotemporal drug resistance score.

10. The intelligent early warning system for recurrence after liver transplantation for downstaging treatment of liver cancer according to claim 1, characterized in that, The postoperative early warning system includes: If the spatiotemporal drug resistance score is greater than the preset high-risk threshold, a high-risk warning is required. If the spatiotemporal drug resistance score is greater than or equal to the preset low-risk threshold and less than or equal to the preset high-risk threshold, a medium-risk warning is required. If the spatiotemporal drug resistance score is less than the preset low-risk threshold, a low-risk warning is required.

Images