Prognostic risk prediction method and device for liver transplantation of liver cancer, electronic equipment and medium

Abstract

The invention provides a prognosis risk prediction method and device for liver cancer liver transplantation, electronic equipment and a medium, and relates to the technical field of prognosis risk prediction.The method comprises the steps that a liver cancer liver transplantation block set of a target patient is obtained; respectively inputting each liver cancer liver transplantation block in the liver cancer liver transplantation block set into the single-block liver cancer liver transplantation scoring model to obtain a liver cancer liver transplantation score corresponding to each liver cancer liver transplantation block output by the single-block liver cancer liver transplantation scoring model; determining a target liver cancer liver transplantation block according to the liver cancer liver transplantation score corresponding to each liver cancer liver transplantation block and each liver cancer liver transplantation block in the liver cancer liver transplantation block set; and inputting the target liver cancer liver transplantation block into a liver cancer liver transplantation region prognosis analysis model to obtain a liver cancer liver transplantation prognosis recurrence risk prediction result of the target patient output by the liver cancer liver transplantation region prognosis analysis model. The accuracy of risk prediction is improved based on feature structure analysis.

Description

Technical Field

[0001] This invention relates to the field of prognostic risk prediction technology, and in particular to a method, device, electronic device, and medium for predicting the prognostic risk of liver transplantation for liver cancer. Background Technology

[0002] The extreme shortage of donor livers, the high incidence of postoperative complications (such as graft rejection, biliary complications, and recurrence of the primary disease), and significant individual differences in prognosis remain the core bottlenecks restricting the improvement of liver transplantation efficacy. The lack of accurate prognostic assessment tools and the failure to intervene in a timely manner in some high-risk patients lead to a significant reduction in survival rates.

[0003] Currently, in existing technologies, the prediction and assessment of recurrence after liver transplantation usually rely on models such as the Model for End-Stage Liver Disease (MELD) score, Child-Pugh classification (a clinically used grading standard for quantitatively assessing liver reserve function in patients with cirrhosis), and static statistical models (such as Cox regression, also known as the Proportional Hazards Model, used for survival analysis). However, none of these models consider the spatial distribution of liver lesions (such as hepatocellular carcinoma lesions) and fail to effectively model the spatial distribution, leading to biases in the assessment of risks such as recurrence after liver transplantation for hepatocellular carcinoma and vascular complications.

[0004] Therefore, there is an urgent need for a prognostic risk prediction method for liver transplantation in patients with liver cancer to improve the accuracy of assessment. Summary of the Invention

[0005] This invention provides a method, device, electronic device, and medium for predicting the prognostic risk of liver transplantation for liver cancer. It addresses the shortcomings of existing technologies that fail to consider the spatial distribution of liver lesions (such as liver cancer lesions) and lack effective modeling of this spatial distribution, leading to biased assessments of risks such as recurrence and vascular complications after liver transplantation. The invention introduces a single-block liver cancer transplantation scoring model to identify the target liver cancer transplantation block in a local region of the target patient from a set of liver cancer transplantation blocks. Furthermore, it introduces a regional prognostic analysis model to perform characteristic structural analysis of the target liver cancer transplantation block on the target liver cancer transplantation lesion, obtaining a prognostic risk prediction result for the target patient's liver cancer transplantation recurrence. Based on characteristic structural analysis, the accuracy of predicting the risk of liver cancer recurrence after liver transplantation is improved.

[0006] This invention provides a method for predicting the prognostic risk of liver transplantation for liver cancer, comprising the following steps.

[0007] Obtain the liver transplant block set of the target patient with liver cancer; wherein the liver transplant block set of the target patient with liver cancer includes at least two liver transplant blocks with liver cancer.

[0008] Each liver cancer transplantation block in the liver cancer transplantation block set is input into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model. The single-block liver cancer transplantation scoring model is trained based on liver cancer transplantation block samples and is a model for predicting and calculating the liver cancer transplantation score of the liver cancer transplantation block.

[0009] The target liver cancer transplantation block is determined based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks.

[0010] The target liver cancer transplantation block is input into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of recurrence of liver cancer transplantation in the target patient. The liver cancer transplantation region prognostic analysis model is trained based on the recurrence risk score samples of the liver cancer transplantation block. The model is based on the analysis of the characteristic structure of the liver cancer transplantation lesion in the target patient's liver cancer transplantation block, and thus predicts the risk of the target patient based on the characteristic structure.

[0011] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer is provided, which obtains a set of liver transplantation blocks for liver cancer in a target patient, including: obtaining liver transplantation images of liver cancer in the target patient; and segmenting the liver transplantation images to obtain a set of liver transplantation blocks for liver cancer in the target patient.

[0012] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer involves inputting each liver cancer transplantation block in a set of liver cancer transplantation blocks into a single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model. The method includes: inputting each liver cancer transplantation block in the set of liver cancer transplantation blocks into a feature extraction module in the single-block liver cancer transplantation scoring model to obtain a feature extraction map corresponding to each liver cancer transplantation block output by the feature extraction module; and inputting the feature extraction map corresponding to each liver cancer transplantation block into a single-block liver cancer transplantation scoring model. The conditional random field (CRF) in the single-block liver cancer transplantation scoring model is used to obtain the feature processing results corresponding to each liver cancer transplantation block. The feature extraction maps corresponding to each liver cancer transplantation block are then input into the global pooling module in the single-block liver cancer transplantation scoring model to obtain the pooling results corresponding to each liver cancer transplantation block output by the global pooling module. Finally, the feature processing results and pooling results corresponding to each liver cancer transplantation block are input into the result fusion module in the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the result fusion module.

[0013] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer includes determining a target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks. The method comprises: sorting all liver cancer transplantation scores to determine a set of liver cancer transplantation scores; determining candidate liver cancer transplantation scores from the set of liver cancer transplantation scores according to a preset number of liver cancer transplantation scores; and selecting the corresponding liver cancer transplantation block as the target liver cancer transplantation block from the set of liver cancer transplantation blocks based on the candidate liver cancer transplantation scores.

[0014] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer includes inputting a target liver cancer transplantation block into a liver cancer transplantation regional prognostic analysis model to obtain the predicted risk of liver cancer recurrence for the target patient. The method comprises: inputting the target liver cancer transplantation block into a feature transformation network within the liver cancer transplantation regional prognostic analysis model to obtain a feature vector output by the feature transformation network; inputting the feature vector into a graph structure feature network within the liver cancer transplantation regional prognostic analysis model to obtain a feature structure graph output by the graph structure feature network; and inputting the feature structure graph into a prediction network within the liver cancer transplantation regional prognostic analysis model to obtain the predicted risk of liver cancer recurrence for the liver cancer transplantation output by the prediction network.

[0015] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer includes inputting a feature structure map into a prediction network in a prognostic analysis model for liver transplantation of liver cancer, and obtaining the prediction result of the recurrence risk of liver cancer transplantation output by the prediction network. The method comprises: inputting the feature structure map into a regularization module in the prediction network to obtain a regularization result output by the regularization module; inputting the regularization result into an attention module in the prediction network to obtain a target feature structure map output by the attention module; and inputting the target feature structure map into a risk prediction module in the prediction network to obtain the prediction result of the recurrence risk of liver cancer transplantation output by the risk prediction module.

[0016] According to the present invention, a method for predicting the prognostic risk of liver transplantation for liver cancer includes the following steps in training a prognostic analysis model for the liver transplantation region: obtaining samples of recurrence risk scores for liver cancer transplantation blocks; inputting the samples of recurrence risk scores for liver cancer transplantation blocks into a basic prognostic analysis model to obtain a basic feature structure map output by the basic prognostic analysis model; optimizing the basic prognostic analysis model if the basic feature result map does not conform to the preset structure map, until the basic feature result map conforms to the preset structure map, thereby obtaining a prognostic analysis model for the liver transplantation region of liver cancer.

[0017] The present invention also provides a prognostic risk prediction device for liver transplantation in patients with liver cancer, comprising the following modules.

[0018] The set determination module is used to obtain the set of liver transplant blocks for liver cancer in the target patient; wherein the set of liver transplant blocks for liver cancer in the target patient includes at least two liver transplant blocks for liver cancer.

[0019] The block scoring module is used to input each liver cancer transplantation block in the liver cancer transplantation block set into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model. The single-block liver cancer transplantation scoring model is trained based on liver cancer transplantation block samples and is a model for predicting and calculating the liver cancer transplantation score of liver cancer transplantation blocks.

[0020] The block determination module is used to determine the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the liver cancer transplantation block set.

[0021] The risk prediction module is used to input the target liver cancer transplantation block into the liver cancer transplantation region prognostic analysis model, and obtain the predicted risk of liver cancer recurrence in the target patient output by the liver cancer transplantation region prognostic analysis model. The liver cancer transplantation region prognostic analysis model is trained based on the recurrence risk score samples of the liver cancer transplantation block. The model is based on the analysis of the characteristic structure of the liver cancer transplantation lesion in the target patient's liver cancer transplantation block, and thus performs risk prediction for the target patient based on the characteristic structure.

[0022] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for predicting the prognostic risk of liver transplantation for liver cancer.

[0023] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting the prognostic risk of liver transplantation for liver cancer.

[0024] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a prognostic risk prediction method for liver transplantation of liver cancer as described above.

[0025] This invention provides a method, device, electronic device, and medium for predicting the prognostic risk of liver transplantation for liver cancer. The method involves acquiring a set of liver transplantation blocks from a target patient, wherein the set includes at least two liver transplantation blocks. Each liver transplantation block in the set is input into a single-block liver transplantation scoring model to obtain the corresponding liver transplantation score for each liver transplantation block output by the single-block model. The single-block liver transplantation scoring model is trained based on liver transplantation block samples and is a model for predicting and calculating the liver transplantation score for each liver transplantation block. The target liver cancer transplantation block is determined based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks. The target liver cancer transplantation block is input into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of liver cancer recurrence for the target patient. The liver cancer transplantation region prognostic analysis model is trained based on the liver cancer transplantation block recurrence risk score sample. The model is based on the characteristic structure of the liver cancer transplantation lesion of the target patient analyzed by the target liver cancer transplantation block, and thus predicts the risk of the target patient based on the characteristic structure. The technical solution of this invention addresses the shortcomings of existing technologies that fail to consider the spatial distribution of liver lesions (such as hepatocellular carcinoma lesions) and lack effective modeling of this spatial distribution, leading to biases in the assessment of risks such as recurrence and vascular complications after liver transplantation for hepatocellular carcinoma. This invention introduces a single-block liver transplantation scoring model to identify the target liver transplantation block in a local region of the target patient from a set of liver transplantation blocks. Furthermore, it introduces a regional prognostic analysis model for liver transplantation to analyze the characteristic structure of the target liver transplantation block, thereby obtaining a prediction result of the patient's prognostic risk of recurrence after liver transplantation. Based on the characteristic structure analysis, the accuracy of predicting the risk of recurrence after liver transplantation for hepatocellular carcinoma is improved. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating the prognostic risk prediction method for liver transplantation in liver cancer provided by the present invention.

[0028] Figure 2 This is a schematic diagram illustrating the accuracy of liver transplantation site identification for liver cancer provided by the present invention.

[0029] Figure 3 This is a schematic diagram illustrating the performance evaluation of the single-block liver cancer liver transplantation scoring model provided by the present invention.

[0030] Figure 4 This is a schematic diagram of the confusion matrix provided by the present invention.

[0031] Figure 5 This is a schematic diagram of the prognostic risk prediction device for liver transplantation in liver cancer provided by the present invention.

[0032] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0034] The following is combined with Figure 1 The present invention describes a method for predicting the prognostic risk of liver transplantation for liver cancer. This method is applicable to the prognostic analysis of liver cancer. The subject executing this method can be an electronic device or a prognostic risk prediction device for liver transplantation for liver cancer installed in the electronic device. This prognostic risk prediction device for liver transplantation for liver cancer can be implemented by software, hardware, or a combination of both. Figure 1 This is a flowchart illustrating the prognostic risk prediction method for liver transplantation in liver cancer provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps 101, 102, 103 and 104.

[0035] Step 101: Obtain the liver transplant block set of the target patient's liver cancer.

[0036] In this step, the target patient's liver cancer liver transplant block set includes at least two liver cancer liver transplant blocks.

[0037] In one specific implementation, obtaining a set of liver transplant blocks for liver cancer in a target patient includes: acquiring images of liver transplants for liver cancer in the target patient; and segmenting the images of liver transplants for liver cancer in the target patient to obtain a set of liver transplant blocks for liver cancer in the target patient.

[0038] In this step, the liver transplant images for liver cancer refer to relevant images obtained from the target patient's liver transplant.

[0039] Specifically, the process involves acquiring liver transplant images of the target patient with liver cancer; then segmenting and dividing these images, which can be done by using a sliding window to perform multiple non-overlapping or overlapping segmentations, thereby obtaining a set of liver transplant blocks that includes at least two liver transplant regions.

[0040] For example, the size of the liver transplant block obtained from the cutting process can be a small patch of 224x224 pixels, but this embodiment does not limit this.

[0041] In one specific implementation, when performing the cutting process, the liver transplant image of the target patient with liver cancer is usually magnified to about 10 times before cutting to ensure that each liver transplant block contains sufficiently detailed disease cell information. At the same time, during the cutting process, the entire liver transplant image with liver cancer is traversed in a left-to-right and top-to-bottom order. This embodiment does not limit this process.

[0042] In one specific implementation, after the set of liver transplantation blocks for liver cancer is obtained, the coordinate position (x, y) of each liver transplantation block in the set of liver cancer transplantation blocks is recorded to facilitate the use of subsequent models.

[0043] Step 102: Input each liver cancer transplantation block in the liver cancer transplantation block set into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model.

[0044] In this step, the single-block liver cancer liver transplantation scoring model is trained based on liver cancer liver transplantation block samples. The single-block liver cancer liver transplantation scoring model is a model for predicting and calculating the liver cancer liver transplantation score of the liver cancer liver transplantation block.

[0045] Among them, a single-block liver cancer liver transplantation scoring model can be, for example, ResNet34. ResNet34 consists of multiple residual blocks, each containing two 3×3 convolutional layers and skip connections to solve the gradient vanishing problem in deep network training.

[0046] Specifically, after obtaining each liver cancer transplantation block in the liver cancer transplantation block set, each liver cancer transplantation block is sequentially input into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to the liver cancer transplantation block output by the single-block liver cancer transplantation scoring model.

[0047] In one specific embodiment, before inputting each liver cancer transplantation block in the liver cancer transplantation block set into the single-block liver cancer transplantation scoring model, the format of each liver cancer transplantation block in the liver cancer transplantation block set is further converted, for example, it can be converted to red, green, blue (RGB) format, and the converted liver cancer transplantation blocks are standardized (standardization can include the mean and standard deviation of ImageNet, which is a common normalization method in image preprocessing). The set of liver cancer transplantation blocks with standardized liver cancer transplantation blocks is then input into the single-block liver cancer transplantation scoring model. This embodiment does not limit this step.

[0048] In one specific implementation, each liver cancer transplantation block in the liver cancer transplantation block set is input into a single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model. This includes: inputting each liver cancer transplantation block in the liver cancer transplantation block set into the feature extraction module of the single-block liver cancer transplantation scoring model to obtain the feature extraction map corresponding to each liver cancer transplantation block output by the feature extraction module; and inputting the feature extraction map corresponding to each liver cancer transplantation block into the single-block liver cancer transplantation scoring model. The Conditional Random Field (CRF) in the liver transplantation scoring model is used to obtain the feature processing results corresponding to each liver transplantation block of hepatocellular carcinoma. The feature extraction maps corresponding to each liver transplantation block of hepatocellular carcinoma are then input into the global pooling module in the single-block liver transplantation scoring model to obtain the pooling results corresponding to each liver transplantation block of hepatocellular carcinoma output by the global pooling module. The feature processing results and pooling results corresponding to each liver transplantation block of hepatocellular carcinoma are then input into the result fusion module in the single-block liver transplantation scoring model to obtain the liver transplantation score corresponding to each liver transplantation block of hepatocellular carcinoma output by the result fusion module.

[0049] In this step, one liver transplantation block for liver cancer corresponds to multiple feature extraction maps, but this embodiment does not limit this.

[0050] Specifically, each liver cancer transplantation block in the liver cancer transplantation block set is input into the feature extraction module of the single-block liver cancer transplantation scoring model. The first layer of the feature extraction module is a 7×7 convolutional layer with a stride of 2, which halves the size of the liver cancer transplantation block. Then, it passes through a 3×3 max pooling layer to further halves the size. Following this are four stages of residual block groups: The first residual module extracts local texture features through three repeated residual units, focusing on microstructures such as nuclear chromatin distribution and cell membrane boundaries. The second residual module is activated after downsampling, using four residual units to capture mesoscale tissue architecture features, identifying histological patterns such as gland formation and matrix distribution. The third residual module extracts macroscopic contextual features through six residual units, modeling complex spatial relationships such as tumor-stromal interactions. The fourth residual module finally uses three residual units to generate high-level semantic features, encoding pathological discrimination patterns related to tumor malignancy, resulting in the feature extraction map corresponding to each liver cancer transplantation block.

[0051] For the feature maps obtained in the previous step, the feature extraction maps corresponding to each liver transplantation block of hepatocellular carcinoma are input into the Conditional Random Field (CRF) in the single-block liver transplantation scoring model to obtain the feature processing results corresponding to each liver transplantation block of hepatocellular carcinoma output by the CRF. The feature extraction maps corresponding to each liver transplantation block of hepatocellular carcinoma are then input into the global pooling module in the single-block liver transplantation scoring model. The global pooling module compresses each feature extraction map to obtain the pooling results corresponding to each liver transplantation block of hepatocellular carcinoma output by the global pooling module. The pooling results can be, for example, 512-dimensional feature vectors. Finally, the feature processing results and pooling results corresponding to each liver transplantation block of hepatocellular carcinoma are input into the result fusion module in the single-block liver transplantation scoring model. The result fusion module performs a bitwise summation of the feature processing results and pooling results corresponding to each liver transplantation block of hepatocellular carcinoma, and finally obtains the liver transplantation score corresponding to each liver transplantation block of hepatocellular carcinoma output by the result fusion module.

[0052] For example, when the feature extraction maps corresponding to each liver transplantation block of hepatocellular carcinoma are input into the CRF, the CRF first constructs a univariate potential function. A fully connected layer maps the feature vectors of the feature extraction map to an initial tumor / non-tumor probability distribution, establishing a preliminary diagnosis for each independent block. A spatially constrained bivariate potential function is then introduced: a Gaussian kernel distance matrix is ​​constructed based on the block coordinates of the feature extraction map to model the label consistency of adjacent tissue blocks. Simultaneously, a learnable feature interaction layer captures deep pathological feature associations across blocks. Furthermore, iterative optimization is performed based on the energy function using a mean-field approximation, synchronously updating the probability estimates of adjacent blocks in each iteration, thus improving the accuracy of the CRF's output feature processing results.

[0053] Step 103: Determine the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the liver cancer transplantation block set.

[0054] In one specific implementation, determining the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the liver cancer transplantation block set includes: sorting all liver cancer transplantation scores to determine a liver cancer transplantation score set; determining candidate liver cancer transplantation scores from the liver cancer transplantation score set according to a preset number of liver cancer transplantation scores; and selecting the corresponding liver cancer transplantation block as the target liver cancer transplantation block from the liver cancer transplantation block set based on the candidate liver cancer transplantation scores.

[0055] Specifically, after obtaining the liver transplantation score corresponding to each liver cancer transplantation block, all liver cancer transplantation scores are sorted to determine the liver cancer transplantation score set; candidate liver cancer transplantation scores are determined from the liver cancer transplantation score set according to the preset number of liver cancer transplantation scores; in the liver cancer transplantation block set, the corresponding liver cancer transplantation block is selected as the target liver cancer transplantation block according to the candidate liver cancer transplantation scores.

[0056] For example, if the preset number of liver cancer liver transplantation scores is 20, the liver cancer liver transplantation blocks corresponding to the top 20 liver cancer liver transplantation scores are selected from the liver cancer liver transplantation block set as the target liver cancer liver transplantation blocks. This embodiment does not limit this.

[0057] For example, the trained single-block liver transplant scoring model for liver cancer was validated using 1190 WSI images from 757 post-transplant liver cancer patients across five hospitals. Figure 2 This is a schematic diagram illustrating the accuracy of liver transplantation site identification for liver cancer provided by the present invention, as shown below. Figure 2 As shown, the training model for single-block liver cancer transplantation scoring is used to analyze WSI images to determine the accuracy of tumor and non-tumor identification in WSI images. The right side shows the pre-obtained baseline true values ​​for judging whether the liver cancer transplantation block is tumor or non-tumor, with colors ranging from light to dark as 0.00, 0.25, 0.50, 0.75, and 1.00. Figure 2 In the study, the accuracy rate for identifying non-tumor conditions in liver transplantation sites for liver cancer reached 0.985 (false positive rate 0.015), and the accuracy rate for identifying tumor conditions reached 0.932 (false positive rate 0.068). Figure 3 This is a schematic diagram illustrating the performance evaluation of the single-block liver transplantation scoring model for hepatocellular carcinoma provided by this invention, as shown below. Figure 3As shown, the area under the curve (AUC) of the single-block liver transplantation scoring model for liver cancer in the WSI image for each liver transplantation block is 0.9904. Figure 3 The horizontal axis represents the false alarm rate in the evaluation process of the single-block liver cancer liver transplantation scoring model, the vertical axis represents the true positive rate in the evaluation process of the single-block liver cancer liver transplantation scoring model, the diagonal dashed line represents random guessing, and the blue curve represents the area under the curve of the liver cancer liver transplantation score corresponding to each liver cancer liver transplantation block evaluated by the single-block liver cancer liver transplantation scoring model.

[0058] Step 104: Input the target liver cancer liver transplantation block into the liver cancer liver transplantation region prognostic analysis model to obtain the predicted recurrence risk of liver cancer in the target patient output by the liver cancer liver transplantation region prognostic analysis model.

[0059] In this step, the liver cancer liver transplantation region prognostic analysis model is trained based on the recurrence risk score samples of liver cancer liver transplantation blocks. The liver cancer liver transplantation region prognostic analysis model is a model that analyzes the characteristic structure of liver cancer liver transplantation lesions in target patients based on the target liver cancer liver transplantation blocks, and then predicts the risk of target patients based on the characteristic structure.

[0060] In one specific implementation, the target liver cancer transplantation block is input into a liver cancer transplantation regional prognostic analysis model to obtain the predicted risk of liver cancer recurrence after transplantation for the target patient, output by the liver cancer transplantation regional prognostic analysis model. This includes: inputting the target liver cancer transplantation block into a feature transformation network in the liver cancer transplantation regional prognostic analysis model to obtain a feature vector output by the feature transformation network; inputting the feature vector into a graph structure feature network in the liver cancer transplantation regional prognostic analysis model to obtain a feature structure graph output by the graph structure feature network; and inputting the feature structure graph into a prediction network in the liver cancer transplantation regional prognostic analysis model to obtain the predicted risk of liver cancer recurrence after transplantation output by the prediction network.

[0061] In this step, the feature transformation network can be, for example, CONtrastive learning from Captions for Histopathology (CONCH), which is a visual language foundation model for histopathology. This embodiment does not limit this to a specific model.

[0062] Specifically, after obtaining the target liver transplantation block for hepatocellular carcinoma, the target liver transplantation block is input into the feature transformation network in the hepatocellular carcinoma liver transplantation region prognostic analysis model. Based on the feature transformation module, the target liver transplantation block is transformed to obtain a multi-dimensional feature vector output by the feature transformation network, which can be, for example, a 512-dimensional feature vector, denoted as the feature vector. , Representing the One data input, This indicates the number of liver transplant blocks for the target liver cancer, but this embodiment does not limit this number.

[0063] Furthermore, after obtaining the multidimensional feature vector, it is input into the graph structure feature network in the hepatocellular carcinoma liver transplantation region prognostic analysis model. The graph structure feature network then performs a spatial neighborhood search operation, calculating the Euclidean distance between the block center point and all other blocks based on the block center point position in the full-slice coordinate system of the target hepatocellular carcinoma liver transplantation block. The k-nearest neighbor algorithm is used to precisely select the 8 blocks with the smallest spatial distance as direct neighbors for each core node. Subsequently, a weighted undirected graph connection is constructed, establishing bidirectional edges between the core node and its neighbors to ensure spatial proximity, thus obtaining the feature structure graph output by the graph structure feature network. Feature structure diagram Used for prognostic model analysis. That is, the feature structure map is finally input into the prediction network of the hepatocellular carcinoma liver transplantation region prognostic analysis model to obtain the prediction result of the risk of recurrence after hepatocellular carcinoma liver transplantation output by the prediction network.

[0064] In one specific implementation, the feature structure map is input into the prediction network of the liver transplant prognosis analysis model for liver cancer, and the prediction network outputs the prediction result of the risk of recurrence after liver transplantation for liver cancer. This includes: inputting the feature structure map into the regularization module in the prediction network to obtain the regularization result output by the regularization module; inputting the regularization result into the attention module in the prediction network to obtain the target feature structure map output by the attention module; and inputting the target feature structure map into the risk prediction module in the prediction network to obtain the prediction result of the risk of recurrence after liver transplantation for liver cancer output by the risk prediction module.

[0065] In this step, the regularization module includes two domain regularized graph network layers (GIN Block1) and one dropout regularized graph network layer (GIN Block2). The attention module can be, for example, an attention block, but this embodiment does not limit it.

[0066] Since the target patients' liver cancer transplantation blocks are derived from data prediction and analysis from different hospitals (different data sources), it is necessary to consider the distribution differences between these data sources. Therefore, GINBlock is constructed based on the GIN graph neural network layer. Its core lies in aggregating neighborhood node information and fusing it with its own features through a multilayer perceptron (MLP). Compared to traditional graph neural networks (GNNs) that rely on mean or max pooling aggregation methods, GIN employs a summation-based computation mechanism. This design enhances information transmission capabilities, enabling the model to obtain a more expressive graph representation.

[0067] Specifically, for each feature structure graph The results are processed through a regularization module, an attention module, and a risk prediction module to obtain the prediction results of the recurrence risk after liver transplantation for liver cancer in the target patients.

[0068] For example, the representation of a node in a feature structure graph can be expressed as shown in formula (1).

[0069] (1) In formula (1), Represents a node exist Features of the layer Represents a node The set of neighboring nodes, express Layer application of multilayer perceptron, when In the case of (i.e., the initial layer) Represents a node initial feature vector For each feature Both methods use Batch Normalization (BN) to accelerate training and alleviate the gradient explosion / vanishing problem. They also combine Regularization (IN) to handle differences in content from different data sources and independently standardize each sample, ignoring statistical information between samples. Finally, they use the Rectified Linear Unit (RELU) activation function to achieve non-linear mapping.

[0070] In the GIN Block2 module, Dropout regularization is introduced to reduce co-adaptation relationships between neurons during training to prevent overfitting of deep neural networks. However, it is not used during testing to ensure the integrity and accuracy of the prognostic analysis model for liver transplantation regions of liver cancer.

[0071] The advantage of this setting is that the core idea of ​​regularization is to combine IN and BN, which overcomes the performance difference problem between the training domain and the testing domain of traditional normalization methods. It can eliminate the problem that the output results of the prognostic analysis model of liver cancer liver transplantation area are inaccurate due to the difference in the risk score of liver cancer recurrence in liver transplantation blocks of the same patient in different hospitals. It effectively alleviates the subjective differences in the assessment of surgical conditions by different doctors.

[0072] Finally, a Graph Attention Network (GAT) is used to apply a learnable linear transformation to each node, projecting the original 512-dimensional feature vector into a high-dimensional latent space. During the attention coefficient calculation phase, feature-level interaction operations are performed for each pair of adjacent nodes: the transformed features of the source and target nodes are concatenated, and a single-layer feedforward neural network is used to generate the original attention score. This score is non-linearized using the LeakyReLU activation function and then normalized among neighboring nodes using the softmax function, ultimately generating an interpretable attention weight distribution. Finally, global pooling is used to aggregate the features of all nodes in the graph, and a linear mapping layer is applied to achieve the final prognostic score, G-TRS prediction.

[0073] In G-TRS calculations, this score corresponds to the prognostic risk prediction of liver cancer recurrence after liver transplantation, as output by the liver transplantation regional prognostic analysis model for liver cancer.

[0074] For example, given that a single target patient may have multiple target hepatocellular carcinoma liver transplant blocks, these images need to be integrated to achieve patient-level prognostic analysis. The hepatocellular carcinoma liver transplant region prognostic analysis model obtains the predicted recurrence risk of hepatocellular carcinoma liver transplant for the same patient by averaging the prognostic scores G-TRS corresponding to all target hepatocellular carcinoma liver transplant blocks in the image representation stage and performing risk assessment through recurrence-free survival (RFS). Its mathematical expression is as follows: formula (2).

[0075] (2) In formula (2), Indicates the patient's first Global representation of a block Indicates the number of floors. Indicates the total number of blocks. This represents the multilayer perceptron mapping function for predicting G-TRS.

[0076] The advantage of this setup is that it aligns the RFS time: considering the subjective differences in surgical condition assessments among different physicians, the record of recurrence-free survival (RFS) time may be biased, thus affecting the accuracy of the prediction results of recurrence risk after liver transplantation for liver cancer. Therefore, an RFS branch is constructed to predict the RFS time for each block. A supervised learning loss with observation state constraints is provided, as shown in Equation (3).

[0077] (3) In formula (3), Represents a node In the last layer Features Represents global features. This indicates that the multi-layer attention mechanism maps the structure graph to the RFS prediction value. The function is represented as 1 when a recurrence event is observed, and 0 otherwise.

[0078] In one specific implementation, the training of the prognostic analysis model for liver transplantation of hepatocellular carcinoma includes the following steps: obtaining a sample of recurrence risk scores for liver transplantation blocks of hepatocellular carcinoma; inputting the sample of recurrence risk scores for liver transplantation blocks of hepatocellular carcinoma into the basic prognostic analysis model to obtain the basic feature structure map output by the basic prognostic analysis model; if the basic feature result map does not conform to the preset structure map, optimizing the basic prognostic analysis model until the basic feature result map conforms to the preset structure map, thereby obtaining the prognostic analysis model for liver transplantation of hepatocellular carcinoma.

[0079] In this step, the prognostic analysis model for liver transplantation regions of liver cancer is defined as GraphNet.

[0080] Specifically, a sample of recurrence risk scores for liver transplantation blocks of liver cancer is obtained; the sample of recurrence risk scores for liver transplantation blocks of liver cancer is input into the basic prognostic analysis model to obtain the basic feature structure map output by the basic prognostic analysis model; if the basic feature result map does not conform to the preset structure map, the basic prognostic analysis model is optimized until the basic feature result map conforms to the preset structure map, thus obtaining the prognostic analysis model for the liver transplantation region of liver cancer.

[0081] In one specific implementation, for example, the HS-1+SL dataset, YA dataset, HS-2 dataset, and CY dataset can be used to train, test, and apply the hepatocellular carcinoma (HCC) liver transplantation regional prognostic analysis model. The accuracy of the HCC liver transplantation regional prognostic analysis model in predicting the risk of recurrence after HCC liver transplantation is determined to be 0.77, while the prediction accuracy of the existing Milan standard is 0.62 and the prediction accuracy of the University of California, San Francisco (UCSF) standard is 0.64. Therefore, the HCC liver transplantation regional prognostic analysis model of this application is superior to the existing Milan standard and the University of California, San Francisco (UCSF) standard.

[0082] In one specific implementation, during the process of training a liver cancer liver transplantation region prognostic analysis model based on a basic prognostic analysis model, the target loss function for model training is defined as shown in formula (4).

[0083] (4) In formula (4), This represents the negative log-likelihood loss in prognostic analysis. This represents the RFS time alignment loss corresponding to formula (3); and correspond and The weighting factor for the loss; These represent different survival states, where 1 indicates that the event has occurred (e.g., a case of death), and 0 indicates that the event has not occurred. Represents the survival time of the model simulation; This represents the actual survival time of the corresponding case; The risk prediction results represent the model.

[0084] In one specific implementation, Figure 4 This is a schematic diagram of the confusion matrix provided by the present invention, as shown below. Figure 4 As shown, the confusion matrix obtained by using the HS-1+SL dataset, YA dataset, HS-2 dataset, and CY dataset to determine the prognostic analysis model of liver transplantation region for liver cancer, the Milan criterion, and the UCSF criterion for predicting the risk of recurrence after liver transplantation for liver cancer is illustrated. Figure 4In the table, (1) represents the confusion matrix of the prediction results of the risk of recurrence after liver transplantation for hepatocellular carcinoma based on the HS-1+SL dataset using the UCSF standard, with a prediction accuracy of 0.64; (2) represents the confusion matrix of the prediction results of the risk of recurrence after liver transplantation for hepatocellular carcinoma based on the HS-1+SL dataset using the Milan standard, with a prediction accuracy of 0.62; (3) represents the confusion matrix of the prediction results of the prediction results of the regional prognostic analysis model for liver transplantation for hepatocellular carcinoma based on the HS-1+SL dataset, with a prediction accuracy of 0.77; (4) represents the confusion matrix of the prediction results of the risk of recurrence after liver transplantation for hepatocellular carcinoma based on the CY dataset using the UCSF standard, with a prediction accuracy of 0.69; (5) represents the confusion matrix of the prediction results of the risk of recurrence after liver transplantation for hepatocellular carcinoma based on the CY dataset using the Milan standard, with a prediction accuracy of 0.68; and (6) represents the confusion matrix of the prediction results of the regional prognostic analysis model for liver transplantation for hepatocellular carcinoma based on the CY dataset, with a prediction accuracy of 0. .75; (7) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the HS-2 dataset using the UCSF standard, with a prediction accuracy of 0.58; (8) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the HS-2 dataset using the Milan standard, with a prediction accuracy of 0.54; (9) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the HS-2 dataset using the regional prognostic analysis model for liver cancer after liver transplantation, with a prediction accuracy of 0.74; (10) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the YA dataset using the UCSF standard, with a prediction accuracy of 0.54; (11) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the YA dataset using the Milan standard, with a prediction accuracy of 0.48; (12) represents the confusion matrix of the prediction results of the recurrence risk of liver cancer after liver transplantation based on the YA dataset using the regional prognostic analysis model for liver cancer after liver transplantation, with a prediction accuracy of 0.68. Figure 4 The presentation of the confusion matrix shows that the regional prognostic analysis model for liver cancer transplantation is superior to the Milan criteria and UCSF criteria in predicting the risk of recurrence after liver cancer transplantation. This embodiment does not limit this aspect.

[0085] This invention provides a method for predicting the prognostic risk of liver transplantation for liver cancer. The method involves obtaining a set of liver transplantation blocks for a target patient, wherein the set includes at least two liver transplantation blocks. Each liver transplantation block in the set is input into a single-block liver transplantation scoring model to obtain the corresponding liver transplantation score for each liver transplantation block output by the single-block model. The single-block liver transplantation scoring model is trained based on liver transplantation block samples and is a model for predicting and calculating the liver transplantation score for each liver transplantation block. Based on each liver transplantation block... The target liver cancer transplantation block is determined by the liver cancer transplantation score corresponding to the liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks. The target liver cancer transplantation block is input into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of liver cancer recurrence for the target patient. The liver cancer transplantation region prognostic analysis model is trained based on the liver cancer transplantation block recurrence risk score samples. The model is based on the characteristic structure of the liver cancer transplantation lesion in the target patient, and thus predicts the risk of the target patient based on the characteristic structure. Based on the above embodiments, the technical solution of the present invention addresses the shortcomings of existing technologies that fail to consider the spatial distribution of liver lesions (such as hepatocellular carcinoma lesions) and fail to effectively model the spatial distribution, leading to biases in the assessment of risks such as recurrence and vascular complications in liver transplantation for hepatocellular carcinoma. The present invention introduces a single-block liver transplantation scoring model to identify the target liver transplantation block in a local region of the target patient from a set of liver transplantation blocks for hepatocellular carcinoma. Furthermore, it introduces a regional prognostic analysis model for liver transplantation to analyze the characteristic structure of the target liver transplantation block for hepatocellular carcinoma lesions in the target patient, thereby obtaining a prediction result of the recurrence risk of liver transplantation for the target patient. Based on the characteristic structure analysis, the accuracy of predicting the risk of recurrence in liver transplantation for hepatocellular carcinoma is improved.

[0086] The prognostic risk prediction device for liver transplantation of liver cancer provided by the present invention will be described below. The prognostic risk prediction device for liver transplantation of liver cancer described below can be referred to in correspondence with the prognostic risk prediction method for liver transplantation of liver cancer described above.

[0087] Figure 5 This is a schematic diagram of the prognostic risk prediction device for liver transplantation in liver cancer provided by the present invention, with reference to... Figure 5 As shown, the prognostic risk prediction device 500 for liver transplantation in liver cancer includes: a set determination module 501, a block scoring module 502, a block determination module 503, and a risk prediction module 504.

[0088] The set determination module 501 is used to obtain the set of liver transplantation blocks for liver cancer of the target patient; wherein the set of liver transplantation blocks for liver cancer of the target patient includes at least two liver transplantation blocks.

[0089] The block scoring module 502 is used to input each liver cancer transplantation block in the liver cancer transplantation block set into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model; wherein, the single-block liver cancer transplantation scoring model is obtained by training based on liver cancer transplantation block samples, and the single-block liver cancer transplantation scoring model is a model for predicting and calculating the liver cancer transplantation score of the liver cancer transplantation block.

[0090] The block determination module 503 is used to determine the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the liver cancer transplantation block set.

[0091] The risk prediction module 504 is used to input the target liver cancer transplantation block into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of liver cancer recurrence in the target patient. The liver cancer transplantation region prognostic analysis model is trained based on the recurrence risk score samples of the liver cancer transplantation block. The liver cancer transplantation region prognostic analysis model is a model that analyzes the characteristic structure of the liver cancer transplantation lesion in the target patient based on the target liver cancer transplantation block, and then performs risk prediction for the target patient based on the characteristic structure.

[0092] In one example embodiment, the set determination module 501 is specifically used to: acquire liver transplant images of the target patient with liver cancer; and segment the liver transplant images to obtain a set of liver transplant blocks of the target patient with liver cancer.

[0093] In one example embodiment, the block scoring module 502 is specifically used for: inputting each liver cancer transplantation block in the liver cancer transplantation block set into the feature extraction module of the single-block liver cancer transplantation scoring model to obtain the feature extraction map corresponding to each liver cancer transplantation block output by the feature extraction module; inputting the feature extraction map corresponding to each liver cancer transplantation block into the conditional random field of the single-block liver cancer transplantation scoring model to obtain the feature processing result corresponding to each liver cancer transplantation block output by the conditional random field; inputting the feature extraction map corresponding to each liver cancer transplantation block into the global pooling module of the single-block liver cancer transplantation scoring model to obtain the pooling result corresponding to each liver cancer transplantation block output by the global pooling module; and inputting the feature processing result and pooling result corresponding to each liver cancer transplantation block into the result fusion module of the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the result fusion module.

[0094] In one example embodiment, the block determination module 503 is specifically used to: sort all liver cancer liver transplantation scores to determine a set of liver cancer liver transplantation scores; determine candidate liver cancer liver transplantation scores from the set of liver cancer liver transplantation scores according to a preset number of liver cancer liver transplantation scores; and select the corresponding liver cancer liver transplantation block as the target liver cancer liver transplantation block from the set of liver cancer liver transplantation blocks according to the candidate liver cancer liver transplantation scores.

[0095] In one example embodiment, the risk prediction module 504 is specifically used for: inputting the target liver cancer liver transplantation block into the feature transformation network in the liver cancer liver transplantation region prognostic analysis model to obtain the feature vector output by the feature transformation network; inputting the feature vector into the graph structure feature network in the liver cancer liver transplantation region prognostic analysis model to obtain the feature structure map output by the graph structure feature network; and inputting the feature structure map into the prediction network in the liver cancer liver transplantation region prognostic analysis model to obtain the liver cancer liver transplantation prognostic recurrence risk prediction result output by the prediction network.

[0096] In one example embodiment, the risk prediction module 504 inputs the feature structure map into the prediction network of the liver cancer liver transplantation region prognostic analysis model to obtain the liver cancer liver transplantation prognostic recurrence risk prediction result output by the prediction network. Specifically, it is used to: input the feature structure map into the regularization module in the prediction network to obtain the regularization result output by the regularization module; input the regularization result into the attention module in the prediction network to obtain the target feature structure map output by the attention module; and input the target feature structure map into the risk prediction module in the prediction network to obtain the liver cancer liver transplantation prognostic recurrence risk prediction result output by the risk prediction module.

[0097] In one example embodiment, the device further includes a model training module. The model training module is configured to: acquire samples of recurrence risk scores for liver transplantation sites of hepatocellular carcinoma; input the samples of recurrence risk scores for liver transplantation sites of hepatocellular carcinoma into a basic prognostic analysis model to obtain a basic feature structure map output by the basic prognostic analysis model; and optimize the basic prognostic analysis model if the basic feature result map does not conform to the preset structure map, until the basic feature result map conforms to the preset structure map, thereby obtaining a prognostic analysis model for the liver transplantation region of hepatocellular carcinoma.

[0098] The apparatus of this embodiment can be used to execute the method of any embodiment in the side embodiment of the prognostic risk prediction method for liver transplantation of liver cancer. Its specific implementation process and technical effects are similar to those in the side embodiment of the prognostic risk prediction method for liver transplantation of liver cancer. For details, please refer to the detailed description in the side embodiment of the prognostic risk prediction method for liver transplantation of liver cancer, which will not be repeated here.

[0099] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a prognostic risk prediction method for liver transplantation of liver cancer. This method includes: acquiring a set of liver transplantation blocks for the target patient; wherein the set of liver transplantation blocks for the target patient includes at least two liver transplantation blocks; inputting each liver transplantation block in the set of liver transplantation blocks into a single-block liver transplantation scoring model to obtain the liver transplantation score corresponding to each liver transplantation block output by the single-block liver transplantation scoring model; wherein the single-block liver transplantation scoring model is trained based on liver transplantation block samples, and the single-block liver transplantation scoring model predicts and calculates the liver transplantation risk of the liver transplantation block. The model for liver transplantation scoring is used; the target liver cancer transplantation block is determined based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks; the target liver cancer transplantation block is input into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of liver cancer recurrence for the target patient output by the liver cancer transplantation region prognostic analysis model; wherein, the liver cancer transplantation region prognostic analysis model is trained based on the liver cancer transplantation block recurrence risk score sample, and the liver cancer transplantation region prognostic analysis model is based on the characteristic structure of the liver cancer transplantation lesion of the target patient analyzed by the target liver cancer transplantation block, and thus the risk prediction of the target patient is based on the characteristic structure.

[0100] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the prognostic risk prediction method for liver transplantation of liver cancer provided by the above methods. The method includes: obtaining a set of liver transplantation blocks for liver cancer in a target patient; wherein the set of liver transplantation blocks for liver cancer in the target patient includes at least two liver transplantation blocks; inputting each liver transplantation block in the set of liver transplantation blocks into a single-block liver transplantation scoring model to obtain the liver transplantation score corresponding to each liver transplantation block output by the single-block liver transplantation scoring model; wherein the single-block liver transplantation scoring model is obtained by training based on liver transplantation block samples. The single-block liver cancer liver transplantation scoring model is a model for predicting and calculating the liver cancer transplantation score of a liver cancer transplantation block. It determines the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and the liver cancer transplantation blocks in the set of liver cancer transplantation blocks. The target liver cancer transplantation block is then input into the liver cancer transplantation region prognostic analysis model to obtain the predicted recurrence risk of the target patient's liver cancer transplantation. The liver cancer transplantation region prognostic analysis model is trained based on samples of liver cancer transplantation block recurrence risk scores. It is a model that analyzes the characteristic structure of the target patient's liver cancer transplantation lesion based on the target liver cancer transplantation block, thereby predicting the target patient's risk based on the characteristic structure.

[0102] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for predicting the prognostic risk of liver transplantation for liver cancer provided by the methods described above. This method includes: obtaining a set of liver transplantation blocks for a target patient; wherein the set of liver transplantation blocks for the target patient includes at least two liver transplantation blocks; inputting each liver transplantation block in the set of liver transplantation blocks into a single-block liver transplantation scoring model to obtain a liver transplantation score corresponding to each liver transplantation block output by the single-block liver transplantation scoring model; wherein the single-block liver transplantation scoring model is trained based on liver transplantation block samples. The model is a prediction and calculation model for liver transplantation scores of liver cancer transplantation blocks. Target liver cancer transplantation blocks are determined based on the liver transplantation scores corresponding to each liver cancer transplantation block and the liver cancer transplantation blocks in the set of liver cancer transplantation blocks. The target liver cancer transplantation block is then input into the liver cancer transplantation region prognostic analysis model to obtain the predicted recurrence risk of the target patient's liver cancer transplantation. The liver cancer transplantation region prognostic analysis model is trained based on samples of recurrence risk scores from liver cancer transplantation blocks. This model analyzes the characteristic structure of the liver cancer transplantation lesions in the target patient based on the target liver cancer transplantation block, thereby predicting the risk of the target patient based on the characteristic structure.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units 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. Those skilled in the art can understand and implement this without any creative effort.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the prognostic risk of liver transplantation for liver cancer, characterized in that, include: Obtain a set of liver transplantation blocks for liver cancer in a target patient; wherein the set of liver transplantation blocks for liver cancer in the target patient includes at least two liver transplantation blocks for liver cancer. Each liver cancer transplantation block in the set of liver cancer transplantation blocks is input into a single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model; wherein, the single-block liver cancer transplantation scoring model is obtained by training based on liver cancer transplantation block samples, and the single-block liver cancer transplantation scoring model is a model for predicting and calculating the liver cancer transplantation score of the liver cancer transplantation block; The target liver cancer transplantation block is determined based on the liver cancer transplantation score corresponding to each liver cancer transplantation block and each liver cancer transplantation block in the set of liver cancer transplantation blocks. The target liver cancer transplantation block is input into the liver cancer transplantation region prognostic analysis model to obtain the predicted risk of recurrence of liver cancer in the target patient. The liver cancer transplantation region prognostic analysis model is trained based on samples of recurrence risk scores for liver cancer transplantation blocks. This model is based on analyzing the characteristic structure of the liver cancer transplantation lesion in the target patient using the target liver cancer transplantation block, thereby predicting the risk of recurrence based on this characteristic structure.

2. The method for predicting the prognostic risk of liver transplantation for liver cancer according to claim 1, characterized in that, The acquisition of the liver transplant block set for the target patient with liver cancer includes: Obtain images of liver transplantation for liver cancer from the target patient; The liver transplant image of the liver cancer is segmented to obtain the set of liver transplant blocks of the target patient.

3. The prognostic risk prediction method for liver transplantation in liver cancer according to claim 1, characterized in that, The step of inputting each liver cancer transplantation block in the set of liver cancer transplantation blocks into a single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model includes: Each liver cancer liver transplantation block in the liver cancer liver transplantation block set is input into the feature extraction module in the single-block liver cancer liver transplantation scoring model to obtain the feature extraction map corresponding to each liver cancer liver transplantation block output by the feature extraction module. The feature extraction maps corresponding to each liver cancer liver transplantation block are respectively input into the conditional random field in the single-block liver cancer liver transplantation scoring model to obtain the feature processing results corresponding to each liver cancer liver transplantation block output by the conditional random field. The feature extraction maps corresponding to each liver cancer liver transplantation block are respectively input into the global pooling module in the single-block liver cancer liver transplantation scoring model to obtain the pooling results corresponding to each liver cancer liver transplantation block output by the global pooling module. The feature processing results and pooling results corresponding to each liver cancer liver transplantation block are respectively input into the result fusion module in the single-block liver cancer liver transplantation scoring model to obtain the liver cancer liver transplantation score corresponding to each liver cancer liver transplantation block output by the result fusion module.

4. The method for predicting the prognostic risk of liver transplantation for liver cancer according to claim 1, characterized in that, The step of determining the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each of the liver cancer transplantation blocks and each of the liver cancer transplantation blocks in the set of liver cancer transplantation blocks includes: All the liver transplant scores for liver cancer are sorted to determine the set of liver transplant scores for liver cancer. Candidate liver cancer transplant scores are determined from the set of liver cancer transplant scores based on the preset number of liver cancer transplant scores. In the set of liver cancer transplantation blocks, the corresponding liver cancer transplantation block is selected as the target liver cancer transplantation block based on the candidate liver cancer transplantation score.

5. The method for predicting the prognostic risk of liver transplantation for liver cancer according to claim 1, characterized in that, The step of inputting the target liver cancer liver transplantation block into the liver cancer liver transplantation region prognostic analysis model to obtain the liver cancer liver transplantation region prognostic prediction result of the target patient's liver cancer liver transplantation risk output by the liver cancer liver transplantation region prognostic analysis model includes: The target liver cancer liver transplantation block is input into the feature transformation network in the liver cancer liver transplantation region prognostic analysis model to obtain the feature vector output by the feature transformation network. The feature vector is input into the graph structure feature network in the liver transplantation region prognostic analysis model for liver cancer to obtain the feature structure map output by the graph structure feature network. The feature structure map is input into the prediction network of the liver transplantation prognostic analysis model for liver cancer, and the prediction result of the recurrence risk of liver cancer after liver transplantation is obtained from the prediction network.

6. The prognostic risk prediction method for liver transplantation in liver cancer according to claim 5, characterized in that, The step of inputting the feature structure map into the prediction network of the liver transplantation prognostic analysis model for liver cancer, and obtaining the prediction result of the recurrence risk of liver cancer after liver transplantation output by the prediction network, includes: The feature structure map is input into the regularization module in the prediction network to obtain the regularization result output by the regularization module; The regularization result is input into the attention module in the prediction network to obtain the target feature structure map output by the attention module. The target feature structure diagram is input into the risk prediction module in the prediction network to obtain the prediction result of the risk of recurrence after liver transplantation of liver cancer output by the risk prediction module.

7. The method for predicting the prognostic risk of liver transplantation for liver cancer according to any one of claims 1-6, characterized in that, The training of the liver transplantation region prognostic analysis model for liver cancer includes the following steps: Obtain the recurrence risk score sample of the liver transplantation block for liver cancer; The recurrence risk score samples of the liver transplantation block of liver cancer are input into the basic prognostic analysis model to obtain the basic feature structure diagram output by the basic prognostic analysis model. If the basic feature result map does not conform to the preset structure map, the basic prognostic analysis model is optimized until the basic feature result map conforms to the preset structure map, thus obtaining the prognostic analysis model for the liver transplantation region of liver cancer.

8. A prognostic risk prediction device for liver transplantation in patients with liver cancer, characterized in that, include: A set determination module is used to obtain a set of liver transplantation blocks for liver cancer in a target patient; wherein the set of liver transplantation blocks for liver cancer in the target patient includes at least two liver transplantation blocks for liver cancer. The block scoring module is used to input each liver cancer transplantation block in the liver cancer transplantation block set into the single-block liver cancer transplantation scoring model to obtain the liver cancer transplantation score corresponding to each liver cancer transplantation block output by the single-block liver cancer transplantation scoring model; wherein, the single-block liver cancer transplantation scoring model is obtained by training based on liver cancer transplantation block samples, and the single-block liver cancer transplantation scoring model is a model for predicting and calculating the liver cancer transplantation score of the liver cancer transplantation block; The block determination module is used to determine the target liver cancer transplantation block based on the liver cancer transplantation score corresponding to each of the liver cancer transplantation blocks and each of the liver cancer transplantation blocks in the liver cancer transplantation block set. The risk prediction module is used to input the target liver cancer transplantation block into the liver cancer transplantation region prognostic analysis model to obtain the liver cancer transplantation region prognostic recurrence risk prediction result of the target patient output by the liver cancer transplantation region prognostic analysis model; wherein, the liver cancer transplantation region prognostic analysis model is trained based on the liver cancer transplantation block recurrence risk score sample, and the liver cancer transplantation region prognostic analysis model is a model that analyzes the characteristic structure of the liver cancer transplantation lesion of the target patient based on the target liver cancer transplantation block, thereby predicting the risk of the target patient based on the characteristic structure.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the prognostic risk prediction method for liver transplantation of liver cancer as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the prognostic risk prediction method for liver transplantation of liver cancer as described in any one of claims 1 to 7.

Images