CN112562855A

CN112562855A - Hepatocellular carcinoma postoperative early recurrence risk prediction method

Info

Publication number: CN112562855A
Application number: CN202011511024.5A
Authority: CN
Inventors: 黄炳升; 吴大盛; 王霁朏; 冯仕庭; 李子平; 孙美丽; 谢传淼; 王铭宇; 袁程朗; 张洪源; 黄增安; 罗梓欣; 邱峥轩; 吴煜轩; 易宸源; 李毅
Original assignee: Shenzhen University
Current assignee: Shenzhen University
Priority date: 2020-12-18
Filing date: 2020-12-18
Publication date: 2021-03-26
Anticipated expiration: 2040-12-18
Also published as: CN112562855B

Abstract

The application discloses a method for predicting early recurrence risk after hepatocellular carcinoma surgery, which comprises the following steps: acquiring a to-be-predicted image and pathological information corresponding to the to-be-predicted image; determining a feature vector corresponding to the image to be predicted based on a preset segmentation model; and determining a prediction result corresponding to the image to be predicted based on the feature vector, the pathological information and a preset prediction model. According to the method and the device, after the image to be predicted and the pathological information are obtained, the feature vector corresponding to the image to be predicted is determined through the segmentation model, the feature vector and the pathological information are learned through the prediction model, and a prediction result is obtained, so that the dependence of early recurrence prediction after hepatocellular carcinoma operation on doctor experience is reduced, and the accuracy of early recurrence prediction after hepatocellular carcinoma operation can be improved.

Description

Hepatocellular carcinoma postoperative early recurrence risk prediction method

Technical Field

The application relates to the technical field of biomedical engineering, in particular to a hepatocellular carcinoma postoperative early recurrence risk prediction method.

Background

Hepatocellular carcinoma (HCC) is a common malignant tumor, accounts for 70-85% of the total load of liver cancer worldwide, and seriously affects life and life safety of patients. The treatment of HCC employs tumor surgical resection and chemotherapy to eradicate cancer cells as much as possible, reducing the rate of cancer recurrence. However, there is still a significant proportion of patients who relapse after surgery. Researches show that portal vein invasion and liver cancer cell metastasis are two most important factors for predicting relapse, but the relapse is also influenced by factors such as chronic hepatitis, liver cirrhosis, tumor number and tumor size, and the like, and the reasons are complex. Clinical judgment of the risk of recurrence in the liver usually depends on the experience judgment of a plurality of risk factors accumulated by a doctor for a long time, and the accuracy of recurrence prediction cannot be ensured.

Disclosure of Invention

The technical problem to be solved by the present application is to provide a method for predicting risk of early recurrence after hepatocellular carcinoma surgery, aiming at the deficiencies of the prior art.

In order to solve the above technical problem, a first aspect of the embodiments of the present application provides a method for predicting risk of early recurrence after hepatocellular carcinoma surgery, the method including:

acquiring a to-be-predicted image and pathological information corresponding to the to-be-predicted image;

determining a feature vector corresponding to the image to be predicted based on a preset segmentation model;

and determining a prediction result corresponding to the image to be predicted based on the feature vector, the pathological information and a preset prediction model.

The hepatocellular carcinoma postoperative early recurrence risk prediction method is characterized in that the image to be predicted is a portal CT image, and the image to be predicted carries a hepatocellular carcinoma region.

The method for predicting early recurrence risk after hepatocellular carcinoma operation, wherein the determining the feature vector corresponding to the image to be predicted based on the preset segmentation model specifically comprises:

acquiring a plurality of tumor slice images corresponding to the image to be predicted;

determining candidate feature vectors corresponding to the tumor slices respectively based on the segmentation model;

and determining the feature vector corresponding to the image to be predicted based on all the acquired candidate feature vectors.

The method for predicting risk of early recurrence after hepatocellular carcinoma operation, wherein the determining candidate feature vectors corresponding to the respective tumor slices based on the segmentation model specifically includes:

inputting the tumor slice image into the segmentation model for each tumor slice image, and determining a candidate feature image corresponding to the tumor slice image through the segmentation model;

and converting the candidate feature map into candidate feature vectors by adopting global average pooling so as to obtain the candidate feature vectors corresponding to the tumor slices respectively.

The prediction method for early recurrence risk after hepatocellular carcinoma operation is described, wherein the vector dimensions of the candidate feature vectors are the same; the determining, based on all the obtained candidate feature vectors, the feature vector corresponding to the image to be predicted specifically includes:

selecting a target characteristic vector from all candidate characteristic vectors, and taking the candidate characteristic vectors except the target characteristic vector from all candidate characteristic vectors as reference characteristic vectors;

for each target vector element in the target feature vector, selecting a reference vector element corresponding to the target vector element from each reference feature vector;

determining feature vector elements according to the target vector elements and the corresponding reference vector elements;

and taking the feature vector formed by each determined feature vector element as the feature vector corresponding to the image to be predicted.

The method for predicting the early recurrence risk after hepatocellular carcinoma operation comprises the following steps of determining a prediction result corresponding to a to-be-predicted image based on the feature vector, pathological information and a preset prediction model:

splicing the characteristic vector and the pathological information to obtain an input vector;

and inputting the input vector into a preset prediction model, and determining a prediction result corresponding to the image to be predicted according to the prediction model.

The prediction method for early recurrence risk after hepatocellular carcinoma operation is described, wherein the segmentation model comprises an encoder and a decoder; the feature vector is determined based on an output term of an active layer in the encoder, wherein the active layer is connected to and located after a last convolutional layer in the encoder.

The method for predicting the risk of early recurrence after hepatocellular carcinoma operation is described, wherein the pathological information comprises demographic information and pathological examination information, and the pathological information is stored in a vector form.

A second aspect of embodiments of the present application provides a computer-readable storage medium storing one or more programs, which are executable by one or more processors to implement the steps of the method for predicting risk of early recurrence after hepatocellular carcinoma as described in any one of the above.

A third aspect of the embodiments of the present application provides a terminal device, including: a processor, a memory, and a communication bus; the memory has stored thereon a computer readable program executable by the processor;

the communication bus realizes connection communication between the processor and the memory;

the processor, when executing the computer readable program, implements the steps of the method for predicting risk of early recurrence after hepatocellular carcinoma surgery as described in any one of the above.

Has the advantages that: compared with the prior art, the application provides a method for predicting the risk of early recurrence after hepatocellular carcinoma surgery, which comprises the following steps: acquiring a to-be-predicted image and pathological information corresponding to the to-be-predicted image; determining a feature vector corresponding to the image to be predicted based on a preset segmentation model; and determining a prediction result corresponding to the image to be predicted based on the feature vector, the pathological information and a preset prediction model. According to the method and the device, after the image to be predicted and the pathological information are obtained, the feature vector corresponding to the image to be predicted is determined through the segmentation model, the feature vector and the pathological information are learned through the prediction model, and a prediction result is obtained, so that the dependence of early recurrence prediction after hepatocellular carcinoma operation on doctor experience is reduced, and the accuracy of early recurrence prediction after hepatocellular carcinoma operation can be improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without any inventive work.

Fig. 1 is a flowchart of a method for predicting risk of early recurrence after hepatocellular carcinoma surgery provided in the present application.

Fig. 2 is a schematic diagram of a segmentation model in the method for predicting the risk of early recurrence after hepatocellular carcinoma surgery provided by the present application.

FIG. 3 is a comparison of ROC curves for a predictive model based on training data.

FIG. 4 is a ROC curve comparison of predictive models based on validation data.

Fig. 5 shows the results of the survival analysis of training data.

Fig. 6 shows the result of the survival analysis of the verification data.

Fig. 7 is a schematic structural diagram of a terminal device provided in the present application.

Detailed Description

The present application provides a method for predicting risk of early recurrence after hepatocellular carcinoma surgery, and in order to make the purpose, technical scheme and effect of the present application more clear and definite, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.

It will be understood by those within the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

The inventor finds that hepatocellular carcinoma (HCC) is a common malignant tumor, accounts for 70-85% of the total load of liver cancer worldwide, and seriously affects the life and life safety of patients. The treatment of HCC employs tumor surgical resection and chemotherapy to eradicate cancer cells as much as possible, reducing the rate of cancer recurrence. However, there is still a significant proportion of patients who relapse after surgery. Researches show that portal vein invasion and liver cancer cell metastasis are two most important factors for predicting relapse, but the relapse is also influenced by factors such as chronic hepatitis, liver cirrhosis, tumor number and tumor size, and the like, and the reasons are complex. Clinically, the judgment of the recurrence risk in the liver usually depends on the long-term accumulated experience judgment of doctors on various risk factors, and lacks guidance of accurate data.

Medical imaging examination is an important means for HCC screening, efficacy monitoring and prognosis. The information contained in the imaging images may reflect pathological, pathophysiological, or even genomic or proteomic information. The common recurrence related indexes such as the size, the number, the edge characteristics, the satellite nodules, the macroscopic blood vessel invasion and the like of the tumor can be accurately measured and judged in the traditional imaging images. Microvascular invasion (MVI) has been proven to be an independent risk factor for HCC postoperative early recurrence and poor overall survival rate, however, MVI and HCC pathological grading are only one of the factors affecting HCC postoperative recurrence, and other factors such as oncogene, tumor cell function, liver function condition, liver fibrosis degree and the like also have a great influence on HCC prognosis.

Computed Tomography (CT) is a commonly used medical imaging technique in HCC diagnosis and preoperative planning. Currently, the liver lesion prediction work based on medical images is mainly completed by analyzing and interpreting medical images by human experts, and prognosis is predicted only by limited image feature analysis, and clinical examination information such as tumor genomes and proteomics is lacked, so that the prediction effect is poor.

In order to solve the above problem, in the embodiment of the present application, a to-be-predicted image and pathological information corresponding to the to-be-predicted image are acquired; determining a feature vector corresponding to the image to be predicted based on a preset segmentation model; and determining a prediction result corresponding to the image to be predicted based on the feature vector, the pathological information and a preset prediction model. According to the method and the device, after the image to be predicted and the pathological information are obtained, the feature vector corresponding to the image to be predicted is determined through the segmentation model, the feature vector and the pathological information are learned through the prediction model, and a prediction result is obtained, so that the dependence of early recurrence prediction after hepatocellular carcinoma operation on doctor experience is reduced, and the accuracy of early recurrence prediction after hepatocellular carcinoma operation can be improved.

The following further describes the content of the application by describing the embodiments with reference to the attached drawings.

The present embodiment provides a method for predicting risk of early recurrence after hepatocellular carcinoma surgery, as shown in fig. 1, the method includes:

and S10, acquiring the image to be predicted and the pathological information corresponding to the image to be predicted.

Specifically, the patient corresponding to the image to be predicted is the same as the patient corresponding to the pathological information, in other words, the acquiring of the image to be predicted and the pathological information corresponding to the image to be predicted refers to acquiring the image to be predicted of the patient and the pathological information of the patient, and taking the acquired pathological information as the pathological information corresponding to the image to be predicted.

The prediction image to be predicted can be acquired by a terminal device configured with the hepatocellular carcinoma postoperative early recurrence risk prediction method, can be acquired by an external device and sent to the terminal device, and can also be stored locally in the terminal device. The image to be predicted is a CT image, and the image to be predicted carries a hepatocellular carcinoma region, wherein the CT image can be a portal-stage CT image, an arterial-stage CT image and the like. In a specific implementation manner of this embodiment, the CT image may be a portal CT image, wherein a CT scanning layer thickness of the CT image is 1 mm.

The pathology information includes demographic information and pathology examination information and is stored in a vector form. The demographic information includes gender and age, the pathology examination information includes HBsAg (1+, 0-), HBV-DNA (IU/mL) (≧ 100, 1; < 100, 0), AFP continuous numerical variables, grade variables (> 400, 1; < 400,0), Child grade (A, B) grade variables, total bilirubin (μmol/L) (continuous numerical variables, grade variables < 34.2, 1; 34.2-51.3, 2 > 51.3, 3), albumin (g/L) (continuous numerical variables, grade variables > 35, 1; 28-35, 2, < 28, 3), PT time(s) (continuous numerical variables), PT extended time(s) (grade variables <4, 1; 4-6, 2; > 6, 3), ALT (U/L) (continuous numerical variables, grade <40, 0; 40, 1) GGT (U/L) (continuous numerical variables, a level variable <50, 0; more than or equal to 50, 1).

And S20, determining the feature vector corresponding to the image to be predicted based on a preset segmentation model.

In particular, the segmentation model is trained for locating a tumor region in the image to be predicted. The input item of the segmentation model is a to-be-predicted image, and the output item of the segmentation model is a tumor region in the to-be-predicted image, wherein the tumor region is determined based on the image characteristics corresponding to the to-be-predicted image learned by the segmentation model. It can be understood that, the segmentation model learns the image characteristics of the image to be predicted, and locates the image to be predicted through the image characteristics to obtain the tumor region corresponding to the image to be predicted.

In an implementation manner of this embodiment, since the CT image includes a plurality of tumor slice layers carrying tumor regions, it is necessary to acquire a plurality of tumor slice layers in the image to be predicted based on the feature vector determined by the image to be predicted, and learn each tumor slice layer respectively to obtain a candidate feature map corresponding to each tumor slice layer. Correspondingly, the determining the feature vector corresponding to the image to be predicted based on the preset segmentation model specifically includes:

In particular, the several tumor slices are all comprised within the image to be predicted. The number of tumor slice images may include all tumor slice layers in the image to be predicted, i.e., each tumor slice layer in the image to be predicted is contained within the number of tumor slice images; alternatively, the plurality of tumor slice images may include a portion of the tumor slice layer in the image to be predicted, and the portion of the tumor slice layer in the image to be predicted is not included in the plurality of tumor slice images, for example, the top-most layer of the tumor slice, the bottom-most layer of the tumor slice, and the like are not included in the plurality of tumor slice images. In one implementation of this embodiment, the tumor slice layers may include all tumor slice layers in the image to be predicted, in other words, for each tumor slice layer in the image to be predicted, the tumor slice layer is included in the tumor slice layers, and for each tumor slice layer in the tumor slice layers, the tumor slice layer is a tumor slice layer in the image to be predicted.

The candidate feature vector is determined based on the segmentation model and the tumor slice image, and is used for reflecting feature information of a tumor area carried by the tumor slice image. In one implementation, the determining, based on the segmentation model, the candidate feature vector corresponding to each tumor slice specifically includes:

Specifically, the tumor slice is an input item of the segmentation model, and the candidate feature map is a feature map of a tumor slice learned by the segmentation model. The segmentation model comprises an encoder and a decoder; the candidate feature map is an output item of an active layer in the encoder, wherein the active layer is connected with and located after a last convolutional layer in the encoder. It is understood that the candidate feature map is obtained by an active layer from an output item of a last convolutional layer in the encoder, and the active layer may be configured with a Relu activation function, so that the abstraction degree and the characterization capability of image information carried by the candidate feature map may be improved, and thus the accuracy of determining a prediction result based on the candidate feature map may be improved.

In one implementation manner of this embodiment, as shown in fig. 2, the segmentation model may adopt a U-Net structure, the segmentation model may include an encoder and a decoder, the encoder includes a plurality of down-sampling modules, the decoder includes a plurality of up-sampling modules, the plurality of down-sampling modules correspond to the plurality of up-sampling modules one-to-one, and each down-sampling module communicates with its corresponding up-sampling module through jumping connection. The down-sampling module comprises a down-sampling layer and a residual convolution block, the down-sampling layer is connected with the residual convolution block, and the output item of the down-sampling layer is the input item of the residual convolution block. The up-sampling module comprises an up-sampling layer and a residual convolution block, the up-sampling layer is connected with the residual convolution block, and the output item of the up-sampling layer is the input item of the residual convolution block. The residual convolution block comprises a first convolution unit, a second convolution unit and a convolution layer, wherein the first convolution unit and the second convolution unit are cascaded, an input item of the first convolution unit is an input item of the residual convolution block, and an input item of the second convolution unit is an output item of the first convolution unit; and the input item of the first convolution layer is the input item of the residual convolution block, and the output item of the second convolution unit and the output item of the convolution layer are subjected to residual connection to be used as the output item of the residual convolution block. The first convolution unit and the second convolution unit each include a 3 × 3 convolution layer and an activation function layer, wherein the activation function layer may be configured with a relu activation function, and the convolution layer may be a 3 × 3 convolution layer.

In this embodiment, the segmentation network may adopt cross entropy as a loss function during training, Adam as an optimizer, and a basic learning rate (lr) is 10^-5. Screening by using a Dice similarity coefficient on the verification image set to obtain an optimal segmentation model, wherein the computing formula of the Dice similarity coefficient can be as follows:

the method comprises the steps that A represents a labeling area corresponding to a training image, B represents a prediction area predicted by a preset model corresponding to a segmentation model, wherein the model structure of the preset model is the same as the model result of the segmentation model, the preset model is different from the segmentation model in that the preset model is configured with initial model parameters, and the segmentation model is configured with trained model parameters.

After the candidate feature maps corresponding to the respective tumor slices are obtained, the candidate feature maps may be converted into candidate feature vectors using global average pooling. For example, the image scale of the candidate feature map is 32 × 512, and the dimension of the candidate feature vector corresponding to the candidate feature map is 1 × 512. In addition, the vector dimension of each candidate feature vector in the candidate feature vectors corresponding to the tumor slices is the same, for example, the dimension of one candidate feature vector in the candidate feature maps is 1 × 512, and then the dimension of any candidate feature vector in the candidate feature vectors is 1 × 512.

In an implementation manner of this embodiment, the determining, based on all the obtained candidate feature vectors, a feature vector corresponding to the to-be-predicted image specifically includes:

Specifically, the dimension of the feature vector corresponding to the image to be predicted is the same as the dimension of the candidate feature vector, the target feature vector is any one of the candidate feature vectors, and the reference feature vector is a candidate feature vector of a target feature vector in the candidate feature vectors. In other words, after the candidate feature vectors are obtained, the candidate feature vectors are divided into two vector sets, a first vector set of the two vector sets includes one candidate feature vector, and a second vector set of the two vector sets includes all the candidate feature vectors except the first vector set of the candidate feature vectors. For example, the candidate feature vectors include a candidate feature vector a, a candidate feature vector B, and a candidate feature vector C, where the candidate feature vector a is the target feature vector, and then the candidate feature vector B and the candidate feature vector C are both reference feature vectors.

The reference vector element refers to a vector element whose element position in the reference feature vector is the same as the element position of the target vector element, and for any target vector element in the target feature vector, one reference vector element corresponding to the target vector element exists in the reference feature vector, because the dimension of the reference feature vector is the same as that of the target feature vector, and thus for each element position in the target feature vector, one reference element position exists in the reference feature vector and is the same as that of the element position. For example, the target feature vector is (a, B, C, D), the reference feature vector is (a, B, C, D), and then the reference vector element corresponding to the target vector element a is the reference vector element a, the reference vector element corresponding to the target vector element B is the reference vector element B, the reference vector element corresponding to the target vector element C is the reference vector element C, and the reference vector element corresponding to the target vector element D is the reference vector element D.

For each target vector element, after all reference vector elements corresponding to the target vector element are obtained, determining an average value of the target vector element and all reference vector elements corresponding to the target vector element, taking the determined average value as a feature vector element, and taking an element position corresponding to the target vector element as an element position corresponding to the feature vector element. Therefore, after the feature vector elements corresponding to the target vector elements are obtained, all the obtained feature vector elements are arranged according to the corresponding element positions, so that the feature vector corresponding to the image to be predicted is obtained. Of course, in practical applications, when determining the feature vector elements based on the target vector elements and all the reference vector elements corresponding thereto, the maximum value of the target vector elements and all the reference vector elements corresponding thereto may also be selected, and the selected maximum value is used as the feature vector element; or, the median of the target vector element and all the corresponding reference vector elements is selected, which is not described here.

In one implementation of this embodiment, since there is a portion of the tumor slices in each tumor slice of the CT image that do not contain information with resolution (e.g., the uppermost and lower slices of the tumor), redundant slices exist in the tumor slices, and the number of slices containing information with resolution in the tumor slices is greater than the number of slices that do not contain information with resolution; therefore, after the candidate feature vectors corresponding to the tumor slice images are obtained, the obtained candidate feature vectors can be filtered to remove the candidate feature vectors carrying the tumor feature information, so that the candidate feature vectors corresponding to the tumor slice images containing the resolution information in the tumor slice images can be reserved, and the candidate feature vectors corresponding to the tumor slice images not containing the resolution information in the tumor slice images are removed.

Based on this, before determining the feature vector corresponding to the image to be predicted based on all the obtained candidate feature vectors, the method includes:

performing cluster analysis on all the obtained candidate feature vectors to obtain a plurality of candidate feature vector classes;

selecting a candidate feature vector class with the largest number of images in a plurality of candidate feature vector classes as a target feature image class;

and taking the candidate feature vector included in the target feature graph as the candidate feature vector corresponding to the image to be predicted.

Specifically, the cluster analysis is configured to divide the candidate feature vectors into a plurality of candidate feature vector classes, where each of the candidate feature vector classes includes at least one candidate feature vector of the candidate feature vectors, and candidate feature vectors included in any two candidate feature vector classes in each of the candidate feature vector classes are different from each other. For example, the candidate feature vector classes include a candidate feature vector class a and a candidate feature vector class B, the candidate feature vector class a includes a candidate feature vector a and a candidate feature vector B, the candidate feature vector class B includes a candidate feature vector c and a candidate feature vector d, and then the candidate feature vector a, the candidate feature vector B, the candidate feature vector c and the candidate feature vector d are different from each other.

In an implementation manner of this embodiment, the performing cluster analysis on all the obtained candidate feature vectors may specifically perform cluster analysis on all the obtained candidate feature vectors by using a K-means clustering algorithm, and the process of cluster analysis may specifically be: dividing a plurality of candidate feature vectors into K groups of candidate feature vector groups, randomly selecting one candidate feature vector from each candidate feature vector group as a clustering center to obtain K clustering centers, then calculating Euclidean distances between each candidate feature vector and each clustering center, allocating each candidate feature vector to the nearest clustering center, and using the clustering center and the candidate feature vector allocated to the clustering center as a candidate feature vector category. In addition, each candidate feature vector is assigned, the cluster center of the cluster is recalculated according to the candidate feature vector in the cluster, and the process is repeated until a certain termination condition is met. Wherein the termination condition may be that no (or minimum number) candidate eigenvectors are reassigned to different clusters, no (or minimum number) cluster centers are changed again, the square sum of errors is locally minimal, etc. In a specific implementation of this embodiment, the number of the clustering clusters is 2, and accordingly, two candidate feature vector classes can be obtained by performing clustering analysis on a plurality of candidate feature vectors.

After a plurality of candidate eigenvector classes are obtained, the number of the candidate eigenvectors included in each candidate eigenvector class is respectively obtained, and a target candidate eigenvector class is selected according to the number corresponding to each obtained candidate eigenvector class, wherein the target candidate eigenvector class is the candidate eigenvector class including the largest number of the candidate eigenvectors in the plurality of candidate eigenvector classes. In this embodiment, the number of candidate eigenvector classes is two, and the two candidate eigenvectors are respectively a candidate eigenvector class with more tumor region features and a candidate eigenvector with less tumor region features; candidate feature vectors in the target candidate feature vector class are candidate feature vectors with more tumor region features, so that the candidate feature vectors with less tumor region features in a plurality of candidate feature vectors can be removed, the characterization of the left slices to the tumor is gathered in a feature distribution space, and redundant slice features are removed, and the slice features of each patient after aggregation are more consistent. This is because, in each tumor slice image of the CT image, there is a portion of the tumor slice image that does not contain information with resolution (e.g., the uppermost slice of the tumor), so that there are redundant slices in the tumor slice image, and the number of slices containing information with resolution in the tumor slice image is greater than the number of slices that do not contain information with resolution, so that by using cluster analysis and selecting a candidate eigenvector class containing a large number of candidate eigenvectors, the candidate eigenvectors corresponding to the tumor slice image containing information with resolution in the tumor slice image can be retained, and the candidate eigenvectors not containing information with resolution in the tumor slice image can be removed.

And S30, determining a corresponding prediction result of the target user based on the feature vector, the pathological information and a preset prediction model.

Specifically, the prediction model is a trained network model, and the prediction model is used for predicting a prediction result corresponding to a picture to be predicted, wherein the prediction result includes early recurrence or early non-recurrence. It can be understood that the prediction result corresponding to the image to be predicted is early relapse or early relapse-free, and the prediction result includes a prediction probability, and when the prediction result is early relapse, the prediction probability is early relapse probability; when the prediction result is that the early stage does not recur, the prediction probability is the early stage does not recur probability. In the embodiment, when recurrence prediction is performed on a to-be-predicted image, the feature vector determined based on the segmentation model is used as an input item, so that the accuracy of prediction categories can be improved. In addition, the early recurrence refers to recombination within a preset time after operation, and the early non-recurrence refers to non-recombination within a preset time after operation, wherein the preset time is preset, for example, 2 years and the like.

In an implementation manner of this embodiment, the prediction model selects a logistic regression model as a recurrence prediction model, is constructed by using a logistic regression algorithm in a scinit-leann machine learning library, and determines parameters of a classifier through a grid optimization strategy. The grid optimization parameter range is as follows:

1) regularization strength: 0.01,0.05,0.1,0.3,0.4,0.5,0.6,0.8,1.0, 1.2;

2) maximum number of iterations: 100,150,200,400,500, respectively;

3) the condition for stopping the iterative optimization is as follows: 1e-5,1e-4,1e-3,1 e-2.

Furthermore, the predictive model uses leave-one-out cross-validation to verify model performance: randomly selecting one sample as a test set, training the model by using the rest samples, and then testing the test set to predict the recurrence probability of the liver cancer of the sample of the test set. The process is repeated until all samples are traversed. The final model was evaluated using AUC, specificity, sensitivity, positive predictive value, negative predictive value. Model performance between different phase data and different methods was analyzed using the Delong test. In addition, in this embodiment, a model integration method is used, in which an integration model is used, the integration model includes a plurality of parallel prediction network models, each prediction model is used to predict verification data, and an average value of prediction results of the prediction models is taken as a prediction result of the integration model.

In an implementation manner of this embodiment, the segmentation model and the prediction model may be cascaded, and the cascaded model obtained by the cascading is used as the segmentation model, so that after the image to be predicted and the pathological information corresponding to the image to be predicted are obtained, the prediction result corresponding to the image to be predicted may be determined by the cascaded model. Certainly, the cascade model may include a conversion module and a clustering module, where the conversion module is configured to convert the candidate feature map into candidate feature vectors, and the clustering module is configured to cluster all the obtained candidate feature vectors. In other words, the input item of the cascade model is the image to be predicted and the pathological information corresponding to the image to be predicted, the output item of the cascade model is the prediction result, and the cascade model can perform the process of the prediction method for early recurrence risk after hepatocellular carcinoma operation provided by this embodiment.

In order to further illustrate the method for predicting the risk of early recurrence after hepatocellular carcinoma surgery provided in this example, specific experiments are given below as a basis.

The experiment was based on segmentation models constructed separately from arterial phase data and portal phase data, and the model performance of the two segmentation models is shown in table 1. The segmentation result of the portal phase is dsc-0.488, the segmentation result of the portal phase is dsc-0.868, and the segmentation result of the portal phase is superior to that of the portal phase (p-0.017 by the mann whitney test), because the image data of the portal phase is clearer, the tumor boundary is more obvious, the image noise is less, and therefore the image feature expression capability extracted by the model is better. Thus, the segmentation model in the present embodiment is obtained by training based on portal period data.

TABLE 1 deep segmentation model Performance

2) Prediction model

Based on the depth segmentation characteristics of the patients, adding pathological information (demographic information and pathological examination) of the patients as supplementary characteristics, constructing a liver cancer postoperative early recurrence prediction model, and comparing the model with a clinical macroscopic image model, wherein the macroscopic image symptom information comprises continuous numerical variables with the maximum diameter mm, grade variables (less than 3mm, 3-5mm, and more than 5mm), flat scan density intensity (0 uniform and 1 nonuniform), Intratumoral vascular endothelial vessels (0 absent and 1 present), peritumoral annular enhanced rim enhancement (0 absent and 1 present), peritumoral hepatic parenchymal enhanced peripheral hepatic enhancement (0 absent and 1 present), envelope Tumor capsule (0 absent and 1 present), Tumor edge Tumor markers (0 smooth and 1 not smooth), and Cirrhosis (0 absent and 1 present).

As a result of the experiment with the training data shown in fig. 3 and table 2, the depth segmentation feature model is better than the macroscopic image model, and after clinical information is added as supplement, the performance of the model can reach AUC 0.782. As shown in fig. 4 and table 3, the performance of the prediction model provided in this embodiment in the verification data, in which the prediction model added with the pathological information can achieve the result that AUC is 0.744, shows the excellent performance and generalization capability of the method of this embodiment.

TABLE 2 comparison of Performance of recurrence prediction models on training data

TABLE 3 comparison of Performance of recurrence prediction models on validation data

3) Survival analysis results

The predicted value of the depth migration model is used as a risk factor, the highest point of the Jordan index on the ROC curve is used as a risk factor layering threshold, risk degree grouping is respectively carried out on the patients at two centers, the survival analysis result is shown in fig. 5 and fig. 6, and the log-rank test in the K-M method has significant difference (the internal p is less than 0.001, and the external p is 0.045). The result shows that the invention has good early relapse prediction capability.

Based on the method for predicting risk of early recurrence after hepatocellular carcinoma surgery, the present embodiment provides a computer-readable storage medium, which stores one or more programs that can be executed by one or more processors to implement the steps of the method for predicting risk of early recurrence after hepatocellular carcinoma surgery as described in the above embodiments.

Based on the method for predicting risk of early recurrence after hepatocellular carcinoma surgery, the present application also provides a terminal device, as shown in fig. 7, which includes at least one processor (processor) 20; a display screen 21; and a memory (memory)22, and may further include a communication Interface (Communications Interface)23 and a bus 24. The processor 20, the display 21, the memory 22 and the communication interface 23 can communicate with each other through the bus 24. The display screen 21 is configured to display a user guidance interface preset in the initial setting mode. The communication interface 23 may transmit information. The processor 20 may call logic instructions in the memory 22 to perform the methods in the embodiments described above.

Furthermore, the logic instructions in the memory 22 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product.

The memory 22, which is a computer-readable storage medium, may be configured to store a software program, a computer-executable program, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes the functional application and data processing, i.e. implements the method in the above-described embodiments, by executing the software program, instructions or modules stored in the memory 22.

The memory 22 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal device, and the like. Further, the memory 22 may include a high speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, may also be transient storage media.

In addition, the specific processes loaded and executed by the storage medium and the instruction processors in the terminal device are described in detail in the method, and are not stated herein.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. A method for predicting risk of early recurrence after hepatocellular carcinoma surgery, comprising:

2. The method for predicting the risk of early recurrence after hepatocellular carcinoma operation as claimed in claim 1, wherein the image to be predicted is portal CT image, and the image to be predicted carries hepatocellular carcinoma region.

3. The method according to claim 1, wherein the determining the feature vector corresponding to the image to be predicted based on the preset segmentation model specifically comprises:

4. The method of claim 3, wherein the determining candidate feature vectors corresponding to the respective tumor slices based on the segmentation model specifically comprises:

5. The method of predicting risk of early recurrence after hepatocellular carcinoma as claimed in claim 3, wherein the vector dimension of several candidate feature vectors is the same; the determining, based on all the obtained candidate feature vectors, the feature vector corresponding to the image to be predicted specifically includes:

6. The method according to claim 1, wherein the determining the prediction result corresponding to the image to be predicted based on the feature vector, pathological information and a predetermined prediction model specifically comprises:

7. The method for predicting risk of early recurrence after hepatocellular carcinoma as claimed in any one of claims 1 to 6, wherein the segmentation model comprises an encoder and a decoder; the feature vector is determined based on an output term of an active layer in the encoder, wherein the active layer is connected to and located after a last convolutional layer in the encoder.

8. The method for predicting risk of early recurrence after hepatocellular carcinoma as claimed in claim 1, wherein the pathological information comprises demographic information and pathological examination information, and the pathological information is stored in vector form.

9. A computer readable storage medium storing one or more programs, which are executable by one or more processors to perform the steps of the method for predicting risk of early recurrence after hepatocellular carcinoma as described in any one of claims 1 to 8.

10. A terminal device, comprising: a processor, a memory, and a communication bus; the memory has stored thereon a computer readable program executable by the processor;

the processor, when executing the computer readable program, implements the steps of the method for predicting risk of early recurrence after hepatocellular carcinoma as described in any one of claims 1-8.