A liver cancer prognosis analysis method, system and device based on multi-task learning

Abstract

The application provides a liver cancer prognosis analysis method, system and device based on multi-task learning, which comprises the following steps: obtaining a liver cancer case data set; constructing a liver cancer prognosis analysis model based on a multi-task neural network; training the liver cancer prognosis analysis model based on the multi-task neural network by using the liver cancer case data set to obtain a trained liver cancer prognosis analysis model based on the multi-task neural network; obtaining liver cancer case data to be predicted and inputting the liver cancer case data to be predicted into the trained liver cancer prognosis analysis model based on the multi-task neural network to obtain the predicted survival state of the liver cancer case at each time point and estimate the postoperative survival time; and performing risk factor screening and risk assessment based on the trained liver cancer prognosis analysis model. The liver cancer prognosis analysis method based on multi-task learning is adopted, so that the problem of low accuracy of the existing liver cancer prognosis analysis is solved from the root.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing and machine learning technology, specifically relating to a method, system and device for liver cancer prognostic analysis based on multi-task learning. Background Technology

[0002] Hepatocellular carcinoma (HCC) is a primary malignant tumor of the liver, mainly developing from chronic hepatitis and cirrhosis. HCC is the 7th most common cancer and the 2nd leading cause of cancer-related death worldwide, causing more than 800,000 deaths annually. Most HCC patients are diagnosed at an intermediate or advanced stage. According to the Barcelona staging system (BCLC), transarterial chemoembolization (TACE) is the preferred treatment for intermediate-stage (stage B) patients. However, due to the complex etiology of liver cancer and its high heterogeneity within the population, the overall survival after TACE in intermediate-stage patients varies greatly, ranging from 11 to 45 months, indicating a poor prognosis. Therefore, analyzing postoperative survival in HCC is crucial for the personalized treatment of liver cancer patients, providing clinicians with information for postoperative health management decisions.

[0003] With the increasing volume of modern medical data, artificial intelligence models have been introduced to process this data and extract valuable information. Currently, most established liver cancer prognostic models, both domestically and internationally, rely on traditional statistical models, which are not well-suited to handling censoring issues in survival data, resulting in poor predictive performance. Multi-task learning is an important research area in current artificial intelligence models. It primarily reduces prediction errors for each task by simultaneously learning multiple tasks and utilizing shared representations from related tasks.

[0004] Therefore, how to conduct accurate prognostic analysis of liver cancer and improve prediction accuracy is an urgent problem to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, system, and device for liver cancer prognostic analysis based on multi-task learning. This invention, based on multi-task learning, combines clinical data on liver cancer, deep neural network technology, and multi-task learning technology, thereby fundamentally solving the problem of low accuracy in existing liver cancer prognostic analyses.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for prognostic analysis of liver cancer based on multi-task learning, characterized by the following steps:

[0008] S1. Obtain a liver cancer case dataset; the liver cancer case dataset is represented as {(X i T i δ i)|i=1,2,...,n},where X i Let T be the feature vector of the i-th liver cancer case; i Let δ be the survival time of the i-th liver cancer case; i Let δ represent the censored state of the i-th liver cancer case. i A value of 0 indicates that the case data is censored, meaning the patient remained alive until the end of the follow-up period. i =1 indicates that the case data is complete, meaning the patient died before the end of the follow-up period;

[0009] The characteristics of the liver cancer cases include at least the following: etiology and type, alpha-fetoprotein, hemoglobin, platelets, white blood cells, aspartate aminotransferase, lactate dehydrogenase, albumin, total bilirubin, C-reactive protein, hepatic encephalopathy, tumor diameter, lesion location, number of intrahepatic lesions, new lesions, portal vein invasion, portal vein invasion, hepatic vein invasion, invasion of ductus venosus or atrium, ascites, and distant metastasis. The types include: HBV, HCV, and no viral infection.

[0010] S2. Construct a prognostic analysis model for liver cancer based on a multi-task neural network, including:

[0011] The survival time conversion divides the survival time of liver cancer cases into J time points, where J is the maximum survival time of the included cases. The survival time of each case is then converted into its survival status at these J time points, represented by "1" for death, "0" for survival, and "?" for unknown. After conversion, the survival time of each case is transformed into J survival statuses, denoted by y. i = [y1, y2, ..., y J The transformed label vector is represented by ]; the feature vector X of liver cancer cases is used to represent the label vector. i and the transformed label vector y i As input to a multi-task neural network;

[0012] The liver cancer prognostic analysis model based on a multi-task neural network includes a three-layer structure: an input layer, a parameter sharing layer, and a multi-task output layer; the input layer receives the feature vector X of the liver cancer case. i and the transformed label vector y iThe input layer and the parameter sharing layer are fully connected. The parameter sharing layer consists of multiple layers, and the multi-task parameter sharing method uses a hard parameter sharing mechanism, meaning that the layers in the sharing layer are connected through a fully connected mechanism. The output part of the network begins to branch, including J output tasks. Each output task corresponds to two nodes, which correspond to the probability of death or survival, respectively. When the survival probability is greater than the death probability, it means that the case is alive at this time point. Otherwise, the J output tasks correspond to the survival status at J time points. The survival time of the patient is predicted by the survival status at these J time points. In addition, a dropout layer and a softmax layer are added before the output layer. The dropout layer improves the generalization ability of the model, and the softmax layer maps the output to (0,1), making the output result a probability.

[0013] S3. Using the liver cancer case dataset, train the liver cancer prognostic analysis model based on multi-task neural network to obtain a trained liver cancer prognostic analysis model based on multi-task neural network.

[0014] Specifically, during model training, the objective function of the liver cancer prognostic analysis model based on a multi-task neural network consists of two loss functions: survival loss and ranking loss, as follows:

[0015]

[0016] Where α and β are positive constants used to adjust the weights of survival loss and ranking loss, θ is the network parameter, and λ is the regularization parameter;

[0017] The calculation methods for survival loss and ranking loss are as follows:

[0018] (1) Survival loss

[0019] Uncensored data is labeled with two states: alive and dead. However, due to censoring, the survival status of censored cases is known at some time points, but unknown at others. Therefore, the survival loss consists of two parts, calculated as follows:

[0020] L cen (θ)=Lo bs (y ij ,p(X i |θ))+λL unlabel (p(X i |θ))

[0021] Among them, y ij p(X) represents the actual label value of case i at time point j. i |θ) represents the label value for case prediction, L obs L represents the loss of labeled data. unlabelThis represents the loss of unlabeled data;

[0022] For labeled data, which is a binary classification problem, the cross-entropy loss function is used, and the calculation formula is as follows:

[0023]

[0024] s represents the number of labeled tasks, and p1 is the predicted probability of death at that point in time;

[0025] For unlabeled data, the entropy is calculated as follows:

[0026]

[0027] k is the total number of tasks, p0 is the probability of being still alive at that point in time. The higher the probability of belonging to a certain state, the lower the corresponding entropy.

[0028] (2) Ranking loss

[0029] Once the survival status changes to death, i.e., the label is "1", it cannot be changed again. Therefore, the survival probability of the same case at different time points is a non-increasing structure. A ranking loss is added to the objective function to constrain this non-increasing trend. The ranking loss is a pairwise ranking loss used to calculate the ranking error between the surviving instance and the predicted value.

[0030] The ranking error is calculated as follows for the survival probability of case i at any two time points j and h:

[0031] R(y i ,p i )=[(y ij -y ih )-(p ij -p ih )] 2

[0032] y ij y represents the actual label value of case i at time point j. ih p represents the actual label value of case i at time point h. ij p represents the probability that liver cancer case i will die at time point j. ih This represents the probability of liver cancer case i dying at time point h. As the time point moves forward, the patient's probability of death gradually increases.

[0033] The ranking loss is defined as follows:

[0034]

[0035]

[0036] When pij >p ih And y ij <y ih The time-ordering loss adds a penalty term, which ensures that once the survival label becomes "1", it will not become "0" even if the probability of death in the later task is higher than that in the earlier task.

[0037] S4. Obtain the liver cancer case data to be predicted, and input the liver cancer case data to be predicted into the trained liver cancer prognostic analysis model based on multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimate the postoperative survival time based on the predicted survival status at each time point.

[0038] S5. Obtain the liver cancer case data to be predicted, and based on the trained multi-task neural network-based liver cancer prognostic analysis model, perform risk factor screening and risk assessment, specifically including:

[0039] S51. Add random noise to each feature of the liver cancer case data;

[0040] S52. Based on the trained multi-task neural network-based liver cancer prognostic analysis model, calculate the change value of the consistency index of the evaluation index before and after adding noise. Each feature is calculated 50 times, and the average change value is taken as the importance score.

[0041] S53. Sort all features by importance score and screen out high-risk factors; the higher the importance score, the higher the correlation with the disease.

[0042] S54. Using the aforementioned high-risk factors, cases are divided into different risk groups, and the log-rank test is used to measure the differences between the different risk groups.

[0043] Furthermore, after step S1, the data preprocessing of the liver cancer case dataset is also included, specifically: removing abnormal data and filling in missing data, and finally normalizing the data.

[0044] Furthermore, in step S3, the C-index and Brief score from the survival analysis are used as evaluation metrics for the model.

[0045] A liver cancer prognostic analysis system based on multi-task learning, characterized in that the product recommendation system executes the liver cancer prognostic analysis method based on multi-task learning as described in claim 1, including: a data acquisition module, a liver cancer prognostic analysis model construction module, a liver cancer prognostic analysis model training module, a liver cancer prognostic analysis module, and a risk factor screening and risk assessment module;

[0046] The data acquisition module acquires a dataset of liver cancer cases.

[0047] The liver cancer prognostic analysis model construction module constructs a liver cancer prognostic analysis model based on a multi-task neural network.

[0048] The liver cancer prognostic analysis model training module uses the liver cancer case dataset to train the liver cancer prognostic analysis model based on a multi-task neural network, thereby obtaining a trained liver cancer prognostic analysis model based on a multi-task neural network.

[0049] The liver cancer prognosis analysis module acquires liver cancer case data to be predicted and inputs the liver cancer case data into the trained liver cancer prognosis analysis model based on a multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimates the postoperative survival time based on the predicted survival status at each time point.

[0050] The risk factor screening and risk assessment module acquires liver cancer case data to be predicted, and performs risk factor screening and risk assessment based on the trained multi-task neural network-based liver cancer prognostic analysis model.

[0051] The present invention also proposes a computer device, the device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0052] This invention discloses a method, system, and device for liver cancer prognostic analysis based on multi-task learning. This invention, by combining liver cancer clinical data, deep neural network technology, and multi-task learning technology, fundamentally solves the problem of low accuracy in existing liver cancer prognostic analyses.

[0053] Compared with existing technologies, it has the following beneficial effects:

[0054] 1. The liver cancer prognostic analysis method based on multi-task learning of the present invention predicts the postoperative survival time of liver cancer patients by establishing a liver cancer prognostic analysis model based on multi-task neural network and conducts prognostic risk assessment for patients, thereby improving the accuracy of liver cancer prognostic analysis, helping clinicians to assess the prognostic status of patients, and providing clinical decision-making for postoperative health management.

[0055] 2. The liver cancer prognostic analysis method based on multi-task learning of the present invention introduces multi-task learning into the prognostic analysis model of liver cancer, discretizes the survival time of liver cancer cases, transforms the survival time regression problem into a multi-task classification problem, solves the censoring problem in liver cancer prognostic data, and further improves the accuracy of liver cancer prognostic analysis. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of a liver cancer prognostic analysis method based on multi-task learning provided in an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram illustrating the construction and training of a liver cancer prognostic analysis model based on multi-task learning, as provided in an embodiment of the present invention.

[0059] Figure 3 This is a schematic diagram illustrating the transformation of lifespan provided in an embodiment of the present invention.

[0060] Figure 4 This is a schematic diagram showing the ranking results of the importance scores of features provided in an embodiment of the present invention.

[0061] Figure 5 This is a schematic diagram illustrating the results of hazard factor screening and risk assessment provided in an embodiment of the present invention.

[0062] Figure 6 This is a schematic diagram of a liver cancer prognostic analysis system based on multi-task learning provided in an embodiment of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present 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 the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0065] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0066] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0067] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0068] This invention provides a method for prognostic analysis of liver cancer based on multi-task learning. For example... Figure 1 As shown, a prognostic analysis method for liver cancer based on multi-task learning includes the following steps S1 to S5.

[0069] S1. Obtain a liver cancer case dataset; the liver cancer case dataset is represented as {(X i T i δ i )|i=1,2,...,n},where X i Let T be the feature vector of the i-th liver cancer case; i Let δ be the survival time of the i-th liver cancer case; i Let δ represent the censored state of the i-th liver cancer case. i A value of 0 indicates that the case data is censored, meaning the patient remained alive until the end of the follow-up period. i =1 indicates that the case data is complete, meaning the patient died before the end of the follow-up period.

[0070] The characteristics of liver cancer case data can be represented as: X i =[X i1 X i2 , ...X iM ]; where M is the number of features included; the original labels for liver cancer case data are: {(T i δ i )|i=1,2,...,n}.

[0071] The characteristics of the liver cancer cases include: gender, age, etiology and type, alpha-fetoprotein, hemoglobin, platelets, white blood cells, aspartate aminotransferase, lactate dehydrogenase, albumin, total bilirubin, C-reactive protein, hepatic encephalopathy, tumor diameter, lesion location, number of intrahepatic lesions, new lesions, portal vein invasion, portal vein invasion, hepatic vein invasion, invasion of ductus venosus or atrium, ascites, and distant metastasis. The types include: HBV, HCV, and no viral infection.

[0072] In one specific embodiment, the dataset used is derived from a case study of mid-stage liver cancer postoperative follow-up data from a cancer hospital.

[0073] After step S1, the data preprocessing of the liver cancer case dataset is also included, specifically: removing abnormal data and filling in missing data, and finally normalizing the data.

[0074] The process of constructing and training a liver cancer prognostic analysis model based on multi-task learning is as follows: Figure 2 As shown.

[0075] S2. Construct a prognostic analysis model for liver cancer based on a multi-task neural network, including:

[0076] The conversion of survival time, specifically as follows: Figure 3 As shown, the survival time of liver cancer cases is divided into J time points, where J is the maximum survival time of the included cases. The survival time of each case is converted into the survival status at these J time points, represented by "1" for death, "0" for survival, and "?" for unknown. After conversion, the survival time of each case is transformed into J survival statuses, denoted by y. i = [y1, y2, ..., y J The transformed label vector is represented by ]; the feature vector X of liver cancer cases is used to represent the label vector. i and the transformed label vector y i As input to a multi-task neural network.

[0077] Furthermore, in the discretization of survival time, the original label, i.e., the continuous survival time, is transformed into the survival state under a series of discrete time intervals, thus transforming the regression problem into a multi-task classification problem.

[0078] The liver cancer prognostic analysis model based on a multi-task neural network includes a three-layer structure: an input layer, a parameter sharing layer, and a multi-task output layer; the input layer receives the feature vector X of the liver cancer case. i and the transformed label vector y iThe input layer and the parameter sharing layer are fully connected. The parameter sharing layer consists of multiple layers, and the multi-task parameter sharing method uses a hard parameter sharing mechanism, meaning that the layers in the sharing layer are connected through a fully connected mechanism. The output part of the network begins to branch, including J output tasks. Each output task corresponds to two nodes, which correspond to the probability of death or survival, respectively. When the survival probability is greater than the death probability, it means that the case is alive at this time point. Otherwise, the J output tasks correspond to the survival status at J time points. The survival time of the patient is predicted by the survival status at these J time points. In addition, a dropout layer and a softmax layer are added before the output layer. The dropout layer improves the generalization ability of the model, and the softmax layer maps the output to (0,1), making the output result a probability.

[0079] S3. Using the liver cancer case dataset, train the liver cancer prognostic analysis model based on a multi-task neural network to obtain a trained liver cancer prognostic analysis model based on a multi-task neural network.

[0080] The model's objective function consists of two loss functions: survival loss and ranking loss. Considering the non-increasing survival probabilities among unlabeled tasks and among different tasks for the same case due to censoring in the case data, a regularized loss function is introduced for unlabeled tasks. For the survival probabilities among different tasks for the same case, a ranking loss is introduced to constrain this non-increasing structure.

[0081] Specifically, during model training, the objective function of the liver cancer prognostic analysis model based on a multi-task neural network consists of two loss functions: survival loss and ranking loss, as follows:

[0082]

[0083] Where α and β are positive constants used to adjust the weights of survival loss and ranking loss, θ is the network parameter, and λ is the regularization parameter;

[0084] The calculation methods for survival loss and ranking loss are as follows:

[0085] (1) Survival loss

[0086] Uncensored data is labeled with two states: alive and dead. However, due to censoring, the survival status of censored cases is known at some time points, but unknown at others. Therefore, the survival loss consists of two parts, calculated as follows:

[0087] L cen (θ)=L obs (y ij ,p(X i |θ))+λL unlabel(p(X i |θ))

[0088] Among them, y ij p(X) represents the actual label value of case i at time point j. i |θ) represents the label value for case prediction, L obs L represents the loss of labeled data. unlabel This represents the loss of unlabeled data;

[0089] For labeled data, which is a binary classification problem, the cross-entropy loss function is used, and the calculation formula is as follows:

[0090]

[0091] s represents the number of labeled tasks, and p1 is the predicted probability of death at that point in time;

[0092] For unlabeled data, the entropy is calculated as follows:

[0093]

[0094] k is the total number of tasks, and p0 is the probability of being alive at that point in time. The higher the probability of belonging to a certain state, the lower the corresponding entropy.

[0095] (2) Ranking loss

[0096] Once the survival status changes to death, i.e., the label is "1", it cannot be changed again. Therefore, the survival probability of the same case at different time points is a non-increasing structure. A ranking loss is added to the objective function to constrain this non-increasing trend. The ranking loss is a pairwise ranking loss used to calculate the ranking error between the surviving instance and the predicted value.

[0097] The ranking error is calculated as follows for the survival probability of case i at any two time points j and h:

[0098] R(y i ,p i )=[(y ij -y ih )-(p ij -p ih )] 2

[0099] y ij y represents the actual label value of case i at time point j. ih p represents the actual label value of case i at time point h. ij p represents the probability that liver cancer case i will die at time point j. ihThis represents the probability of liver cancer case i dying at time point h. As the time point moves forward, the patient's probability of death gradually increases.

[0100] The ranking loss is defined as follows:

[0101]

[0102]

[0103] When p ij >p ih And y ij <y ih The time-ordering loss adds a penalty term, which ensures that once the survival label becomes "1", it will not become "0" even if the probability of death in the later task is higher than that in the earlier task.

[0104] In step S3, the C-index and Brief score from survival analysis are used as evaluation metrics for the model. The C-index is the consistency index, representing the proportion of case pairs whose predictions are consistent with the actual outcomes; it ranges from 0 to 1, with 1 being the optimal value. The Brief score represents the mean squared distance between the observed survival status and the predicted survival probability; it also ranges from 0 to 1, with 0 being the best possible value.

[0105] In one specific embodiment, to verify the effectiveness of the trained hepatocellular carcinoma prognostic analysis model based on a multi-task neural network, the dataset used was derived from a mid-term postoperative follow-up data instance of hepatocellular carcinoma from a certain oncology hospital. The preprocessed hepatocellular carcinoma case data was divided into a training set and a test set in a 4:1 ratio. The data in the test set was input into the trained hepatocellular carcinoma prognostic analysis model based on a multi-task neural network. The C-index and Brief score in survival analysis were used as evaluation indicators. Compared with traditional statistical methods, the present invention has improved performance.

[0106] S4. Obtain the liver cancer case data to be predicted, and input the liver cancer case data to be predicted into the trained liver cancer prognostic analysis model based on multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimate the postoperative survival time based on the predicted survival status at each time point.

[0107] S5. Obtain the liver cancer case data to be predicted, and based on the trained multi-task neural network-based liver cancer prognostic analysis model, perform risk factor screening and risk assessment, specifically including:

[0108] S51. Add random noise to each feature of the liver cancer case data;

[0109] S52. Based on the trained multi-task neural network-based liver cancer prognostic analysis model, calculate the change value of the consistency index C-index before and after adding noise. Each feature is calculated 50 times, and the average change value is taken as the importance score.

[0110] S53. Rank all features by importance score to identify high-risk factors; a higher importance score indicates a higher correlation with the disease; specific results are as follows: Figure 4 As shown.

[0111] Through steps S51-S53, high-risk factors are screened based on the trained multi-task neural network-based liver cancer prognostic analysis model.

[0112] S54. Using the aforementioned high-risk factors, cases are divided into different risk groups, and the log-rank test is used to measure the differences between the different risk groups.

[0113] In one specific embodiment, the results of hazard factor screening and risk assessment are as follows: Figure 5 As shown, the top four most important features were selected as high-risk factors. The vertical column represents the top four features, and the horizontal column represents the risk assessment on different datasets.

[0114] This invention discloses a method, system, and device for liver cancer prognostic analysis based on multi-task learning. By combining liver cancer clinical data, deep neural network technology, and multi-task learning technology, this invention fundamentally solves the problems of censoring and low accuracy in existing liver cancer prognostic data. This invention establishes a liver cancer prognostic analysis model based on a multi-task neural network to predict the postoperative survival time of liver cancer patients and conduct prognostic risk assessments, improving the accuracy of liver cancer prognostic analysis. This helps clinicians assess patients' prognostic status, allowing them to evaluate the suitability of appropriate treatments based on predicted postoperative survival time and providing clinical decision-making support for postoperative health management of high-risk patients.

[0115] In one specific embodiment, the results of hazard factor screening and risk assessment are as follows: Figure 5 As shown, the top four most important features were selected as high-risk factors. The vertical column represents the top four features, and the horizontal column represents the risk assessment on different datasets.

[0116] This invention discloses a method, system, and device for liver cancer prognostic analysis based on multi-task learning. By combining clinical data on liver cancer, deep neural network technology, and multi-task learning technology, this invention fundamentally solves the problems of censoring and low prediction accuracy in existing liver cancer prognostic data. This invention predicts the postoperative survival time of liver cancer patients by establishing a liver cancer prognostic analysis model based on a multi-task neural network and performs prognostic risk assessment, improving prediction accuracy and helping clinicians assess patients' prognostic status. Clinicians can use the predicted postoperative survival time to evaluate whether appropriate treatment is suitable for liver cancer patients and provide clinical decision-making for postoperative health management of high-risk patients.

[0117] Figure 6 This invention provides a liver cancer prognostic analysis system based on multi-task learning. For example... Figure 6 As shown, the liver cancer prognostic analysis system based on multi-task learning includes: a data acquisition module, a liver cancer prognostic analysis model construction module, a liver cancer prognostic analysis model training module, a liver cancer prognostic analysis module, and a risk factor screening and risk assessment module.

[0118] The data acquisition module acquires a dataset of liver cancer cases.

[0119] The liver cancer prognostic analysis model construction module constructs a liver cancer prognostic analysis model based on a multi-task neural network.

[0120] The liver cancer prognostic analysis model training module uses the liver cancer case dataset to train the liver cancer prognostic analysis model based on a multi-task neural network, thereby obtaining a trained liver cancer prognostic analysis model based on a multi-task neural network.

[0121] The liver cancer prognosis analysis module acquires liver cancer case data to be predicted and inputs the liver cancer case data into the trained liver cancer prognosis analysis model based on a multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimates the postoperative survival time based on the predicted survival status at each time point.

[0122] The risk factor screening and risk assessment module acquires liver cancer case data to be predicted, and performs risk factor screening and risk assessment based on the trained multi-task neural network-based liver cancer prognostic analysis model.

[0123] The aforementioned liver cancer prognostic analysis system based on multi-task learning can be implemented as a computer program that can run on a computer device.

[0124] The computer device can be a server, which can be a standalone server or a server cluster consisting of multiple servers.

[0125] Computer devices include processors, memory, and network interfaces connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0126] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform a liver cancer prognostic analysis method based on multi-task learning.

[0127] This processor provides computing and control capabilities to support the operation of the entire computer device.

[0128] This internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program enables the processor to perform a liver cancer prognostic analysis method based on multi-task learning.

[0129] This network interface is used for network communication with other devices. Those skilled in the art will understand that the above-described computer device structure is only a partial structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements.

[0130] The processor is used to run a computer program stored in the memory, which implements the liver cancer prognostic analysis method based on multi-task learning described in Embodiment 1.

[0131] It should be understood that, in the embodiments of this application, the processor may be a Central Processing Unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0132] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0133] The present invention also provides a storage medium. This storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the multi-task learning-based liver cancer prognostic analysis method described in Embodiment 1.

[0134] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0135] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0136] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0137] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a 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 all or 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, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0139] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for prognostic analysis of liver cancer based on multi-task learning, characterized in that, Including the following steps: S1. Obtain a liver cancer case dataset; the liver cancer case dataset is represented as {(X i T i δ i )|i=1,2,…,n},where X i Let T be the feature vector of the i-th liver cancer case; i Let δ be the survival time of the i-th liver cancer case; i Let δ represent the censored state of the i-th liver cancer case. i A value of 0 indicates that the case data is censored, meaning the patient remained alive until the end of the follow-up period. i =1 indicates that the case data is complete, meaning the patient died before the end of the follow-up period; The characteristics of the liver cancer cases include at least the following: etiology and type, alpha-fetoprotein, hemoglobin, platelets, white blood cells, aspartate aminotransferase, lactate dehydrogenase, albumin, total bilirubin, C-reactive protein, hepatic encephalopathy, tumor diameter, lesion location, number of intrahepatic lesions, new lesions, portal vein invasion, portal vein invasion, hepatic vein invasion, invasion of ductus venosus or atrium, ascites, and distant metastasis. The types include HBV, HCV, and no viral infection. S2. Construct a prognostic analysis model for liver cancer based on a multi-task neural network, including: The survival time conversion divides the survival time of liver cancer cases into J time points, where J is the maximum survival time of the included cases. The survival time of each case is then converted into its survival status at these J time points, represented by "1" for death, "0" for survival, and "?" for unknown. After conversion, the survival time of each case is transformed into J survival statuses, denoted by y. i = [y1, y2, ..., y J The transformed label vector is represented by ]; the feature vector X of liver cancer cases is used to represent the label vector. i and the transformed label vector y i As input to a multi-task neural network; The liver cancer prognostic analysis model based on a multi-task neural network includes a three-layer structure: an input layer, a parameter sharing layer, and a multi-task output layer; the input layer receives the feature vector X of the liver cancer case. i and the transformed label vector y i The input layer and parameter sharing layer are fully connected. The parameter sharing layer consists of multiple layers, and the multi-task parameter sharing method uses a hard parameter sharing mechanism, meaning that the layers in the sharing layer are connected through a fully connected mechanism. The output part of the network begins to branch, including J output tasks. Each output task corresponds to two nodes, which correspond to the probability of death or survival, respectively. When the survival probability is greater than the death probability, it means that the case is alive at this time point; otherwise, the J output tasks correspond to the survival status at J time points. The survival time of the patient is predicted by the survival status at these J time points. In addition, a dropout layer and a softmax layer are added before the output layer. The dropout layer improves the generalization ability of the model, and the softmax layer maps the output to (0,1), making the output result a probability. S3. Using the liver cancer case dataset, train the liver cancer prognostic analysis model based on multi-task neural network to obtain a trained liver cancer prognostic analysis model based on multi-task neural network. Specifically, during model training, the objective function of the liver cancer prognostic analysis model based on a multi-task neural network consists of two loss functions: survival loss and ranking loss, as follows: Where α and β are positive constants used to adjust the weights of survival loss and ranking loss, θ is the network parameter, and λ is the regularization parameter; The calculation methods for survival loss and ranking loss are as follows: (1) Survival loss Uncensored data is labeled with two states: alive and dead. However, due to censoring, the survival status of censored cases is known at some time points, but unknown at others. Therefore, the survival loss consists of two parts, calculated as follows: THE cen (θ)=L obs (y ij ,p(X i |θ))+λL unlabel (p(X i |θ)) Among them, y ij p(X) represents the actual label value of case i at time point j. i |θ) represents the label value for case prediction, L obs L represents the loss of labeled data. unlabel This represents the loss of unlabeled data; For labeled data, which is a binary classification problem, the cross-entropy loss function is used, and the calculation formula is as follows: s represents the number of labeled tasks, and p1 is the predicted probability of death at that point in time; For unlabeled data, the entropy is calculated as follows: k is the total number of tasks, p0 is the probability of being still alive at that point in time. The higher the probability of belonging to a certain state, the lower the corresponding entropy. (2) Ranking loss Once the survival status changes to death, i.e., the label is "1", it cannot be changed again. Therefore, the survival probability of the same case at different time points is a non-increasing structure. A ranking loss is added to the objective function to constrain this non-increasing trend. The ranking loss is a pairwise ranking loss used to calculate the ranking error between the surviving instance and the predicted value. The ranking error is calculated as follows for the survival probability of case i at any two time points j and h: R(y i ,p i )=[(y ij -y ih )-(p ij -p ih )] 2 y ij y represents the actual label value of case i at time point j. ih p represents the actual label value of case i at time point h. ij p represents the probability that liver cancer case i will die at time point j. ih This represents the probability of liver cancer case i dying at time point h. As the time point moves forward, the patient's probability of death gradually increases. The ranking loss is defined as follows: When p ij >p ih And y ij <y ih The time-ordering loss adds a penalty term, which ensures that once the survival label becomes "1", it will not become "0" even if the probability of death in the later task is higher than that in the earlier task. S4. Obtain the liver cancer case data to be predicted, and input the liver cancer case data to be predicted into the trained liver cancer prognostic analysis model based on multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimate the postoperative survival time based on the predicted survival status at each time point. S5. Obtain the liver cancer case data to be predicted, and based on the trained multi-task neural network-based liver cancer prognostic analysis model, perform risk factor screening and risk assessment, specifically including: S51. Add random noise to each feature of the liver cancer case data; S52. Based on the trained multi-task neural network-based liver cancer prognostic analysis model, calculate the change value of the consistency index of the evaluation index before and after adding noise. Each feature is calculated 50 times, and the average change value is taken as the importance score. S53. Sort all features by importance score and screen out high-risk factors; the higher the importance score, the higher the correlation with the disease. S54. Using the aforementioned high-risk factors, cases are divided into different risk groups, and the log-rank test is used to measure the differences between the different risk groups.

2. The method according to claim 1, characterized in that, After step S1, the data preprocessing of the liver cancer case dataset is also included, specifically: removing abnormal data and filling in missing data, and finally normalizing the data.

3. The method according to claim 1, characterized in that, In step S3, the C-index and Brief score from the survival analysis are used as evaluation metrics for the model.

4. A liver cancer prognostic analysis system based on multi-task learning, characterized in that, The liver cancer prognostic analysis system executes the liver cancer prognostic analysis method based on multi-task learning as described in claim 1, including: a data acquisition module, a liver cancer prognostic analysis model construction module, a liver cancer prognostic analysis model training module, a liver cancer prognostic analysis module, and a risk factor screening and risk assessment module. The data acquisition module acquires a dataset of liver cancer cases. The liver cancer prognostic analysis model construction module constructs a liver cancer prognostic analysis model based on a multi-task neural network. The liver cancer prognostic analysis model training module uses the liver cancer case dataset to train the liver cancer prognostic analysis model based on a multi-task neural network, thereby obtaining a trained liver cancer prognostic analysis model based on a multi-task neural network. The liver cancer prognosis analysis module acquires liver cancer case data to be predicted and inputs the liver cancer case data into the trained liver cancer prognosis analysis model based on a multi-task neural network to obtain the predicted survival status of the liver cancer case at each time point, and estimates the postoperative survival time based on the predicted survival status at each time point. The risk factor screening and risk assessment module acquires liver cancer case data to be predicted, and performs risk factor screening and risk assessment based on the trained multi-task neural network-based liver cancer prognostic analysis model.

5. A computer device, characterized in that, The device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 3.

Images