CN115565669A - Cancer survival analysis method based on GAN and multitask learning - Google Patents

Cancer survival analysis method based on GAN and multitask learning Download PDF

Info

Publication number
CN115565669A
CN115565669A CN202211240631.1A CN202211240631A CN115565669A CN 115565669 A CN115565669 A CN 115565669A CN 202211240631 A CN202211240631 A CN 202211240631A CN 115565669 A CN115565669 A CN 115565669A
Authority
CN
China
Prior art keywords
survival
patient
cancer
training
task
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202211240631.1A
Other languages
Chinese (zh)
Other versions
CN115565669B (en
Inventor
邱航
阳旭菻
杨萍
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202211240631.1A priority Critical patent/CN115565669B/en
Publication of CN115565669A publication Critical patent/CN115565669A/en
Application granted granted Critical
Publication of CN115565669B publication Critical patent/CN115565669B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention belongs to the technical field of medical information, and particularly relates to a cancer survival analysis method based on GAN and multitask learning, wherein a GAN network is used for data enhancement, and the characteristics, survival time and outcome type of a cancer patient with outcome are input into the GAN network for training to generate a large amount of non-deletion survival data; constructing a multi-task learning cancer survival analysis model based on soft parameter sharing, wherein a plurality of different tasks respectively predict the probability of different outcomes of a patient at each moment in a future period of time; inputting the characteristics of the cancer patient to be analyzed into the constructed survival analysis model, and outputting the probability of different outcomes in the future.

Description

Cancer survival analysis method based on GAN and multitask learning
Technical Field
The invention belongs to the technical field of medical information, and particularly relates to a cancer survival analysis method based on GAN and multitask learning.
Background
Accurate prognosis prediction for cancer patients facilitates physicians to optimize treatment measures, improve patient prognosis, and reduce patient disease burden. Medically, prognosis generally refers to the use of a patient's characteristics to predict the probability of outcome over a period of time. The fatality is usually death, recurrence or aggravation. Survival analysis is an analytical method often used in the prognosis of cancer. One key to survival analysis is the presence of censoring data, which indicates that the patient has not had an outcome event during the study. The survival analysis model does not directly predict the survival time of a patient, but rather predicts the probability distribution of the survival time of the patient.
Traditionally, cox Proportional Hazards (CPH) are often used for cancer survival analysis studies. There are two assumptions for CPH: 1) Proportional risk assumption: the risk ratio between different patients is a fixed value and does not change with time. 2) The log linear assumption is: the characteristics of the patient are linearly related to the logarithm of the patient risk. However, real survival data hardly satisfies the linear proportional risk condition. With the continuous development of deep learning in recent years, more and more learners apply the structures such as the fully-connected neural network, the convolutional neural network, the cyclic neural network, the graph neural network and the like to the research of cancer survival analysis. In addition, some scholars apply semi-supervised, self-supervised, active learning, and multitask learning methods to the field of cancer survival analysis.
Currently, the existing cancer survival analysis methods have the following disadvantages. Firstly, the method comprises the following steps: in the cancer survival analysis research, patients are frequently deleted, but the existing survival analysis method cannot handle the condition of high deletion. Secondly, the method comprises the following steps: cancer survival analysis methods using multi-task learning are all based on hard parameter sharing, which is mainly suitable for scenes with close connection between tasks. However, in cancer survival analysis, the variability between different tasks is large and task-to-task conflicts are even possible. Thirdly, the method comprises the following steps: the existing survival analysis model can predict the short-term outcome incidence of cancer patients more accurately, but the prediction capability of the existing survival analysis model on the long-term outcome incidence of cancer patients still needs to be improved.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention provides a cancer survival analysis method based on GAN and multitask learning, and aims to solve the problem that the existing survival analysis method cannot handle high deletion rate.
The technical scheme adopted by the invention is as follows:
a GAN and multitask learning based cancer survival analysis method comprising the steps of:
step 1: acquiring survival data of a cancer patient, forming a survival data set of the cancer patient, and taking part of the survival data in the survival data set as a training set;
step 2: performing data enhancement on Survival data in a training set based on a trained survivval-GAN model;
and step 3: building a cancer survival analysis model based on multi-task learning, and training the cancer survival analysis model based on the enhanced training set data; searching out the optimal hyper-parameter of the cancer survival analysis model by using a grid search method and matching with five-fold cross validation, and retraining the cancer analysis model by using the optimal hyper-parameter;
and 4, step 4: inputting the characteristics of the cancer patient to be analyzed into the constructed cancer survival analysis model, and obtaining the probability of different outcomes of the cancer patient at each moment in a future period of time.
According to the invention, data is enhanced based on the survivval-GAN model, so that a large amount of non-deleted Survival data can be generated, the sample size is expanded, and the accuracy and robustness of model prediction are enhanced.
Preferably, the survival data of the cancer patient comprises patient characteristics, observation time and outcome type of last follow-up time; if the last follow-up visit time of the patient does not have any outcome, the observation time is the deletion time of the patient; and if the last follow-up time of the patient is ended, the observation time is the survival time of the patient.
Preferably, the step 2 comprises the following steps:
step 2.1: dividing the survival data of the cancer patients in the training set into two groups with deletion and fate according to whether the obtained survival data has fate, and respectively recording the number of the two groups;
step 2.2: training a survivval-GAN model based on Survival data of occurrence outcome;
step 2.3: searching out the optimal hyper-parameter of the Survival-GAN model by using a grid search method and matching with five-fold cross validation, and retraining the Survival-GAN model by using the optimal hyper-parameter;
step 2.4: randomly selecting K real survival times from a training set sample and respectively pairing the K real survival times with K different outcomes; sequentially inputting the K pairing results into a survivval-GAN model to generate K Survival data with occurrence outcome;
step 2.5: n2 self-increment K, i.e. N2= N2+ K, N2 representing the number of survival data with outcome;
since K Survival data can be generated after each round of survivval-GAN model training, the Survival data after one round of training is equal to K plus the Survival data in input; i.e. N2= N2+ K.
Step 2.6: judging whether N2 is smaller than N1, if not, directly ending; if yes, returning to the step 2.4 to continue execution until N2 is larger than N1; where N1 represents the number of the deleted data.
Preferably, the Survival-GAN model comprises a generator and a discriminator;
the generator comprises a full-connection network, and the number of the full-connection layers and the number of neurons in each layer of the full-connection network are hyper-parameters;
the arbiter is a multitask full-connection network, and the number of the full-connection layers and the number of neurons in each layer of the arbiter are hyper-parameters;
the discriminator comprises three tasks, the first task being for determining whether the input patient characteristic is true or generated by the discriminator; the second task predicts an outcome type based on the survival data; the third task predicts a time-to-live based on the survival data.
Preferably, the training step of the survivval-GAN model is as follows:
setting hyper-parameters of the generator: the dimensionality output by Embedding, the dimensionality of random noise, the number of layers of all-connected layers, the number of neurons of each layer, the learning rate and an optimizer;
setting the hyper-parameters of the discriminator: the number of layers of the full connection layer, the number of neurons of each layer, the learning rate and the optimizer;
setting other hyper-parameters: the number of training rounds and the batch _ size, wherein the batch _ size is the number of training samples grabbed by one training;
data splicing: randomly acquiring m noise data from standard normal distribution, splicing the labels of the input m real survival data with the noise data after the labels are coded by an Embedding layer to obtain data C i ；
Calculate total loss of generator:
L G ＝L G1 +L G2 +L G3 ；
in the formula: l is G Represents the total loss of the generator, L G1 、L G2 And L G3 All represent a loss function;
updating training parameters of a generator: updating training parameters of the generator based on the total loss function of the generator and a preset learning rate;
calculate total loss of arbiter:
L D ＝L D1 +L D2 +L D3 ；
in the formula: l is D Represents the total loss of the discriminator, L D1 、L D2 And L D3 All represent a loss function;
updating training parameters of the discriminator: updating the training parameters of the discriminator based on the total loss function of the discriminator and a preset learning rate;
end of generator and arbiter training: and judging whether the number of training rounds reaches the specified number of times, if so, finishing the training of the generator and the discriminator, and if not, continuing to execute the training of the discriminator and the generator until the number of training rounds meets the specified number of times.
Preferably, the loss function L G1 For bringing the generator-generated features and the true features closer together, we mean:
Figure BDA0003884113070000031
in the formula: MES is the mean square loss function expressed as
Figure BDA0003884113070000032
Representing the mean square loss of q and p of the input; MSE (G (C) i ),x i ) To generate a patient characteristic G (C) i ) And true patient characteristics x i The mean square error of (d); MSE (D (G (C)) i ))[1]And 1) output D (G (C) of the first task of the discriminator i ))[1]Mean square error with 1;
the loss function L G2 For reconciling the outcome of the patient characteristic prediction generated by the generator with the outcome of the input, expressed as:
Figure BDA0003884113070000033
in the formula: the expression of the cross entropy loss function of crossEncopy is:
Figure BDA0003884113070000034
Figure BDA0003884113070000035
where h is the predicted probability of K outcomes, class is the true outcome; cross Entrophy (D (G (C) i ))[2]，e i ) Is the output D (G (C)) of the second task of the arbiter i ))[2]And true outcome e i Cross entropy of (d);
the loss function L G3 The input lifetime and the predicted lifetime for the patient feature generated by the generator are consistently expressed as:
Figure BDA0003884113070000041
in the formula: MSE (D (G (C)) i ))[2]，s i ) Is the output D (G (C)) of the third task of the arbiter i ))[3]With real lifeTime of storage s i The mean square error of (c).
Preferably, the loss function L D1 For enabling the discriminator to identify whether the input patient characteristic is true or false, is expressed as:
Figure BDA0003884113070000042
wherein MSE (D (x) i )[1]1) inputting the real patient characteristics x i The mean square error of the output of the first task of the discriminator and 1 is calculated; MSE (D (G (C)) i ))[1]0) patient characteristics G (C) generated for the input generator i ) The mean square error of the output of the first task of the discriminator and 0 is calculated;
the loss function L D2 For enabling the arbiter to accurately predict the outcome type of the patient, expressed as:
Figure BDA0003884113070000043
wherein, cross Entrophy (D (x) i )[2]，e i ) For inputting real patient characteristics x i The output of the second task of the arbiter and e i Cross entropy loss of (d); cross Entrophy (D (G (C) i ))[2]，e i ) Patient characteristics G (C) generated for an input generator i ) The output of the second task of the discriminator and e i Cross entropy loss of (d);
the loss function L D3 For enabling the arbiter to accurately predict the patient's time-to-live, is expressed as:
Figure BDA0003884113070000044
where MSE (D (x) i )[3]，s i ) For inputting real patient characteristics x i The output of the third task of the discriminator and s i The mean square error of (d); MSE (D (G (C)) i ))[3]，s i ) Generated for input generatorsPatient characteristic G (C) i ) The output of the third task of the discriminator and s i The mean square error of (d).
Preferably, the cancer survival analysis model comprises an expert network, a task network, an attention network and an auxiliary task network.
Preferably, the training step of the cancer survival analysis model is as follows:
A. setting a hyper-parameter: setting the number of full-connection layers of a task network, an auxiliary task network, an expert network and an attention network, the number of neurons in each layer, the learning rate, an optimizer, the number of training rounds, the batch _ size, the number of prediction moments and the weight of 4 loss functions;
B. presetting the value of batch _ size as m, wherein the outcome types of the patients are K in total; in the training process of each batch, inputting survival data of m patients into a cancer survival analysis model for training;
C. calculating the loss of the cancer survival analysis model:
total loss function L of cancer survival analysis model s Expressed as:
L s ＝λ 1 ·L s1 +λ 2 ·L s2 +λ 3 ·L s3 +λ 4 ·L s4 ；
in the formula: lambda [ alpha ] 1 ，λ 2 ，λ 3 ，λ 4 Weights, which are 4 loss functions respectively, are hyper-parameters; l is s1 、L s2 、L s3 And L s4 All represent a loss function;
D. based on the total loss function L S And updating the parameter theta of the cancer survival analysis model by the preset optimizer Adam and the learning rate gamma S ：
θ s ＝Adam(L s ，θ s ，γ)；
E. And C, judging whether the training turns of the cancer survival analysis model accord with the specified times, if not, returning to the step B, and storing the cancer survival analysis model until the training turns accord with the specified times.
Preferably, the loss function L s1 Expressed as:
Figure BDA0003884113070000051
in the formula:
Figure BDA0003884113070000052
representing the patient as x i Under the condition of (1), at a time s i Occurrence of e i Probability of outcome P(s) i ，e i |x i )；
Figure BDA0003884113070000053
Is an indication function, and the condition is 1 if satisfied, otherwise, the condition is 0; f j (s i |x i ) The expression of (a) is: f j (s i |x i )＝P(s≤s i ，e i ＝j|x＝x i ) Is expressed in the patient characteristic x i Under the condition that the patient outcome is j and occurs at time s i The probability of the previous;
loss function L s2 Expressed as:
Figure BDA0003884113070000054
in the formula: a. The j，i，p The expression of (a) is:
Figure BDA0003884113070000055
the indicator function is to find the pair of patients (i, p) that can be risk compared; the expression of the η function is:
Figure BDA0003884113070000056
loss function L s3 Expressed as:
Figure BDA0003884113070000057
in the formula: the expression of the Sigmoid function is:
Figure BDA0003884113070000058
Figure BDA0003884113070000059
predicting for the model a probability of occurrence of an outcome j for the ith patient at time t;
Figure BDA00038841130700000510
is the probability that the ith patient actually has an outcome of j at time t;
loss function L s4 Expressed as:
Figure BDA00038841130700000511
in the formula:
Figure BDA00038841130700000512
non-missed patient outcome type prediction for assisted task networks
Figure BDA00038841130700000513
And true outcome type e i Cross entropy loss of (2).
Preferably, dividing part of survival data in the survival data set into a test set, and evaluating the performance of the trained cancer survival analysis model by using the test set, wherein the evaluation index is C-index;
the specific calculation steps of the C-index are as follows:
a. pairing all patients pairwise;
b. if the observation time of the patient A is shorter than that of the patient B and the patient A does not have an outcome, the pairing is excluded; if the situation that the outcome does not occur to both patients in the pairing exists, the pairing is excluded; finally obtaining useful pairs;
c. calculating the pairing number of the predicted result and the actual result in the useful pairing pair, wherein the predicted result and the actual result are consistent;
d. calculating a pairing value:
c-index = concordant/useful pair number.
The beneficial effects of the invention include:
1. according to the invention, data is enhanced based on the survivval-GAN model, so that a large amount of non-deleted Survival data can be generated, the sample size is expanded, and the accuracy and robustness of model prediction are enhanced.
2. Compared with the existing multitask cancer survival analysis model based on hard parameter sharing, the multitask cancer survival analysis model based on soft parameter sharing can better handle the situation that the connection among a plurality of tasks of the survival analysis is not tight, so that the prediction precision is higher.
3. An auxiliary task for distinguishing different outcomes is added on the basis of the original task, so that the accuracy of the cancer survival analysis model is improved.
4. The invention designs a loss function for calculating the difference between the predicted ending occurrence probability and the real ending occurrence probability, thereby improving the accuracy of model prediction.
Drawings
FIG. 1 is a schematic overall flow chart of the present invention.
FIG. 2 is a schematic diagram of the network structure of survivval-GAN.
Fig. 3 is a network architecture diagram of a cancer survival analysis model based on multitask learning.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of embodiments of the present application, generally described and illustrated in the figures herein, may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, as presented in the figures, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.
It is first noted that the fates described herein refer to death, disease recurrence, and disease exacerbations; if the above situation does not occur during the follow-up visit, the ending is not considered to occur.
The above definition is not a limitation of the present invention, and the outcome of the present invention may be death, and if the death does not occur during the follow-up, the outcome is considered to be no occurrence; the specific definition of the outcome can be determined according to the actual situation.
The invention is described in further detail below with reference to figures 1 to 3:
a GAN and multitask learning based cancer survival analysis method comprising the steps of:
step 1: acquiring survival data of a cancer patient, forming a survival data set of the cancer patient, and taking part of the survival data in the survival data set as a training set;
survival data for the cancer patient includes patient characteristics, time of observation, and outcome type for last follow-up time; if the last follow-up visit time of the patient does not have any outcome, the observation time is the deletion time of the patient; and if the last follow-up time of the patient is ended, the observation time is the survival time of the patient.
Assuming a total of survival data for N patients is acquired, survival data set D may be represented as:
Figure BDA0003884113070000071
x is a patient characteristic, typically including patient demographic information, tumor pathology, and treatment. s is the observation time, when the last follow-up did not result in any outcome, s is the deletion time of the patient; when the patient had a certain outcome (e.g., death) at the last follow-up, s is the survival time of the patient. e is the outcome type of the patient at the last follow-up visit, and e =0 when no outcome occurred at the last follow-up visit. Assuming a total of K different outcomes, the range of values for e is: {0,1,2, … K }.
Step 2: performing data enhancement on Survival data in a training set based on a trained survivval-GAN model;
the step 2 comprises the following steps:
step 2.1: dividing the survival data of the cancer patients in the training set into two groups with deletion and fate according to whether the survival data has fate, and respectively recording the number of the two groups;
step 2.2: training a survivval-GAN model based on Survival data of the occurrence outcome;
step 2.3: searching out the optimal hyper-parameter of the Survival-GAN model by using a grid search method and matching with five-fold cross validation, and retraining the Survival-GAN model by using the optimal hyper-parameter;
step 2.4: randomly selecting K real survival times from a training set sample and respectively pairing the K real survival times with K different outcomes; inputting the K pairing results into a survivval-GAN model in sequence to generate K Survival data;
step 2.5: n2 self-increment K, i.e., N2= N2+ K; n2 represents the number of survival data with the outcome;
step 2.6: judging whether N2 is smaller than N1, if not, directly ending; if yes, returning to the step 2.4 to continue execution until N2 is larger than N1; where N1 represents the number of the deleted data.
Referring to fig. 2, the survivval-GAN model includes a generator and an arbiter;
the generator is composed of a full-connection network, and the number of layers of full-connection layers and the number of neurons in each layer are both hyper-parameters; the tag of the survival data for the occurrence of the outcome consists of two parts: outcome type (e) and time to live(s); the ending type and the survival time are respectively subjected to Embedding, and then are spliced with random noise (Z) and then are input into a generator; the input of the discriminator is the false patient characteristic and the real patient characteristic output by the generator;
the discriminator is a multi-task full-connection network, and the number of layers of full-connection layers and the number of neurons in each layer are both hyper-parameters; the discriminator has three tasks: the first task is to determine whether the input patient characteristics are true or generated by the generator; the second task is to predict the outcome type using survival data; the third task is to use the survival data to predict survival time. The outputs of these three tasks are represented by G1, G2 and G3, respectively.
The training steps of the survivval-GAN model are as follows:
setting hyper-parameters of the generator: the dimensionality of the output of Embedding, the dimensionality of random noise, the number of layers of a full-connection layer, the number of neurons of each layer, a learning rate and an optimizer (Adam, SGD and the like);
setting the hyper-parameters of the discriminator: the number of layers of the full connection layer, the number of neurons of each layer, learning rate and an optimizer (Adam, SGD and the like);
setting other hyper-parameters: the number of training rounds (epoch) and the batch _ size, which is the number of training samples captured in one training;
data splicing: let the value of batch _ size be m; randomly acquiring m noise data Z from standard normal distribution 1 ，Z 2 ，...Z m And m real survival data: (x) 1 ，s 1 ，e 1 )，(x 2 ，s 2 ，e 2 )，...，(x m ，s m ，e m ). C is used for data after labels of real data are subjected to Embedding and are spliced with noise data i Represents;
calculate total loss of generator:
L G ＝L G1 +L G2 +L G3 ；
in the formula: l is G Represents the total loss of the generator, L G1 、L G2 And L G3 All represent a loss function;
loss function L G1 For bringing the generator-generated features and the true features closer together, we mean:
Figure BDA0003884113070000081
in the formula: MES is expressed as the mean square loss function
Figure BDA0003884113070000082
Representing the mean square loss of q and p of the input; MSE (G (C) i )，x i ) To generate a patient characteristic G (C) i ) And true patient characteristics x i The mean square error of (d); MSE (D (G (C)) i ))[1]And 1) output D (G (C) of the first task of the discriminator i ))[1]Mean square error with 1;
the loss function L G2 For reconciling the outcome of the patient characteristic prediction generated by the generator with the outcome of the input, expressed as:
Figure BDA0003884113070000091
in the formula: the expression of the cross entropy loss function of Cross Encopy is:
Figure BDA0003884113070000092
Figure BDA0003884113070000093
where h is the predicted probability of K outcomes, class is the true outcome; cross Entrophy (D (G (C) i ))[2]，e i ) Is the output D (G (C)) of the second task of the arbiter i ))[2]And true outcome e i Cross entropy of (d);
the loss function L G3 The input lifetime and the predicted lifetime for the patient feature generated by the generator are consistently expressed as:
Figure BDA0003884113070000094
in the formula: MSE (D (G (C)) i ))[2]，s i ) Is the output D (G (C)) of the third task of the arbiter i ))[3]And the real life time s i The mean square error of (d).
Updating training parameters of a generator: the SGD used by the optimizer, assuming a learning rate of α. By theta G Representing the parameters of the generator training. Each one of which isBatch θ G The update of (1) is:
θ G ＝SGD(L G ，θ G ，α)；
calculate total loss of arbiter:
L D ＝L D1 +L D2 +L D3 ；
in the formula: l is D Represents the total loss of the discriminator, L D1 、L D2 And L D3 All represent loss functions;
the loss function L D1 For enabling the discriminator to identify whether the input patient characteristic is true or false, is expressed as:
Figure BDA0003884113070000095
wherein MSE (D (x) i )[1]1) inputting the real patient characteristics x i The mean square error of the output of the first task of the discriminator and 1 is calculated; MSE (D (G (C)) i ))[1]0) patient characteristics G (C) generated for the input generator i ) The mean square error of the output of the first task of the discriminator and 0 is calculated;
the loss function L D2 For enabling the arbiter to accurately predict the outcome type of the patient, expressed as:
Figure BDA0003884113070000096
wherein, cross Entrophy (D (x) i )[2]，e i ) For inputting real patient characteristics x i The output of the second task of the arbiter and e i Cross entropy loss of (d); cross Entrophy (D (G (C) i ))[2]，e i ) Patient characteristics G (C) generated for an input generator i ) The output of the second task of the discriminator and e i Cross entropy loss of (d);
the loss function L D3 For enabling the arbiter to accurately predict the patient's time-to-live, is expressed as:
Figure BDA0003884113070000097
where MSE (D (x) i )[3]，s i ) For inputting real patient characteristics x i The output of the third task of the discriminator and s i The mean square error of (d); MSE (D (G (C)) i ))[3]，s i ) Patient characteristics G (C) generated for an input generator i ) The output of the third task of the discriminator and s i The mean square error of (d).
Updating training parameters of the discriminator: let the learning rate be β, adam used by the optimizer. By theta D Parameters representing discriminant training; each batch θ D The update of (1) is:
θ D ＝Adam(L D ，θ D ，β)；
end of generator and arbiter training: and judging whether the number of training rounds reaches the specified number of times, if so, finishing the training of the generator and the discriminator, and if not, continuing to execute the training of the discriminator and the generator until the number of training rounds meets the specified number of times.
And step 3: building a cancer survival analysis model based on multi-task learning, and training the cancer survival analysis model based on the enhanced training set data; searching out the optimal hyper-parameter of the cancer survival analysis model by using a grid search method and matching with five-fold cross validation, and retraining the cancer analysis model by using the optimal hyper-parameter;
the cancer survival analysis model comprises an expert network, a task network, an attention network and an auxiliary task network, wherein the expert network, the task network, the attention network and the auxiliary task network all belong to a fully-connected neural network, and the number of layers of the fully-connected layer and the number of neurons of each layer are hyper-parameters. A total of K outcomes, each outcome corresponding to an independent task network; the output of the K task networks is the probability of a cancer patient to have K different outcomes at a future time, where T max The longest survival time for the patients in the training set. The auxiliary task network is the outcome of predicting the patient characteristics of the inputs, which may help the model to betterThe different outcomes are well distinguished. The output of the attention mechanism network is the weight of the K +2 expert networks. The outputs of the K +2 expert networks are multiplied by the outputs of the attention network respectively and then added, and then the products are input into the task network and the auxiliary task network. The K task networks and the 1 auxiliary task network share K +2 expert networks.
The training steps of the cancer survival analysis model are as follows:
A. setting a hyper-parameter: setting the number of full-connection layers and the number of neurons of each layer of the task network, the auxiliary task network, the expert network and the attention network, learning rate, optimizers (Adam, SGD and the like), the number of training rounds, batch _ size, the number of prediction moments and the weight of 4 loss functions;
B. assuming a value of batch size of m, the outcome type of the patient is a total of K. Each batch of training requires inputting survival data for m patients into a multitask learning based cancer survival analysis model for training.
C. Calculating the loss of the cancer survival analysis model:
total loss function L of cancer survival analysis model s Expressed as:
L s ＝λ 1 ·L s1 +λ 2 ·L s2 +λ 3 ·L s3 +λ 4 ·L s4 ；
in the formula: lambda 1 ，λ 2 ，λ 3 ，λ 4 Weights, which are 4 loss functions respectively, are hyper-parameters; l is s1 、L s2 、L s3 And L s4 All represent loss functions;
loss function L s1 Is operative to cause the model to learn a general representation of the joint distribution of the occurrence times and the outcome events of the outcome, L s1 Expressed as:
Figure BDA0003884113070000111
in the formula:
Figure BDA0003884113070000112
representing the patient as x i Under the condition of (1), at a time s i Occurrence of e i Probability of outcome P(s) i ，e i |x i )；
Figure BDA0003884113070000113
Is an indication function, and the condition is 1 if satisfied, otherwise, the condition is 0; f j (s i |x i ) The expression of (a) is: f j (s i |x i )＝P(s≤s i ，e i ＝j|x＝x i ) Is expressed in the patient characteristic x i Under the condition of (2), the patient outcome is j and occurs at time s i The probability of the previous;
loss function L s2 The effect of (a) is to make the survival time of patients with higher probability of occurrence of outcome predicted by the model smaller than that of patients with lower probability of occurrence of outcome, i.e. to improve the discriminative power of the model, L s2 Expressed as:
Figure BDA0003884113070000114
in the formula: a. The j，i，p The expression of (a) is:
Figure BDA0003884113070000115
the indicator function is to find the pair of patients (i, p) that can be risk compared; the expression of the η function is:
Figure BDA0003884113070000116
loss function L s3 The effect of (1) is to make the predicted end probability of the model closer to the real end probability, namely to improve the calibration capability, L, of the model s3 Expressed as:
Figure BDA0003884113070000117
in the formula: the Sigmoid function is expressed as:
Figure BDA0003884113070000118
Figure BDA0003884113070000119
predicting for the model a probability of occurrence of an outcome j for the ith patient at time t;
Figure BDA00038841130700001110
is the probability that the ith patient actually has an outcome of j at time t;
loss function L s4 Is operative to enable the model to accurately predict the outcome, L, of the patient s4 Expressed as:
Figure BDA00038841130700001111
in the formula:
Figure BDA00038841130700001112
non-missed patient outcome type prediction for assisted task networks
Figure BDA00038841130700001113
And true outcome type e i Cross entropy loss of (c).
D. Parameters of the model are updated. Assuming a learning rate of γ, an optimizer of Adam, and a model parameter of θ S Then theta for each batch S The update of (1) is:
θ s ＝Adam(L s ，θ s ，γ)；
E. and C, judging whether the training turns of the cancer survival analysis model meet the specified times, if not, returning to execute the step B, and storing the cancer survival analysis model until the training turns meet the specified times.
Dividing part of survival data in the survival data set into a test set, and evaluating the performance of the trained cancer survival analysis model by using the test set, wherein the evaluation index is C-index; the survival data set may be partitioned into a training set and a test set in a 4: 1 ratio.
The specific calculation steps of the C-index are as follows:
a. pairing all patients pairwise;
b. if the observation time of the patient A is shorter than that of the patient B and the patient A does not have an outcome, the pairing is excluded; if the condition that the outcome does not occur to both patients in the pairing exists, the pairing is excluded; finally obtaining useful pairs;
c. calculating the pairing number of the predicted result and the actual result in the useful pairing pair, wherein the predicted result and the actual result are consistent;
d. calculating a pairing value:
c-index = concordant/useful pair number.
And 4, step 4: inputting the characteristics of the cancer patient to be analyzed into the constructed cancer survival analysis model, and obtaining the probability of different outcomes of the cancer patient at each moment in a future period of time.
According to the invention, data is enhanced based on the survivval-GAN model, so that a large amount of non-deleted Survival data can be generated, the sample size is expanded, and the accuracy and robustness of model prediction are enhanced.
The invention uses the GAN network to enhance data, and inputs the characteristics, survival time and outcome type of the cancer patients with outcome into the GAN network for training, thereby generating a large amount of non-deleted survival data; further, a multi-task learning cancer survival analysis model based on soft parameter sharing is established, a plurality of different tasks respectively predict the probability of different outcomes of the patient at each moment in a future period of time, and an auxiliary task for distinguishing the different outcomes is added on the basis of the original task; then, adding a loss function, wherein the loss function is the product of the mean square error of the predicted ending probability and the real ending probability at each moment and Sigmoid (current moment); finally, inputting the survival data after data enhancement into a multi-task learning cancer survival analysis model based on soft parameter sharing for training, wherein the output of the model is the probability of different outcomes of the patient in a future period of time; if the maximum observation time of the patient in the training sample is s max Then the cancer survival method based on multitask learning can predict the future s of the patient max The probability of different outcomes occurring within the range.
The above-mentioned embodiments only express the specific embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for those skilled in the art, without departing from the technical idea of the present application, several changes and modifications can be made, which are all within the protection scope of the present application.

Claims (10)

1. A cancer survival analysis method based on GAN and multitask learning is characterized by comprising the following steps:
step 1: acquiring survival data of the cancer patient, forming a survival data set of the cancer patient, and taking part of the survival data in the survival data set as a training set;
step 2: performing data enhancement on Survival data in a training set based on a trained survivval-GAN model;
and step 3: constructing a cancer survival analysis model based on multi-task learning, and training the cancer survival analysis model based on the enhanced training set data; searching out the optimal hyper-parameter of the cancer survival analysis model by using a grid search method and matching with five-fold cross validation, and retraining the cancer analysis model by using the optimal hyper-parameter;
and 4, step 4: inputting the characteristics of the cancer patient to be analyzed into the constructed cancer survival analysis model, and obtaining the probability of different outcomes of the cancer patient at each moment in a future period of time.
2. The method of claim 1, wherein the survival data of the cancer patient includes patient characteristics, observation time, and outcome type of last follow-up time; if the last follow-up time of the patient does not have any outcome, the observation time is the deletion time of the patient; and if the last follow-up time of the patient is ended, the observation time is the survival time of the patient.
3. The method of claim 1, wherein the step 2 comprises the steps of:
step 2.1: dividing the survival data of the cancer patients in the training set into two groups with deletion and fate according to whether the obtained survival data has fate, and respectively recording the number of the two groups;
step 2.2: training a survivval-GAN model based on Survival data of the occurrence outcome;
step 2.3: searching out the optimal hyper-parameter of the Survival-GAN model by using a grid search method and matching with five-fold cross validation, and retraining the Survival-GAN model by using the optimal hyper-parameter;
step 2.4: randomly selecting K real survival times from a training set sample and respectively pairing the K real survival times with K different outcomes; inputting the K pairing results into a survivval-GAN model in sequence to generate K Survival data with occurrence outcomes;
step 2.5: n2 self-increment K, i.e. N2= N2+ K, N2 representing the number of survival data with outcome;
step 2.6: judging whether N2 is smaller than N1, if not, directly ending; if yes, returning to the step 2.4 to continue execution until N2 is larger than N1; where N1 represents the number of the deleted data.
4. The GAN and multitask learning based cancer Survival analysis method according to claim 1, wherein the survivval-GAN model comprises a generator and a discriminator;
the generator comprises a full-connection network, and the number of the full-connection layers and the number of neurons in each layer of the full-connection network are hyper-parameters;
the arbiter is a multitask full-connection network, and the number of the full-connection layers and the number of neurons in each layer of the arbiter are hyper-parameters;
the discriminator comprises three tasks, the first task being for determining whether the input patient characteristic is true or generated by the generator; the second task predicts an outcome type based on the survival data; the third task predicts a time-to-live based on the survival data.
5. The method for GAN and multitask learning based cancer Survival analysis according to claim 1, wherein the survivval-GAN model is trained by the following steps:
setting hyper-parameters of the generator: the dimensionality output by Embedding, the dimensionality of random noise, the number of layers of all-connected layers, the number of neurons of each layer, the learning rate and an optimizer;
setting the hyper-parameters of the discriminator: the number of layers of the full connection layer, the number of neurons of each layer, the learning rate and the optimizer;
setting other hyper-parameters: the number of training rounds and the batch _ size, wherein the batch _ size is the number of training samples grabbed by one training;
data splicing: randomly acquiring m noise data from standard normal distribution, splicing the labels of the input m real survival data with the noise data after the labels are coded by an Embedding layer to obtain data C i ；
Calculate total loss of generator:
L G ＝L G1 +L G2 +L G3 ；
in the formula: l is G Represents the total loss of the generator, L G1 、L G2 And L G3 All represent a loss function;
updating training parameters of a generator: updating training parameters of the generator based on the total loss function of the generator and a preset learning rate;
calculate total loss of arbiter:
L D ＝L D1 +L D2 +L D3 ；
in the formula: l is D Represents the total loss of the discriminator, L D1 、L D2 And L D3 All represent a loss function;
updating training parameters of the discriminator: updating the training parameters of the discriminator based on the total loss function of the discriminator and a preset learning rate;
end of generator and arbiter training: and judging whether the number of training rounds reaches the specified number of times, if so, finishing the training of the generator and the discriminator, and if not, continuing to execute the training of the discriminator and the generator until the number of training rounds meets the specified number of times.
6. The method of claim 1, wherein the cancer survival analysis model comprises an expert network, a task network, an attention network and an auxiliary task network.
7. The method of claim 1, wherein the training of the cancer survival analysis model comprises:
A. setting a hyper-parameter: setting the number of full-connection layers of a task network, an auxiliary task network, an expert network and an attention network, the number of neurons in each layer, the learning rate, an optimizer, the number of training rounds, the batch _ size, the number of prediction moments and the weight of 4 loss functions;
B. presetting the value of batch _ size as m, wherein the outcome types of the patients are K in total; in the training process of each batch, inputting survival data of m patients into a cancer survival analysis model for training;
C. calculating the loss of the cancer survival analysis model:
total loss function L of cancer survival analysis model s Expressed as:
L s ＝λ 1 ·L s1 +λ 2 ·L s2 +λ 3 ·L s3 +λ 4 ·L s4 ；
in the formula: lambda [ alpha ] 1 ，λ 2 ，λ 3 ，λ 4 Weights, which are 4 loss functions respectively, are hyper-parameters; l is s1 、L s2 、L s3 And L s4 All represent a loss function;
D. based on the total loss function L S And a preset optimizer Adam and learning rate gamma updateParameter θ of cancer survival analysis model S ：
θ s ＝Adam(L s ，θ s ，γ)；
E. And C, judging whether the training turns of the cancer survival analysis model meet the specified times, if not, returning to execute the step B, and storing the cancer survival analysis model until the training turns meet the specified times.
8. The method of claim 7, wherein the loss function L is a function of the survival of cancer s1 Expressed as:
Figure FDA0003884113060000031
in the formula:
Figure FDA0003884113060000032
representing the patient as x i Under the condition of (1), at a time s i Occurrence of e i Probability of outcome P(s) i ，e i |x i )；
Figure FDA00038841130600000310
Is an indication function, and the condition is 1 if satisfied, otherwise, the condition is 0; f j (s i |x i ) The expression of (a) is: f j (s i |x i )＝P(s≤s i ，e i ＝j|x＝x i ) Is expressed in the patient characteristic x i Under the condition that the patient outcome is j and occurs at time s i The probability of the previous;
loss function L s2 Expressed as:
Figure FDA0003884113060000033
in the formula: a. The j，i，p The expression of (a) is:
Figure FDA00038841130600000311
the indicator function is to find the pair of patients (i, p) that can be risk compared; the expression of the η function is:
Figure FDA0003884113060000034
loss function L s3 Expressed as:
Figure FDA0003884113060000035
in the formula: the Sigmoid function is expressed as:
Figure FDA0003884113060000036
Figure FDA0003884113060000037
predicting for the model a probability of occurrence of an outcome j for the ith patient at time t;
Figure FDA0003884113060000038
is the probability that the ith patient actually has an outcome of j at time t;
loss function L s4 Expressed as:
Figure FDA0003884113060000039
in the formula:
Figure FDA0003884113060000041
non-abrogated patient outcome type predicted for assisted task networks
Figure FDA00038841130600000410
And true outcome type e i Cross entropy loss of (2).
9. The method of claim 5, wherein the loss function L is a function of the survival of cancer G1 Expressed as:
Figure FDA0003884113060000042
in the formula: MES is the mean square loss function expressed as
Figure FDA0003884113060000043
Representing the mean square loss of q and p of the input; MSE (G (C) i )，x i ) To generate a patient characteristic G (C) i ) And true patient characteristics x i The mean square error of (d); MSE (D (G (C)) i ))[1]And 1) output D (G (C) of the first task of the discriminator i ))[1]Mean square error with 1;
the loss function L G2 Expressed as:
Figure FDA0003884113060000044
in the formula: the expression of the cross entropy loss function of crossEncopy is:
Figure FDA0003884113060000045
Figure FDA0003884113060000046
where h is the predicted probability of K outcomes, class is the true outcome; cross Entrophy (D (G (C) i ))[2]，e i ) Is the output D (G (C)) of the second task of the arbiter i ))[2]And true outcome e i Cross entropy of (d);
the loss function L G3 Expressed as:
Figure FDA0003884113060000047
in the formula: MSE (D (G (C)) i ))[2]，s i ) Is the output D (G (C)) of the third task of the arbiter i ))[3]And the real life time s i The mean square error of (d).
10. The method of claim 5, wherein the loss function L is a function of the survival of cancer D1 Expressed as:
Figure FDA0003884113060000048
wherein MSE (D (x) i )[1]1) inputting the real patient characteristics x i The mean square error of the output of the first task of the discriminator and 1 is calculated; MSE (D (G (C)) i ))[1]0) patient characteristics G (C) generated for the input generator i ) The mean square error of the output of the first task of the discriminator and 0 is calculated;
the loss function L D2 Expressed as:
Figure FDA0003884113060000049
wherein, cross Entrophy (D (x) i )[2]，e i ) For inputting real patient characteristics x i The output of the second task of the arbiter and e i Cross entropy loss of (d); cross Entrophy (D (G (C) i ))[2]，e i ) Patient characteristics G (C) generated for an input generator i ) The output of the second task of the discriminator and e i Cross entropy loss of (d);
the loss function L D3 Expressed as:
Figure FDA0003884113060000051
where MSE (D (x) i )[3]，s i ) Is composed ofInputting real patient characteristics x i The output of the third task of the discriminator and s i The mean square error of (d); MSE (D (G (C)) i ))[3]，s i ) Patient characteristics G (C) generated for an input generator i ) The output of the third task of the discriminator and s i The mean square error of (c).
CN202211240631.1A 2022-10-11 2022-10-11 Cancer survival analysis method based on GAN and multitask learning Active CN115565669B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211240631.1A CN115565669B (en) 2022-10-11 2022-10-11 Cancer survival analysis method based on GAN and multitask learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211240631.1A CN115565669B (en) 2022-10-11 2022-10-11 Cancer survival analysis method based on GAN and multitask learning

Publications (2)

Publication Number Publication Date
CN115565669A true CN115565669A (en) 2023-01-03
CN115565669B CN115565669B (en) 2023-05-16

Family

ID=84744408

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211240631.1A Active CN115565669B (en) 2022-10-11 2022-10-11 Cancer survival analysis method based on GAN and multitask learning

Country Status (1)

Country Link
CN (1) CN115565669B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117409968A (en) * 2023-10-27 2024-01-16 电子科技大学 Hierarchical attention-based cancer dynamic survival analysis method and system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050197982A1 (en) * 2004-02-27 2005-09-08 Olivier Saidi Methods and systems for predicting occurrence of an event
CN110660478A (en) * 2019-09-18 2020-01-07 西安交通大学 Cancer image prediction and discrimination method and system based on transfer learning
CN111640510A (en) * 2020-04-09 2020-09-08 之江实验室 Disease prognosis prediction system based on deep semi-supervised multitask learning survival analysis
CN112687327A (en) * 2020-12-28 2021-04-20 中山依数科技有限公司 Cancer survival analysis system based on multitask and multi-mode

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050197982A1 (en) * 2004-02-27 2005-09-08 Olivier Saidi Methods and systems for predicting occurrence of an event
CN110660478A (en) * 2019-09-18 2020-01-07 西安交通大学 Cancer image prediction and discrimination method and system based on transfer learning
CN111640510A (en) * 2020-04-09 2020-09-08 之江实验室 Disease prognosis prediction system based on deep semi-supervised multitask learning survival analysis
CN112687327A (en) * 2020-12-28 2021-04-20 中山依数科技有限公司 Cancer survival analysis system based on multitask and multi-mode

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117409968A (en) * 2023-10-27 2024-01-16 电子科技大学 Hierarchical attention-based cancer dynamic survival analysis method and system
CN117409968B (en) * 2023-10-27 2024-05-03 电子科技大学 Hierarchical attention-based cancer dynamic survival analysis method and system

Also Published As

Publication number Publication date
CN115565669B (en) 2023-05-16

Similar Documents

Publication Publication Date Title
WO2021203796A1 (en) Disease prognosis prediction system based on deep semi-supervised multi-task learning survival analysis
Widiputra et al. Multivariate CNN‐LSTM Model for Multiple Parallel Financial Time‐Series Prediction
CN109036553A (en) A kind of disease forecasting method based on automatic extraction Medical Technologist's knowledge
CN111243736B (en) Survival risk assessment method and system
CN112766496B (en) Deep learning model safety guarantee compression method and device based on reinforcement learning
CN114547974A (en) Dynamic soft measurement modeling method based on input variable selection and LSTM neural network
Wang et al. Diabetes Risk Analysis Based on Machine Learning LASSO Regression Model
CN110838364A (en) Crohn disease prediction method and device based on deep learning hybrid model
WO2021237917A1 (en) Self-adaptive cognitive activity recognition method and apparatus, and storage medium
CN115565669A (en) Cancer survival analysis method based on GAN and multitask learning
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN113674862A (en) Acute renal function injury onset prediction method based on machine learning
Kuo et al. Using stacked denoising autoencoder for the student dropout prediction
Tiruneh et al. Feature selection for construction organizational competencies impacting performance
CN112149355A (en) Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model
CN115171896A (en) System and method for predicting long-term death risk of critically ill patient
CN114444654A (en) NAS-oriented training-free neural network performance evaluation method, device and equipment
Christodoulou et al. Improving the performance of classification models with fuzzy cognitive maps
CN114048837A (en) Deep neural network model reinforcement method based on distributed brain-like map
CN112232557A (en) Switch machine health degree short-term prediction method based on long-term and short-term memory network
Zhang et al. Assessing the memory ability of recurrent neural networks
CN113035363B (en) Probability density weighted genetic metabolic disease screening data mixed sampling method
CN115829036B (en) Sample selection method and device for text knowledge reasoning model continuous learning
CN111539306B (en) Remote sensing image building identification method based on activation expression replaceability
Abdel-Kader Prediction of Lung Cancer Using Supervised Machine Learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant