CN-122000016-A - Mesothelioma prediction method based on CGAN-SVDD-DBN-ELM-BP

CN122000016ACN 122000016 ACN122000016 ACN 122000016ACN-122000016-A

Abstract

The invention provides a mesothelioma prediction method based on CGAN-SVDD-DBN-ELM-BP, which comprises the steps of firstly, obtaining detection data of mesothelioma patients and carrying out data preprocessing, wherein the data preprocessing comprises filling missing values, deleting abnormal values and normalizing, secondly, generating more patient samples based on mesothelioma definite patient data and mesothelioma related characteristics by using condition generation countermeasure network generation, solving the imbalance problem among the data, thirdly, carrying out abnormal detection on the mesothelioma data after data enhancement by using a support vector data description method, establishing a minimum hypersphere containing as many training samples as possible, removing abnormal data samples outside the hypersphere, and finally, training a DBN-ELM-BP deep learning algorithm by using the mesothelioma data after removing the abnormal data, wherein the algorithm combines the problems of DBN unsupervised characteristic extraction, the rapid learning speed of ELM and the local minimum value generation, effectively reduces the convergence speed and error caused by parameter random initialization, and improves the prediction performance. And classifying the samples by using the trained classifier and outputting a mesothelioma prediction result.

Inventors

XIE SHUANGBO
ZHANG BIN
LUO ZHE
GU XIAOMENG

Assignees

湖南科技学院

Dates

Publication Date: 20260508
Application Date: 20241108

Claims (6)

1. A method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP, comprising: Obtaining detection data of a mesothelioma patient and carrying out data preprocessing, wherein the data preprocessing comprises filling in missing values, deleting abnormal values and normalizing; generating more patient samples based on mesothelioma-confirmed patient data and mesothelioma-related features using a condition generation countermeasure network, thereby balancing the sample data; and performing anomaly detection on mesothelioma data subjected to data enhancement by a support vector data description method, establishing a minimum hypersphere containing as many training samples as possible, and eliminating abnormal data samples outside the hypersphere. The DBN-ELM-BP deep learning algorithm is based on the combination of DBN unsupervised feature extraction, ELM rapid learning speed and generalization capability, solves the problems of low convergence speed and local minimum sinking caused by parameter random initialization, effectively reduces training errors and generalization errors, and improves prediction performance. And classifying the samples by using the trained classifier and outputting a mesothelioma prediction result.
2. The method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP according to claim 1, wherein the steps of obtaining mesothelioma data and preprocessing comprise data preprocessing by filling in missing values, deleting abnormal values, normalizing and the like.
3. The method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP according to claim 1, wherein the data enhancement using the conditional generation countermeasure network comprises the conditional generation countermeasure network structure including 2 deep neural networks of a generator and a discriminator, and gradually generating a false sample similar to a real sample through a reciprocal countermeasure game between the 2 neural networks. The condition generation countermeasure network combines supervised learning and semi-supervised learning, allows additional information to be used as input of a generator in the process of generating the sample, guides the generation process of the sample, and solves the problem of unbalance between data.
4. The method for mesothelioma prediction based on CGAN-SVDD-DBN-ELM-BP according to claim 1, wherein the using of the detected outlier data and the culling of outlier objects comprises the steps of first performing a calculation in a low-dimensional feature space, then mapping the input samples to a high-dimensional space by a kernel function, and finally constructing a minimum hypersphere containing as many training sample points as possible in the high-dimensional space, wherein the sample points on the hypersphere are support vectors. Sample points within the hypersphere are normal and sample points outside the hypersphere are abnormal data samples.
5. The method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP according to claim 1, wherein the training DBN-ELM-BP deep learning algorithm combines the capabilities of DBN unsupervised feature extraction, ELM rapid learning speed and generalization, solves the problems of slow convergence speed and local minimum sinking caused by parameter random initialization, effectively reduces training errors and generalization errors, and improves the prediction performance. And classifying the samples by using the trained classifier and outputting a mesothelioma prediction result.
6. The method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP according to claim 1, wherein the data preprocessing is characterized in that data enhancement is performed on an countermeasure network by using condition generation, balance among the data is guaranteed, anomaly detection is performed by using support vector data description after the data enhancement, anomaly data are removed, and a DBN-ELM-BP algorithm is trained for mesothelioma prediction after the anomaly data are removed.

Description

Mesothelioma prediction method based on CGAN-SVDD-DBN-ELM-BP Technical Field The invention belongs to the field of intersection of deep learning and medical diagnosis, and particularly relates to a mesothelioma prediction method based on CGAN-SVDD-DBN-ELM-BP. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. Malignant mesothelioma is a rare invasive malignancy that originates from the mesothelial layer of the serosal surface, such as the pleura and peritoneum. Typically, mesothelioma is closely associated with asbestos contact. In the traditional medical diagnostic mode, patient test data is dependent upon manual processing by an experienced physician. However, in order to achieve the purposes of early discovery, early diagnosis, early treatment, improved diagnosis efficiency and reduced treatment cost by systematically analyzing the characteristics and the internal relations of mesothelioma patients, effective diagnosis, prevention and control of mesothelioma diseased people are required, and deep learning provides an effective technical means for achieving the purpose. Currently, the following problems exist for mesothelioma prediction: 1. the identification of abnormal data samples is not accurate enough, so that the accuracy of mesothelioma prediction is not high enough. 2. The data sample size of mesothelioma patient is not large enough, and meanwhile, the malignant mesothelioma patient sample is less than the normal sample, so that the problem of unbalanced data exists, the training of a deep learning model is difficult, and the accuracy of mesothelioma prediction is not high enough. 3. The neural network can usually submit the prediction accuracy based on a large number of training samples, however, the mesothelioma belongs to rare diseases, and the data sample size of the patient is usually smaller, so that the prediction accuracy of the mesothelioma is adversely affected. Disclosure of Invention In order to solve the problems, the invention provides a mesothelioma prediction method based on condition generation, antagonism network-support vector data description-deep confidence network-extreme learning machine-back propagation, which can be based on a data set consisting of relevant characteristics obtained by mesothelioma patient detection, wherein in the first step, data enhancement is carried out on small samples through CGAN, in the second step, abnormal data samples are identified through SVDD, and finally, mesothelioma prediction results are given through training of a DBN-ELM-BP deep learning model. Based on the technology of the invention, mesothelioma patients can perform result evaluation at any time through early blood examination and computer tomography without being limited by sites and time, can also provide auxiliary reference for clinicians, lighten the workload of doctors and improve the mesothelioma identification efficiency. According to some embodiments, the present invention employs the following technical solutions: a method for predicting mesothelioma based on CGAN-SVDD-DBN-ELM-BP, comprising: Obtaining detection data of a mesothelioma patient and carrying out data preprocessing, wherein the data preprocessing comprises filling in missing values, deleting abnormal values and normalizing; Generating more patient samples based on the diagnosed mesothelioma patient data and mesothelioma-related features using CGAN, thereby balancing the mesothelioma sample and normal sample data; performing abnormality detection on mesothelioma patient data enhanced by CGAN data through SVDD, establishing a minimum hypersphere containing as many training samples as possible, and eliminating abnormal data samples outside the hypersphere; Training a DBN-ELM-BP deep learning model by using mesothelioma patient samples, and performing unsupervised pre-training and ELM fine tuning of the DBN, and completing training of the model by combining a back propagation algorithm. And predicting the mesothelioma sample by using the trained DBN-ELM-BP deep learning model and outputting a predicted result. Further, the method further comprises the steps of acquiring data and carrying out data preprocessing, wherein the data preprocessing comprises the steps of filling missing values, namely filling missing parts by using the median or average value of the corresponding missing characteristic attributes, and removing obviously abnormal parts. Further, CGAN generated more mesothelioma patient samples, allowing mesothelioma patients to equilibrate with normal samples as much as possible. Assuming that the input sample of the generator G is noise z, the true sample is x, the data distribution p (x) is obeyed, c represents the input additional sample information, G (x|c) represents the generated sample, the generated sample serves as the input of the discriminator D, and the output of the discriminator D is a