Multi-subtype Classification Of Non-small Cell Lung Cancer Based On Radiomics

Lung cancer is the most common type of cancer and the most lethal of all cancer types,accounting for more than a quarter?26%?.Non-small cell lung cancer?NSCLC?accounts for about 85%-90%of the incidence rate of lung cancer.NSCLC can be divided into four subtypes:squamous cell carcinoma,adenocarcinoma,large cell carcinoma and not otherwise specified.It is of great clinical value to realize the accurate classification of non-small cell lung cancer subtypes by analyzing computed tomography images based on radiomics.However,at present,the existing studies are mainly focused on the classification studies between squamous cell carcinoma and adenocarcinoma,while the multi-subtype classification studies of NSCLC based on radiomics are very few.Therefore,it is of great clinical significance to explore a sensitive and effective feature set for multi-subtype classification of NSCLC and to construct a high-precision multi-subtype classification model that can realize all subtypes of NSCLC.This paper studies to verify the feasibility of radiomics to solve the multi-subtype classification of NSCLC,find the effective radiomics feature group to solve specific problems and construct the multi-subtype classification model based on radiomics.The main contents include the following aspects:?1?An effective radiomics feature group for the classification of multi-subtypes of NSCLC was proposed.On the LUNG1/LUNG3 data set,1409 radiomics features were extracted from CT images and 12 kinds of filtered images,including 14 types of shape features,18 types of intensity features and 75 types of texture features.Combining with the classification model,features were analyzed and screened to form an effective radiomics feature group that finally contains 247 features and is sensitive to the multi-subtype classification of NSCLC,so as to provide a priori reference for feature extraction to solve such problems.?2?To verify the feasibility of the radiomics method in solving the problem of multi-subtype classification of NSCLC.An SLS radiomics model for the classification of multi-subtypes of NSCLC was constructed.In combination with the synthetic minority oversampling technique,L_2,1 normal form and support vector machine algorithm,the four subtypes of NSCLC were successfully differentiated on a small data set and verified by an independent test set,which laid a feasible foundation for subsequent experiments.?3?An ensemble learning model is proposed to optimize the SLS model.By comparing the classification effects of logistic regression,support vector machine,adaptive boosting,k-nearest neighbors,decision tree,gaussian-NB,bagging,gradient boosting and quadratic discriminant analysis,and finally combining the three dominant classifiers of support vector machine,adaptive boosting and quadratic discriminant analysis through soft-voting strategy,an ensemble classification model was constructed based on SLS model,which effectively improved the accuracy of multi-subtype classification of NSCLC.