Correlation Between MRI Imaging Features And Texture Analysis And Molecular Typing Of Breast Cancer

Objective: The aim of this study is to determine whether the features of MRI image and texture analysis of breast cancer are associated with its molecular subtypes,and explore these features' feasibility and value in predicting molecular subtypes of breast cancer.Methods: Retrospective analysis was performed on 97 patients with breast cancer who met the inclusion criteria in First Affiliated Hospital of Wannan Medical College from February 2013 to July 2019.According to the immunohistochemical results,breast masses are divided into 4 types,Luminal A,Luminal B,HER-2 overexpression and triple negative(TN).First,the clinicopathological characteristics of four molecular subtypes were analysed.Chi-square and Fisher's exact probability methods were used to analyze the differences of the MRI image features(maximum diameter,shape,edge,boundary lesions,strengthening method and dynamic enhancement curve type)between the four kinds of molecular subtypes and Luminal A versus non-Luminal A,Luminal B versus non-Luminal B,HER-2 expression versus non-HER-2 expression,TN versus non-TN of difference.Then the breast cancer lesions on the fat-pressed T2 WI,DWI and DCE images were segmented manually to extract the texture features.After dimensionality reduction by LASSO algorithm,14 texture parameters were screened out,then different statistical test methods,including Kruskal-wallis test and mann-whitney U test and t test,were used to analyze the differences of texture parameters in four molecular subtypes,Luminal A versus non-luminal A,Luminal B versus non-luminal B,HER-2 overexpression versus non-HER-2 overexpression,TN versus non-TN.Finally,classification models based on support vector machine(SVM)were established based on the image features and texture parameters with statistical significance obtained above.ROC for each model was drawn and the area under ROC curve(AUC)was calculated to obtain the diagnostic performance.Results: Among the patients included in this study,Luminal A was 26,Luminal B was 43,HER-2 overexpression was 14 and TN was also 14.There were no statistically significant differences(P > 0.05)in the clinicopathological characteristics among patients with different molecular subtypes of breast cancer.Between four molecular subtypes,there were statistically significant differences(P < 0.05)in the overall distribution of 3 MRI image features and 9 texture parameters.Comparison between two groups based on different molecular subtypes: Luminal A versus non-luminal A,MRI image features were not statistically significant and six texture parameters were statistically significant.Luminal B versus non-luminal B,one image feature and four texture parameters had statistical significance.HER-2 overexpression versus non-HER-2 overexpression had no image features and only one texture parameter had statistical significance.TN versus non –TN,three image features and six texture parameters had statistical significance.Based on the SVM binary classification model,the optimal AUC of Luminal A versus non-luminal A,Luminal B versus non-luminal B,her-2 versus nonher-2,and t TN versus non-TN were 0.78,0.73,0.68 and 0.91,respectively.Conclusion: 1.There is a certain correlation between MRI image features and molecular typing of breast cancer,especially for TN breast cancer,but there is little correlation between Luminal A type and non-Luminal A type,HER-2 overexpression type and non-HER-2 overexpression type.2.MRI texture analysis of breast cancer is closely related to molecular type,especially Luminal A and TN breast cancer.3.The SVM classification model based on the image feature and texture analysis of MRI has great value in differentiating different molecular classification of breast cancer,which can be applied in clinical non-invasive prediction of breast cancer molecular typing in the future,effectively guiding treatment and prognosis evaluation.