Integration Of DCE-MRI And T2-weighted Imaging Radiomic Features To Predict Histologic Grade In Breast Cancer

Breast cancer is one of the most common types of cancer,as well as the leading cause of death among female cancer patients.At present,it has become a consensus that the treatment of breast cancer needs to follow the principle of individual treatment based on comprehensive treatment.The histological grade is one of the major prognostic factors used in breast cancer treatment.Investigating the molecular biology of breast cancer through non-invasive and repeatable medical imaging has become a research focus.Breast T2-weighted image is one of the standard MRI scan routines.Since the signal strength is directly related to potential lesion morphology,it is usually used to eliminate the breast cyst,lymph nodes and other benign breast lesions and improve the specificity.DCE-MRI is the most sensitive imaging diagnostic technology for breast cancer detection.It not only clearly reflects the form of the abnormal strengthening signals in the mammary gland,but also reflects the changes of the internal structure of the diseased tissue.In clinical practice,diagnosis using DCE-MRI images with assistant of T2WI has been widely applied.However,few academic studies have been done to combine these two.The purpose of this study is to predict the histological grade of breast cancer based on T2-weighted images and dynamic contrast-enhance magnetic resonance images(MRI).A dataset of 167 invasive breast cancer cases,which has underwent preoperative breast MRI with a 3.0T scanner,was collected.Among these cases,95 ones were diagnosed as high-grade malignant(Grade 3)invasive breast cancer,while 72 were low-grade malignant(Grade 1 and 2).Semi-automatic lesion segmentation was performed on each T2WI and DCE-MRI,in which,textural,morphological and statistical features were extracted.In order to fully explore the significance of the joint research of two kinds of parameters,comparison trials were designed in terms of single-parameter and multi-parameter of MRI image.First is the single-parameter-based breast cancer histological grading prediction study,which analyses the histological grading ability of breast cancer by T2WI and DCE-MRI images respectively.1)For T2WI,the correlation between T2WI image features and breast cancer histological grade was explored by statistical methods.Then,the imaging features were used to predict the histological grade.Pearson Correlation Coefficient was used to screen out the highly correlated features.The Lasso Method was used to select features.The support vector machine model was designed to predict the high and low grade of breast cancer histology.With the leave-one-out cross-validation method,the receiver operating characteristic curve(ROC)was drawn and the area under the corresponding curve(AUC)was calculated.Meanwhile,the sensitivity,specificity,FI-Measure and other indicators were calculated for each prediction model.2)For DCE-MRI,the prediction abilities of different sequences were compared,while the study method of single sequence was basically the same as that of T2WI.In addition,this paper only focuses on the pre-contrast,the third and fifth post-contrast sequences.Second is the study based on multi-parameter image features.Single and multiple sequences of DCE-MRI were combined with T2WI sequences respectively.In order to fully investigate the joint research method of multi-parameter image features,the predictive models were designed in three aspects:1)Feature fusion methods including concatenation,canonical correlation analysis(CCA)and discriminant correlation analysis(DCA);2)Decision fusion methods including weighting method and learning method;3)Multi-parameter dictionary learning including unsupervised and supervised dictionary learning.The result indicated that the characteristics of DCE-MRI and T2WI images had a certain correlation with the histological grade of breast cancer.Especially the morphological and texture features such as the maximum radius of the lesion and Gray-Level Non-Uniformity had significant influence on the histological grade of breast cancer.By comparing single-parameter and multi-parametric MRI results,it was found that T2WI image features contributed to the prediction results.In other words,the accuracy of the histological grade prediction was improved by combining the two imaging features.In addition,the prediction model of T2WI and the third enhancement sequence through CCA had an AUC of 0.834±0.032,which was significantly higher than the single parameter prediction results(AUC=0.737±0.039,0.751±0.038).The significance was tested by Bootstrap with a result of 0.018.Therefore,it was proved that combined T2WI and DCE-MRI imaging features can improve the predictive ability of identifying breast cancer histological grader.Multi-parametric MRI features could be utilized as promising biomarkers to predict histological grade of invasive breast cancer.