Multiparametric Magnetic Resonance Imaging Based Radiomics For Prediction Of Histological Information Of Breast Cancer

Breast cancer is the most common cancer in women.It has drawn widespread attention due to its high morbidity and mortality.At present,the principles of personalized and scientific precision diagnosis and treatments have gradually become the focus.Breast cancer histological information is the basis for precision diagnosis and treatments of breast cancer.Breast cancer molecular subtypes,Ki-67 expression,histological grade and lymph node metastasis status are closely linked with variations in breast cancer clinical manifestations,tumor proliferation status and therapeutic efficacy,and reflect the survival and prognosis of breast cancer.These four histological indicators provide valuable references for the clinical diagnosis and treatment decision-making of breast cancer.Magnetic Resonance Imaging(MRI)is one of the most important breast imaging techniques,with multiparametric imaging modalities,including dynamic contrast-enhanced MRI(DCE-MRI),T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI).Multiparametric MRI is able to provide non-invasive and high-resolution morphological and functional tumor information,and plays an important role in breast cancer diagnosis and treatment.Radiomics is able to extract and analyze quantitative features from medical images to construct prediction models that can provide valuable diagnostic and prognostic information.Most of the traditional researches only focus on single parametric MRI,and most studies only focus on one or two histological indicators.This study combined multiparametric MRI to predict the molecular subtype,histological grade,Ki-67 expression and lymph node metastasis of breast cancer.Specific research contents are as follows:(1)Processing and analysis of histological information and imaging data: This retrospective analysis enrolled 150 patients who were diagnosed with invasive breast cancer.Basic information such as age and menopause as well as histological information including molecular subtype,histological grade,Ki-67 expression and lymph node metastasis of each patient were collected.Statistical analysis was performed on the patients' histological information using variance analysis and chi-square test.Multiparametric images of DCE-MRI,T2 WI and DWI were obtained and preprocessing analyses were carried out.Breast tumor areas were segmented and the statistical,morphological,and texture features of images were extracted.(2)Breast cancer histological information prediction models based on single parametric image features and multiparametric image features: The single feature analysis,single parametric image multi-feature prediction model study,and multiparametric image feature prediction model study were performed for each histological information prediction task.Support vector machine(SVM)algorithm and statistical methods were used in single feature analysis to explore the correlation between image features and histological information labels.The optimal image feature subset was obtained using a support vector machine recursive feature elimination(SVM-RFE)algorithm and a single parametric image prediction model based on SVM was created using the training set of each parameter imaging series.The prediction models for all parameter imaging series were fused using the probabilistic averaging method,the probabilistic voting method,and the probabilistic model optimization method.The performance of prediction was evaluated by calculating the area under the ROC curve(AUC).(3)Breast cancer histological information prediction model based on deep multiparametric image features: Convolutional neural networks were used to fuse multi parametric image features,and fully connected neural networks were used to classify histological information.Fully connected neural network models based on single parametric image features and convolutional neural network models based on deep multi parametric image features were constructed to predict breast cancer histological information.Our results showed prediction models based on multi parametric image features discriminated among the Luminal A,Luminal B,HER-2,and Basal-like subtypes with the best AUC values of 0.7995,0.7279,0.7375 and 0.7925,respectively.Prediction models based on multiparametric image features discriminated among the histological grade,Ki-67 expression and lymph node metastasis with the best AUC values of 0.8017?0.7718 and 0.7083,respectively.Prediction models based on deep multi parametric image features discriminated among the Luminal A,Luminal B,HER-2,and Basal-like subtypes with the best AUC values of 0.7219?0.6141?0.7347 and 0.7479,respectively.Prediction models based on deep multi parametric image features discriminated among the histological grade,Ki-67 expression and lymph node metastasis with the best AUC values of 0.7830?0.6560 and 0.6233,respectively.Therefore,conducting radiomics-ralated research using the combination of multiparametric MRI(DCE-MRI,T2 WI and DWI)to improve the performance of single parametric imaging model in predicting breast cancer histological information is of great significance for the diagnosis and treatment decision-making of breast cancer.