Intratumoral And Peritumoral Radiomics Based On Multi-parameter MRI For Prediction Of Molecular Subtypes In Breast Cancer

Purpose:This study aim to investigate the diagnostic efficacy of predicting the molecular subtypes in breast cancer based on T₂WI,DWI,DCE-MRI radiomics.And exploring the impact of different peri-tumoural areas on the diagnostic efficacy of the radiomics model.The diagnostic value of conventional MRI signs,radiomics scores and nomogram in identifying the molecular subtypes of breast cancer was also compared.Materials and Methods1.Research SubjectsFemale breast cancer patients who attended the Second Hospital of the Second Hospital of Dalian Medical University from January 2019 to December 2021 were collected after postoperative pathological confirmation,and 254 patients were finally included.2.Scanning InstrumentsGE Signa HDxt 1.5T（USA）,GE Discovery MR 750W 3.0T（USA）and Siemens Verio 3.0T（Germany）scanners were used with scanning sequences including:T₁WI,T₂WI,DWI and DCE-MRI.3.MRI features analysisMRI features were analysed and recorded as follows:tumor location,tumor length,fibroglandular tissue（FGT）composition staging,background parenchymal enhancement（BPE）,tumor shape and margin,mass-like enhancement features,time-intensity curve（TIC）staging,peritumor edema,and T₂WI signal level.4.Radiomics analysisA Volume of Interest（VOI）was obtained by manually outlining layer by layer in T2WI,DWI and DCE-MRI（T1-FS+C）images using 3D Slicer software to cover the tumor.The VOI of the peritumor region was automatically expanded in 3D space from the tumor boundary to 3 mm,6 mm,9 mm and 12 mm,respectively.1130 quantitative radiomics features were extracted from each model,and all the extracted features were imported into the Darwin Research Platform and randomly divided into training and validation groups according to the ratio of 7:3.Percentile and Minimum redundancy maximum relevance（m RMR）were used to select the top 10 variables most relevant to the categorical variables,and a logistic regression classifier was used to build the imaging histology model.The area under curve（AUC）was calculated to assess the predictive performance of the model.5.Construction of the nomogramThree groups of independent predictors of molecular subtypes of breast cancer were screened by univariate and multivariate logistic regression,and the radiomics score（Rad-score,RS）was calculated by LASSO regression and used to construct the nomogram.6.Statistical methodsSPSS 23.0 software and R software（version 4.1.2）were used for statistical analysis.Count data were analyzed using the chi-square test or Fisher’s exact probability method,and measurement data were analyzed using the Kruskal-Wallis test.Results1.General information analysis254 breast cancer patients were included in this study,including 49 TNBC,148HR+/HER2-,and 57 HER2+,with a mean age of 53.89±11.36 years.The differences in tumor size,pathological staging,lymph node metastasis,and Ki-67 expression were significant in the three groups of molecular subtypes of breast cancer（P<0.05）.2.Analysis of MRI signs of three groups of molecular subtypes of breast cancerTumor length diameter,margin enhancement,and peritumoral edema were independent risk factors for predicting TNBC.Tumor length diameter,burr,and perineural edema were independent risk factors for predicting HR+/HER2-type breast cancer.Peritumoral edema is an independent risk factor for predicting HER2+breast cancer.The AUC（95%CI）for identifying TNBC,HR+/HER2-,and HER2+based on conventional MRI signs were:0.78（0.70-0.85）,0.74（0.66-0.82）,and 0.58（0.50-0.67）,respectively.3.Predictive models for molecular subtypes of breast cancer based on T₂WI,DWI and DCE-MRI tumor radiomics featuresThe predictive models based on T₂WI,DWI and DCE-MRI（T₁-FS+C）tumor radiomics features,the best prediction efficacy of T₁-FS+C radiomics features was found in the prediction TNBC,HR+/HER2-,and HER2+tasks,with AUCs of 0.80,0.78,and0.76 for the training set and 0.78,0.79,and 0.72 for the validation set,respectively.4.Molecular subtypes prediction model for breast cancer based on DCE-MRI peri-tumoral radiomics featuresFor the prediction of TNBC,HR+/HER2-,and HER2+tasks,the highest AUC was found for the 6-mm perineural model（training set AUC:0.82,0.79,and 0.76,respectively,and validation set AUC:0.80,0.80,and 0.78,respectively）,followed by the 3-mm perineural model,and the 9-mm and 12-mm perineural models had sequentially lower prediction efficacy.Moreover,after combining the tumor and perineural features,the highest AUC was found for the combined 6-mm perineural model（training set AUC:0.92,0.86,and 0.84,respectively,and validation set AUC:0.85,0.84,and 0.82,respectively）.The best efficacy of each radiomics model was used to identify TNBC.5.Establishment and evaluation of the nomogramIndependent predictors of molecular subtypes of the three breast cancer groups in conventional MRI features with their respective Rad-scores were used to construct nomogram with AUCs of 0.848,0.834,and 0.802 for the TNBC,HR+/HER2-,and HER2+training sets,respectively;and AUCs of 0.910,0.909,and 0.845 for the validation set,respectively.Conclusion1.Conventional MRI signs including tumor length,burr,marginal enhancement, and peritumoral edema are helpful in identifying the molecular subtypes of breast cancer.2.The radiomics model of T₁-FS+C was more effective than T₂WI and DWI in predicting molecular subtypes of breast cancer,especially in identifying triple-negative breast cancer.3.The choice of peri-tumor size of the breast cancer tumor can influence the efficacy of the imaging model in predicting molecular staging,with 6 mm perimeter being the best.4.The integration of conventional MRI features and nomogram of intratumoral and peritumoral Rad-score can further improve the diagnostic efficacy of predicting the molecular staging of breast cancer.