| Background Breast cancer is the first high-risk cancer affecting women ’s health worldwide.It is highly heterogeneous,and there are huge differences in the progression,treatment and prognosis of different molecular subtypes of breast cancer.In the era of precision treatment,traditional imaging examinations mainly use magnetic resonance morphology to help diagnose and evaluate treatment responses,and cannot observe the heterogeneity of tumors by the naked eye.Texture analysis can extract texture features to reflect the characteristics of malignant tumor microenvironment,providing a non-invasive diagnostic method for clinical practice.Objective Texture analysis was performed on breast dynamic enhanced magnetic resonance imaging(MRI)images to study the feasibility and value of using texture features to identify molecular typing and axillary lymph node status in breast cancer patients.The diagnostic efficacy of the prediction model based on MRI texture features combined with machine learning algorithms was preliminarily discussed.Methods Retrospectively collected 75 cases of breast cancer patients who met the inclusion and exclusion criteria in the Yichang First People’ s Hospital.All patients underwent dynamic contrast-enhanced magnetic resonance imaging before surgery.The results of pathological immunohistochemistry were divided into four molecular subtypes : luminal A,luminal B,HER-2 overexpression and triple-negative breast cance(TNBC).According to axillary lymph node status,they were divided into lymph node metastasis and non-lymph node metastasis groups.The general data of patients were analyzed by 2 test and variance analysis.The DICOM images with the most significant enhancement of dynamic contrast-enhanced MRI lesions in the collected cases were imported into the post-processing platform(Darwin Research Platform),and the region of interest(ROI)was delineated along the edge of the lesion.The molecular subtypes and lymph node status were taken as the two-category indicators to extract the feature parameters.The feature parameters were reduced by standardized and optimal feature filters.The independent sample t test or Mann-Whitney U test was used to identify the optimal texture parameters with statistical differences between different groups.The area under the receiver operating characteristic(ROC)curve(AUC)was used to evaluate the diagnostic efficacy of texture analysis.In addition,a logistic regression classification model was constructed based on dynamic enhanced MRI texture features,and the ROC curve was drawn to evaluate the diagnostic efficacy of the model for different molecular subtypes and lymph node status.Results(1)Among the 75 patients with breast cancer,there were 11 cases of luminal A,36 cases of luminal B,14 cases of HER-2 overexpression and 14 cases of TNBC,24 cases with lymph node metastasis and 51 cases without lymph node metastasis.There was no significant difference in MRI enhancement,pathological type,lymph node status,age and menopausal status among patients with different molecular types of breast cancer(P > 0.05).(2)Based on the comparison between the two groups of different molecular types,the three texture parameters between the luminal A group and the non-luminal A group were statistically significant.Among them,the texture feature 3D＿glcm＿CS had the highest differential diagnosis efficiency(AUC = 0.753,P = 0.008).When 3D＿glcm＿CS > 0.439,the sensitivity of the diagnosis of luminal A breast cancer was 54.55 %,and the specificity was 90.62 %.There were statistically significant differences in two texture parameters between luminal B and non-luminal B groups.The texture feature o＿glcm＿CS had the best differential efficacy(AUC = 0.652,P = 0.024).When o＿glcm＿CS > 0.169,the sensitivity and specificity of diagnosing luminal B breast cancer were 72.22 % and 64.10 %,respectively.There were five texture parameters with statistical significance between HER-2overexpression group and non-HER-2 overexpression group,and the optimal features were the most in all groups.The five texture features were lbp-2D＿glrlm＿RE,s＿glszm＿DE,3D＿gldm＿SDLGLE,3D＿glszm＿LGLZE,3D＿glszm＿SALGLE,and their AUC values were 0.756,0.810,0.793,0.799 and 0.819,respectively(P < 0.01).Among them,3D＿glszm＿SALGLE had the best differential efficacy(AUC = 0.819,P < 0.001).When 3D＿glszm＿SALGLE≤-0.460 was used to diagnose HER-2 overexpressing breast cancer,the sensitivity was 92.86 % and the specificity was 70.49 %.There was only one texture feature difference between TNBC and non-TNBC groups.When w-LH＿ngtdm＿B >-0.255 was used to diagnose TNBC,the sensitivity was 92.86 %,the specificity was 39.34 %,and the AUC was0.653.(3)Comparison between the two groups based on lymph node status : three texture features were statistically significant between lymph node metastasis and non-lymph node metastasis groups.Among them,w-LH＿glcm＿Imc1 had the highest differential diagnostic efficiency(AUC = 0.694,P = 0.006).When w-LH＿glcm＿Imc1 >-0.049 was used to diagnose axillary lymph node metastasis,the sensitivity was 79.17 % and the specificity was56.86 %.(4)A binary classification model based on logistic regression was constructed by feature screening and dimension reduction.The AUCs of the best models for identifying luminal A and non-luminal A,luminal B and non-luminal B,HER-2 overexpression and non-HER-2 overexpression,TNBC and non-TNBC,lymph node metastasis and non-lymph node metastasis test groups were 0.92,0.83,0.83,0.72 and 0.91,respectively.Among them,the predictive models for identifying luminal A and non-luminal A,lymph node metastasis and non-lymph node metastasis had better diagnostic performance,and the AUC values were all above 0.9.Conclusion(1)Dynamic enhanced MRI texture analysis can effectively predict the molecular typing and axillary lymph node status of breast cancer before operation,especially for the differential diagnosis of HER-2 overexpressing breast cancer.(2)The logistic regression classification model based on dynamic enhanced MRI texture analysis has a good performance in the identification of breast cancer molecular typing and lymph node status.It can be used in clinical non-invasive prediction of breast cancer molecular typing and axillary lymph node status,and provide a theoretical reference for clinicians to make preoperative decisions. |
PDF Full Text Request