Clinical Diagnosis And Prediction Of Breast Cancer Based On Mammography Image Radiomics Analysis

For breast cancer patients,axillary lymph node metastasis is the main pathway for cancer cell metastasis.Preoperative axillary lymph node metastasis in breast cancer patients has a great significance of tumor staging,treatment planning and prognosis evaluation.In the clinical,the way of the axillary lymph node metastasis status both has some deficiencies,whether based on sentinel lymph node biopsy or intraoperative axillary lymph node biopsy.For most BIRADS 4 patients,there is no accurate way to classify them and the treatment methods used by BIRADS 4a and BIRADS 4c are quite different.BIRADS 4a and BIRADS 4c are also large differences in the benign and malignant tumors.The estrogen receptor for breast cancer can be used to classify patients into two categories.The first are Luminal A and Luminal B,the treatments are endocrine therapy;other is HER-2 overexpression and Basal-like,HER-2 overexpression uses targeted drug therapy and Basal-like type uses radiotherapy.It is a great clinical significance to explore the factors that affect the benign or malignant tumors in normal tissues of breast cancer patients.The main work and innovation of this paper consists of the following parts:First,147 patients were selected in this study(110 and 37 in training and validation sets,83 and 64 in axillary lymph node metastasis positive and negative,respectively),and 299 image features were extracted from tumor tissue,including first-order features,texture features,shape features and wavelet transform features.Ten features were obtained by the LASSO algorithm to establish feature signatures.The performance of the feature signatures was verified by the support vector machine.The AUC of the training and validation sets were 0.875(95% CI,0.689-0.891)and 0.800(95% CI,0.664-0.832),respectively.Combined feature signatures with age,T stage,tumor location,ultrasound diagnosis of axillary lymph node status,estrogen receptor status,and progesterone receptor status,a multivariate logistic regression model was established.The C-Index of the training and validation sets were respectively 0.779(95% CI,0.752-0.793)and 0.809(95% CI,0.794-0.833).Second,this paper selected 244 BIRADS 4 patients(183 and 61 in training and validation sets,103 and 141 benign and malignant tumors,respectively).257 features were extracted based on tumor area,including first-order features,texture features,wavelet transform features.LASSO algorithm was used to obtain five features that were used to establish feature signature.The AUC of SVM model in the training and validation sets were 0.862(95% CI,0.825-0.928)and 0.690(95% CI,0.586-0.712),respectively.The C-Index of logistic regression model in the training and validation sets were 0.772(95% CI,0.701-0.836)and 0.676(95% CI,0.576-0.721),respectively.Third,97 patients were selected(72 and 25 in training and validation sets,48 and 49 in estrogen receptor positive and negative,respectively).257 features were extracted based on the tumor area,including first-order features,texture features,wavelet transform features.Six features were obtained by LASSO algorithm and these features were used to establish feature signature.The performance of feature signatures was show in AUC of SVM model,and the AUC of the training and validation sets were 0.857(95% CI,0.758-0.889)and 0.676(95% CI,0.633-0.725),respectively.The C-Index of logistic regression model in the training and validation sets were 0.819(95% CI,0.782-0.846)and 0.680(95% CI,0.662-0.709),respectively.Fourth,792 patients were selected in this paper(594 and 198 training sets and test sets,396 and 396 malignant tumors and normal subjects,respectively).The tissues that get rid of the tumors were obtained after segmentation and 424 features were extracted from the image.Including the texture features and wavelet transform features.Difference analysis used to select seven features indicating significant differences between normal and patient groups.Seven characteristics were used to establish a multivariate logistic regression model.The C-Index of the training and validation sets were 0.935(95% CI,0.886-0.950)and 0.917(95% CI,0.868-0.945),respectively.In summary,the radiomics method was used to accurately evaluate the axillary lymph node metastasis status,the benign or malignant in BIRADS 4 patients,estrogen receptor status and the factors affecting the benign or malignant tumors in normal tissues of breast cancer patients has all show a great performance.