Radiomics Based Tumor Lesion Segmentation And Prognosis Analysis In Breast Cancer

Radiomics research is of great potential and value in quantifying tumor heterogeneity and assisting clinical decision-making.In the clinical diagnosis and treatment of breast cancer,whether lymph node metastasis is important for the development of surgical process.However,there is no safe and effective method for predicting lymph node metastasis.In this paper,we applied Radiomics method to quantifying the lung CT images for the prediction of lymph node metastasis in breast cancer.We collected 456 cases of breast cancer from Guangdong People's Hospital.We proposed a semi-automatic contrast enhancement random walk method for the CT breast cancer segmentation.Our proposed algorithm first extracted local breast lesion using manual assistant and maximum region finding.Then,it used the modified Histogram Equalization with Iterative-Filling to enhance the contrast.Finally,it used the four seeds random walk to segment the lesions.We used the segmented lesions and extract 592 features from each case.Through these quantitative image intensity features,shape and size features,texture features and wavelet features of tumor images,the model of lymph node metastasis basic prediction in breast cancer is constructed using LASSO Logistic Regression(LLR).The result of LLR is the Rad-score.Multivariable logistic regression analysis starting with the clinical information and Rad-score is developed into a nomogram as the final individualized prediction model.From the above program,we get the flowing results.The AUC of the training set is 0.712 and the AUC of the validation set is 0.737 on the LLR basic model.And compared with different models,the value of F1 is the best one 0.705.On the nomogram,all the data are used to take individualized prediction that the C-index is0.721(95% CI: 0.676-0.768).The good performance on the calibration curve for all the data provides valuable clinical information for real cases.