The Clinical Application Research Of ADC Map Based Radiomics In Predicting Histological Grade Of Invasive Breast Carcinoma Of No Specific Type

Purpose: To investigate the value of radiomics based on ADC map in predicting histological grade of invasive breast carcinoma of no specific type.Method: A total of 152 patients with MRI examination and pathologically confirmed invasive breast carcinoma of no specific type were retrospectively enrolled in our hospital,and sampled according to 7: 3 ratio,the patients were randomly divided into training cohorts(n = 108,54 cases in grade ?/?,54 cases in grade ?)and testing cohorts(n = 44,22 cases of grade ?/?,22 cases of grade ?).The ADC images acquired by each patient on the Siemens 3.0 T MR scanner are used for image preprocessing,full tumor segmentation and feature extraction through AK and ITKSNAP software.The radiomics features that have the greatest correlation and the least redundancy with the histological grade in the training cohorts were selected to construct radiomic signature,three simple radiomics models including support vector machine(SVM),decision trees(Tree),and logistic regression were established.Single-factor Logistic regression analysis was performed on clinical and imaging signs,and the characteristics of p <0.1 were retained,these features combined with the radiomic signature score of the best simple radiomics model to establish a combined model.The performance of the model was assessed by the area under curve(AUC)of the receiver operating characteristic curve and verify its classification performance in the testing cohorts.the Delong test was used to compare AUC between models.Decision curve analysis and calibration curve were applied to assess the clinical usefulness.Result:1.The AUC of Logistic regression,SVM,Tree and combined model in the training cohorts were 0.692(95% CI: 0.592-0.791),0.696(95% CI: 0.597-0.795),0.832(95% CI: 0.757-0.907),0.904(95% CI: 0.851-0.956),and the AUC in the test cohorts were 0.577(95% CI: 0.402-0.752),0.613(95% CI: 0.442-0.783),0.777(95% CI: 0.649-0.904),0.807(95% CI: 0.676-0.937);2.The difference between the AUC of Logistic regression and SVM was not statistically significant(P > 0.05).The AUC of Tree was significantly higher than that of Logistic regression and SVM(P <0.05).The AUC of the combined model of image signs and Tree model was higher than that of the Tree(P > 0.05);In the analysis of decision curve,the net benefit of Tree and combination model is higher than Logistic regression and SVM;Conclusion:1.The radiomics model based on ADC map can predict the histological grade of invasive breast carcinoma of no specific type;2.Different machine learning algorithm models have different classification performance and clinical application value for invasive breast carcinoma of no specific type histological classification.The best classifier is Tree,and its prediction performance can improved by combining imaging signs.