Evaluation of radiological features for breast tumour classification in clinical screening with machine learning methods

摘要: Objective:: In this work, methods utilizing supervised and unsupervised machine learning are applied to analyze radiologically derived morphological calculated kinetic tumour features. The features extracted from dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) time-course data. Material:: DCE-MRI data of the female breast obtained within UK Multicenter Breast Screening Study. group patients imaged in study is selected on basis an increased genetic risk for developing cancer. Methods:: k-means clustering self-organizing maps (SOM) signal structure terms visualization. We employ k-nearest neighbor classifiers (k-nn), support vector machines (SVM) decision trees (DT) classify using a computer aided diagnosis (CAD) approach. Results:: Regarding techniques, according indicating benign malignant characteristics observed limited extend. approaches classified with 74% accuracy providing area under receiver-operator-characteristics (ROC) curve (AUC) 0.88 (SVM). Conclusion:: It was found that contour wash-out type (WOT) determined by radiologists lead best SVM classification results. Although fast uptake early time-point measurements important feature malignant/benign tumours, our results indicate might be considered as important.