Breast cancer nuclei segmentation and classification based on a deep learning approach (Q2042989)

The paper is concerned with image segmentation of breast cancer data (images) and involves a two-class classification problem. The overall design process realizing segmentation and leading to classification: image preprocessing, semantic segmentation, cell nuclei detection and instance segmentation, feature extraction and selection, and classification. Each phase uses in a systematic way a slew of well-known techniques encountered in image processing, namely U-neural network (for segmentation purposes), LASSO (feature selection) and a collection of commonly used classifiers (LDA (linear discriminant analysis), QDA (quadratic discriminant analysis), SVM (support vector machine), RF (random forests), NB (naive Bayes), KNN (k-nearest neighbors), and RPART (recursive partitioning and regression trees). The performance evaluation of classifiers involves a leave-one-out and k-fold cross-validation. The detailed experimental results obtained for real-world data are reported; it is demonstrated with the SVM classifier producing the best resulting among all classifiers under investigation.