Fuzzy Clustering of Histopathological Images Using Deep Learning Embeddings

Abstract

Metric learning is a machine learning approach that aims to learn a new distance metric by increasing (reducing) the similarity of examples belonging to the same (different) classes. The output of these approaches are embeddings, where the input data are mapped to improve a crisp or fuzzy classification process. The deep metric learning approaches regard metric learning, implemented by using deep neural networks. Such models have the advantage to discover very representative nonlinear embeddings. In this work, we propose a triplet network deep metric learning approach, based on ResNet50, to find a representative embedding for the unsupervised fuzzy classification of benign and malignant histopathological images of breast cancer tissues. Experiments computed on the BreakHis benchmark dataset, using Fuzzy C-Means Clustering, show the benefit of using very low dimensional embeddings found by the deep metric learning approach.