Transferência de conhecimento utilizando aprendizado profundo para classificação de imagens histopatológicas

Abstract

In the present work, the authors make an investigation about the use of knowledge transfer applied to the classification of histopathological images. Knowledge transfer is carried out from deep neural networks previously trained for the recognition of large multi-class data sets that have had good results in the classification of the set for which they were trained. Recent work analyzes that these deep learning architectures can extract relevant information from images aiming for improvements in the classifications of other classifiers that will later use the output vectors generated by these networks as training and test data. Based on the results obtained by other works, we reproduced the technique of knowledge transfer for the classification of histopathological images using different architectures for the extraction of characteristics and different classifiers. The use of knowledge transfer on the database of histopathological images showed promising results and can be well explored in order to bring improvements to the automatic classification of this database.