In Search of the Optimal Set of Indicators when Classifying Histopathological Images

Abstract

There is currently a large amount of histopathological images due to the intensive prevention screening programs worldwide. This fact overloads the pathologists' tasks. Hence, there is a connected high need for a quantitative image-based evaluation of digital pathology slides. The current work extracts 76 numerical features from 357 histopathological images and focuses on the selection of the most valuable features that conducts to a smaller data set on which a SVM classifier achieves a better prediction. The gain in accuracy is of over 4% more than in the situation when the entire data set was used. The paper also indicates a subset of the attributes that proved to be the most informative with respect to 4 feature selection approaches.