An ensemble of learned features and reshaping of fractal geometry-based descriptors for classification of histological images

Classification of histology images has been the focus of plenty researchers in computer vision. Recently, the most common approaches for this task consist of applying deep learning through CNN models. However, there are some limitations to the use of CNN in the context of histological image classification such as the need for large datasets and the difficulty to implement a generalized model able to classify different types of histology tissue. In this work, an ensemble model based on handcrafted fractal features and deep learning that consists of combining the classification of the ResNet-50 model and the classification of local and global handcrafted features by applying the sum rule is proposed. Fractal geometry concepts are used to obtain handcrafted local and global features from different histological datasets. The local features are reshaped into a matrix in order to compose a feature image. Four different reshaping procedures are evaluated, wherein each generates a representation model of fractal features which is given as input to a CNN model. Another CNN architecture receives as input the original image. After associating the results of both CNN models with the classification of the handcrafted local and global features using machine learning approaches, accuracy rates that range from 88.45% up to 99.77% on five datasets were obtained. Moreover, the model was able to classify images from datasets with different resolutions and imbalanced classes with few training epochs. In general, the proposed method is able to provide results that are compatible with the state-of-the-art in histology image classification.