Breast Cancer Detection in Histological Images Using Radiomic Features and Fully Connected Neural Networks.

Breast cancer requires an early and accurate diagnosis for effective treatment. The examination of biopsy specimens can assist in the detection of tumor tissue, through the analysis of histological properties. Deep neural networks can be used to better identify underlying characteristics of breast cancer. In this study, we propose an approach that uses radiomic features extracted from regions and voxels of images of biological samples to differentiate breast tumor tissue from healthy tissue using a Fully Connected Neural Network. Our experiments, based on k-fold cross-validation, achieved overall accuracy of 91.5%, and sensitivity of 90.0% for a region-based approach and accuracy of 74.3% and sensitivity of 84.0% for a voxel-based approach. The individual classification of voxels allows us to identify healthy and tumor regions in the images. Such results demonstrate the potential of this approach as a tool to assist pathologists and improve the diagnosis of breast cancer.