| Abstract: |
Tumor heterogeneity is a characteristic of many cancers. Radiomics is emerging as a non-invasive approach to quantify the image-based phenotypic heterogeneity within cancers and use that heterogeneity to predict outcomes. In this chapter, we present an overview of radiomics analysis applied to gynecologic cancers. We highlight some of the challenges in performing radiomics analysis including image segmentation, feature extraction, as well as ensuring robustness and reproducibility of the extracted features across different datasets. While deep learning has been emerging as an approach for circumventing some of the afore-mentioned issues by eliminating need for designing features for extraction, the problem of extracting features that are reproducible across datasets still remains and possibly extracting segmentations or at least region of interests enclosing tumors for extracting features still remains. This chapter will focus on the general issue of obtaining robust radiomics features which concerns both traditional and deep learning-based radiomics measures. We suggest some potential solutions to address those issues using example applications using non-deep learning based radiomics analysis. © 2019 by Taylor & Francis Group, LLC. |
