Files

Download

Download Full Text (527 KB)

Presentation date

2021

College, Institute, or Department

Radiation Oncology

Abstract

Medical imaging is an essential component of clinical cancer treatment, with Computer Tomography scans being the most popular type of imaging used for such diagnostic purposes. Currently, the analysis of these scans is mainly done to interpret areas of concern, as well as to determine the size of the lesion or tumor. Experimentation with these scans and their corresponding computer programs has led to recent success in quantitative imaging analysis. This type of analysis is redefining the role of medical imaging as a new source of biomarker data. Radiomics is a method of high-throughput extraction on hundreds of features encrypted in these scans and images based on the delineation of boundary of a 3D volume of interest (VOI), called segmentation. These features include the shape, and the first-, second-, and higher-order statistics of a 3D VOI. The features provide a comprehensive and quantitative description of a tumor’s phenotype and are more in-depth than qualitative descriptors from doctors. This is an active area of research because of the advantages for biomarker development, which focus on risk assessment, treatment response predictions, and the relationship between image features and genomics. The inconsistency and variability of segmentation and extraction negatively affects the robustness of the predictive model and its ability to generate with a different dataset. Patient positioning and image acquisition also affect each feature to varying degrees by introducing different perturbations such as image rotation. In this study, we attempt to evaluate the uncertainties introduced by the image perturbation (e.g., rotation or translation) as well as image resampling

Keywords

Radiomics, medical imaging

Uncertainty Analysis of Radiomic Features in Normal and Pancreatic Cancer Patients
COinS