Click here to


Are you sure ?

Yes, do it No, cancel

Introduction to Radiomics

L Court1*, X Fave2*, S Zhou3*, (1) UT MD Anderson Cancer Center, Houston, TX, (2) University of California San Diego Moores Cancer Ctr, La Jolla, CA, (3) UT MD Anderson Cancer Center, Houston, Texas


(Wednesday, 8/1/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

The central hypothesis of radiomics is that quantitative features measured from routine medical images are related to the tumor phenotype and provide important information for personalized medicine. Radiomics researchers apply a variety of techniques to calculate and use this information, such as feature extraction methodologies and machine learning, including deep learning. In this session, which is the first in the Radiomics Certificate Course, we will provide an introduction to the field of radiomics, considerations when calculating quantitative image features, and the necessary statistical methods to relate the features to clinical endpoints. Subsequent sessions will describe machine learning techniques, including deep learning, as they apply to radiomics research.

When using quantitative image features for radiomics, the initial workflow involves segmenting an image, applying image processing, fine-tuning parameters for the radiomics features you wish to measure, and extracting them. Each step is dependent on image modality and research goals. Additionally the majority of features are not yet linked to biological characteristics which can make selecting appropriate image processing and feature parameters challenging. In this presentation, we will describe basic principles and guidelines for each step required to calculate radiomics features. We will highlight how changes in texture matrix parameters, segmentation and image processing techniques can affect radiomics features. We will also discuss the pros and cons of calculating features from multiple imaging modalities.

A variety of statistical tools are prevalent and useful for understanding the dynamics of radiomics features and relating them to clinical endpoints in medical research. When the primary outcomes are time-to-events, such as progression-free survival and overall survival, the conventional survival analysis and relevant methods can be applied as a benchmark to evaluate the goodness of fit of machine learning techniques. We will briefly review the standard statistical approaches, including parametric and semi-parametric survival regression models, penalized regressions, dynamic prediction, statistical model selection and measures of assessment, with discussion of their assumptions, advantages and limitations.

Learning Objectives:
1. To introduce the goals and objectives of radiomics research
2. To describe the current status of radiomics research
3. To understand the workflow when using quantitative image features for radiomics research
4. To understand the key statistical techniques used in radiomics

Funding Support, Disclosures, and Conflict of Interest: This work was partially funded by the NCI



Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email