Radiomics is a technique in medical image analysis. It is about extracting and analyzing features in medical images. In this post, I am going to provide a brief note on the basic aspects of radiomics, providing some learning materials.


  • Radiomics often involves (classical) statistical learning and/or deep learning. You need to master the basic ideas in machine learning if you want to conduct research in this area. Accordingly, a minimum level of statistics is also required.
  • You need to know something about the DICOM standard, so as to process real-world medical imaging data. You may want to pick your favorite DICOM viewer.
  • You need to master python. It would be helpful if you are familiar with the following packages, although you can learn them whenever necessary:
    • numpy, pandas, for basic data manipulation;
    • matplotlib, simpleITK, opencv-python (cv2), for image processing;
    • pydicom, for DICOM data import;
    • torch, scikit-learn and any other machine learning packages.

Historical background and personal remarks

The name radiomics is obviously inspired by the ‘omics’ era in the early 2000’s, and in fact, it was first introduced in a 2012 paper by Philippe Lambin. It is a very bright idea, yet in retrospect, it is actually very intuitive (like many other great ideas). The foundations of radiomics root deeply in decades of discoveries in digital image processing and representation learning.

Although radiomics introduce a whole set of pipeline in image data processing, the core concept lies in the feature extraction operation. In this process, standardization and harmonization are vital to the feature stability and generalization ability.

Learning materials

There are many materials for introductory level radiomics learning. I am not providing a complete list of the materials for you to study, rather, I am only showing you what kinds of materials could be beneficial in your learning.

Software and platforms

  • The most important tool in radiomics study is pyradiomics, whose documentations are here. For the most part, pyradiomics is compliant with the IBSI standard (see below), and is one of the most popular radiomics tool used by the scientific community. Using pyradiomics, you can customize your own working pipelines. This is what we do in our research.
  • LIFEx is a well-known radiomics tool. It has a graphic interface and is rather easy to use. However, it provides relatively small flexibility to customize.
  • The Radiomics software developed by Dr. Jie Tian.

Other resources

  • As I have stated, standardization and harmonization are two key issues in radiomics. Without them, the generalization ability of radiomic models are questionable. The Imaging Biomarker Standardization Initiative is an organization devoted to develop standardized image biomarkers (features) for radiomics analysis. The works of this collaboration are highly valued by the community, and the collaboration is still continuing. You can find more about IBSI in the other blog post on this site.
  • You are encouraged to explore the GitHub topic page of #radiomics for related code and projects.