Machine Learning and Radiomics in Radiation Oncology

Medical imaging constitutes a cornerstone of clinical practice. Especially in Radiation Oncology, the use of computed tomography (CT), magnetic resonance imaging (MRI) and positron emission tomography (PET) is essential for therapy decisions, therapy planning, therapy delivery, and patients' follow-up.
Multiple studies were able to demonstrate that quantitative algorithm-based analysis of such imaging data can provide additional information extending beyond the qualitative assessment of the images. Such information has been used for tumor characterization such as prediction of molecular aberrations (e.g. mutations) or prediction of patients' outcome.
The underlying field of "Radiomics" can be considered as a two-step process with (1) extraction of quantitative imaging features, and (2) incorporation of these features into mathematical models to predict clinical endpoints. A variety of predefined radiomics features can be extracted from a given volume of interest. Applied to the analysis of a malignant tumor, these features describe intensity statistics, determine the spatial configuration of the tumor's shape or quantify the texture of the tumor. Following the complexity of the resulting feature numbers, dimensionality reduction methods and machine learning techniques are necessary to achieve maximally accurate prediction of clinical endpoints.
Besides the extraction of predefined features, neural networks can be applied directly to image analysis. This allows for the generation of novel imaging features and the direct prediction of endpoints by the neural network.

The Radiomics Workflow [1]:

Our research topics:

Topic 1:
Application of machine learning and Radiomics for improved prognostic assessment of oncological patients.

Topic 2:
Application of machine learning and Radiomics for non-invasive tumor characterization.

Topic 3:
Application of machine learning and Radiomics for the prediction of radiation therapy-dependent side effects and therapy response.

Reference
[1] Peeken JC, Nüsslin F, and Combs SE. “Radio-oncomics” - The potential of radiomics in radiation oncology. Strahlenther Onkol 2017;193:767–79. doi:10.1007/s00066-017-1175-0.