Project 2: Predictive modeling for radiomics biomarkers in solid cancers

Background

Cancer is a heterogeneous disease regarding etiology, pathogenesis, treatment response and prognosis. For optimizing treatment strategies on a patient individual level, identification of prognostic and predictive biomarkers is essential. Imaging biomarkers are of special interest as they provide spatial information and are acquired non-invasively. Radiomics is a large throughput method for quantification of medical images with comprehensive and quantitative analysis of areas involved by malignant cancer and regions of normal tissue. Such quantitative radiomic features have been shown as promising imaging biomarkers for prediction of overall survival or treatment response.

Our group has developed a fully DICOM compatible radiomic implementation, which has been standardized in a multicenter study. Using this implementation, we have shown that head and neck squamous cell cancer characterized by more heterogeneous CT density has a higher probability of local recurrence. Further analyses indicated that adding metabolic information from 18F-FDG PET did not improve the prediction model.

For an introduction to our work, please watch this video!

Working hypothesis

  1. The current results obtained for HNSCC using our software implementation can be expanded to other solid tumors irrespective of imaging modality (CT, MRI, PET).
  2. Tumor specific outcome prediction models can be developed to predict local tumor control and overall survival based on pre- and post-therapeutic imaging (predictive response assessment)

Specific aims for this research project

Within this WP we aim to establish a comprehensive software suite that integrates radiomic feature analysis and outcome prediction models:

  1. To establish the methodology for radiomics analyses (semi-automated image segmentation, feature extraction and analysis) using longitudinal imaging studies (delta radiomics).
  2. To implement a ML-based predictive modelling framework with automated algorithms to predict clinically relevant outcome measures from imaging features. 
  3. To adapt the radiomics implementation to general medical imaging modalities, e.g. incident light skin microscopy pictures (WP Early detection of skin cancer).

 

Figure radiomics