Background

OPTIMAM1 developed mathematical simulation tools to allow virtual clinical trials. Using these tools we have successfully completed observer studies to measure the impact of various factors on cancer detection. However, these tools have not yet been used to study and optimise the latest designs of imaging system.
 
We plan to study the following:

  • Improved breast dose levels and X-ray quality settings
  • Different designs of DBT with different tube motions, angular ranges, intervals and reconstruction algorithms
  • Synthetic 2D mammograms created from images acquired for DBT, which may enable DBT alone to replace 2D mammography in screening (Gur et al 2012)
OPTIMAM2 requires good working relationships with the major X-ray manufacturers, e.g. to get access to raw data of emerging modalities and cooperation in reprocessing and reconstruction of test images. We have the support and cooperation of all the major manufacturers (GE, Philips, Siemens, Hologic, IMS, Agfa Healthcare) for this project.
 
Many of our work-packages use human observer studies to assess cancer detection. In OPTIMAM1 we used statistical techniques known as receiver operating characteristics (ROC) and free response receiver operating characteristics (FROC) and alternative FROC (AFROC) to analyse our results and test for statistical significance (e.g. Warren et al, 2012). We will use the same methodology in OPTIMAM2 and Professor Chakraborty, a world-leading expert in analysis of this type will continue to provide advice. These tests evaluate improvement in terms of both detection and reduced unnecessary recalls (false positives).
 
In OPTIMAM2 we will develop and use mathematical tools known as model observers. While their development is challenging they have the potential to allow parameters to be tested with a level of detail that would be impractical with human observers.
 
In OPTIMAM1 we relied on manufacturers' DBT reconstruction software for our simulations. To allow a greater choice of parameters we will also use in-house software developed by the group in Leuven. This can employ either simple filtered back projection (FBP) or more complex iterative approaches.