QIBA DCE-MRI Analysis Algorithm Validation Specification and Testing Daniel Barboriak M.D. Duke University Medical Center
Overview Problems: Lack of standardized analysis Quantitative imaging results not comparable across labs Multiple causes Different models Different implementations Hinders broad acceptance CT perfusion example Goh V et al. Radiology 2007, 242:777.
Purpose of this initiative To aid in the process of DCE-MRI analysis standardization by developing standard set of DICOM images simulating DCE-MRI images To develop a framework for evaluation of DCE-MRI analysis methods
Simulated data Good: Ideally, can test software to see that it performs as designed (verification, not validation) Bad: Simulated data is not realistic Challenge is to create simulated data appropriate to the use case.
Measuring performance Challenging: what is the proper metric? Robust vs. accurate Realistic view: all software fails at extremes. Simulations can define the areas of “parameter space” for which results are inaccurate and/or unreliable.
Benefits of this initiative Primary: evaluation of analysis tools for the clinical trial outlined in the use case Secondary: Provide test data useful to analysis software developers before release Support FDA approval of analysis package Compare analysis methods after release
Action plan phases Phase I Review and next steps Define acceptance criteria for software packages (don’t look for perfection) Phase II Software package analysis Phase III An analysis framework
Phase I Review progress that has already has been made ( mri_test_images/index.php?id=1)
Steps Define a “vascular input function” Choose a model and create tissue simulations from the vascular function Convert to signal intensities Convert to DICOM images
Phase I Next steps Analysis model Synthetic T1 mapping images Ratio image bias correction Noise addition
Phase I Software output problems Figure of merit Acceptance criteria