Meeting Banner
Abstract #1091

Tracer-Kinetic Model-Driven Registration Improves Data-Driven Tumour Sub-Segmentation of DCE-MRI Data

Giovanni Alessandro Buonaccorsi1, Caleb Roberts1, James P. B. O'Connor1, Chris J. Rose1, Susan Cheung1, Yvonne Watson1, Alan Jackson2, Gordon C. Jayson3, Geoff J. M. Parker1

1ISBE, University of Manchester, Manchester, United Kingdom; 2WMIC, University of Manchester, Manchester, United Kingdom; 3Cancer Research UK Dept of Medical Oncology, Christie Hospital, Manchester, United Kingdom


Using DCE-MRI data from ten patients enrolled in a trial of the VEGF inhibitor bevacizumab we performed tracer-kinetic model-driven registration (TKMDR) followed by cross-visit tumour sub-segmentation. In 7 of the 9 evaluable post-registration data sets, TKMDR altered the number of retained principal components (PCs). We present an example in which a cluster in an unregistered segmentation result could have been misinterpreted as a treatment effectwe demonstrate the removal of the cluster by TKMDR and provide evidence that TKMDR improved the data structure in the PC space by eliminating patterns that were attributable to motion corruption.