Presentation at the October 2015 Radiomics meeting by Paul Kinahan, Matt Nyflot, Larry Pierce, Art Chaovalitwongse, Hannah Linden
Paul Kinahan, from U. Washington, discussed the need for improvements in standards, tools, and best practices. Because big data/small N analyses are prone to overfitting, it is critical that results are tested against independent data sets by independent investigators. Disturbingly, when PET or CT phantoms, as well as DICOM digital reference objects (DRO) are imaged or analyzed across sites, there are high amounts of variability. He stressed that one cannot do quantitative imaging that is reproducible unless there are established reference standards against which tools can be assessed. Thus there is a critical need, being addressed by AAPM, the RSNA’s QIBA and the NCIs QIN to develop such reference standards. This has not been sufficiently addressed in radiomics, which extracts features that may not be obvious and have yet to be tested across sites.
Radiomics 2015 saw us return to the Hilton on Clearwater Beach, after last year’s successful meeting with the AAPM in Houston. Keeping count, Radiomics Retreats began in 2009 with an in-house retreat of Moffitt and MAASTRO investigators, and continued to be held on campus through 2011. In 2012, we moved the venue to the Clearwater Hilton, and opened registration to about 70 interested investigators. The meeting was at the same place in 2013, and slightly larger. In 2014, at the invitation of the AAPM research committee, we were hosted by John Hazle and Rivka Colen at MD Anderson, and this included an expansion of attendees to approximately 90 scientists. As a result of the increased exposure, this meeting was the largest yet, with over 100 attendees from 5 countries (China, Netherlands, UK, Canada, Germany) and 27 institutions.
Thanks to Matthieu Hatt and Fei Yang
Researchers should cite this work as follows: