MGH Martinos Center for Biomedical Imaging 2017 | Page 39

ized biological feedback would enable clini-
cians to identify those patients for whom
therapy is ineffective and develop an alter-
native approach to treatment. In this study,
Larimer and colleagues demonstrated that a
novel Granzyme B PET imaging agent they
designed was able to distinguish between
responders and non-responders in murine
models of cancer immunotherapy. In this way,
it could provide the much-needed biomarker.
Wednesday of the meeting, RSNA announced
that Mahmood had joined its Board of Direc-
tors as the liaison for international affairs.
Mahmood has long been a dedicated member
of the radiology community. In addition to
sitting on and chairing boards and commit-
tees at the Clinical Center of the NIH and the
Society of Nuclear Medicine and Molecular
Imaging, for example, he has served as chair
of the RSNA Research & Education (R&E)
Foundation Grant Program and as a member
The awards weren’t the only exciting pieces of the R&E Foundation Board of Trustees.
of news to emerge from Chicago. On the
‘Crowds Cure Cancer’ at RSNA
Can crowdsourcing provide us with a cure for
cancer? During an experiment at the RSNA
annual meeting in late November, researchers
with the Martinos Center and elsewhere came
a step closer to finding out.
Advances in artificial intelligence and machine
learning have broad potential for cancer pre-
vention, diagnosis, monitoring and treatment,
especially when combined with radiological
images. The Cancer Imaging Archive—a large
repository of publicly available images—is a
rich resource that could be used to this end.
But the images in the archive often lack the
necessary labels. This is where the experiment
came in.
During RSNA, the Center’s Jayashree Kalpa-
thy-Cramer and colleagues recruited attend-
Jayashree Kalpathy-Cramer. Photo by
ees to help label tumors by marking their
Caroline Magnain.
largest diameter—using computers set up at
the “Crowds Cure Cancer” booth or online they could help in developing better, AI-based
using their laptops, tablets and phones. They tools for cancer diagnosis and prognosis.
included approximately 450 cases with the
goal of securing three annotations per case.
“We are still analyzing the results but are quite
optimistic that they will be very useful. The
By the end of the meeting, participation had organizers believe that this would be a great
far outstripped their expectations. All cases way to get all the cases in the Cancer Imaging
were annotated by at least five people each. Archive annotated. And these annotations are
More than 300 were annotated by at least six. really useful for people wishing to make use of
the publicly available data—for further image
The attendees’ efforts made the experiment analysis, to combine with non-imaging data,
a success. Ultimately, Kalpathy-Cramer says, or for training algorithms to locate tumors.”
36