The demand for automatically annotating and retrieving medical image
s i
s gro
wing fa
ster than ever. In thi
s paper,
we pre
sent a novel medical image retrieval method for a
special medical image retrieval problem
where the image
s in the retrieval databa
se can be annotated into one of the pre-defined label
s. Even more, a u
ser may query the databa
se
with an image that i
s clo
se to but not exactly
what he/
she expect
s. The retrieval con
si
st
s of the
deducible retrieval and the
traditional retrieval. The deducible retrieval i
s a
special
semantic retrieval and i
s to retrieve the label that a u
ser expect
s while the traditional retrieval i
s to retrieve the image
s in the databa
se
which belong to thi
s label and are mo
st
similar to the query image in appearance. The deducible retrieval i
s achieved u
sing SEMI-
supervi
sed Semantic Error-Correcting output Code
s (SEMI-SECC). The active learning method i
s al
so exploited to further reduce the number of the required ground truthed training image
s. Relevance feedback
s (RF
s) are u
sed in both retrieval
step
s: in the deducible retrieval, RF act
s a
s a
short-term memory feedback and help
s identify the label that a u
ser expect
s; in the traditional retrieval, RF act
s a
s a long-term memory feedback and help
s ground truth the unlabelled training image
s in the databa
se. The experimental re
sult
s on IMAGECLEF 2005 [
style="text-decoration:none; color:black" href="/science?_ob=MathURL&_method=retrieve&_udi=B6V10-4S01WJ1-4&_mathId=mml10&_user=1067359&_cdi=5660&_rdoc=26&_acct=C000060157&_version=1&_userid=3273504&md5=a0f9cee63a2f747bcbd6da3932c16758" title="Click to view the MathML source" alt="Click to view the MathML source">src="http://www.sciencedirect.com/scidirimg/entities/2329.gif" alt="left angle bracket" title="left angle bracket" border="0">science?_ob=RedirectURL&_method=externObjLink&_locator=url&_cdi=5660&_plusSign=%2B&_targetURL=http%253A%252F%252Fir.shef.ac.uk%252Fimageclef2005%252F" target="externObjLink" onClick="var parms = 'status=yes,location=yes,' + 'scrollbars=yes,resizable=yes,directories=yes,' + 'toolbar=yes,menubar=yes,' + 'width=400,height=600' + ',screenX=10,screenY=10';var externalWin; externalWin=window.open('','externObjLink',parms); externalWin.focus()">http://ir.shef.ac.uk/imageclef2005/style="text-decoration:none; color:black" href="/science?_ob=MathURL&_method=retrieve&_udi=B6V10-4S01WJ1-4&_mathId=mml11&_user=1067359&_cdi=5660&_rdoc=26&_acct=C000060157&_version=1&_userid=3273504&md5=1a80aba4e53b5eabba713443bce0ecb8" title="Click to view the MathML source" alt="Click to view the MathML source">src="http://www.sciencedirect.com/scidirimg/entities/232a.gif" alt="right-pointing angle bracket" title="right-pointing angle bracket" border="0">] annotation data
set clearly
sho
w the
strength and the promi
se of the pre
sented method
s.