<s>
GemIdent	B-Application
is	O
an	O
interactive	O
image	O
recognition	O
program	O
that	O
identifies	O
regions	B-Algorithm
of	I-Algorithm
interest	I-Algorithm
in	O
images	O
and	O
photographs	O
.	O
</s>
<s>
For	O
example	O
,	O
color	O
image	B-Algorithm
segmentation	I-Algorithm
of	O
:	O
</s>
<s>
GemIdent	B-Application
also	O
packages	O
data	O
analysis	O
tools	O
to	O
investigate	O
spatial	O
relationships	O
among	O
the	O
objects	O
identified	O
.	O
</s>
<s>
GemIdent	B-Application
was	O
developed	O
at	O
Stanford	O
University	O
by	O
Adam	O
Kapelner	O
from	O
June	O
,	O
2006	O
until	O
January	O
,	O
2007	O
in	O
the	O
lab	O
of	O
Dr.	O
Peter	O
Lee	O
under	O
the	O
tutelage	O
of	O
Professor	O
Susan	O
Holmes	O
.	O
</s>
<s>
Hence	O
,	O
GemIdent	B-Application
works	O
well	O
when	O
identifying	O
cells	O
in	O
IHC-stained	O
tissue	O
imaged	O
via	O
automated	O
light	O
microscopy	O
when	O
the	O
nuclear	O
background	O
stain	O
and	O
membrane/cytoplasmic	O
stain	O
are	O
well-defined	O
.	O
</s>
<s>
identification	O
of	O
regions	B-Algorithm
of	I-Algorithm
interest	I-Algorithm
in	O
the	O
images	O
.	O
</s>
<s>
These	O
scores	O
are	O
then	O
used	O
to	O
build	O
a	O
random	B-Algorithm
forest	I-Algorithm
machine-learning	O
classifier	O
which	O
will	O
then	O
classify	O
pixels	O
in	O
any	O
given	O
image	O
.	O
</s>
<s>
These	O
training-classifying-retraining-reclassifying	O
iterations	O
(	O
considered	O
interactive	O
boosting	B-Algorithm
)	O
can	O
result	O
in	O
a	O
highly	O
accurate	O
segmentation	B-Algorithm
.	O
</s>
<s>
This	O
example	O
illustrates	O
GemIdent	B-Application
's	O
ability	O
to	O
find	O
multiple	O
phenotypes	O
in	O
the	O
same	O
image	O
:	O
the	O
raw	O
microscopic	O
image	O
of	O
a	O
stained	O
lymph	O
node	O
(	O
top	O
left	O
)	O
from	O
the	O
Kohrt	O
study	O
,	O
a	O
superimposed	O
mask	O
showing	O
the	O
pixel	O
classification	O
results	O
(	O
top	O
right	O
)	O
,	O
and	O
finally	O
the	O
image	O
marked	O
with	O
the	O
centroids	O
of	O
the	O
objects	O
of	O
interest	O
-	O
the	O
cancer	O
nuclei	O
(	O
in	O
green	O
stars	O
)	O
,	O
the	O
T-cells	O
(	O
in	O
yellow	O
stars	O
)	O
,	O
and	O
non-specific	O
background	O
nuclei	O
(	O
in	O
cyan	O
stars	O
)	O
.	O
</s>
<s>
The	O
histogram	B-Algorithm
displays	O
the	O
distribution	O
of	O
distances	O
from	O
T-cells	O
to	O
neighboring	O
cancer	O
cells	O
.	O
</s>
<s>
The	O
open	O
PDF	B-Application
document	O
is	O
the	O
autogenerated	O
report	O
of	O
the	O
analysis	O
which	O
includes	O
a	O
thumbnail	O
view	O
of	O
the	O
entire	O
lymph	O
node	O
,	O
counts	O
and	O
Type	O
I	O
error	O
rates	O
for	O
all	O
phenotypes	O
,	O
as	O
well	O
as	O
a	O
transcript	O
of	O
the	O
analyses	O
performed	O
.	O
</s>
