67
OCTOBER 16-18, 2011 • PORTLAND, OR
P31
Automated Classifcation of Leukemia Using
Cytometric Fingerprinting
Richard D. Schretzenmair, Nicholas Kurtzman, Neisan
Sabet, Jonni S. Moore, Wade T. Rogers
Dept. Pathology & Laboratory Medicine, University of
Pennsylvania, Philadelphia, PA, USA
Introduction
:Flow cytometry is a critical tool in the
primary diagnosis of hematologic malignancies. In
order to better discriminate subtle differences in
malignancies,panels are evolving towards more
parameters in fewer tubes to simplify data acquisition
and elucidate marker correlations that are diffcult
to discern otherwise. However, higher dimensional
measurement increases the complexity of data
analysis, lengthening the time to results. We aim to
develop computational methods, amenable to high
throughput technologist independent analysis, that
mine such high-dimensional datasets for patterns
that discriminate hematologic malignancies
Materials
and Methods
: Patient samples,(blood and/or bone
marrow) sent for routine diagnosis to the Clinical
FlowCytometry laboratory, were prepared according
to standard procedures. Routine clinical panels were
run and diagnoses confrmed by hematopathologists.
Concurrently, cells were stained with a second single
tube panel consisting of CD3, CD10, CD13, CD15,
CD19,CD33, CD34, CD45, Hla-Dr and DAPI for
viability and data was acquired on a Beckman Coulter
Gallios fow cytometer. Data analysis was carried
out with the R/Bioconductor package fowFP. After
automated gating to exclude debris and non-viable
cells,fngerprints were computed for each sample.
Fingerprint-based classifcation of AML, ALL, MDS
and normal blood/marrow was evaluated using leave-
one-out cross validation by measuring a distance
metric of a test sample fngerprint to models of each
class constructed from all fngerprints excluding
the test sample.
Results
: Accurate classifcation
(88%sensitivity, n=24) of AML vs. ALL and Normal
was achieved. Classifcation of ALL bone marrow
samples was perfect but only involved limited
numbers (N=2). The authors are grateful to Beckman
Coulter Inc. for support.
POSTER ABSTRACTS