GRAPHICS PROCESSING UNIT ACCELERATED MEDICAL IMAGING

spongemintSoftware and s/w Development

Dec 2, 2013 (3 years and 8 months ago)

82 views

GRAPHICS PROCESSING UNIT
ACCELERATED

MEDICAL IMAGING



Sam Van der Jeught

University of Antwerp

Belgium


New Challenges in the European
Area: Young Scientist's 1st
International Baku
Forum

20
-
25
th

may 2013


1

Graphics processing units (GPU) as
low
-
cost supercomputers

GT9800, $40USD on
ebay








Multicore architecture GPU VS single (or
double
-

or quad
-
) core CPU








Parallel programming!

2

Some applications: real
-
time
geometric lens distortion
correction




Go from input A to
output B










Wide
-
angle lens
systems suffer from
barrel distortion










1024x1024p @30 fps









3

Some applications: real
-
time
geometric lens distortion
correction

Apply general recalibration software to distorted image.

Coordinates now have floating point values

Scattered data interpolation resamples pixels onto integer
grid.

Highly time consuming!

Regular 2D interpolation on incoming images

One
-
time scattered data interpolation on

integer grid of desired size

+

4

Some applications: real
-
time
geometric lens distortion
correction


After calibration, our
algorithm also works
on random distortion
(not only barrel)






(+ video)








5

Some applications: real
-
time
geometric lens distortion
correction

6

Some applications: optical
coherence tomography


I Ophthalmology








II
Detect signature
forgeries and hidden
layers in paintings








Interferometric
imaging technique
using infrared light

(non
-
invasive, non
-

destructive, non
-
contact)








7

Replace expensive FPGA DAQ
board with low
-
cost GPU


Data processing
boards can cost up
to >$1000 USD











Real
-
time data
processing can be
achieved with $40
USD GPU










Commercial leaflet “
Santec

inc.











8

Example real
-
time OCT


Real
-
time beating
ant heart at 25 fps
versus 8 fps on CPU







(+video)










9

Example real
-
time OCT

10


Real
-
time GPU
-
based data processing



Considerably faster than CPU



Low
-
cost alternative to data processing boards such
as FPGA’s




Conclusion

11

Thank you

Questions?