Multimodality imaging
FPGA-accelerated
computing
3D Ultrasound imaging
Real-time tumor tracking
Image segmentation
/ shape reconstruction
Visualization
Brain
PET differentiates viable tumors from necrotic masses more accurately
than CT and MRI. CT and MRI depict anatomical information more precisely than
PET. We fuse CT, MRI and PET with the goal of improving gamma-knife surgery
and other radiotherapies/radiosurgeries.
Heart
SPECT and CT images of the heart provide complementary perfusion and structural
information. Cardiac images are 4D (3D space + time), and we developed algorithms
to fuse following both temporal and spatial registration.
click for movie
Body
PET and CT are truly complementary, so much so that medical imaging vendors
manufacture combined PET/CT scanners. We can elastically register body PET and
CT, whether from stand-alone or from combined scanners.
Image registration
Automatic image registration usually is slow. FAIR
and FAIR-II architectures (FAIR stands for fast automatic image registration)
we have developed are allow performing rigid and elastic image registration
in real time. A custom board that implemented FAIR architecture is shown below.
CT Reconstruction
CT image reconstruction speed remains a problem. A reconfigurable system-on-chip
architecture with parallel data-driven deep pipeline execution units was designed
for large-scale parallel computations. This accelerated FDK cone-beam CT reconstruction
by a factor of 60. The architecture can also be adapted to other 3D imaging
applications.
Preprocessing
3D ultrasound is currently the only real-time 3D medical imaging modality
available. Because of its characteristic high level of speckle noise, it is
necessary to perform preprocessing at acquisition rates. We have developed FPGA-based
architectures for real-time preprocessing of 3D ultrasound using median filtering
or anisotropic diffusion filtering.
Coming soon!
Coming soon!
Image segmentation / Shape reconstruction
Medical image segmentation is the process of outlining organ boundaries in medical images, and is useful for organ shape reconstruction, visualization and clinical decision making.
Click for movie
One way to visualize large 3D and 4D (3D shape + time) data sets on desktop PCs is to use the GPU in addition to the CPU. Click below to see an interactive exploration of 4D cardiac ultrasound image using GPU-based computing.
Click for movie