Limits...
GPU-accelerated framework for intracoronary optical coherence tomography imaging at the push of a button.

Han M, Kim K, Jang SJ, Cho HS, Bouma BE, Oh WY, Ryu S - PLoS ONE (2015)

Bottom Line: To help more accurate diagnosis and monitoring of the disease, many researchers have recently worked on visualization of various coronary microscopic features including stent struts by constructing three-dimensional (3D) volumetric rendering from series of cross-sectional intracoronary FD-OCT images.In this paper, we present the first, to our knowledge, "push-of-a-button" graphics processing unit (GPU)-accelerated framework for intracoronary OCT imaging.Our framework visualizes 3D microstructures of the vessel wall with stent struts from raw binary OCT data acquired by the system digitizer as one seamless process.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea.

ABSTRACT
Frequency domain optical coherence tomography (FD-OCT) has become one of the important clinical tools for intracoronary imaging to diagnose and monitor coronary artery disease, which has been one of the leading causes of death. To help more accurate diagnosis and monitoring of the disease, many researchers have recently worked on visualization of various coronary microscopic features including stent struts by constructing three-dimensional (3D) volumetric rendering from series of cross-sectional intracoronary FD-OCT images. In this paper, we present the first, to our knowledge, "push-of-a-button" graphics processing unit (GPU)-accelerated framework for intracoronary OCT imaging. Our framework visualizes 3D microstructures of the vessel wall with stent struts from raw binary OCT data acquired by the system digitizer as one seamless process. The framework reports the state-of-the-art performance; from raw OCT data, it takes 4.7 seconds to provide 3D visualization of a 5-cm-long coronary artery (of size 1600 samples x 1024 A-lines x 260 frames) with stent struts and detection of malapposition automatically at the single push of a button.

No MeSH data available.


Related in: MedlinePlus

CUDA profiler timeline for FD-OCT image reconstruction.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4400174&req=5

pone.0124192.g006: CUDA profiler timeline for FD-OCT image reconstruction.

Mentions: We parallelize the data transfers between the host and the device and kernel executions [20]. Fig 6 illustrates that the data transfers from the host to the device are asynchronous (Memcpy HtoD [async]), which allows the kernel functions to execute concurrently with the data transfers. Using two GTX 680 GPUs, our algorithm processes 1.5 MA-lines in every second where we use 2048 A-lines (2 polarization channels×1024 A-lines) of raw binary data to reconstruct one FD-OCT image. Our algorithm is faster than the best GPU accelerated FD-OCT image reconstruction algorithms [19, 20]. In Zhang and Kang [19], the full-range 1024-pixel FD-OCT is the closest one to our FD-OCT system, and the LIFFT-D (linear spline interpolation with FFT and numerical dispersion compensation) algorithm is the closest one to our algorithm. Because we need both directions of data transfers, we should compare the 2-way limited speed by PCI-E. Then, their algorithm reports 672 KA-lines/second while ours reports 809.5 KA-lines/second. To compare with Jian et al.’s algorithm with 2.24 MA-lines/second [20], we revised our algorithm to execute similarly to theirs like performing FFT only once, and our revised algorithm reports 2.27 MA-lines/second.


GPU-accelerated framework for intracoronary optical coherence tomography imaging at the push of a button.

Han M, Kim K, Jang SJ, Cho HS, Bouma BE, Oh WY, Ryu S - PLoS ONE (2015)

CUDA profiler timeline for FD-OCT image reconstruction.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4400174&req=5

pone.0124192.g006: CUDA profiler timeline for FD-OCT image reconstruction.
Mentions: We parallelize the data transfers between the host and the device and kernel executions [20]. Fig 6 illustrates that the data transfers from the host to the device are asynchronous (Memcpy HtoD [async]), which allows the kernel functions to execute concurrently with the data transfers. Using two GTX 680 GPUs, our algorithm processes 1.5 MA-lines in every second where we use 2048 A-lines (2 polarization channels×1024 A-lines) of raw binary data to reconstruct one FD-OCT image. Our algorithm is faster than the best GPU accelerated FD-OCT image reconstruction algorithms [19, 20]. In Zhang and Kang [19], the full-range 1024-pixel FD-OCT is the closest one to our FD-OCT system, and the LIFFT-D (linear spline interpolation with FFT and numerical dispersion compensation) algorithm is the closest one to our algorithm. Because we need both directions of data transfers, we should compare the 2-way limited speed by PCI-E. Then, their algorithm reports 672 KA-lines/second while ours reports 809.5 KA-lines/second. To compare with Jian et al.’s algorithm with 2.24 MA-lines/second [20], we revised our algorithm to execute similarly to theirs like performing FFT only once, and our revised algorithm reports 2.27 MA-lines/second.

Bottom Line: To help more accurate diagnosis and monitoring of the disease, many researchers have recently worked on visualization of various coronary microscopic features including stent struts by constructing three-dimensional (3D) volumetric rendering from series of cross-sectional intracoronary FD-OCT images.In this paper, we present the first, to our knowledge, "push-of-a-button" graphics processing unit (GPU)-accelerated framework for intracoronary OCT imaging.Our framework visualizes 3D microstructures of the vessel wall with stent struts from raw binary OCT data acquired by the system digitizer as one seamless process.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea.

ABSTRACT
Frequency domain optical coherence tomography (FD-OCT) has become one of the important clinical tools for intracoronary imaging to diagnose and monitor coronary artery disease, which has been one of the leading causes of death. To help more accurate diagnosis and monitoring of the disease, many researchers have recently worked on visualization of various coronary microscopic features including stent struts by constructing three-dimensional (3D) volumetric rendering from series of cross-sectional intracoronary FD-OCT images. In this paper, we present the first, to our knowledge, "push-of-a-button" graphics processing unit (GPU)-accelerated framework for intracoronary OCT imaging. Our framework visualizes 3D microstructures of the vessel wall with stent struts from raw binary OCT data acquired by the system digitizer as one seamless process. The framework reports the state-of-the-art performance; from raw OCT data, it takes 4.7 seconds to provide 3D visualization of a 5-cm-long coronary artery (of size 1600 samples x 1024 A-lines x 260 frames) with stent struts and detection of malapposition automatically at the single push of a button.

No MeSH data available.


Related in: MedlinePlus