Wei-Min Tseng1, Teng-Yi Huang2,
Yi-Ru Lin3, Ming-Ting Wu4
1Department of Electrical
Engineering, National Taiwan
University of Science & Technology, Taipei, Taiwan; 2Department
of Electrical Engineering, National Taiwan University of Science &
Technology, Taipei, Taiwan; 3Department of Electronic Engineering,
National Taiwan University of Science & Technology, Taipei, Taiwan; 4Department
of Radiology, Kaohsiung Veterans General Hospital, Kao-Hsiung, Taiwan
This study attempts to accelerate pixel-by-pixel non-linear fitting with modern parallel computation on graphic processing units. Levenberg-Marquardt algorithm was implemented into GPU-based toolbox compatible to MATLAB environment. The time-intensity curve of each pixel is distributed to each thread of parallel computation. The four-GPU system can compute up to 960 curve-fitting in parallel. The toolbox was tested with a 7-slice pulmonary DCE perfusion dataset to reconstruct the pulmonary blood volume maps. The proposed GPU-fitting toolbox reduced the total computation time from ~20 minutes to ~0.5 minute.