A Computational Approach to Real-Time Image Processing for Serial Time-Encoded Amplified Microscopy
Minoru Oikawa,Daisuke Hiyama,Ryuji Hirayama,Satoki Hasegawa,Yutaka Endo,Takahisa Sugie,Norimichi Tsumura,Mai Kuroshima,Masanori Maki,Genki Okada,Cheng Lei,Yasuyuki Ozeki,Keisuke Goda,Tomoyoshi Shimobaba
DOI: https://doi.org/10.1117/12.2211646
2016-01-01
Abstract:High-speed imaging is an indispensable technique, particularly for identifying or analyzing fast-moving objects. The serial time-encoded amplified microscopy (STEAM) technique was proposed to enable us to capture images with a frame rate 1,000 times faster than using conventional methods such as CCD (charge -coupled device) cameras. The application of this high-speed STEAM imaging technique to a real-time system, such as flow cytometry for a cell-sorting system, requires successively processing a large number of captured images with high throughput in real time. We are now developing a high-speed flow cytometer system including a STEAM camera. In this paper, we describe our approach to processing these large amounts of image data in real time. We use an analog-to-digital converter that has up to 7.0G samples/s and 8 -bit resolution for capturing the output voltage signal that involves grayscale images from the STEAM camera. Therefore the direct data output from the STEAM camera generates 7.0G byte/s continuously. We provided a field -programmable gate array (FPGA) device as a digital signal pre-processor for image reconstruction and finding objects in a microfluidic channel with high data rates in real time. We also utilized graphics processing unit (GPU) devices for accelerating the calculation speed of identification of the reconstructed images. We built our prototype system, which including a STEAM camera, a FPGA device and a GPU device, and evaluated its performance in real-time identification of small particles (beads), as virtual biological cells, flowing through a microfluidic channel.