Fast Calcium Trace Extraction for Large-Field-of-View Miniscope

Recent advancement in miniaturized calcium imaging microscope enables monitoring the cell activity from hundreds of neurons simultaneously in vivo. However, extracting calcium traces from a large population of cells in real time under a strict resource and energy constraint remains a challenge. To overcome it, we design a customized accelerator on a low-power FPGA for fast calcium trace extraction. This enables us to extract calcium traces from hundreds of cells in real time with a short and deterministic latency. In addition, we propose a series of techniques, including the region segmentation, the double buffering and the fast forward mechanisms, to further reduce the latency with minimal overhead. Applying these techniques, our implementation can achieve real-time calcium trace extraction for a maximum of 1024 cells from a 512×512 calcium video with sub-ms latency, which is promising in support of closed loop neurofeedhack applications.