On-line SEUDO for real-time cell recognition in Calcium Imaging

Closed-loop neuroscience experimentation, where recorded neural activity is used to modify the experiment on-the-fly, is critical for deducing causal connections and optimizing experimental time. A critical step in creating a closed-loop experiment is real-time inference of neural activity from streaming recordings. One challenging modality for real-time processing is calcium imaging (CI) via multi-photon microscopy. CI enables the recording of activity in large populations of neurons however, often requires batch processing of the video data to extract single-neuron activity from the fluorescence videos. Extremely few methods attempt real-time CI processing, and are often susceptible to spurious noise in the image statistics and difficulties in separating overlapping components. In this work we use the recently proposed robust estimator---Sparse Emulation of Unused Dictionary Objects (SEUDO) algorithm---as a basis for a new on-line frame-by-frame processing algorithm that simultaneously identifies neurons in the fluorescence video and infers their time traces in a way that is robust to as-yet unidentified neurons.
Real-time SEUDO (real-SEUDO) initializes the set of known fluorescing components in the video as empty, and iteratively identifies new components (i.e., Regions Of Interest, ROIs) from the streaming CI videos by running the SEUDO and analyzing the residual activity. SEUDO returns the robust estimate of each neuron's activity, providing on-line updates as to the individual component trim-traces. To ensure that real-SEUDO is capable of real-time updates, the processing time of each frame must be less than the sampling period. To achieve this goal, we developed fast C-based implementations of all the algorithms that can run SUEDO and the cell detection stage at >30Hz. We find that the overall quality of real-SEUDO matches offline algorithms (e.g., CNMF), and outperformes the current on-line approach (OnACID).