Enabling Optical Methods for Next-Generation Neural Prostheses

Optical methods present interesting new opportunities for brain computer interfaces (BCIs) and closed-loop experiments because of their capability to densely monitor and stimulate in-vivo large neural populations across weeks with single cell resolution. For instance, combining optical methods for recording (two-photon imaging of calcium indicators) and perturbing (optogenetics) neural ensembles opens the door to exciting closed-loop experiments, where the stimulation pattern can be determined based on the recorded activity and/or the behavioral state. However, the adoption of such tools for BCIs is currently hindered by the lack of algorithms that track neural activity in real-time. In a typical closed-loop experiment, the monitored/perturbed regions of interest (ROIs) have been preselected by analyzing offline a previous dataset from the same field of view. Monitoring the activity of a ROI, which usually corresponds to a soma, typically entails averaging the fluorescence over the corresponding ROI, resulting in a signal that is only a proxy for the actual neural activity and which can be sensitive to motion artifacts and drifts, as well as spatially overlapping sources, background/neuropil contamination, and noise. Furthermore, by preselecting the ROIs, the experimenter is unable to detect and incorporate new sources that become active later during the experiment or track changes in neuronal morphology, which prevents the execution of truly closed-loop experiments.

In the first portion of this talk I will present an Online, single-pass, algorithmic framework for the Analysis of Calcium Imaging Data (OnACID). The framework is highly scalable with minimal memory requirements, as it processes the data in a streaming fashion one frame at a time, while keeping in memory a set of low dimensional sufficient statistics and a small minibatch of the last data frames. Every frame is processed in four sequential steps: i) The frame is registered against the previous denoised (and registered) frame to correct for motion artifacts. ii) The fluorescence activity of the already detected sources is tracked. iii) Newly appearing neurons are detected and incorporated to the set of existing sources. iv) The fluorescence trace of each source is denoised and deconvolved to provide an estimate of the underlying spiking activity. I will present the results of applying OnACID to several large-scale (90-350GB) mouse and zebrafish larvae in-vivo datasets. OnAcid can find and track tens of thousands of neurons faster than real-time, and outperforms state of the art algorithms benchmarked on multiple manual annotations using a precision-recall framework. 

In the second portion of the talk, I will present an application of brain optical imaging to unveil coding properties and feedback mechanisms implemented by neurons in the cerebellum, a brain area implied in motor control and in the production of agile movement sequences. By monitoring across days the same neuronal populations of mice undergoing associative learning I will show that a predictive signal about the upcoming movement is widely available at the input stage of the cerebellar cortex, as required by forward models of cerebellar control.    

In the last section of the talk, I will discuss my plans to develop all-optical neural prostheses interfacing with the cerebellum to recover lost motor function in the central nervous system because of injury or disease.