Calcium imaging-based brain-computer interface in freely behaving mice

Brain-computer interfaces (BCIs) are a powerful tool in both clinical applications and basic science research. Traditionally, BCIs rely on electrical signals collected by electrode arrays, which usually suffer from instability due to electrode movements and cell loss [1]. Recent approaches to address this issue continuously update BCI decoders [2]. By reliably identifying the same cells over days, using $Ca^{+2}$ imaging as BCI input may circumvent these limitations, and enable simultaneous recording from large neuronal populations over extended periods. Nonetheless, $Ca^{+2}$ imaging data typically requires offline processing to extract single cells’ activity [3]. Emerging real-time processing techniques offer novel BCI implementations with short-latency closed-loop feedback [4-5], yet few have demonstrated their techniques using in vivo imaging. Here we present a novel $Ca^{+2}$ imaging-based BCI we developed using miniaturized fluorescence microscopes for freely behaving mice. Our BCI enables the activation of external hardware based on real-time detection of specified activity patterns. Our system is composed of several hardware and software components, allowing customization according to experimental needs (Fig 1, A-B). The hardware components include a commercial head-mounted miniature fluorescence microscope (Inscopix, California), an overhead camera for tracking mouse behavior (The Imaging Source, Germany), and an Arduino-controlled system for feedback (e.g., reward administration) within the behavioral arena. The software components consist of Python code for online processing of neuronal recordings, field of view (FOV) alignment (Fig 1, A), extraction of fluorescence traces (Fig 1, C), detection of single-cells' $Ca^{+2}$ events (Fig 1, D), detection of neuronal activity patterns, and live tracking of the mouse’s position. Unlike classical BCIs, our system allows longitudinal readout of neuronal activity from tens to hundreds of the same neurons in freely behaving mice. With this system, we trained both head-fixed and freely behaving mice (Fig 1, E) to attain water rewards upon activation of the same neuronal pattern from hippocampal CA1. Under both conditions, mice maintained a stable level of reward throughout multiple recording days (Fig 1, F-H). Our results established the ability to activate a BCI with the long-term dynamics of neural codes.