Classification of A Pressing Movement by Background Removed Single- photon Calcium Image Sequences

Calcium imaging, with its capability to record hundreds or even thousands of neurons simultaneously, has been widely applied in the fields of neuroscience and neural decoding. Current neural decoding experiments utilizing calcium imaging depend on offline processing algorithms to extract neuronal signals from the acquired calcium imaging data for training the decoder. However, the computational speed of the offline algorithms typically limits the processing efficiency of this approach. While a few studies have employed two-photon calcium imaging data for neural decoding with some success, no comparable studies have been conducted on single-photon data, which exhibits more complex background fluorescence. In this study, we utilized single-photon calcium image sequences acquired during a mouse lever-pressing experiment, improved image quality through the application of a background removal algorithm, and achieved successful classification of mouse lever-pressing behavior using a 3D residual network for decoding. Our findings demonstrate that the utilization of the background removal algorithm yields higher decoding accuracy compared to using raw images. Our study offers an efficient workflow for decoding tasks relying on single-photon calcium images and possesses extensive applications in real-time closed-loop experiments, such as those involving brain-machine interfaces.