Neurons identification of single-photon wide-field calcium fluorescent imaging data

Abstract

Tracking neurons by analyzing their calcium imaging data has enabled biological scientists to better understand the structure and working principle of the nervous system. Several algorithms have been proposed for neurons identification, but most of them become less effective when processing data recorded by single-photon wide-field fluorescence microscopes due to low signal-to-noise ratio (SNR). Moreover, defocus blur, which is common in in vivo imaging, and interference of other biological structures near the neurons have brought greater challenges. In the face of these issues, we have presented an improved method based on the extended constrained nonnegative matrix factorization (CNMF-E) framework to better identify the spatial locations and temporal activities of the neurons. To obtain more appropriate spatial components, we have introduced regularizations into the optimization problem and applied more morphological processing. For more precise temporal components, we have performed a piecewise baseline adjustment on the neurons’ fluorescence traces and suppressed the overestimated signals caused by the estimation error of background fluctuations. Our approach has been tested on the mouse brain cortex recorded by the Real-time, Ultra-large-Scale imaging at High-resolution (RUSH) macroscope. Due to the lack of existing datasets similar to the current imaging conditions, we have manually labeled some neurons and compared the results qualitatively, which show that our method has identified the neurons more accurately compared with the original CNMF-E method.