HOTARU: Automatic sorting system for large scale calcium imaging data

Currently, calcium imaging allows long-term recording of large-scale neuronal activity in diverse states. However, it remains difficult to extract neuronal dynamics from recorded imaging data. In this study, we propose an improved constrained nonnegative matrix factorization (CNMF)-based algorithm and an effective method to extract cell shapes with fewer false positives and false negatives through image processing. We also show that the evaluation metrics obtained during image and signal processing can be combined and used for false-positive cell determination. For the CNMF algorithm, we combined cell-by-cell regularization and baseline shrinkage estimation, which greatly improved its stability and robustness. We applied these methods to real data and confirmed their effectiveness. Our method is simpler and faster, detects more cells with lower firing rates and signal-to-noise ratios, and enhances the quality of the extracted cell signals. These advances can improve the standard of downstream analysis and contribute to progress in neuroscience.