Anomaly Detection Using Autoencoders for Miniscope Calcium Imaging Data Analysis

Deep learning methods are widely used for diverse data analysis tasks, with neuroscience emerging as a particularly promising application area. Machine learning techniques offer the capability to uncover patterns in data that classical statistical methods may overlook. In this study, we applied a machine learning method, an autoencoder designed for anomaly detection. It was applied to analyze data describing the cellular activity of neural circuits in the hippocampus of freely moving mice. Input data consisted of calcium traces obtained via miniature fluorescence microscopy (miniscope). These traces represent the temporal dependencies of fluorescence intensity from a genetically encoded calcium indicator, a proxy for neuronal activity. Presented artificial neural network effectively distinguished between normal conditions and those induced by acute stress modeling. These distinctions were supported by observing an increase in the calcium signal data reconstruction error value.