Dissecting neuronal circuit function and dysfunction in Alzheimer's disease mouse models: from conventional calcium imaging analysis to machine learning-based approaches.

In vivo calcium imaging is a powerful technique for the monitoring activity of large neuronal populations in the intact brain. Two-photon microscopy provides subcellular resolution in head-fixed animals, and miniaturized fluorescence microscopy (miniscope) enables recordings of neuronal network activity in freely behaving animals. Combined with genetically encoded calcium indicators, these methods have revealed important new information about the functioning of neuronal circuits and enabled investigation of neuronal network dysfunction in mouse models of Alzheimer's disease. Typically, such imaging data are analyzed using event-based and statistical approaches, which have been highly informative but fail to capture the higher-order spatiotemporal structure of circuit activity. Here we propose that dimensionality reduction and neuronal manifold analysis, machine learning, and artificial intelligence (AI)-based methods provide powerful data-driven approaches, enabling feature extraction, detection of disease-associated network states, and simultaneous analysis of neural network activity and behavior. Crucially, beyond describing pathology, these approaches could offer a more sensitive and complementary readout for screening candidate therapeutic strategies in neurological disease models. Here, we summarize neuronal activity alterations identified across Alzheimer's disease mouse models using classical and AI-based calcium imaging analyses, and outline future perspectives for AI-driven approaches in this field, from interpretable architectures to multimodal integration of activity and behavior.