Classifying Calcium Imaging Dynamics with Deep Learning: Multi-Frequency Analysis through Quantile-Based Time-Series Network Representations

To address the limitations of calcium imaging data, we propose a segmentation-agnostic deep learning framework that integrates Quantile-Based Time-Series Network (QTN) representations with convolutional neural networks to classify neuronal dynamics across multiple spatial resolutions and acquisition frequencies. By transforming fluorescence traces into compact, fixed-size matrices derived from quantile transitions, the method standardizes inputs across recordings while markedly reducing dimensionality and computational cost. Several QTN variants were systematically evaluated, demonstrating strong and consistent classification performance across both whole-image and grid-based preprocessing strategies. Notably, the framework maintained high accuracy under reduced temporal resolution and controlled noise perturbations, confirming that discrimination arises from meaningful temporal patterns rather than artifacts. This study establishes a robust, scalable, and generalizable approach for analyzing calcium imaging dynamics, paving the way for efficient, segmentation-independent characterization of neuronal activity in pharmacological and systems neuroscience applications.