Stars2Cells: Astrometric Tracking of Neurons Across Imaging Sessions

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Chronic calcium imaging is vital for understanding neural plasticity and learning over time.
• Accurate identification of individual neurons across multiple recording sessions is a prerequisite for longitudinal studies.
• Existing neuron registration tools struggle with repeated recording sessions, limiting longitudinal research.

Purpose of the Study:

• To introduce Stars2Cells (S2C), a novel computational pipeline for robust longitudinal neuron tracking in chronic calcium imaging.
• To overcome the limitations of current registration methods that degrade under repeated recording sessions.
• To enable high-resolution analysis of neural population dynamics and individual neuron stability over time.

Main Methods:

• Developed Stars2Cells (S2C), a pipeline inspired by astrometric plate-solving.
• Represented neuron local geometry using four-dimensional quad descriptors invariant to rotation, translation, and scaling.
• Employed descriptor-space matching, Random Sample Consensus (RANSAC) verification, and Hungarian assignment for neuron matching.
• Validated S2C on a synthetic benchmark of 1,262 paired runs with varying neuron counts and perturbation conditions.

Main Results:

• S2C achieved a pooled F1 score of 98.4% on a synthetic benchmark, significantly outperforming standard ROI-based matching (36.0%).
• Applied to dorsomedial striatum (DMS) imaging during fentanyl self-administration, S2C revealed near-complete single-neuron turnover masked by conserved population activity.
• Demonstrated that S2C can uncover representational drift invisible to bulk photometry, requiring same-cell tracking.

Conclusions:

• Stars2Cells (S2C) provides a highly accurate and robust solution for longitudinal neuron tracking in chronic calcium imaging.
• S2C enables the study of neural plasticity and representational drift at the single-neuron level.
• The pipeline is user-friendly, distributed as a GUI-driven standalone application for macOS and Windows, requiring no specialized setup.