DeepCINAC: A Deep-Learning-Based Python Toolbox for Inferring Calcium Imaging Neuronal Activity Based on Movie...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Developmental Neuroscience

Background:

• Two-photon calcium imaging is crucial for studying neuronal dynamics via indicator fluorescence.
• Current computational tools struggle with detecting fluorescence transients from highly synchronous neurons in dense regions like the developing CA1 hippocampus.
• Existing analytical tools lack sufficient benchmark measurements for validation.

Purpose of the Study:

• To develop advanced computational tools for reliable calcium transient detection in densely packed, synchronous neurons during early development.
• To create a benchmark for evaluating the performance of new analytical methods in calcium imaging.
• To introduce DeepCINAC, an open-source toolbox for processing and evaluating calcium imaging data.

Main Methods:

• Developed a graphical user interface (GUI) for precise manual detection of calcium transients.
• Utilized a convolutional neural network (CNN) with attention and a bidirectional long-short term memory (LSTM) network for automated analysis.
• Compared the performance of the CNN-LSTM method against CaImAn using F1 scores.

Main Results:

• The CNN-LSTM method achieved human-level performance in detecting calcium transients.
• DeepCINAC demonstrated a superior F1 score compared to CaImAn for inferring neural activity in the developing CA1.
• The method successfully identified activity originating from GABAergic neurons automatically.

Conclusions:

• DeepCINAC provides a robust, user-friendly, and open-source solution for analyzing complex calcium imaging data.
• The developed toolbox enables accurate inference of neural activity, particularly in challenging datasets from developing brain regions.
• DeepCINAC facilitates performance evaluation and offers automated identification of specific neuronal populations, such as GABAergic neurons.