A Deep Learning Approach for the Classification of Neuronal Cell Types

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning in Biology

Background:

• Traditional neuron classification from extracellular recordings relies on spike shape and firing patterns, primarily distinguishing excitatory and inhibitory neurons.
• Existing methods face limitations in robustness and the ability to identify specific neuronal subtypes.
• High-density Multi-Electrode Arrays (MEAs) offer rich spatial information that can potentially enhance classification accuracy.

Purpose of the Study:

• To develop a more robust method for classifying excitatory and inhibitory neurons using spatial information from high-density MEAs.
• To investigate the potential of classifying subtypes of excitatory and inhibitory neurons.
• To leverage machine learning, specifically Convolutional Neural Networks (CNNs), for neural classification.

Main Methods:

• Generation of a large simulated dataset of action potentials from detailed neural models.
• Extraction of spike features from simulated recordings using a high-density MEA model.
• Application of Convolutional Neural Networks (CNNs) for classifying simulated neuronal cell types.

Main Results:

• The forward modeling and machine learning approach achieved high accuracy (>= 92.15%) in distinguishing excitatory and inhibitory neuron spikes.
• The CNN-based method demonstrated capability in classifying different neuronal subtypes to a certain extent.
• Simulated results indicate robustness and potential for real-world application.

Conclusions:

• The developed approach offers a robust alternative for classifying neural cell types from extracellular recordings.
• Increased detail in neural models and availability of high-density recordings will further enhance the viability of this method.
• This technique holds promise for advancing our understanding of neural circuits by enabling finer classification of neuronal populations.