Classification of Hippocampal Ripples: Convolutional Neural Network Learns Episode-Specific Changes.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Memory Research

Background:

• The hippocampus is crucial for memory, processing spatiotemporal details of episodic experiences.
• Previous research identified episode-dependent diversity in hippocampal CA1 neuron firing patterns (ripple firings).
• The waveform diversity of ripple firings may vary with the specific type of episode experienced.

Purpose of the Study:

• To test the hypothesis that ripple firing waveform diversity is dependent on the type of episode experienced.
• To develop a method for classifying ripple firings into categories corresponding to different episodes.
• To identify specific ripple waveform features that are representative of each episode category.

Main Methods:

• Constructed a convolutional neural network (CNN) to classify ripple firings from freely moving rats.
• Trained the CNN to categorize ripple firings into five classes: four distinct episodes and a pre-episode baseline.
• Applied Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize CNN's focus areas within ripple waveforms.
• Utilized t-distributed Stochastic Neighbor Embedding (t-SNE) to map ripple waveforms into a 2D feature space for analysis.

Main Results:

• The developed CNN successfully classified hippocampal ripple firings into the five defined categories.
• Grad-CAM analysis highlighted specific partial ripple waveforms that the CNN utilized for classification.
• Analysis of Grad-CAM-identified waveforms in the t-SNE space suggested these partial waveforms are category-representative.

Conclusions:

• Convolutional neural networks can effectively classify hippocampal ripple firings based on experienced episodes.
• Specific partial ripple waveforms identified by Grad-CAM may serve as neural correlates for different episodic experiences.
• This approach offers a potential method for decoding neural representations of memory episodes.