Biomimetic Visual Information Spatiotemporal Encoding Method for In Vitro Biological Neural Networks

Area of Science:

• Neuroscience
• Robotics
• Biomimetic Engineering

Background:

• Biological neural networks (BNNs) integrated with robotic systems show promise for information processing and adaptive learning.
• Current BNN-robotic systems struggle with visual perception due to inefficient sensory encoding.
• A need exists for advanced methods to bridge visual input with BNN processing capabilities.

Purpose of the Study:

• To propose and validate a biomimetic visual information spatiotemporal encoding method for BNNs.
• To enable BNNs to perform image recognition tasks by transforming visual data into neural stimuli.
• To assess the impact of the encoding method on BNN information processing and functional connectivity.

Main Methods:

• Developed an improved delayed phase encoding method to convert high-dimensional images into pulse sequences.
• Utilized convolution, temporal delay, alignment, and compression for stimulus preparation.
• Conducted three stages of unsupervised training on in vitro BNNs using high-density microelectrode arrays (HD-MEAs).
• Decoded neural activity using a logistic regression model to evaluate image recognition performance.

Main Results:

• The proposed encoding method generated separable firing patterns in BNNs for different spatiotemporal stimuli.
• Image recognition accuracy reached 80.33% ± 7.94% after three training stages, a 13.64% improvement over the first stage.
• Unsupervised training led to significant increases in BNN connection number, connection strength, and inter-module participation coefficient.

Conclusions:

• The developed biomimetic encoding method effectively enables BNNs to process visual information for image recognition.
• The method enhances functional connectivity and cross-module information exchange within BNNs.
• This approach represents a significant advancement in integrating biological neural computation with robotic perception.