Adversarial brain multiplex prediction from a single brain network with application to gender fingerprinting

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Brain connectivity networks from MRI reveal functional and structural relationships between brain regions.
• Previous studies on gender differences in brain connectivity focused on limited pairwise relationships, neglecting complex high-order network interactions.
• Existing brain multiplex models require multiple networks, limiting their use with single-network connectomic data.

Purpose of the Study:

• To introduce the Adversarial Brain Multiplex Translator (ABMT), the first framework for predicting brain multiplexes from a single source network.
• To investigate gender differences in the human brain by analyzing high-order brain connectivity.
• To improve the understanding of complex brain network interactions and their relation to gender.

Main Methods:

• Developed ABMT using geometric adversarial learning, comprising a U-Net-like translator, a conditional discriminator, and a multi-layer perceptron classifier.
• The framework predicts target multiplexes from a source network, utilizing skip connections and adversarial training.
• Gender classification accuracy was evaluated using predicted multiplexes against source networks.

Main Results:

• Predicted multiplexes significantly improved gender classification accuracy compared to using only source networks.
• The ABMT framework successfully identified both low-order and high-order gender-specific brain multiplex connections.
• Demonstrated the capability of predicting complex brain multiplexes from single brain networks for enhanced analysis.

Conclusions:

• ABMT is a novel approach for analyzing high-order brain connectivity and gender differences using single brain networks.
• The framework offers a powerful tool for uncovering complex, gender-specific patterns in brain networks.
• Future research can leverage ABMT for deeper insights into neurological and psychological variations across genders.