A Supervised Contrastive Learning-based Analysis of rs-tMRI Data Captures Gender Differences in Nonlinear Functional...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Network Analysis

Background:

• Functional connectivity (FC) commonly uses linear correlation to interpret brain activity.
• Linear methods overlook nonlinear dependencies that may hold crucial information.
• Existing FC approaches may not fully capture complex inter-regional brain dynamics.

Purpose of the Study:

• To develop a deep learning approach for capturing nonlinear temporal relationships between brain networks.
• To identify sex-based differences in brain network interactions.
• To offer a complementary interpretation of functional brain activity beyond linear FC.

Main Methods:

• Utilized a deep learning framework comprising an encoder and a similarity metric.
• Employed independent component analysis (ICA) to estimate time courses.
• Learned domain-specific embeddings to measure nonlinear functional relationships between networks.

Main Results:

• Male subjects showed stronger nonlinear network-network relationships than female subjects.
• The deep learning approach captured intra-network relationships, particularly between cognitive control and visual networks.
• Significant sex differences were observed in these intra-network relationships.

Conclusions:

• The deep learning method effectively captures nonlinear brain network interactions.
• Nonlinear functional relationships differ between sexes, with males exhibiting stronger connections.
• This approach provides complementary insights to traditional FC, highlighting sex-specific network dynamics.