Explaining Deep Learning-Based Representations of Resting State Functional Connectivity Data: Focusing on...

Area of Science:

• Neuroscience
• Machine Learning
• Psychiatric Disorders

Background:

• Resting state functional Magnetic Resonance Imaging (rs-fMRI) provides insights into brain organization in psychiatric disorders.
• High dimensionality of rs-fMRI data necessitates dimensionality reduction for machine learning.
• Variational Autoencoders (VAEs) extract low-dimensional latent representations from rs-fMRI, but interpretation remains challenging.

Conclusions:

• Introduced latent contribution scores to interpret nonlinear patterns identified by VAEs.
• Scores capture changes in rsFC features relative to latent representations.
• Enables explainable deep learning for understanding ASD neural mechanisms.