Kajal Singla, researcher at ScaDS.AI Dresden/Leipzig and Max Planck Institute for Human Cognitive and Brain Sciences presented her latest paper “Learning Latent Spaces for Individualized Functional Neuroimaging with Variational Autoencoders” at the CCN Conference in Amsterdam. Together with Nico Scherf (PI at ScaDS.AI Dresden/Leipzig) and Pierre-Louis Bazin, she introduced a novel deep learning approach that leverages variational autoencoders (VAEs) to model functional Magnetic Resonance Imaging data in subject-specific latent spaces.

Traditional methods (e.g., ICA, diffusion map embedding) capture group-level brain networks but often miss individual-specific differences. The VAE-based framework reconstructs and denoises fMRI data in a low-dimensional latent space, enhancing the separation of signals from distinct functional networks without directly aligning them to specific latent axes. These individualized latent spaces can also be aligned across subjects, enabling meaningful cross-subject comparisons. This approach not only enhances the signal-to-noise ratio but also opens new avenues for personalized fMRI analysis and a deeper understanding of the brain’s functional architecture.