Deep generation of personalized connectomes based on individual attributes

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Individual brain connectomes exhibit unique architectures.
• Variations in connectome are linked to health, cognition, and lifestyle.
• Predicting attributes from connectomes is common, but inferring connectomes from attributes is less explored.

Purpose of the Study:

• To develop and validate a deep model for generating connectome architecture from personal attributes.
• To investigate the influence of various personal factors on connectome generation.
• To explore the utility of generated connectomes in machine learning applications.

Main Methods:

• Utilized a deep learning model to generate connectomes based on age, sex, body phenotypes, cognition, and lifestyle.
• Trained and validated the model on the UK Biobank connectome cohort (N=8,086).
• Employed diffusion MRI and tractography for empirical connectome mapping.

Main Results:

• The model successfully generated connectome architectures that closely matched empirically mapped connectomes.
• Age, sex, and body phenotypes were found to be the most influential factors (approx. 4x more than cognition/lifestyle).
• Cognition showed regionally specific importance, particularly in the association cortex.

Conclusions:

• It is feasible to infer brain connectivity from an individual's personal data.
• Generated connectomes can enhance machine learning model training and reduce prediction errors.
• This approach enables applications like data augmentation and anonymous data sharing in neuroscience research.