Multi-task machine learning reveals the functional neuroanatomy fingerprint of mental processing

Area of Science:

• Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• Understanding the neural mechanisms of mental processing requires precise identification of functional neuroanatomy.
• A key question is whether specific mental processes have a unique 'functional neuroanatomy fingerprint'.
• Current methods struggle to reliably map these distinct neural signatures.

Purpose of the Study:

• To determine if distinct mental processes possess a unique functional neuroanatomy fingerprint.
• To investigate the neural substrates underlying seven representative mental processes.
• To establish a foundation for understanding the neural basis of cognition.

Main Methods:

• Utilized a multi-task deep learning model to analyze functional magnetic resonance imaging (fMRI) data.
• Examined data from 1235 subjects across two independent cohorts (US and China).
• Disentangled the functional neuroanatomy fingerprint for Emotion, Gambling, Language, Motor, Relational, Social, and Working Memory.

Main Results:

• Consistently demonstrated that each of the seven mental processes has a unique functional neuroanatomy fingerprint.
• This fingerprint is defined by specific whole-brain activity weights, indicating regional involvement.
• Achieved high classification accuracy (93%) and AUC (0.99) in distinguishing these fingerprints across datasets and atlases.

Conclusions:

• Each mental process is characterized by a distinct and reliable functional neuroanatomy fingerprint.
• This finding provides a robust neuroanatomical basis for studying the neural mechanisms of mental functions.
• The study validates the use of deep learning in uncovering complex brain-behavior relationships.