TSP-GNN: a novel neuropsychiatric disorder classification framework based on task-specific prior knowledge and graph...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Neuropsychiatric disorders (ND) exhibit altered functional connectivity (FC) during specific cognitive tasks.
• Existing deep learning models often overlook task-specific FC patterns, potentially limiting classification accuracy.
• Understanding brain-behavior relationships requires analyzing neural activity within task contexts.

Purpose of the Study:

• To develop a novel deep learning framework, TSP-GNN, for classifying neuropsychiatric disorders.
• To incorporate prior knowledge of task-specific connectome patterns into graph neural network models.
• To enhance the accuracy of disease classification by focusing on task-relevant brain connectivity.

Main Methods:

• Developed TSP-GNN, a graph neural network (GNN) model.
• Extracted task-specific prior (TSP) connectome patterns from functional connectivity data.
• Validated the model using publicly available datasets of neuropsychiatric disorders.

Main Results:

• Different types of neuropsychiatric disorders display distinct task-specific functional connectivity patterns.
• Utilizing task-specific nodes in TSP-GNN improved classification accuracy compared to whole-brain approaches.
• The study demonstrates the efficacy of integrating prior task-specific connectome information.

Conclusions:

• TSP-GNN represents a novel approach combining task-specific connectome priors with deep learning for ND classification.
• The findings highlight the importance of task context in understanding brain dysfunction in neuropsychiatric diseases.
• This research offers valuable insights into the cognitive mechanisms underlying neuropsychiatric disorders.