IdenBAT: Disentangled representation learning for identity-preserved brain age transformation.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computer Vision

Background:

• Brain age transformation aims to modify brain images to reflect target age-specific features.
• Preserving age-irrelevant attributes during transformation is challenging due to entangled image features.

Purpose of the Study:

• To propose a novel architecture, IdenBAT, for identity-preserved brain age transformation.
• To disentangle image features for selective age-related attribute modification.

Main Methods:

• Developed a novel architecture employing disentangled representation learning (IdenBAT).
• Decomposes image features to preserve individual traits while transforming age characteristics.
• Validated on 2D and 3D brain datasets.

Main Results:

• IdenBAT accurately converts brain images to target ages while preserving individual characteristics.
• The method demonstrates superior performance fidelity compared to state-of-the-art approaches.
• Successful application on both 2D and 3D neuroimaging data.

Conclusions:

• IdenBAT effectively achieves identity-preserved brain age transformation.
• The disentangled representation learning approach overcomes feature entanglement challenges.
• IdenBAT offers a promising solution for realistic brain age manipulation in neuroimaging research.