IdenBAT: Disentangled Representation Learning for Identity-Preserved Brain Age Transformation

TL;DR

This work proposes a novel architecture that employs disentangled representation learning for identity-preserved brain age transformation, called IdenBAT, which adeptly converts input images to target age while retaining individual characteristics accurately.

Abstract

Brain age transformation aims to convert reference brain images into synthesized images that accurately reflect the age-specific features of a target age group. The primary objective of this task is to modify only the age-related attributes of the reference image while preserving all other age-irrelevant attributes. However, achieving this goal poses substantial challenges due to the inherent entanglement of various image attributes within features extracted from a backbone encoder, resulting in simultaneous alterations during the image generation. To address this challenge, we propose a novel architecture that employs disentangled representation learning for identity-preserved brain age transformation called IdenBAT. This approach facilitates the decomposition of image features, ensuring the preservation of individual traits while selectively transforming age-related characteristics to match those of the target age group. Through comprehensive experiments conducted on both 2D and full-size 3D brain datasets, our method adeptly converts input images to target age while retaining individual characteristics accurately. Furthermore, our approach demonstrates superiority over existing state-of-the-art regarding performance fidelity.

Figures (8)

• Figure 1: Visualization of results between subjects' follow-up scans (Ground Truth) and brain age transformation studies (Synthesized Images). (a) shows ground truth images of the longitudinal MRI scans and their difference map (red box). (b) displays synthesized images with age 71 and the difference maps between their and ground truth image with age 66.
• Figure 2: Overall architecture of IdenBAT for brain age transformation. Age transformer $\mathcal{T}$, which consists of encoder $\mathcal{E}$, identity extracting module (IEM), age injecting module (AIM), and generator $\mathcal{G}$ aims to synthesize age-transformed image $\hat{\mathbf{X}}$.
• Figure 3: Detailed architecture of IEM and AIM. DeNorm and CBN refer to the denormalization process and conditional batch normalization, respectively.
• Figure 4: Qualitative comparison of brain aging with different ages on 2D brain images. Difference maps are synthesized by subtracting the input image from the age-converted images.
• Figure 5: Qualitative comparison of brain aging across various ages on 3D brain images.