计算机科学
翻译(生物学)
人工智能
连接体
自然语言处理
特征(语言学)
度量(数据仓库)
语音识别
模式识别(心理学)
口译(哲学)
人工神经网络
作者
Yi Yang,Nan Zhao,Zeyu Zhang,Jing Zhang,Lu Zhang
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-08
卷期号:: 1-5
标识
DOI:10.1109/isbi61048.2026.11515388
摘要
Understanding the complex relationship between structural and functional brain networks is essential for uncovering the mechanisms that govern brain organization. However, most existing studies model this relationship in a unidirectional manner, either predicting functional connectivity (FC) from structural connectivity (SC) or vice versa, while very few approaches enable true bidirectional translation between the two modalities. To address this gap, we introduce Brain-CLIP, a novel framework inspired by the multimodal contrastive learning paradigm exemplified by CLIP, in which heterogeneous modalities are aligned in a shared latent space through semantic correspondence. Similarly, Brain-CLIP aligns SC and FC in a common embedding space to achieve bidirectional translation. Furthermore, we propose an Individual-Preserving CLIP Loss that maintains precise subject-level alignment between SC and FC representations, enabling the model to capture individual-specific connectomic patterns rather than merely population-level correspondences. Comprehensive evaluations on two large-scale datasets, HCP and ADNI, demonstrate that Brain-CLIP substantially outperforms existing methods. (Code is available here: https://github.com/BrainX-Lab/Brain-CLIP)
科研通智能强力驱动
Strongly Powered by AbleSci AI