Spontaneous neural activity observed in resting-state fMRI is characterized by complex spatio-temporal dynamics. Different measures related to local and global brain connectivity and fluctuations in low-frequency amplitudes can quantify individual aspects of these neural dynamics. Even though such measures are derived from the same functional signals, they are often evaluated separately, neglecting their interrelations and potentially reducing the analysis sensitivity. In our study, we present a fusion searchlight (FuSL) framework to combine the complementary information contained in different resting-state fMRI metrics and demonstrate how this can improve the decoding of brain states. Moreover, we show how explainable AI allows us to reconstruct the differential impact of each metric on the decoding, which additionally increases spatial specificity of searchlight analysis. In general, this framework can be adapted to combine information derived from different imaging modalities or experimental conditions, offering a versatile and interpretable tool for data fusion in neuroimaging.

翻译：静息态功能磁共振成像中观测到的自发神经活动具有复杂的时空动力学特征。与局部及全局脑连接性相关的不同测量指标以及低频振幅的波动能够量化这些神经动力学的各个层面。尽管这些指标源自相同的功能信号，它们通常被单独评估，忽略了彼此间的关联，从而可能降低分析灵敏度。在本研究中，我们提出了一种融合搜索灯框架，用于整合不同静息态fMRI指标中包含的互补信息，并证明该方法如何提升大脑状态的解码性能。此外，我们展示了可解释人工智能如何重建各指标对解码过程的差异化贡献，从而进一步增强了搜索灯分析的空间特异性。总体而言，该框架可适用于整合来自不同成像模态或实验条件的信息，为神经影像数据融合提供了一个通用且可解释的工具。