Exploratory temporal ICA based analysis in task and resting-state fMRI

Temporally independent functional modes (TFMs) are functional brain networks identified based on their temporal independence. The rationale behind identifying TFMs is that different functional networks may share a common anatomical infrastructure yet display distinct temporal dynamics. Extracting TFMs usually require a larger number of samples than acquired in standard fMRI experiments, and thus have therefore previously only been performed at the group level. Here, using an ultra-fast fMRI sequence, MESH-EPI, with a volume repetition time of 158 ​ms, we conducted an exploratory study with n ​= ​6 subjects and computed TFMs at the single subject level on both task and resting-state datasets. We identified 6 common temporal modes of activity in our participants, including a temporal default mode showing patterns of anti-correlation between the default mode and the task-positive networks, a lateralised motor mode and a visual mode integrating the visual cortex and the visual streams. In alignment with other findings reported recently, we also showed that independent time-series are largely free from confound contamination. In particular for ultra-fast fMRI, TFMs can separate the cardiac signal from other fluctuations. Using a non-linear dimensionality reduction technique, UMAP, we obtained preliminary evidence that combinations of spatial networks as described by the TFM model are highly individual. Our results show that it is feasible to measure reproducible TFMs at the single-subject level, opening new possibilities for investigating functional networks and their integration. Finally, we provide a python toolbox for generating TFMs and comment on possible applications of the technique and avenues for further investigation.

时间独立功能模式（Temporally Independent Functional Modes, TFMs）是基于时间独立性识别出的功能性脑网络。识别TFMs的核心逻辑在于，不同功能网络可能共享共同的解剖学基础架构，却展现出独特的时间动态特征。提取TFMs通常需要比标准功能磁共振成像（fMRI）实验中获取的样本量更大的数据集，因此以往仅能在组水平开展相关研究。本研究采用超快速fMRI序列MESH-EPI（容积重复时间为158毫秒），对6名受试者进行探索性研究，并在任务态与静息态数据集上完成了单受试者水平的TFMs计算。我们在受试者中识别出6种常见的时间活动模式，包括：呈现默认模式与任务阳性网络间反相关特征的时间默认模式、偏侧运动模式，以及整合视觉皮层与视觉流的视觉模式。与近期报道的其他研究结果一致，我们发现独立时间序列在很大程度上不受混杂因素干扰；尤其对于超快速fMRI而言，TFMs可有效分离心脏信号与其他波动。通过非线性降维技术UMAP，我们获得初步证据表明，TFM模型所描述的空间网络组合具有高度个体特异性。研究结果证实，在单受试者水平测量可重复的TFMs具有可行性，为探索功能网络及其整合机制开辟了新方向。最后，我们提供了用于生成TFMs的Python工具箱，并对该技术的潜在应用场景及未来研究路径进行了阐述。