Core Concepts
The choice of dynamic functional connectivity (dFC) assessment method has a significant impact on the results, with the variability across methods being comparable to the inherent biological variability over time and across subjects.
Abstract
The study aimed to comprehensively assess the analytical flexibility of seven widely used dFC assessment methods: Co-Activation Patterns (CAP), Clustering, Continuous Hidden Markov Model (CHMM), Discrete HMM (DHMM), Sliding Window (SW), Time-Frequency (TF), and Window-less (WL).
The key highlights and insights are:
The overall similarity between the dFC results of different methods ranged from weak to strong, with an average Spearman correlation of 0.38 and high variability (SD = 0.18). This variability was significantly larger than the average variance of pairwise similarities over subjects.
Hierarchical clustering analysis identified three distinct groups of methods based on their similarity: 1) Clustering, CHMM, and DHMM; 2) CAP and WL; 3) SW and TF. These groups differed in their underlying assumptions and advantages.
The variability in dFC estimates across methods was comparable to the expected natural variation over time, emphasizing the significant impact of methodological choices on the results.
Spatial similarity between dFC patterns was higher than temporal similarity, suggesting that capturing the temporal dynamics of functional connectivity may be more challenging than capturing the spatial patterns.
Inter-subject similarity analysis showed that the choice of method can lead to different subject clustering, highlighting the need for multi-analysis approaches to capture the full range of dFC variation.
The study provides an open-source Python toolbox to enable multi-analysis dFC assessment, facilitating the investigation of dFC and the development of new methods.
The findings emphasize the importance of careful method selection, validation, and the use of multi-analysis approaches to enhance the reliability and interpretability of dFC studies.
Stats
"The variance over dFC methods is comparable to the variance over time, with an average ratio of varmethod/vartime = 0.95 (SDmethod/SDtime = 0.97)."
"The functional connections between RSNs such as the FrontoParietal and Ventral Attention networks exhibited equal variance values over time and method, while the functional connections between RSNs such as the Default Mode and Parieto-occipital, and Default Mode and FrontoParietal Networks, and most intra-network connections exhibited higher variation over method than over time."
Quotes
"The variability observed in dFC assessment encourages multi-analysis studies"
"Evalu-ating the overall similarity among the outcomes of various methods offers a perspective on the inter-method similarity relationships. However, it remains unclear to what extent these similarities can be attributed to the spatial and temporal aspects of dFC."