The dynamics of dynamic time warping in fMRI data: A method to capture inter-network stretching and shrinking via warp elasticity

Sir-Lord Wiafe, Ashkan Faghiri, Zening Fu, Robyn L. Miller, Adrian Preda, Vince D. CalhounImaging 2024 · DOI

TL;DR

This paper introduces "warp elasticity," a novel metric derived from Dynamic Time Warping (DTW), to analyze the stretching and shrinking of functional connectivity timescales between brain regions in fMRI data. This method offers a more nuanced understanding of brain dynamics than traditional functional connectivity measures and successfully distinguishes between healthy controls and individuals with schizophrenia.

What Problem Does This Paper Address?

Understanding how different brain regions communicate and coordinate their activity over time is crucial for deciphering brain function. Functional connectivity (FC) analysis, commonly used with fMRI, measures the temporal similarity between brain region signals. However, traditional FC methods often overlook the fact that the timescales of these connections can change dynamically. Brain regions might be coupled for longer or shorter durations, or their coupling might stretch or shrink over time. This temporal flexibility is a significant aspect of brain dynamics that current FC methods, which typically rely on fixed time windows or simple correlations, fail to capture. This gap limits our understanding of how brain networks adapt and how these adaptations might be disrupted in neurological and psychiatric disorders.

Methods

The core of this paper is the introduction of warp elasticity, a metric built upon Dynamic Time Warping (DTW). DTW is an algorithm that aligns two time series by stretching or compressing them to find the optimal correspondence between points, even if their temporal patterns are misaligned.

Here's how warp elasticity is derived:

  1. Dynamic Time Warping (DTW): DTW is applied to pairs of time series representing brain network activity. It computes an "optimal warping path" that minimizes the distance between the aligned time series.
  2. Warp Deviation (WD): The difference between the warping paths of two time series is calculated. This "warp deviation" quantifies the temporal distortion or lag between the signals.
  3. Warp Elasticity (WE): The temporal derivative of the warp deviation is computed. This derivative, termed "warp elasticity," represents the rate of change of the temporal distortion, effectively quantifying the stretching or shrinking of the coupling timescale between brain regions. A positive value indicates stretching, while a negative value indicates shrinking.

The authors also employ k-means clustering to group similar warp elasticity patterns into "warp elasticity (WE) states," allowing for the identification of recurring temporal dynamics.

Datasets

The study utilizes two primary datasets:

  • Human Connectome Project (HCP) dataset: This large dataset (827 subjects) was used to evaluate the noise sensitivity and replicability of the warp elasticity metric.
  • Function Biomedical Informatics Research Network (fBIRN) study: This dataset (160 healthy controls, 151 schizophrenia patients) was used for group comparisons to identify differences in WE states between healthy individuals and those with schizophrenia.

Key Insights and Ideas

  • Beyond Simple Correlation: Traditional FC measures like Pearson correlation and phase synchrony struggle to capture dynamic changes in the duration or speed of brain region couplings. Warp elasticity directly addresses this by quantifying the temporal stretching and shrinking of these couplings.
  • Adaptive Windows: DTW's ability to use adaptive windows is key. It allows the method to capture transient couplings that might be missed by fixed-window approaches.
  • Asymmetric Dynamics: The method can reveal asymmetric dynamics where one brain region's coupling stretches (suggesting prolonged processing) while another's shrinks (suggesting faster processing), or vice-versa. This offers a more nuanced view of inter-network interactions.
  • Robustness to Noise: Warp elasticity demonstrates remarkable resilience to noise in fMRI data, a critical factor for real-world neuroimaging applications.
  • Clinical Signatures: Distinct warp elasticity states and transition patterns were observed between healthy controls and individuals with schizophrenia, suggesting that these temporal dynamics may hold diagnostic or mechanistic value.

Key Results

  • Simulation Accuracy: In simulations, warp elasticity accurately estimated ground truth stretching and shrinking factors, achieving a correlation of 0.88 with the ground truth.
  • Noise Sensitivity: Warp elasticity showed high correlations (around 0.7 to 1) with actual fMRI signals even at low tSNR values, indicating its robustness to noise.
  • Replicability: Bootstrap analysis on the HCP dataset demonstrated high replicability of warp elasticity patterns across different subject subsets.
  • Group Differences (Schizophrenia vs. Controls):
    • Significant differences were found in WE states between SZ and HC groups.
  • Healthy controls spent more time in a state characterized by stretching of visual networks relative to sensory networks, potentially indicating disruptions in visual processing in SZ.
  • Schizophrenia patients showed increased time in states with stretching in higher cognitive areas relative to sensory regions, suggesting prolonged cognitive processing of sensory input.
  • Statistically significant differences were observed in mean dwell times, fraction rates, and transition matrices between the groups, particularly in cluster 2 (associated with visual thought processing).

Limitations and Future Work

  • Visualization Challenges: Visualizing the precise stretching and shrinking windows in actual fMRI time series is complex due to the intricate nature of these phenomena.
  • Interpretational Nuance: The precise significance of specific stretching and shrinking patterns within interconnected brain networks remains an area for further investigation.
  • Clinical Integration: Future research should incorporate additional clinical metrics and subgroup analyses to better understand how factors like psychotropic medications affect these dynamics in schizophrenia.
  • EEG Application: The authors suggest exploring warp elasticity with EEG data, which, due to its high temporal resolution, could reveal even more granular stretching and shrinking patterns.
  • Broader Applications: The method's potential for understanding other dynamic brain processes and disorders warrants further exploration.

Summary updated

Abstract

In neuroimaging research, understanding the intricate dynamics of brain networks over time is paramount for unraveling the complexities of brain function. One approach commonly used to explore the dynamic nature of brain networks is functional connectivity analysis. However, while functional connectivity offers valuable insights, it fails to consider the diverse timescales of coupling between different brain regions. This gap in understanding leaves a significant aspect of brain dynamics unexplored in neuroimaging research. We propose an innovative approach that delves into the dynamic coupling/connectivity timescales of brain regions relative to one another, focusing on how brain region couplings stretch or shrink over time, rather than relying solely on functional connectivity measures. Our method introduces a novel metric called "warping elasticity," which utilizes dynamic time warping (DTW) to capture the temporal nuances of connectivity. Unlike traditional methods, our approach allows for (potentially nonlinear) dynamic compression and expansion of the time series, offering a more intricate understanding of how coupling between brain regions evolves. Through the adaptive windows employed by the DTW method, we can effectively capture transient couplings within varying connectivity timescales of brain network pairs. In extensive evaluations, our method exhibits high replicability across subjects and diverse datasets, showcasing robustness against noise. More importantly, it uncovers statistically significant distinctions between healthy control (HC) and schizophrenia (SZ) groups through the identification of warp elasticity states. These states are cluster centroids, representing the warp elasticity across subjects and time, offering a novel perspective on the dynamic nature of brain connectivity, distinct from conventional metrics focused solely on functional connectivity. For instance, controls spend more time in a warp elasticity state characterized by timescale stretching of the visual domain relative to other domains, suggesting disruptions in the visual cortex. Conversely, patients show increased time spent in a warp elasticity state with stretching timescales in higher cognitive areas relative to sensory regions, indicative of prolonged cognitive processing of sensory input. Overall, our approach presents a promising avenue for investigating the temporal dynamics of brain network interactions in functional magnetic resonance imaging (fMRI) data. By focusing on the elasticity of connectivity timescales, rather than adhering to functional connectivity metrics, we pave the way for a deeper understanding of neuropsychiatric disorders in neuroscience research.
BibTeX
@inproceedings{wiafe2024dynamics,
  author = {Sir-Lord Wiafe and Ashkan Faghiri and Zening Fu and Robyn L. Miller and Adrian Preda and Vince D. Calhoun},
  title = {{The dynamics of dynamic time warping in fMRI data: A method to capture inter-network stretching and shrinking via warp elasticity}},
  booktitle = {Imaging},
  year = {2024},
  doi = {10.1162/imag\_a\_00187},
}
Figure