Moving Beyond Functional Connectivity: Time-Series Modeling for fMRI-Based Brain Disorder Classification

TL;DR

The paper tackles the limitation of static Pearson FC in fMRI-based disease classification by leveraging end-to-end temporal models on raw BOLD signals. It introduces DeCI, a framework that combines Cycle and Drift Decomposition with Channel-Independence to disentangle intra-ROI dynamics from inter-ROI interactions, achieving improved generalization. Across six public datasets, DeCI outperforms FC-based pipelines and other temporal baselines, with strong statistical significance and favorable efficiency. The work demonstrates not only superior accuracy but also better interpretability and robustness, advocating a shift toward end-to-end temporal modeling in fMRI analysis for clinical deployment.

Abstract

Functional magnetic resonance imaging (fMRI) enables non-invasive brain disorder classification by capturing blood-oxygen-level-dependent (BOLD) signals. However, most existing methods rely on functional connectivity (FC) via Pearson correlation, which reduces 4D BOLD signals to static 2D matrices, discarding temporal dynamics and capturing only linear inter-regional relationships. In this work, we benchmark state-of-the-art temporal models (e.g., time-series models such as PatchTST, TimesNet, and TimeMixer) on raw BOLD signals across five public datasets. Results show these models consistently outperform traditional FC-based approaches, highlighting the value of directly modeling temporal information such as cycle-like oscillatory fluctuations and drift-like slow baseline trends. Building on this insight, we propose DeCI, a simple yet effective framework that integrates two key principles: (i) Cycle and Drift Decomposition to disentangle cycle and drift within each ROI (Region of Interest); and (ii) Channel-Independence to model each ROI separately, improving robustness and reducing overfitting. Extensive experiments demonstrate that DeCI achieves superior classification accuracy and generalization compared to both FC-based and temporal baselines. Our findings advocate for a shift toward end-to-end temporal modeling in fMRI analysis to better capture complex brain dynamics. The code is available at https://github.com/Levi-Ackman/DeCI.

Figures (10)

• Figure 1: For each identical backbone (GCN; Transformer), we trained three instances that differ only in input: (i) FC, (ii) raw BOLD signals, and (iii) Shuffled-BOLD that applies the same temporal permutation to every ROI, thus preserving FC while ablating temporal order. The BOLD models outperform FC; Shuffled-BOLD degrades to near-FC, isolating the contribution of temporal dynamics.
• Figure 2: General structure of DeCI, a Channel-Independent framework. We use Linear Convolution networks to extract the $Drift$ pattern, and Nonlinear Squeeze and Excitation Networks (SE-Net) to model the $Cycle$ pattern. The final classification result is the sum of multiple logits generated using $Drift$ and $Cycle$ information extracted from each DeCI Block.
• Figure 3: Average rank of all benchmarks across five metrics.
• Figure 4: Model efficiency comparison on the TaoWu, comparing Accuracy, inference time, and memory footprint.
• Figure 5: Test of resilience to channel noise on PPMI. DeCI is Channel-Independent; others are Channel-Dependent. Gaussian noise with zero mean and varying intensity ($\beta \in {0, 1, \dots, 12}$) is added to the last half of the training channels.