Spatially constrained ICA enables robust detection of schizophrenia from very short resting-state fMRI data

Marlena Duda, Armin Iraji, Judith M. Ford, Kelvin O. Lim, Daniel H. Mathalon, Bryon A. Mueller, Steven G. Potkin, Adrian Preda, Theo G.m. van Erp, Vince D. CalhounmedRxiv 2022 · DOI

Abstract

Resting-state functional network connectivity (rsFNC) has shown utility for identifying characteristic functional brain patterns in individuals with psychiatric and mood disorders, providing a promising avenue for biomarker development. However, several factors have precluded widespread clinical adoption of rsFNC diagnostics, namely a lack of standardized approaches for capturing comparable and reproducible imaging markers across individuals, as well as the disagreement on the amount of data required to robustly detect intrinsic connectivity networks (ICNs) and diagnostically relevant patterns of rsFNC at the individual subject level. Recently, spatially constrained independent component analysis (scICA) has been proposed as an automated method for extracting ICNs standardized to a chosen network template while still preserving individual variation. Leveraging the novel scICA methodology, we investigate the challenge of identifying a minimally sufficient data length for clinical applications of resting-state fMRI (rsfMRI). Using a dataset containing individuals with schizophrenia and controls (M = 310) as well as simulated rsfMRI, we evaluate the robustness of ICN and rsFNC estimates. We found individual estimates from the full-length (5 minute) reference were sufficiently approximated with just 3-3.5 minutes of data (r = 0.85, 0.88, respectively), and significant group-average differences could be approximated with just 2 minutes (r = 0.86). Furthermore, a model trained on 2 minutes of data retained 97-98% classification accuracy relative to the full-length reference. Our results suggest that clinical rsfMRI scans, when decomposed with scICA, could potentially be shortened to just 2-4 minutes without significant loss of information.

Introduction

Resting-state functional MRI (rsfMRI) is a powerful tool for mapping brain connectivity, but it faces a major hurdle in clinical adoption: the "scan length" dilemma. To obtain a reliable "fingerprint" of a patient's brain networks, researchers typically recommend scans lasting 10 to 40 minutes. However, for patients with psychiatric disorders like schizophrenia, sitting perfectly still in a noisy, cramped MRI tube for a long duration is difficult, leading to head motion and distress that corrupts data. Conversely, very short scans are often too noisy to provide stable biomarkers. This study investigates whether spatially constrained ICA (scICA) can bridge this gap by extracting robust connectivity patterns from "minimally sufficient" data lengths of 2 to 4 minutes.

Methods

The study utilized Spatially Constrained Independent Component Analysis (scICA), specifically the MOO-ICAR algorithm. Unlike standard "blind" ICA, scICA uses a spatial template—the "NeuroMark" template consisting of 53 validated brain networks—to guide the decomposition. This template acts as a regularizer, stabilizing results even when input data is short and noisy.

The NeuroMark template consisting of 53 validated intrinsic connectivity networks across seven domains.

The researchers evaluated the pipeline by taking full-length scans and chopping them into segments (1, 2, 3, and 4 minutes) to compare against the full-length data. Data sources included a clinical discovery dataset (310 subjects), an independent validation dataset (129 subjects), and a 30-minute simulated dataset.

Overview of the data preparation, scICA decomposition, and reliability evaluation pipeline.

Key Results

  • Subject-Level Reliability: Individual brain network estimates reached a high robustness threshold (r >= 0.85) at 3 to 3.5 minutes of data.
  • Group-Level Reliability: Significant differences between schizophrenia and control groups were reliably identified with just 2 minutes of data (r = 0.86 compared to the 5-minute reference).
  • Diagnostic Performance: A machine learning model (SVM) trained on just 2 minutes of data retained 97-98% of the classification accuracy of the model trained on the full 5-minute scan. In the external validation set, the 2-minute model achieved an accuracy of 0.68, compared to 0.70 for the 5-minute model.
  • Simulation Consistency: Results from the 30-minute simulated data mirrored clinical findings, confirming that 3-4 minutes is a reliable "sweet spot" even compared against much longer reference lengths.

Discussion and Limitations

The spatial priors in scICA stabilize the solution, allowing for reliable network extraction in a fraction of the time required by seed-based or atlas-based methods. Group-level differences are particularly robust, appearing with as little as 2 minutes of data.

Limitations include the 5-minute clinical reference length; future work should validate these findings against 30-60 minute real-world datasets. Additionally, the simulation used a 2D axial slice, whereas real brains are 3D. Future research will also investigate if scICA can similarly stabilize dynamic functional network connectivity.

Summary updated

Abstract

ABSTRACT Resting-state functional network connectivity (rsFNC) has shown utility for identifying characteristic functional brain patterns in individuals with psychiatric and mood disorders, providing a promising avenue for biomarker development. However, several factors have precluded widespread clinical adoption of rsFNC diagnostics, namely a lack of standardized approaches for capturing comparable and reproducible imaging markers across individuals, as well as the disagreement on the amount of data required to robustly detect intrinsic connectivity networks (ICNs) and diagnostically relevant patterns of rsFNC at the individual subject level. Recently, spatially constrained independent component analysis (scICA) has been proposed as an automated method for extracting ICNs standardized to a chosen network template while still preserving individual variation. Leveraging the novel scICA methodology, which solves the former challenge of standardized neuroimaging markers, we investigate the latter challenge of identifying a minimally sufficient data length for clinical applications of resting-state fMRI (rsfMRI). Using a dataset containing individuals with schizophrenia and controls ( M = 310) as well as simulated rsfMRI, we evaluate the robustness of ICN and rsFNC estimates at both the subject- and group-level, as well as the performance of diagnostic classification, with respect to the length of the rsfMRI time course. We found individual estimates of ICNs and rsFNC from the full-length (5 minute) reference time course were sufficiently approximated with just 3-3.5 minutes of data ( r = 0.85, 0.88, respectively), and significant differences in group-average rsFNC could be sufficiently approximated with even less data, just 2 minutes ( r = 0.86). The results from the shorter clinical data were consistent with the results from the longer simulated data, reliably estimating both individual- and group-level metrics from the full-length (30 minute) reference with just 3-4 minutes of data ( r = 0.85 - 0.88). Furthermore, we found a model trained on 2 minutes of data retained 97-98% classification accuracy relative to that of the full-length reference model. Our results suggest that clinical rsfMRI scans, when decomposed with scICA, could potentially be shortened to just 2-4 minutes without significant loss of individual rsFNC information or classification performance of longer scan lengths.
BibTeX
@misc{duda2022spatially,
  author = {Marlena Duda and Armin Iraji and Judith M. Ford and Kelvin O. Lim and Daniel H. Mathalon and Bryon A. Mueller and Steven G. Potkin and Adrian Preda and Theo G.M. van Erp and Vince D. Calhoun},
  title = {{Spatially constrained ICA enables robust detection of schizophrenia from very short resting-state fMRI data}},
  year = {2022},
  doi = {10.1101/2022.03.17.22271783},
  howpublished = {medRxiv},
}
Figure