Related Experiment Video
Updated: Jun 11, 2026
Measuring Statistical Learning Across Modalities and Domains in School-Aged Children Via an Online Platform and Neuroimaging Techniques
Published on: June 30, 2020
StatLI: A novel statistical approach to the laterality index for task-based fMRI
Sourav Chowdhury1, David Reutens2, Rahul Garg3
1University of Queensland (UQ) - Indian Institute of Technology Delhi (IITD) Research Academy, Indian Institute of Technology Delhi, Hauz Khas, New Delhi, 110 016, Delhi NCT, India; Amar Nath and Shashi Khosla School of Information Technology, Indian Institute of Technology Delhi, Hauz Khas, New Delhi, 110 016, Delhi NCT, India; Centre for Advanced Imaging, The University of Queensland, St Lucia, Brisbane, 4067, Queensland, Australia.
Background:
Accurate determination of lateralization is often critical in neurosurgical decision-making. The laterality index (LI) provides a measure of lateralization by comparing hemispheric activations within specific ROI in task-based fMRI studies. LI may be affected by the statistical threshold, the data SNR, and the ROI and task used.
New Method:
We describe three methods for estimating LI point estimates and confidence intervals, assuming that the activated voxels are known. While limiting direct application to real fMRI analyses, this simplification allowed us to evaluate the statistical properties of different LI estimators. Confidence intervals are calculated (1) as a naïve estimate from interval estimates based on each hemispheric activation; (2) as non-symmetric levels using numerical optimization; and (3) using a transformation-based approach.
Results:
Increasing the SNR and activation blocks improved performance across all methods, decreasing bias % and interval width. The framework was robust across various noise types and ROIs, maintaining a bias below 1 %, and generalized across different lateralization patterns when validated on individual subject maps derived from HCP data.
Comparison With Existing Methods:
A relative hierarchy in accuracy was observed, with the bias being lowest for the Transformed method, followed by the Numeric and the Naïve method. Subsequently, the different threshold-independent methodologies from the literature consistently exhibited high bias and variance.
Conclusions:
This work establishes a theoretical baseline for developing and validating more robust and reliable LI methods. These new LI methods aim to enhance the accuracy of lateralization assessment in both clinical and research applications.
