NeuroStream: an interactive platform for exploratory visualization and harmonization of multicohort brain MRI data

Large-scale neuroimaging studies increasingly integrate brain MRI data from multiple cohorts and acquisition sites. Exploratory visualization and harmonization are essential for identifying batch effects, assessing preprocessing strategies, and preserving biologically meaningful variation. However, existing tools are often cohort-specific, rely on static visualizations, or separate harmonization from data exploration.

We present NeuroStream, an interactive application for real-time visualization and harmonization of multicohort brain MRI quantitative data. Built using the Streamlit framework, NeuroStream enables the exploration of quantitative imaging features derived from structural MRI along with demographic and clinical variables. The platform supports preprocessing and harmonization options, including log transformation, intracranial volume normalization, and ComBat-based batch correction, which can be compared interactively using dynamic visualizations. NeuroStream provides principal component analysis, distributional plots, and group comparison statistics. In addition, the platform supports covariate-adjusted regression-based evaluation of site effects and residual diagnostics, enabling quantitative and visual assessment of cohort-related variability across preprocessing settings. All visualizations presented in this manuscript were generated using transformed example datasets designed to illustrate the functionality of the platform, rather than to report original subject-level measurements, while NeuroStream itself supports direct analysis of user-provided tabular data. Using transformed example datasets derived from BICWALZS and KoGES multicohort MRI data, we demonstrate intuitive inspection of batch-related variance and harmonization outcomes without custom scripting.

NeuroStream is implemented in Python and freely available at https://github.com/NIHxAI/NeuroStream.