Neuroimaging with Sounio

Sounio enables reproducible neuroimaging analysis with built-in uncertainty quantification and provenance tracking.

The Reproducibility Challenge

Neuroimaging research faces significant reproducibility issues:

• Pipeline parameters affect results dramatically
• Processing choices are often not documented
• Statistical thresholds vary between studies
• Uncertainty in measurements propagates unpredictably

Sounio’s Solution

use std::neuroimaging::{fMRI, Voxel, BrainRegion}
use std::epistemic::{Knowledge, Provenance}

struct ActivationMap {
    voxels: Vec<Knowledge<f64>>,
    threshold: f64,
    provenance: Provenance,
}

fn analyze_fmri(
    scan: fMRI,
    pipeline: Pipeline
) -> ActivationMap with IO {
    // Each step tracks provenance
    let preprocessed = preprocess(scan, pipeline.preproc_params)
    let motion_corrected = correct_motion(preprocessed)
    let smoothed = spatial_smooth(motion_corrected, pipeline.fwhm)

    // GLM with uncertainty propagation
    let betas = fit_glm(smoothed, pipeline.design_matrix)

    // Confidence-gated thresholding
    let significant = betas.filter(|v|
        v.value > pipeline.threshold &&
        v.confidence > 0.95  // Only include high-confidence activations
    )

    ActivationMap {
        voxels: significant,
        threshold: pipeline.threshold,
        provenance: betas.provenance,  // Full processing history
    }
}

Pipeline Provenance

Every analysis step is tracked:

fn preprocess(scan: fMRI, params: PreprocessParams) -> fMRI {
    let result = scan
        |> slice_timing_correction(params.tr)
        |> motion_correction(params.reference)
        |> coregistration(params.template)
        |> normalization(params.space)

    // Provenance automatically includes:
    // - Input file checksums
    // - Parameter values used
    // - Software versions
    // - Timestamps

    result
}

Confidence-Gated ROI Analysis

fn roi_analysis(
    activation: ActivationMap,
    atlas: BrainAtlas
) -> Vec<RegionResult> with IO {
    var results = vec![]

    for region in atlas.regions {
        let voxels = activation.voxels_in(region)
        let mean_activation = voxels.mean()

        if mean_activation.confidence > 0.90 {
            results.push(RegionResult {
                region: region.name,
                activation: mean_activation,
                status: "High confidence",
            })
        } else {
            perform IO::warn(
                "Low confidence for " + region.name +
                " (confidence: " + mean_activation.confidence.to_string() + ")"
            )
            results.push(RegionResult {
                region: region.name,
                activation: mean_activation,
                status: "Requires validation",
            })
        }
    }

    results
}

Multi-Site Studies

fn harmonize_sites(
    datasets: Vec<SiteData>
) -> HarmonizedData with IO {
    // Track site-specific effects
    for site in datasets {
        perform IO::log(
            "Site: " + site.name +
            ", Scanner: " + site.scanner_model +
            ", Subjects: " + site.n_subjects.to_string()
        )
    }

    // ComBat harmonization with uncertainty
    let harmonized = combat_harmonize(datasets)

    // Verify harmonization quality
    if harmonized.site_effect.confidence < 0.80 {
        perform IO::warn("Residual site effects may remain")
    }

    harmonized
}

Features for Neuroimaging

• Full provenance: Every processing step documented
• Uncertainty propagation: From raw signal to final statistics
• Confidence gates: Automatic flagging of uncertain results
• BIDS compatibility: Native support for BIDS format
• Reproducible pipelines: Deterministic processing

BIDS Integration

use std::bids::{BidsDataset, load_participants}

fn main() with IO {
    let dataset = BidsDataset::open("./my_study")

    for subject in dataset.participants() {
        let func = dataset.func(subject, "task-rest")
        let anat = dataset.anat(subject, "T1w")

        let result = analyze_subject(func, anat)
        dataset.write_derivative(subject, result)
    }
}

Get Started

sounio new neuro-project --template neuroimaging
cd neuro-project
sounio run examples/fmri_analysis.sio