From Archived Scans to a Completed Neuroimaging Report
Cian Monnin
Research groups collect MRI scans. The scans end up on a hard drive, sometimes for years. File formats are mixed, naming conventions have drifted, the person who ran the scanner has graduated, and the behavioural spreadsheet has been through enough hands that nobody is confident about what was cleaned.
This is a common situation we see at the Douglas Neuroinformatics Platform. A group has data that could answer a real question, but the gap between what they have and an actual analysis feels too wide. There’s no pipeline, no standardised layout, and the team’s strength is the science, not the data engineering.
The bottleneck isn’t the statistics
The statistical methods for relating brain structure to behaviour are well known. The hard part is everything that comes before: getting mixed file formats into a common standard, checking quality on every scan, preprocessing consistently, and putting together a workflow that someone could re-run next year and get the same result.
Groups routinely underestimate this. It’s not a weekend job. It’s months of troubleshooting, often on unfamiliar compute infrastructure, and a misstep early on can quietly affect every result downstream. When the work falls to a trainee learning as they go, reproducibility tends to suffer.
What we actually deliver
We take a group from disorganised data to a finished, reproducible analysis:
A clean, standards-compliant dataset. Whatever came in DICOMs from the scanner, hand-renamed NIfTIs, a mix of both, goes out as a single structured dataset that any tool in the field can work with.
End-to-end preprocessing. Skull stripping, bias correction, template construction, morphometric mapping, all handled with established tools, properly configured for the data. The group doesn’t need to know which software versions clash or which parameters matter for their acquisition.
Statistics that fit the question. Voxel-wise models are a reasonable start, but they’re not always enough. When it makes sense, we use multivariate methods that model brain and behaviour together, picking up distributed patterns that standard approaches bury under multiple-comparison corrections.
Figures and tables for the paper. Brain maps on the group’s own template. Bar charts with confidence intervals. Summary tables with variance and significance. Finished outputs for a manuscript.
A reproducible record. Every step is scripted. The whole analysis reruns from raw data with one config change. When a reviewer asks “what if you control for X,” the answer is a rerun, not a rebuild.
Why come to us
A group that tries to do all of this independently will spend months on work outside their expertise. We’ve done it before, we maintain the infrastructure, and we build every pipeline to be reproducible from day one.
The practical upshot: a dataset collecting dust becomes a completed analysis. Stale data becomes a paper. And the group spends its time on the science instead of fighting the tooling.
Coming soon: scanner-to-derivatives automation
We are building end-to-end services that automate the full path from scanner to pipeline derivatives. Raw DICOMs come off the scanner and flow directly into conversion, quality control, and preprocessing, with analysis-ready outputs produced without manual intervention.
Reproducible research support
If you’d like help developing reproducible research workflows using tools like Nipoppy, Nextflow, nf-core, nf-neuro, or a bespoke pipeline tailored to your data, get in touch at contact@douglasneuroinformatics.ca.