Research Dataset for your site

Every tool on this site can add a scan to a de-identified hemorrhage volumetric dataset that stays in this browser. Review it here, add outcomes when they are known, and export it for REDCap, NIH ScHARe, Parquet, FHIR, OMOP or Convex.

Your site's data, under your site's approvals

Collect and share these data only under your institution's approvals (IRB or ethics committee, data use agreements). Research and education use only; AI volumes are not validated for clinical decisions.

Stays on this computer

Records live in this browser (IndexedDB) for this site only; nothing is uploaded. Clearing the browser's site data deletes them, so download a backup regularly. Each browser and computer has its own dataset; merge them with Import.

De-identified by design

  • Study IDs become pseudonyms with your site key.
  • Dates become hours from onset; age is capped at 90.
  • No images, names, MRNs or free text.
1

Site settings

Used by every tool. Use the same site key on every computer at your site, so the same study ID always gives the same pseudonym and serial scans link up.

—

The key was made at random in this browser. Changing it changes every new pseudonym; records already stored keep theirs. Keep a copy of the key in your site's secure records, separate from the data, and do not send it with the exports.

2

The dataset

One row per scan and tool. Outcomes can be entered here when known; everything else comes from the tools.

3

Export

The whole dataset (or the tool shown above). Every format uses the same variables, described in the data dictionary.

Everything (.zip)

CSV, Parquet, data dictionary, REDCap files, FHIR, OMOP, JSON Lines and a README.

CSV + data dictionary

For NIH ScHARe (Repository upload, mapped to common data elements), R, Python, Stata or Excel.

REDCap

Create a project from the data dictionary, then import the records.

Parquet

Typed columnar file for ScHARe/Terra (BigQuery), pandas, Arrow, DuckDB or Spark.

FHIR R4 and OMOP

FHIR Observations for EHR-side systems; OMOP-style MEASUREMENT rows for OHDSI networks (map concept ids at your site).

JSON Lines and backup

JSON Lines for a Convex table or other databases; the backup restores this dataset in another browser.

4

Import, merge and clear

Merge a backup, a dataset CSV or JSON Lines from another computer at your site. Records with the same record ID are replaced.

?

Using the exports

REDCap

In REDCap: New project → Upload a REDCap Data Dictionary and choose redcap_data_dictionary.csv (forms: scan record, hemorrhage volumes, clinical, outcomes). Then Data Import Tool → upload redcap_records.csv. One REDCap record is one scan; subject_id links serial scans of a subject. Leave the Identifier settings off: there are no direct identifiers.

NIH ScHARe (NIMHD, Terra on Google Cloud)

Upload dataset.csv or dataset.parquet with data_dictionary.csv to a private or controlled collection in the ScHARe Repository and map the variables to common data elements (the dictionary gives each one's NINDS CDE concept), or bring the Parquet file into a Terra workspace. ScHARe's focus on health disparities and AI bias suits comparisons across countries and sites; release beyond your collection follows its review.

Convex

Import the JSON Lines file as a table, then query it from a Convex app (for example a consortium dashboard). Keep it de-identified, and check Convex's business associate agreement and hosting region against your approvals (LGPD and GDPR as well as HIPAA).

npx convex import --table hemorrhage_scans dataset.jsonl
Python and R 
import pandas as pd
df = pd.read_parquet("dataset.parquet")      # or pd.read_csv("dataset.csv")
df.groupby("tool")[["vol_sah_ml", "vol_iph_ml", "vol_ivh_ml"]].describe()
df <- arrow::read_parquet("dataset.parquet")   # or read.csv("dataset.csv")
Federated and secure environments (e.g. Mayo Clinic Platform)

The tools already work in a federated way: each site runs them on its own computers and keeps its own dataset; only the exports you choose leave. A consortium can pool de-identified exports, or keep data at each site and share only summaries. For work inside a secure platform such as Mayo Clinic Platform (data kept behind each partner's walls), the same models can run as containers inside the platform. Check each model's licence: CLAIM, DeepBleed and the Valladolid model are non-commercial.