MRI Scanners by Country: The Datasets That Actually Measure It

Only one source in Datadory's Health Care Equipment pool answers the question at country level: the WHO Medical Devices - Definitions and Policy Repository. It publishes 12 DEVICES-series indicators through the GHO OData API, verified live at roughly 290-540 rows each for about 4,700 rows in total, with observations from 2010 to 2021 (some sourced from OECD 2017-19). Equipment-density measures reach up to roughly 143 WHO Member States per indicator.

Everything else in the pool is a different instrument. The two openFDA APIs are United States regulatory systems - the MAUDE adverse-event feed holds 25,711,469 reports as of the 2026-08-11 load and the Registration & Listing API maps 333,804 establishments as of the 2026-08-10 export - and both count devices marketed in the US, not scanners installed per capita abroad. That distinction decides your whole analysis: WHO measures installed capacity by country, FDA measures who makes what and what fails.

The granularity is one observation per country-year per indicator, so a market-sizing pull is a filter on three axes: indicator code, country and year. Each of the 12 verified indicators carries roughly 290-540 rows - about 4,700 rows in total - which sets the realistic ceiling before you query: you will not get an unbalanced panel, and recent years thin out because the series runs to 2021 with some values sourced from OECD 2017-19.

Two families matter for imaging work. Equipment-density indicators report MRI and CT units per population, covering up to roughly 143 Member States; policy indicators are categorical per country rather than numeric series, recording national regulatory and nomenclature status instead of unit counts. If your model needs a continuous variable, only the first family qualifies.

Coverage differs by indicator, and that asymmetry shapes every ranking you build. Density measures reach up to roughly 143 WHO Member States; each individual indicator carries roughly 290-540 country-year rows, so a single-indicator time series covers far fewer than 143 countries in any given year. The Atlas profiles go wider in kind but not in number: 177 country profiles in the 2022 edition, most describing policy status rather than reporting unit counts.

Three consequences follow. First, state the indicator and year together when you compare countries - 'MRI density, observed 2010-2021' - because mixing vintages manufactures fake gaps. Second, a missing row usually means no reported survey wave, not zero scanners; imputing it flattens real differences between reporting and non-reporting states. Third, small markets drop in and out between waves, so a rank table built on latest-available values silently mixes years unless you carry the observation year as a column.

The join key is the device, not the country, and the sequence runs baseline, then suppliers, then field performance.

Start from the WHO DEVICES series for the demand picture: which countries report MRI density, at what level, over the 2010-2021 window. Then move to supply. The openFDA Device Registration & Listing API held 333,804 establishments worldwide as of the 2026-08-10 export, keyed by FEI number, with nested arrays per listed device product and monthly refresh - foreign facilities that market into the US appear alongside domestic ones, so exporter presence shows up directly.

Finally, field performance. The openFDA Device Adverse Events (MAUDE) API carries 25,711,469 reports as of the 2026-08-11 load, refreshed weekly and split across 365 zipped JSON partition files in bulk. Device entries carry brand_name, model_number and a device_report_product_code joined through an openfda{} block, and event_type separates Death, Injury, Malfunction and Other - FDA cautions that reports are submitted, not validated, so trend direction per product code is the usable signal, not incidence rates.

The premarket bookend is the openFDA Device 510(k) Clearances API, a related record shared from the health-care-supplies slice, which dates clearances so you can tell when a scanner model entered the US market relative to the WHO capacity waves.

Steps 1 and 2 produce the ranking itself; steps 3-5 make it defensible by tying capacity numbers to named suppliers, field evidence and national policy instead of leaving the density column orphaned.