Industries / Health / Healthcare / Imaging centers
Dose and safety
Keeping patients and staff safe in the scan room: holding X-ray doses as low as the task allows, checking machines and staff, and keeping metal away from MRI magnets.
7
Milestones achieved, from 1904 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1904
Harm learned the hard way
Clarence Dally, a glassblower who worked on X-ray tubes and fluorescent screens in Thomas Edison's laboratory from 1896, often used his own hand to check that a tube was working. By 1900 his hands and face were badly damaged. A biopsy of an ulcer on his left hand showed cancer, both his arms were eventually amputated, and he died in October 1904. In a 2019 article, Adrian Thomas places him among many early "x-ray martyrs", whose stories Percy Brown collected in 1936.
LessonThe first X-ray workers found out about the harm through their own bodies, years after the exposure.
Credit
Percy Brown (1936) and Adrian Thomas, who recorded the history
1992
Every mammogram unit inspected
The US Congress passed the Mammography Quality Standards Act (MQSA) on 27 October 1992. Every facility that does mammograms must be accredited, certified by the FDA, surveyed each year by a medical physicist and inspected at least once a year, against federal standards for staff, equipment, radiation dose and records. By November 2026 the FDA counted 9,140 certified facilities with 27,933 accredited units, reporting about 45.2 million mammograms a year, and 84.9% of the 577 inspections so far in its 2027 fiscal year had found nothing out of line.
Credit
US Congress; FDA
1996
A yardstick for ordinary doses
The International Commission on Radiological Protection (ICRP) introduced the "diagnostic reference level" (DRL) in its Publication 73 in 1996. A clinic compares the median dose it gives a sample of typical patients for one kind of exam with a local, national or regional DRL. If its median is higher, it looks for the equipment or procedure to fix. A DRL is not a limit for any one patient.
Credit
ICRP
2001
Pulled into the magnet
In the summer of 2001, at Westchester Medical Center in New York State, the piped oxygen failed while six-year-old Michael Colombini lay sedated in an MRI scanner. A nurse from outside the department handed the anesthesiologist a steel oxygen cylinder, the magnet's always-on field pulled it into the scanner, and it struck the boy, who died two days later. The technologists had not been required to have any MRI safety training. The American College of Radiology (ACR) published MRI safety guidance in 2002, after an investigation of the death led by Emanuel Kanal; later versions divide an MRI suite into zones, lock the inner ones against anyone without MRI training, and require portable metal objects such as oxygen cylinders to be identified in writing as safe or unsafe before they go in.
LessonThe tank was the last of many failures, and training and locked zones are what stop the next one.
Credit
ACR (Emanuel Kanal and colleagues); Tobias Gilk and Robert J. Latino (the account)
2009
A setting nobody checked
In August 2009 Cedars-Sinai Medical Center in Los Angeles learned, after a patient reported hair loss, that a CT scanner used for brain perfusion scans, which show blood flow in suspected stroke, had been set since February 2008 to give about eight times the usual dose. 206 patients were affected. The FDA found at least 50 more at other facilities, on scanners from at least two makers, and in December 2009 asked facilities to check their protocols and makers to review their training. In September 2010 California passed SB 1237, which required from July 2012 that the dose of every CT scan be recorded and put in the report, with displayed doses checked each year by a medical physicist, and from July 2013 that CT facilities be accredited.
LessonNo one compared the dose the machine gave with the dose intended, so the error ran for 18 months.
Credit
California Department of Public Health and FDA (the investigations); state senator Alex Padilla (SB 1237)
2012
Childhood CT and cancer
Mark Pearce and colleagues followed about 178,000 people first scanned by CT before age 22 in Great Britain between 1985 and 2002. They estimated that a total of about 50 mGy might almost triple the risk of leukemia, and about 60 mGy triple the risk of brain cancer. Because both cancers are rare, they put the added risk at about one case of each per 10,000 head scans in children under 10, in the 10 years after the first scan. In 2023 the EPI-CT study, of 658,752 young people scanned in 276 hospitals in nine European countries, found that brain cancer risk rose in step with dose: each 100 mGy added 1.27 times the normal risk, so that 100 mGy a little more than doubled it.
LessonThe risk from one scan is small but real, so each child's scan has to be needed and kept to a low dose.
Credit
Mark S. Pearce and colleagues; Michael Hauptmann and the EPI-CT team
2019
Lead shields set aside
For decades patients had lead shields laid over their reproductive organs or an unborn child during X-rays. On 2 April 2019 the American Association of Physicists in Medicine (AAPM) said the practice should end as routine. The shields can hide parts of the body the doctor needs to see and can confuse the machine's automatic exposure control, which may raise the dose. They also do nothing about the X-rays that scatter inside the body, the main source of dose to organs outside the picture.
LessonA habit kept for decades as plain good sense was dropped once its benefit was weighed against its harms.
Credit
AAPM
Watch for
CT dose weighed against quality
A US quality measure, stewarded by the company Alara Imaging, counts the share of a hospital's adult CT scans that are out of range: either the dose, adjusted for the patient's size, is too high, or the image is too noisy, against thresholds set for the reason the scan was ordered. Medicare's administrator, the Centers for Medicare & Medicaid Services (CMS), adopted it for hospital outpatient departments, voluntary from 2025 and mandatory from 2027. In November 2025, after hospitals and the ACR cited the cost and difficulty of the software, CMS kept it voluntary and said it expected to propose a mandatory start in its next yearly rule. Rebecca Smith-Bindman of the University of California, San Francisco, said an optional measure would lead many hospitals not to report.
Achieved whenreporting becomes mandatory and CMS publishes hospitals' results, or an outside study shows hospitals that report it cutting their out-of-range scans.
Claimed by
CMS, November 2025
Open
01
Audit doses in public scans
A public table of CT doses taken from the file headers of every CT series in the US National Cancer Institute's Imaging Data Commons (IDC), grouped by body region, scanner maker and year, and set beside published national diagnostic reference levels. It reads only dose numbers from files that are already public and de-identified. It opens no image and judges no patient.
The briefCredit · unclaimed02
Check implant MRI labels
A check of whether the MRI safety status that device makers file in the FDA's Global Unique Device Identification Database (GUDID) matches the MRI safety information in their own current labeling, for a random sample of implants. MR Conditional means a device is safe in a scanner only under stated conditions, such as field strength and heating limits. The work uses public records and makers' published documents only.
The briefCredit · unclaimed
Open 01 · the brief
Audit doses in public scans
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of CT doses taken from the file headers of every CT series in the US National Cancer Institute's Imaging Data Commons (IDC), grouped by body region, scanner maker and year, and set beside published national diagnostic reference levels. It reads only dose numbers from files that are already public and de-identified. It opens no image and judges no patient.
Why it is still open
The largest check of this kind that a search found, a 2026 study by S. Yamamoto, read 40 abdominal CT series from the Cancer Imaging Archive. A recorded CTDIvol, the scanner's standard dose index in mGy, survived in 29 of the 40 headers, could be rebuilt in 5 and was missing in 6, and how often it survived differed between makers. A search found no audit across the full public collections.
Built with what exists
IDC's DICOM metadata, which it exports to Google BigQuery and to public Parquet files that the open-source DuckDB database can query without an account; pydicom for spot checks of single files; Python for the tables. OpenREM, a free, open-source dose-monitoring program, already does this for a hospital's own scanners, so the new work is the public, cross-collection view.
First step
Query IDC's Parquet files with DuckDB to count the CT series that carry a CTDIvol value, collection by collection.
Done looks like
every CT series in IDC sorted into "dose recorded", "dose can be rebuilt" and "no dose", by maker; for series with a dose, the median and 75th percentile CTDIvol by body region, set beside at least two countries' national reference levels; code and table public, rerun on each IDC release.
Open 02 · the brief
Check implant MRI labels
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A check of whether the MRI safety status that device makers file in the FDA's Global Unique Device Identification Database (GUDID) matches the MRI safety information in their own current labeling, for a random sample of implants. MR Conditional means a device is safe in a scanner only under stated conditions, such as field strength and heating limits. The work uses public records and makers' published documents only.
Why it is still open
An August 2026 analysis of the July 2026 GUDID release by MedDeviceGuide, a device-regulation website, found that 902,914 of 1,135,724 implantable device records (79.5%) say their labeling contains no MRI safety information, and 15.7% say MR Conditional. It counted the filed values but did not compare them with the labels. A search found no study that has.
Built with what exists
the AccessGUDID download files from the National Library of Medicine; makers' MRI safety pages and instructions for use; a language model to pull the stated MRI conditions out of PDF labels, with every extraction checked by a person. The Note "Seven more flagged" describes a Claude-based reviewer that checked long manufacturing records for a drug maker.
First step
Download the latest AccessGUDID release, draw 50 implant records marked MR Conditional or "no information", and find each maker's MRI labeling.
Done looks like
500 randomly drawn implant records, each checked by two people against the maker's current labeling, with the share that match, the share that disagree and the share that say "no information" while the label gives MRI conditions, all with 95% confidence intervals; the list sent to the FDA's GUDID help desk and published.