rAidiology

Open source, for education and research

Open your own scans in your browser.

CT, MRI and X-ray pictures from your own disc or ZIP, shown on your device, next to seven open research models for X-ray, CT, MRI and mammogram pictures. Nothing is uploaded.

Open a scan Try the sample scan

Coronal CT slice through the chest and abdomen with the lungs, heart, liver, kidneys, spleen, spine and pelvis coloured by organ.
CT · organs
Side-on MRI of the lower spine with each vertebra and disc coloured separately.
MRI · spine
Front-view chest X-ray showing both lungs, the heart and the ribs.
X-ray · chest
CT slice across the chest in a lung window, showing both lungs and the heart.
CT · lungs
X-ray of a child's wrist and forearm bones.
X-ray · wrist
Mammogram picture showing the tissue of one breast.
Mammogram
Side-view X-ray of the lower spine.
X-ray · spine

Pictures from public datasets under CC BY and similar licences. Picture credits

What was built, in numbers

  • 7

    research models

    Seven models trained on public data, each described on its own card

    Where this comes from
  • 4

    kinds of scan

    X-ray, CT, MRI and mammogram

    Where this comes from
  • 115

    body structures outlined on CT, scored on held-out scans

    Structures found in at least 10 of the 89 held-out CT scans

    Where this comes from
  • 25,596

    chest X-rays in the held-out test

    NIH's official test list, kept out of all training

    Where this comes from
  • 4,067

    children's wrist X-rays in the held-out test

    GRAZPEDWRI-DX held-out test pictures with a certain label

    Where this comes from

Scans we cover

X-ray, CT, MRI and mammogram, in one place.

Each kind of scan has a viewer and open research models trained on public data. Pick one to see what the models do and how they scored.

Research results

One result from each research model.

Each row is one result: the dot is the score and the line is its 95% interval, the range the score would most likely fall in on new data.
A filled dot is a held-out test, data the model never saw while training. A hollow dot is development data.

Research results on public test data. They say nothing about your own scan.

AUROC, 0.5 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Chest X-ray research model: Agreement with ChestX-ray14's labels, mean over its 14 label types: 0.818 (95% interval 0.811 to 0.823); Fracture research model (X-ray): Agreement with GRAZPEDWRI-DX's fracture labels, children's wrists: 0.979 (95% interval 0.974 to 0.984); Mammogram research model: Agreement with CMMD's benign/malignant labels, per picture: 0.793 (95% interval 0.774 to 0.811). Research results on public test data.AUROC, 0.5 to 1Held-out testChest X-ray research model:Agreement with ChestX-ray14'slabels, mean over its 14 labeltypes0.818 (0.811 to 0.823)Fracture research model (X-ray):Agreement with GRAZPEDWRI-DX'sfracture labels, children'swrists0.979 (0.974 to 0.984)DevelopmentMammogram research model:Agreement with CMMD'sbenign/malignant labels, perpicture0.793 (0.774 to 0.811)0.50.751coin toss (0.5)AUROC, 0.5 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Chest X-ray research model: Agreement with ChestX-ray14's labels, mean over its 14 label types: 0.818 (95% interval 0.811 to 0.823); Fracture research model (X-ray): Agreement with GRAZPEDWRI-DX's fracture labels, children's wrists: 0.979 (95% interval 0.974 to 0.984); Mammogram research model: Agreement with CMMD's benign/malignant labels, per picture: 0.793 (95% interval 0.774 to 0.811). Research results on public test data.AUROC, 0.5 to 1Held-out testChest X-ray researchmodel: Agreement withChestX-ray14's labels,mean over its 14 labeltypes0.818 (0.811 to 0.823)Fracture researchmodel (X-ray):Agreement withGRAZPEDWRI-DX'sfracture labels,children's wrists0.979 (0.974 to 0.984)DevelopmentMammogram researchmodel: Agreement withCMMD'sbenign/malignantlabels, per picture0.793 (0.774 to 0.811)0.50.751coin toss (0.5)
Dice overlap, 0 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. CT organ outliner: Mean over 115 structures: 0.867 (95% interval 0.854 to 0.878). Research results on public test data.Dice overlap, 0 to 1Held-out testCT organ outliner: Mean over 115structures0.867 (0.854 to 0.878)00.51Dice overlap, 0 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. CT organ outliner: Mean over 115 structures: 0.867 (95% interval 0.854 to 0.878). Research results on public test data.Dice overlap, 0 to 1Held-out testCT organ outliner:Mean over 115structures0.867 (0.854 to 0.878)00.51
Kappa agreement, 0 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Disc grading research model (MRI): Agreement with SPIDER's Pfirrmann grades, held-out test: 0.773 (95% interval 0.664 to 0.853). Research results on public test data.Kappa agreement, 0 to 1Held-out testDisc grading research model(MRI): Agreement with SPIDER'sPfirrmann grades, held-out test0.773 (0.664 to 0.853)00.510 = chance agreementKappa agreement, 0 to 1Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Disc grading research model (MRI): Agreement with SPIDER's Pfirrmann grades, held-out test: 0.773 (95% interval 0.664 to 0.853). Research results on public test data.Kappa agreement, 0 to 1Held-out testDisc grading researchmodel (MRI): Agreementwith SPIDER'sPfirrmann grades,held-out test0.773 (0.664 to 0.853)00.510 = chance agreement
Percent, 0 to 100Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Lung nodule research model (CT): Share of LUNA16's marked nodules matched, averaged over seven false-alarm allowances: 88.0 (95% interval 86.0 to 89.6); Lumbar spine X-ray corner reader: Share of BUU-LSPINE's marked vertebrae matched, side view (IoU 0.5): 97.9 (95% interval 97.3 to 98.4). Research results on public test data.Percent, 0 to 100DevelopmentLung nodule research model (CT):Share of LUNA16's marked nodulesmatched, averaged over sevenfalse-alarm allowances88.0% (86.0 to 89.6%)Lumbar spine X-ray corner reader:Share of BUU-LSPINE's markedvertebrae matched, side view (IoU0.5)97.9% (97.3 to 98.4%)050100Percent, 0 to 100Dot and interval chart. Filled dot: held-out test. Hollow dot: development. Lung nodule research model (CT): Share of LUNA16's marked nodules matched, averaged over seven false-alarm allowances: 88.0 (95% interval 86.0 to 89.6); Lumbar spine X-ray corner reader: Share of BUU-LSPINE's marked vertebrae matched, side view (IoU 0.5): 97.9 (95% interval 97.3 to 98.4). Research results on public test data.Percent, 0 to 100DevelopmentLung nodule researchmodel (CT): Share ofLUNA16's markednodules matched,averaged over sevenfalse-alarm allowances88.0% (86.0 to 89.6%)Lumbar spine X-raycorner reader: Shareof BUU-LSPINE's markedvertebrae matched,side view (IoU 0.5)97.9% (97.3 to 98.4%)050100
Show the numbers
ResultMetricScore95% intervalTest setKind
Chest X-ray research model: Agreement with ChestX-ray14's labels, mean over its 14 label typesAUROC0.8180.811 to 0.823NIH's official test list: 25,596 pictures of 2,797 patientsheld-out test
CT organ outliner: Mean over 115 structuresDice0.8670.854 to 0.878Held-out test: 89 CT scansheld-out test
Disc grading research model (MRI): Agreement with SPIDER's Pfirrmann grades, held-out testkappa0.7730.664 to 0.853SPIDER held-out test: 44 patients, 264 discs, opened onceheld-out test
Fracture research model (X-ray): Agreement with GRAZPEDWRI-DX's fracture labels, children's wristsAUROC0.9790.974 to 0.984GRAZPEDWRI-DX held-out test: 4,067 pictures of 1,176 patients with a certain labelheld-out test
Lung nodule research model (CT): Share of LUNA16's marked nodules matched, averaged over seven false-alarm allowancespercent88.086.0 to 89.6LUNA16: 888 thin-slice LIDC-IDRI scans, five folds, each read by a model that never trained on itdevelopment
Mammogram research model: Agreement with CMMD's benign/malignant labels, per pictureAUROC0.7930.774 to 0.811CMMD: 3,744 pictures of 1,775 patients, five folds split by patientdevelopment
Lumbar spine X-ray corner reader: Share of BUU-LSPINE's marked vertebrae matched, side view (IoU 0.5)percent97.997.3 to 98.4All 2,000 BUU-LSPINE patients, out of folddevelopment

Where the models fall short

  • CT organ outliner. Both kidney cysts score 0.0000 on the test scans. Vertebra C6, the gallbladder, the portal and splenic veins and the adrenal glands are the next weakest.
  • Disc grading research model (MRI). The three current versions have not been scored on held-out test patients; their results are development results.
  • Lung nodule research model (CT). All the nodule-finding numbers are development scores from the same 888 scans used to choose between designs; there is no untouched test.
  • Mammogram research model. Where no lesion outline exists the score is close to a coin toss: version 38 scores 0.5726 on those 1,002 pictures.

See every result

What the viewer does, and what it does not.

It does

  • Open CT, MRI and X-ray discs and ZIPs on this device
  • Let you scroll through slices, zoom and change the brightness
  • Show the 3D sample scan with organs and bones named

It does not

  • Upload your scan
  • Read your report
  • Tell you what is wrong
  • Replace your doctor

Install it as an app

Add rAidiology to your computer or phone from the browser. It then opens offline, with no account.

How to install it

Pricing

Plain pricing.

  • Free

    Free viewer and licence

    Free, because it runs on your device and costs us nothing to run.

  • Pay what you want

    Supporter licence

    Pay what you want, from $3. The app stays free.

  • Paid

    Commercial licence

    For companies building products or internal tools. From $1,500 a year, with a free 60-day evaluation.

    Not for diagnosis or treatment. Not cleared by the FDA or any regulator.

  • Coming, paid

    Cloud anatomy labels (CT)

    Paid per scan. It uploads the scan to be read. Anatomy names only.

Personal use, charities, schools, universities and public research are free under the licence.

See all pricing

Help train bigger research models.

Your support pays for rented GPU time: an A100 80GB is $1.59 an hour, so $15 is about 9 hours. Today training runs on one home graphics card.

Support the research See pricing

LunarWerx LLC is a for-profit company. Payments are not charitable donations and are not tax-deductible.

Your scan stays on your device.

Nothing is uploaded when you open a scan. The pictures are read by your own browser and are gone when you close the tab. If you ever choose to donate a scan to help the research, you see everything first, and what could identify you is removed in your browser.

Read the privacy policy · See or delete a donation

Open source, and open to questions.

The code and the model cards are public. Ask, report a problem or follow the work on Discord.

GitHub Discord

Picture credits
  • CT, organs and bones coloured. TotalSegmentator dataset v2.0.1, CC BY 4.0. Source. Windowed (width 400, level 40), cropped and coloured by LunarWerx, using the dataset's own masks.
  • MRI of the lumbar spine, vertebrae and discs coloured. SPIDER lumbar spine MRI, CC BY 4.0. Source. Middle sagittal slice, contrast stretched, cropped and coloured by LunarWerx, using the dataset's own masks.
  • X-ray of a child's wrist. GRAZPEDWRI-DX v2, CC BY 4.0. Source. 16-bit picture contrast stretched, cropped and resized by LunarWerx.
  • Chest X-ray. NIH ChestX-ray14, No restrictions (NIH), attribution requested. Source. Contrast stretched, cropped and resized by LunarWerx.
  • CT of the chest, lung window. LIDC-IDRI, CC BY 3.0. Source. Windowed (width 1500, level -600), cropped and resized by LunarWerx.
  • Mammogram. CMMD (Chinese Mammography Database), CC BY 4.0. Source. Contrast stretched, cropped and resized by LunarWerx.
  • Side X-ray of the lower spine. Spondylolisthesis Vertebral Landmark Dataset v1 (Reyes, VSB-TU Ostrava), CC BY 4.0. Source. Contrast stretched, cropped and resized by LunarWerx.