Material Insights · Understanding the Material World

AI diagnosis from medical images: Overview

What it does

The job the technology does for people, in a short paragraph

AI diagnosis from medical images uses neural networks to spot signs of disease in scans, photographs and tissue slides. It can flag urgent cases, act as a second reader or, in some screening tasks, decide alone. The first device that gave a screening decision without a clinician was authorised in the United States in 2018 [3, 8]. By September 2026 the FDA had authorised more than 1,600 AI-enabled devices [4].

as of 2026-11-11

How it works

The working principle in plain words, step by step

A digital image is a grid of numbers, one for the brightness or colour of each pixel. A neural network is a stack of simple calculating units, each taking numbers in and passing a weighted result on. To train it, developers show it many images that specialists have already graded, such as 128,000 retinal photographs [2]. After each guess the network's weights are nudged to make the right answer more likely, a process repeated a great many times. Early layers come to respond to edges and spots, later ones to shapes such as small bleeds or dividing cells. One early system judged each pixel from the square patch of image around it [1]. Once trained and tested on new images, the network gives a score or a decision for each scan in seconds.

as of 2026-11-11

Key figures

The handful of numbers that size it up, each with its unit, year and source

FigureValueUnitYearSource
AI-enabled medical devices of all kinds authorised in the US, more than (to September 2026)1600devices2026[4]
Share of US AI device authorisations in radiology, September 1995 to December 2025 (1,094 of 1,430)76.5%2025[6]
Cut in screen-reading workload with AI support in the MASAI breast screening trial (early safety analysis)44.3%2023[5]
IDx-DR: share of more-than-mild diabetic eye disease correctly identified in its trial (observed sensitivity)87.4%2018[3, 9]
Graded retinal images used to train a diabetic eye disease network128,000images2016[2, 7]

as of 2026-11-11. The radiology share is from a 2026 Cureus study of 1,430 authorisations as reported by AuntMinnie [6]; the device total is the FDA's own and covers all fields [4]; the 2016 image count is dated by the Google blog's publication [2]; the MASAI figure is from the trial's prespecified safety analysis after 80,000 women [5].

Where it is used

The main places and products the technology is found in

  • Diabetic eye screening from retinal photographs in primary care [3]
  • Breast cancer screening with mammograms [5]
  • Radiology, the field with most US authorisations of AI devices [6]
  • Pathology: finding cancer cells in tissue slides [1]

as of 2026-11-11

Scale today

How much of it there is now: units in use, made each year and who uses it

Installed
More than 1,600 AI-enabled devices authorised in the United States [4]
Unit
authorised devices
Year
2026
Note
Counts authorised products, not installations or scans read; no world count was found.

as of 2026-11-11

Sources

  1. Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks MICCAI 2013, Lecture Notes in Computer Science 8150, pp. 411 to 418, Springer (D. C. Ciresan, A. Giusti, L. M. Gambardella and J. Schmidhuber, IDSIA, Lugano), 2013
  2. Deep Learning for Detection of Diabetic Eye Disease (29 November 2016) Google Research blog (L. Peng and V. Gulshan), 2016
  3. FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems Healthcare Innovation (HCI Innovation Group), reproducing the FDA news release of 11 April 2018, 2018
  4. Artificial Intelligence-Enabled Medical Devices US Food and Drug Administration, 2026
  5. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI) The Lancet Oncology 24(8), 936 to 944 (K. Lång and others), record at Lund University, 2023
  6. Radiology dominates thirty years of FDA AI device approvals AuntMinnie (Will Morton, 16 July 2026), reporting a study by P. Golshani and others in Cureus (doi 10.7759/cureus.112583), which could not be opened, 2026
  7. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs JAMA 316(22), 2402 to 2410 (V. Gulshan and others), record in the BVS/LILACS database, 2016
  8. De Novo order letter DEN180001: IDx-DR, retinal diagnostic software device US Food and Drug Administration, 2018
  9. De Novo classification request for IDx-DR (decision summary, DEN180001) US Food and Drug Administration, 2018