AI radiology is software in which an AI model reads medical images, such as chest X-rays and CT scans, and reports what it finds before a radiologist sees them. It is also called radiology AI, or AI for medical imaging. It flags signs of tuberculosis (TB), lung nodules, which are small lumps in the lung, and strokes, so that the urgent cases are seen first.
The model is trained on lakhs of past images that radiologists have already reported. It learns which patterns in an image match which findings. This is a kind of computer vision. For each new image, the model gives a score for each possible finding, and it marks where in the image the finding is. If a score is above a set limit, the image is flagged as abnormal.
The model has two main jobs. The first is screening, in places where there is no radiologist. The second is triage in busy hospitals: sorting the queue so that the most urgent scan is read first. For example, a head CT that shows bleeding in the brain moves to the top of the radiologist's list, instead of waiting behind 200 normal chest X-rays.
Take a TB screening camp in a district of Jharkhand. In one day, a mobile van takes 400 chest X-rays, and no radiologist is present. The AI model reads each X-ray in under a minute. It flags 38 of them as likely TB. Those 38 people give a sputum sample the same day, for a molecular test that confirms TB from the germ's genetic material. The other 362 people are told that their X-ray looks normal, and a radiologist reviews their images later. The World Health Organization (WHO) has recommended this kind of computer-aided detection for TB screening since 2021.
The limit is set so that the model misses very few cases. As a result, it also flags some people who do not have TB. That is acceptable, because a confirming test follows. A doctor, not the model, makes the diagnosis.
The model receives images from medical imaging equipment. Its results reach the radiologist in the PACS and imaging AI platform, and the report is stored in the hospital EHR. Because a wrong reading can harm a patient, the model is regulated as software as a medical device (SaMD). In India, the regulator for medical devices is the Central Drugs Standard Control Organisation (CDSCO). Imaging has more approved products than any other area of medical AI. Indian companies built these models for the shortage of radiologists at home, and now sell them in dozens of countries.