Definition
AI in healthcare uses machine learning to support diagnosis from scans, predict patient risk, assist drug discovery and reduce paperwork. It's meant to support clinicians, and it faces strict demands for safety, privacy and evidence.
In simple words
A doctor sees thousands of X-rays in a career. A model can learn from millions. It can help flag what to look at, while the doctor makes the decision and takes the responsibility.
Why it matters
Mistakes here can cost lives. So it's the clearest example of why high accuracy alone isn't enough: testing across groups and hospitals matters.
Key terms
- Diagnosis
- Working out which illness a patient has.
- Screening
- Checking many people who seem healthy in order to find the few who need treatment early.
- Medical imaging
- Pictures of the inside of the body, such as X-rays, CT scans and photos of the back of the eye.
- Clinician
- A doctor, nurse or other trained person who treats patients.
- Clinical trial
- A careful study that tests whether a treatment or tool really helps patients.
- Second reader
- A second check of a scan. AI is often used as a second reader beside a doctor.
Explained in detail
Why medicine turned to AI
Healthcare has two problems that AI is suited to. The first is a shortage of specialists. In many parts of India a person may live hundreds of kilometres from the nearest eye surgeon or radiologist. The second is volume. Modern hospitals produce huge numbers of scans, and a tired doctor reading the two-hundredth X-ray of the day can miss a small sign. A model does not tire, and a trained model can be copied to a clinic anywhere.
Screening: finding patients early
The clearest success so far is screening from images. Diabetes can slowly damage the back of the eye and cause blindness, but treatment works well if the damage is found early. Models trained on many thousands of eye photos graded by specialists can spot the signs. In 2018 the medicines regulator of the United States approved the first AI system allowed to make this screening decision without a specialist looking at the image. Eye hospitals in India, including Aravind and Sankara Nethralaya, took part in research on such systems, which let a health worker with a camera screen patients in a small town.
Tuberculosis is another example. Software that reads chest X-rays for signs of the disease, some of it developed in India, is used in screening programmes, and the World Health Organization has recommended such tools as an option where trained readers are scarce.
Other uses
| Area | What AI does | Who stays in charge |
|---|---|---|
| Reading scans | Marks areas that look unusual | The radiologist |
| Discovering medicines | Predicts the shapes of proteins and suggests promising molecules | Scientists, then clinical trials |
| Hospital wards | Warns that a patient's readings suggest they may get worse | Nurses and doctors |
| Wearable devices | Notices an irregular heartbeat | The wearer and their doctor |
| Paperwork | Drafts notes from a consultation | The doctor, who checks them |
Why caution is essential
A model learns from the patients and machines in its training data. Move it to a hospital with a different scanner, or to patients of different ages or backgrounds, and its accuracy can fall without anyone noticing. Models for skin conditions trained mostly on lighter skin have performed worse on darker skin. For this reason a medical AI tool must be tested like a medicine, in real clinics, on the kinds of patients it will serve, before it is trusted. Health data is also among the most private information a person has, and it must be collected with consent and stored securely.
How it works
| Use | Example |
|---|---|
| Imaging | Flagging signs of disease in scans |
| Protein structure | AlphaFold predicts how proteins fold |
| Drug discovery | Shortlisting candidate molecules |
| Paperwork | Drafting clinical notes |
| Risk | Predicting which patients may worsen |
Try it yourself
One X-ray's journey
Follow a scan through a hospital that uses AI carefully.
The AI's job or the doctor's job?
In this hospital, AI supports the doctors. Decide who should do each task.
Marking areas of a scan that are worth a closer look
Drafting clinical notes from a visit
Deciding if a model tested at another hospital is safe to use here
Making the final diagnosis
Choosing a treatment with the patient
Shortlisting candidate molecules for a new medicine
Common mistakes
- Trusting a model's headline accuracy without seeing which patients it was tested on.
- Using health data without consent and legal checks.
Check yourself
Summary
- AI supports diagnosis, research and paperwork.
- Testing across groups and sites is essential.
- Clinicians stay responsible.
Finished reading?