AI in Healthcare: How Could It Change Care?
- , by SANUSq Research team
- 13 min reading time
Faster diagnoses, scans read in seconds, less paperwork: AI in healthcare comes with big promises. But how much of that has actually reached the clinic, and how much is still headline? Here is a calm look at what artificial intelligence already does well in medicine, where it still falls short, and why your doctor remains firmly in charge.
What does "AI in healthcare" actually mean?
Most of what gets called artificial intelligence in healthcare is machine learning: computer models trained on very large collections of medical data, which learn to spot patterns that a person would take far longer to see. That ranges from software that flags a shadow on a chest X-ray to systems that summarise records. What they share is the ability to process a great deal of information quickly.
In a widely cited review, Topol (2019) described AI's impact on medicine at three levels: for clinicians, mainly through fast and accurate image interpretation; for health systems, through better workflow and the potential to reduce medical errors; and for patients, by helping them make sense of their own health data. He also set out the current limitations, including bias, privacy and security, and a lack of transparency.
It helps to know that "AI" is not one thing. Several families of tools do quite different jobs:
- Deep learning, which is particularly strong at analysing images.
- Natural language processing, which reads, codes and summarises medical text.
- Predictive models, which estimate a person's risk of a particular outcome from their records.
- Generative AI, the newest arrival, which produces written answers to questions.
Each has its uses and its blind spots, so it helps to know which kind of tool is being discussed.
Where AI is already making a difference
Medical imaging
By some distance, the most advanced area is medical imaging. Deep learning systems can now pick up subtle signs on X-rays, CT scans, photographs of the back of the eye and images of skin lesions, in some focused tasks with accuracy comparable to specialists. The aim is not to replace the radiologist but to give them a tireless second pair of eyes.
Screening for diabetic eye disease from retinal photographs is one of the most mature examples.
Gulshan et al. (2016) trained a deep learning algorithm on more than 128,000 retinal images graded by ophthalmologists, then tested it on two separate sets of images. It detected referable diabetic retinopathy with high sensitivity and specificity. The authors stressed that further research was needed to see whether using it in real clinics would improve care and outcomes.
A systematic review and meta-analysis by Liu et al. (2019) found that, in the studies that compared deep learning models and health-care professionals on the same images, their diagnostic performance was broadly equivalent. However, few studies tested the models on outside data, and poor reporting was common, which limits how far the results can be relied upon.
Some of the strongest evidence so far comes from breast screening, tested in a large randomised trial.
In the Swedish MASAI trial, reported by Gommers et al. (2026), more than 105,000 women were randomly assigned to AI-supported mammography screening or standard double reading by two radiologists. The AI-supported group had a non-inferior rate of cancers appearing between screens, higher sensitivity (80.5% vs 73.8%) and the same specificity, while radiologists' reading workload was reduced. In the AI group, radiologists still read every examination.
Paperwork and the running of care
A second area is the organisation of care. A large share of clinicians' time goes on administration: writing up notes, coding, scheduling. AI could, in principle, give clinicians more of that time back.
In a randomised crossover trial with 160 outpatient clinicians, Chowdhury et al. (2026) compared two "ambient" AI scribes, which listen to a consultation and draft the clinical note. Both tools improved clinicians' satisfaction with their workflow and reduced burnout scores, although time spent on records outside working hours was largely unchanged.
Tools in patients' hands
Finally, AI is reaching patients directly through tracking apps and wearables that measure heart rate or sleep. Used sensibly, they can encourage an interest in your own health, but they are not a substitute for medical advice.
Beyond imaging: promising but younger
Imaging is simply the most mature field. Other areas are moving too. In drug discovery, AI helps researchers sift through vast numbers of candidate molecules far faster than before. Predictive models try to estimate, from a patient's record, the chance that a particular complication will develop. And personalised medicine aims to match treatment more closely to each person's profile.
These directions are real but generally younger and less well validated than imaging, and still have to prove themselves in everyday practice.
The benefits of AI in healthcare come with limits
Enthusiasm needs tempering, and the researchers in the field are often the first to say so. The same concerns come up again and again in the major reviews, including the wide-ranging overview by Yu, Beam and Kohane (2018), which set out the technical, economic, legal and social challenges facing medical AI.
- Biased data. A model trained on one population can make more mistakes with others. It is only as good as the data it learned from.
- The "black box". Some models give an answer without a clear explanation of why, which is a real problem when health is at stake.
- Privacy. These systems run on health data, some of the most sensitive information there is, so security and confidentiality matter enormously.
- Validation. Doing well on a research dataset is no guarantee of working well in a real hospital. Moving into practice needs rigorous testing.
Obermeyer et al. (2019) examined a widely used commercial algorithm that helped US health systems decide which patients should receive extra care. At the same risk score, Black patients were considerably sicker than White patients. The bias arose because the algorithm predicted healthcare costs rather than illness, and less had historically been spent on Black patients' care.
Wong et al. (2021) tested a proprietary sepsis prediction model already in use at hundreds of US hospitals on almost 38,500 hospital stays. It missed about two-thirds of patients who developed sepsis while raising alerts for 18% of all patients, and the authors concluded that it predicted sepsis poorly.
A systematic review by Nagendran et al. (2020) looked at studies comparing deep learning with clinicians in medical imaging. Very few were randomised trials or prospective studies, most were at high risk of bias, and yet the majority claimed in their abstracts that AI performed at least as well as clinicians.
In other words, an impressive result in the lab is not yet a proven benefit at the bedside.
A special case: generative AI and chatbots
Since conversational assistants arrived, many people now type their symptoms into a chatbot. They can explain medical terms and help you prepare questions for an appointment. But they carry a specific risk: they can produce wrong answers in a confident, convincing tone, often called "hallucinations". They do not know your medical history, they cannot examine you and they carry no responsibility for the outcome.
Singhal et al. (2023) evaluated a large language model on a range of medical question sets, including questions people search for online. A medically tuned version answered encouragingly, but when doctors assessed its long-form answers it remained inferior to clinicians, and the authors highlighted key gaps that need better evaluation before clinical use.
These models have developed quickly since, and the evidence is still catching up. A chatbot should never be used to diagnose yourself or decide on treatment alone.
Will AI replace doctors?
It is the question that always comes up, and on current evidence the answer is no. The picture emerging is one of augmentation, not replacement. AI takes on what it does well, such as handling volume and spotting patterns, and leaves people with what it cannot do: overall clinical judgement, understanding the context of someone's life, listening, and shared decision-making. As Topol (2019) put it, the open question is whether these gains will be used to strengthen the relationship between patient and doctor or to erode it.
A doctor connects symptoms to a history, a personality and a social situation, and takes responsibility for the decision. That is what AI cannot do.
Why adoption takes time
AI is moving fast in research, but its arrival in everyday care is slower than the headlines suggest, and that is reassuring rather than disappointing. A health tool has to be clinically validated, authorised by regulators, built into existing systems and, above all, trusted by clinicians who are accountable for its use. Each step rightly takes time: in medicine, caution is a safeguard. Real progress will be measured by proven benefit to patients, not by the number of algorithms.
What it means for you
For most patients, AI will stay largely invisible, working in the background in a radiology department or a laboratory. Where it does become visible, in apps, wearables and health chatbots, the right approach is discernment. These tools can inform and alert; they do not diagnose. A worrying symptom, an unexpected reading or an app's recommendation are all things to discuss with a healthcare professional, not with a screen.
There is also a genuine opportunity. Used well, AI could make care more accessible by easing pressure on specialists and helping regions short of health professionals, provided the benefits reach everyone, not only those who already have the best access to care.
Frequently asked questions
Is AI already used in hospitals?
Yes, mainly in imaging, such as radiology and screening, and in administrative tasks like drafting clinical notes. Its use is still being rolled out gradually and under regulation, because success in research is not enough on its own: each tool needs to be checked in real practice.
Can I trust a diagnosis made by AI?
AI can support a diagnosis, but the decision belongs to your doctor. These tools can be wrong, particularly when their training data are biased, and they do not always explain their reasoning. They are meant to assist clinicians, not replace them.
Are health chatbots safe to use?
They can be helpful for understanding medical terms or preparing questions for an appointment, but they can also give confident answers that are wrong. Do not use them to diagnose yourself or to start, stop or change any treatment.
Do consumer health apps use AI?
Many do, for example in sleep, heart rate and activity tracking. They can help you take an active interest in your health, but they are not diagnostic devices and not a substitute for seeing a professional.
Is my health data safe?
It is one of the central issues in the field. Health data are highly sensitive and, in Europe, strictly protected by law. Before using an app, it is sensible to check how it stores and uses your information and who it shares it with.
Will AI replace my doctor?
No. The trend is towards augmentation rather than replacement: AI helps doctors with specific tasks, while judgement, listening and responsibility for decisions stay with people.
References
- Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nature Medicine. 2019;25(1):44–56. PMID 30617339
- Yu KH, Beam AL, Kohane IS. Artificial intelligence in healthcare. Nature Biomedical Engineering. 2018;2(10):719–731. PMID 31015651
- Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016;316(22):2402–2410. PMID 27898976
- Liu X, Faes L, Kale AU, et al. A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis. Lancet Digital Health. 2019;1(6):e271–e297. PMID 33323251
- Gommers J, Hernström V, Josefsson V, et al. Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial. Lancet. 2026;407(10527):505–514. PMID 41620232
- Chowdhury A, Casey M, Wilson J, et al. Comparing ambient scribes: a randomized crossover clinical trial addressing ambient scribe technologies' impact on physician burnout. Journal of the American Medical Informatics Association. 2026;33(5):990–999. PMID 41729180
- Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366(6464):447–453. PMID 31649194
- Wong A, Otles E, Donnelly JP, et al. External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients. JAMA Internal Medicine. 2021;181(8):1065–1070. PMID 34152373
- Nagendran M, Chen Y, Lovejoy CA, et al. Artificial intelligence versus clinicians: systematic review of design, reporting standards, and claims of deep learning studies. BMJ. 2020;368:m689. PMID 32213531
- Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature. 2023;620(7972):172–180. PMID 37438534
Read more on the blog
Watch on YouTube
Explore more on our channel, @SANUSqHealthSupplements.
We wish you good health on your terms.
The team at SANUSq.
PS. Curious what's currently on offer? Take a look at our latest promotions.
The health information in this article is provided for educational purposes only. Consult your healthcare professional before making any medical decisions.
Regulatory Notice
The information on our websites, blogs, and emails is provided for informational purposes only and has not been evaluated by the EMA, EFSA, or FDA. It is not intended to replace medical advice provided by your healthcare professional, and is not intended to diagnose, treat, cure, or prevent any disease. Our products are intended for adults aged 18 and over. While the vitamins and supplements mentioned may offer various health benefits, supplements and dietary changes should be considered part of an overall health plan and not a substitute for professional medical treatment. Only a qualified healthcare professional can provide personalized advice and treatment plans based on your individual health needs and medical history; you should consult your healthcare professional before taking any product(s) if you are pregnant or breastfeeding.