How AI is used in Singapore healthcare
Singapore healthcare organisations use AI for tasks such as analysing retinal images and helping staff prepare documents. Each system has a defined purpose and a particular review process, and the tools vary from clinic to clinic. Clinical teams remain responsible for safe care, including checking limitations and deciding how to use the results.
What does clinical AI do?
Most AI in a clinic is a pattern-recognition tool trained on a large set of past examples. A model that has seen many thousands of retinal photographs learns what a healthy retina and a damaged one look like. Shown a new photograph, it gives a score, and a threshold on that score turns into a flag. The model works by recognising patterns that were consistent in its training data, and it has no understanding of eyes.
That description covers most of the AI in use today. Image reading, risk scoring, queue prediction and the newer text tools that draft or summarise notes all work from patterns in past data. Such a tool is fast and consistent, and it does not tire.
The limits follow from the same fact. Performance depends on training data, design, testing and the setting in which a tool is used. An AI system can miss a finding or flag something that is normal. Healthcare teams need a defined workflow for validation, monitoring and handling uncertainty.
If you are reading about a healthcare AI tool, check whether it is a research project, a limited deployment or routine service. Also check what it is approved or intended to do and who is responsible for reviewing or acting on its output.
Examples of AI use in Singapore
SELENA+ is a Singapore-developed deep-learning system for analysing retinal photographs. Synapxe describes its role in supporting diabetic eye screening and identifying possible diabetic retinopathy, glaucoma and age-related macular degeneration. The screening pathway determines which findings require further assessment; a screening output is not a complete eye examination.
Imaging is another area of AI development. Software may mark regions of an X-ray for attention, but implementation differs across institutions and products. Ask the imaging service what tools it uses and how its staff review the report rather than assuming every scan follows one national AI workflow.
Staff also use AI in three quieter ways.
- Risk support. AI systems can estimate risk using recorded observations and results. Their clinical usefulness depends on local validation and how staff respond; a risk estimate cannot include a symptom or history that was never recorded.
- Administrative support. Prediction and automation tools can assist planning and routine staff work. The effect on an individual waiting time depends on the service; use of AI does not guarantee a shorter wait.
- Notes and paperwork. In December 2024, NUHS reported that more than 2,000 staff had used its secure RUSSELL-GPT platform. Generative tools can assist with text tasks, but clinical documentation needs appropriate checking before use.
The amount of automation and human review differs by tool and clinical pathway. The important questions are whether the system is validated for its task, whether errors can be detected and who is accountable for acting on the result.
How shared health records support care
Shared records help clinicians see information from different care settings. The National Electronic Health Record, supported by Synapxe, holds a summary of key records contributed by healthcare providers. It may not give a complete account of every visit or every private app entry.
Healthier SG, launched in 2023, asks residents to enrol with one family doctor and build a health plan with them. That plan lives in the same connected system. Access to relevant records can support decision-making, but incomplete, outdated or incorrect data can also affect an AI output. You can help by keeping your care team informed about medicines, diagnoses and care received elsewhere.
Questions about data and your visit
Data handling depends on the institution, the system and applicable healthcare and data-protection requirements. Different rules can apply to public agencies and private organisations. Ask the institution what data a tool processes, whether it sends that data elsewhere and who controls access.
For most visits, you will notice little difference. Staff take the photograph or X-ray the same way, and the changes come later in the process. A screening result may arrive sooner once a tool has cleared the normal images. A waiting time might be shorter. A doctor may glance at a summary that an AI tool drafted for them. Some institutions tell patients that AI assists with reading their images, and it is reasonable to ask.
Limits that still require clinical judgement
There are four established limits.
- No examination. An image reader sees the image. It does not see the patient walk in, feel the abdomen or hear the cough.
- No history. A risk score uses what is in the record. Anything not recorded, including a family history mentioned in passing, is invisible to it.
- Errors and uncertainty. AI tools vary in how they represent uncertainty, and some have confidence scores or abstention rules. Any of them can produce incorrect output. A convincing summary or a high model score is not a guarantee of clinical accuracy.
- Training data. A model trained with most of its data from one population can perform worse on another. Local testing matters, which is one reason Singapore has built some of its own tools.
Call 995 for possible emergencies such as chest pain with sweating or breathlessness, severe difficulty breathing, sudden one-sided weakness or speech difficulty, collapse, or bleeding that will not stop. A sudden severe headache or fever with severe illness also needs urgent assessment. For suicidal thoughts, call 1767 or 1771 for immediate support; call 995 if you cannot stay safe. See a GP or polyclinic about other new, persistent or worsening symptoms instead of relying on an app.
What patients can ask
AI supports some healthcare tasks in Singapore, including retinal-image analysis and staff text work. Deployment is specific to the institution and tool. Patient benefit depends on validation, integration into care and appropriate oversight, and the AI label on its own does not show any of those.
New applications remain under evaluation. Look for published results from the service and population involved, including errors and patient outcomes, before treating a product claim as established practice.
You can ask an institution how it handles your records. If you have a symptom, bring it to a doctor.
A practical next step
You can ask whether AI is involved in your care, what it does and how staff check the result. For questions about access to your records or AI data use, contact the healthcare institution.
Sources and further reading
These references explain the guidance and research discussed above. For advice about your own health, speak to a qualified healthcare professional.
Read next
Health apps, and what happens to the data
A plain guide to the data behind the screen
Asking an AI chatbot about your health: what it is good for, and what it is not
A plain guide to what the machine does