At Praxium Labs we build this for Nepali businesses every month; this is the field-tested version. Rural Nepali clinics have X-ray machines but rarely radiologists. Computer vision can't replace doctors but it can triage — flagging probable abnormalities for urgent specialist review and giving the on-site provider a higher-confidence read on routine cases.
Chest X-ray triage
TB remains a meaningful public-health burden in Nepal. WHO-validated AI chest X-ray models (qXR, CAD4TB, Lunit INSIGHT CXR) achieve TB-detection sensitivity 90–98% with specificity 70–90% on field-quality images. In Nepali rural clinics the workflow is: technician takes the chest X-ray, AI flags as "TB-probable / TB-unlikely / other-pathology", positive cases go to radiologist (often tele-consultation) for confirmation.
Diabetic retinopathy from smartphone
Adapter lenses (Peek Acuity, EyeWatch, etc.) turn a smartphone into a retinal camera. AI models classify retinopathy severity (no DR, mild, moderate, severe, proliferative) with accuracy comparable to junior ophthalmologists. Useful for screening campaigns in rural districts where ophthalmologist presence is rare.
Dermatology triage
Skin-lesion photos classified by AI for "likely benign", "consult dermatologist", "urgent referral" can substitute for the 6-month wait for dermatology at Kathmandu tertiary hospitals. Accuracy on photographic-quality images is acceptable for triage but not for diagnosis. Always frame outputs as triage, never diagnosis.
Where CV does NOT belong in Nepali healthcare today
- Diagnostic decisions without human confirmation: regulatory and ethical line
- Treatment recommendations: always the doctor's call
- Pathology slides: high-stakes, low data volume in Nepal
- Mental health diagnosis: wildly out of scope
Regulatory framing
Nepal Medical Council has not yet issued AI-medical-device specific rules. CE-marked / FDA-cleared AI imaging products are routinely used in tertiary hospitals; for rural / public-health deployments the typical framework is: (a) the AI is a "screening assistive technology" not a diagnostic device, (b) every positive flag is reviewed by a credentialed practitioner, (c) the deployment is documented to the relevant district health office.
Data and partnerships
Training data for Nepali-specific models is scarce. Practical paths: (1) use globally-trained models and validate locally on a few thousand cases, (2) partner with a teaching hospital (Patan Academy, BPKIHS, Tribhuvan University Teaching Hospital) for annotated data, (3) participate in WHO / global TB-screening initiatives that bundle Nepali data into international datasets. For related context, see our AI for Disaster Response in Nepal: Earthquake, Landslide, Flood post.
Deployment economics
- X-ray AI: commercial licences ~$1–3 per X-ray (NPR 130–400), volume discounts
- Self-hosted alternative: open-source models (TorchXRayVision) free to use but require local GPU and validation
- Smartphone CV: on-device inference free; bandwidth for upload-to-cloud cheaper but assumes connectivity
- Cost per screening (X-ray, TB): NPR 200–500 fully loaded (licence + radiologist confirmation)
- Cost per case detected: dramatically lower than traditional sputum testing in low-prevalence settings
Frequently asked questions
Are Nepali hospitals already using this?
A handful of tertiary hospitals in Kathmandu use CE-marked X-ray AI products. Rural deployment is mostly donor-funded pilots — World Bank, USAID, WHO Global Fund. Mainstream adoption is 2–5 years out, faster if NMC issues supportive guidance.
Can a small NGO actually deploy this?
Yes — a well-funded health NGO can deploy a TB-screening program across 5–20 clinics for under NPR 1 crore over 18 months, including hardware, AI licences, and training. Several already do.
What about offline / no-internet operation?
Models can run on-device for some use cases (smartphone-based screening, simple X-ray triage on a local server). Cloud-hosted models require connectivity which is unreliable in rural Nepal. Always design for offline-first.
Will doctors push back?
Properly framed as a "second opinion" or triage tool, doctors mostly welcome AI. Pushback comes when AI is framed as replacing clinical judgement or used to justify reducing physician roles.
Is patient consent needed?
Yes — informed consent for AI-assisted screening is standard. Nepali clinical-research ethics committees expect explicit disclosure that AI is involved. Patient data must remain in-country if Nepali Privacy Act applies (sensitive health data category).
Who can build this in Nepal?
Praxium Labs — Nepal's AI and automation consultancy, based in Lalitpur — designs and builds the systems described in this guide for Nepali businesses and for international teams hiring from Nepal. Start a project or see all services.