All projects

Public Health

PCOS Screening Support

Getting the right patients to the right clinic sooner.

PCOS is badly under-diagnosed, and the reason is structural. Confirming it needs imaging and laboratory work that exists at district level, while first contact happens at a primary health centre or with a health worker in the field. Women with clear symptoms go years without a referral. This is a referral aid — not a diagnosis — built to close that gap.

Line drawing of a field intake form of ticked indicators feeding a double helix, with a referral arrow to a district facility
Status — research prototype

This one is not deployed and should not be. It goes nowhere near a patient until it has been validated with a clinical partner and cleared by an ethics board. We are listing it because the problem is real and the approach is sound, not because it is ready. Any accuracy figure quoted before that validation would be meaningless.

Key Capabilities

Works from indicators a field worker already records
Routine vitals, blood values, cycle history and visible symptoms
Flags who should be prioritised for confirmatory testing
Scores a partially completed record rather than rejecting it
No new equipment, no laboratory, no extra visit
Designed for a non-specialist to operate
Interpretable models evaluated alongside the primary one
Built to report recall on the positive class, the metric a screen lives by
Single-patient scoring for use inside a consultation

How It Works

1

Intake

The indicators from a routine visit, as already written down

2

Prepare

Gaps handled so an incomplete record still scores

3

Screen

A likelihood, not a verdict, for this patient

4

Prioritise

Who to send for imaging and hormone work first

5

Confirm

A clinician at district level makes the diagnosis

Who It's For

Primary Health Centres
ASHA & ANM Field Programmes
State Health Departments
Women & Child Development
Medical Colleges & Research Partners

Intended Impact

  • Points limited specialist capacity at the patients most likely to need it
  • Turns a routine visit into a referral signal, with no new equipment
  • Shortens the years between first symptom and first referral
  • Earlier intervention on a condition linked to infertility and diabetes
  • Usable by a health worker with fifteen minutes per patient

Limits & Compliance

  • A referral aid, never a diagnosis — a clinician decides
  • Not a certified medical device and not validated for clinical use
  • Trained on one dataset, not a state-representative population
  • Any deployment touching patient records needs data-protection compliance and ethics approval
The metric that matters

For a referral screen, overall accuracy is close to meaningless — a model that calls everyone healthy can score well and refer nobody. The number to hold us to is recall on the positive class: of the women who do have it, how many did we flag. That is what we intend to publish once there is an honest evaluation behind it.

Deployment Models

Mobile (Offline Capable)
On-Premise
Hybrid (State Data Centres)

We are looking for a clinical partner

What this needs next is not more engineering — it is a hospital or medical college willing to validate it against real diagnoses. If that could be you, we would like to talk.