From Maps to Households: How AI-NETRA's Three Parts Work

AI-NETRA is a mapping and planning platform that turns the data a tuberculosis programme already collects into decisions that frontline workers, district officers, and programme managers can act on. It does not stop at the district or block. It ranks risk down to small neighbourhoods, so a screening plan points to a real place rather than a general area. Three parts work as one process. One part decides where screening should happen. A second decides where each collected sample should go for testing. A third tracks progress against defined targets.

Where to screen

The first part, AI-ACF, uses routine notification data from Nikshay, along with population, environment, and building information, and feedback from frontline workers. It scores each small area of the district for tuberculosis risk, ranks those areas by expected yield, and groups the highest-priority areas into field-ready plans.

District supervisors see a single view of the district, ranked by expected yield, which can be sent to a frontline worker's device. In the Pune evaluation, informal residential settlements, the areas most likely to be missed by block-level planning, accounted for 48 percent of the highest-priority clusters.

For the ASHA or tuberculosis health visitor doing the screening, the output is not a heat map. It is a named, ordered list of places to visit, in the local language, with satellite imagery, turn-by-turn directions, and photographs of the location. Over a two-week field evaluation across four tuberculosis units, AI-guided teams screened 66,095 people across 65 high-priority neighbourhoods and confirmed 10 laboratory-verified pulmonary cases, at about three times the rate per person screened seen under standard practice.

Where to send samples

Finding a person who needs a test is incomplete if the sample then waits too long, or travels too far, to reach a laboratory that can process it. AI-DNO models Pune's testing network of 49 devices across 43 laboratories, handling more than 53,000 tests, and finds a better route for each sample given laboratory capacity and distance.

Without moving existing devices or buying new equipment, redesigning referral routes alone reduced the average distance a sample travels by about 25 percent, eased pressure on the district's only over-capacity laboratory, and is projected to save about Rs 22 lakh a year.

How to track progress

Most tuberculosis programmes already track the indicators that matter: how many people are examined, how many cases are notified, how productive each cluster is, and how efficient screening is. That information is often spread across records that take time to reconcile. AI-KPI brings these indicators into one system, keeps the numbers and sources behind every figure, and can answer, in ordinary language, questions about a unit's progress toward its TB Mukt Bharat target.

How data is handled

AI-NETRA uses deidentified Nikshay records and open mapping layers, including building footprints, street data, and slum deprivation polygons. Data flows into the system only. Nothing is shared with external providers. The platform is designed to align with the Digital Personal Data Protection Act, 2023, and with recognised standards for medical device quality, risk management, and artificial intelligence governance.

From data to door-to-door screening

The sequence is straightforward. Programme data is taken in. A risk score is produced for each neighbourhood. An expert-guided system groups areas into ranked clusters. Frontline review confirms the plan. The confirmed plan is issued as a door-to-door list, and each collected sample is routed to a capable laboratory by AI-DNO. The plan is rebuilt each cycle as new data arrives.

What comes next

Further work aims to extend screening guidance to areas with little or no notification history. Those areas can be missed by a model that learns mainly from past notifications. Environmental and population signals are being explored as a complementary input. This extension is still directional and needs further validation. More evidence on the platform's performance is available on the Pune Pilot Study page. Interested programmes may also request a demonstration.