Active Flagship program

Case-finding, fixed at the reading bottleneck

Afghanistan carries one of its region’s heaviest tuberculosis burdens, and chest X-ray is the most sensitive screening tool it has — when someone reads the film. This program exists so that every screening film gets a competent read.

Under 15 — human-read

every film goes to a radiologist; no software clears a child’s film

15 and over — software may read first

WHO’s conditional recommendation for CAD in TB screening and triage; abnormal and ambiguous films still go to a radiologist

WHO’s line at fifteen, drawn to scale across the ages. Software may take the first read only above it, and only for screening and triage; below it, and for anything that is not TB, a radiologist reads.2

About 73,000: people estimated to fall ill with TB in Afghanistan each year (WHO, 2020 estimate). 1
Over 10,000: estimated TB deaths in Afghanistan every year (WHO EMRO). 1
15 and over: the age band in which WHO recommends CAD software for TB screening and triage of chest X-rays. Below it, every film is human-read. 2

The films are already being taken

Chest radiography is the most sensitive screening tool for pulmonary tuberculosis in WHO’s screening toolkit2 — it finds disease before symptoms are reported, which is exactly what case-finding needs. Afghan clinics own the machines and take the films; the estimated incidence numbers3 say the disease is there to find. What is missing, at almost every site, is the reading. Screening programs stall not at exposure but at interpretation.

That makes TB the right first program for a reading platform: the burden is heavy, the screening tool is the one modality clinics already produce, and — uniquely among the reads this platform handles — the evidence supports software carrying the routine first pass.

Architecture

The reading cascade

The TB program’s reading cascade. The dashed path is disabled today: automated clearance of CAD-confident normals exists in the architecture and stays off until it passes shadow validation on local films under a signatory medical director. Until then — and always, for children and non-TB patterns — humans read, and a licensed radiologist signs every report the service delivers.

CAD first read — the workhorse: WHO-recommended computer-aided detection software scores every screening film for signs of pulmonary TB. Since 2021, WHO’s screening guidelines state that such software may be used in place of human readers for screening and triage of digital chest X-rays in people aged 15 and over — a conditional recommendation, for a validated class of products. It is the only reading tier that scales without a payroll, and it aligns this program with how TB screening is actually run today.
Human adjudication — everything CAD cannot clear: Abnormal films, ambiguous scores, every patient under 15, and any film suggesting non-TB disease escalate to a licensed radiologist, who reads and signs. This is also where academic partner programs plug in: adjudication, quality assurance on samples of cleared films, and co-authorship on what the data shows.
Contracted surge capacity — the quiet backstop: When volume outruns the first two tiers, contracted teleradiology capacity stands by as a paid backstop, funding permitting. It borrows no authority and headlines nothing — it exists so a screening queue never silently stalls.

Limits

What we will not claim

CAD here is a triage tool, not a diagnostician. “Our AI diagnoses TB” is a sentence you will not read on this site: a positive screen is followed by bacteriological confirmation, per WHO guidance.

WHO’s recommendation covers people aged 15 and over, for screening and triage only. It does not cover children, and it does not cover distinguishing non-TB disease — the human tier exists precisely for what CAD is not validated to do.

WHO recommends a class of tools; it has not evaluated or endorsed Partaw, and nothing here implies otherwise.

We will publish performance figures when we have peer-reviewed results on our own films — and none before. Today we have an architecture and a study design, not accuracy claims.

The study

Questions only this setting can answer

We have found no prospective description of WhatsApp-based teleradiology TB case-finding in rural Afghanistan in the literature — the study we are designing with academic partners would be the first. Among the questions it is positioned to answer:

How well do CAD abnormality thresholds, calibrated on digital radiographs, transfer to smartphone photographs of physical films — the image reality of most of the world?
What is the true yield of CAD-first triage in a rural Afghan screening population?
How should the under-15 pathway perform, and what does pediatric case-finding look like when every child’s film is human-read?
Does a signed report returned over WhatsApp measurably reduce loss to follow-up between screening and confirmation?

Partner on the TB program Authorship, advisory & dataset offers

References

1 World Health Organization, Regional Office for the Eastern Mediterranean — Afghanistan tuberculosis programme. WHO estimated roughly 73,000 people fell ill with TB in Afghanistan in 2020; WHO EMRO reporting puts annual TB deaths above ten thousand. www.emro.who.int/afg/programmes/stop-tuberculosis.html
2 World Health Organization (2021), “WHO consolidated guidelines on tuberculosis. Module 2: Screening — systematic screening for tuberculosis disease.” Conditional recommendation: in individuals aged 15 years and older, computer-aided detection software may be used in place of human readers for interpreting digital chest X-rays for TB screening and triage. www.who.int/publications/i/item/9789240022676
3 World Health Organization, Global Tuberculosis Report — Afghanistan country profile (estimated incidence on the order of 180–190 per 100,000). www.who.int/teams/global-tuberculosis-programme/tb-reports