Study finds AI could lower TB screening costs in rural Philippines

A study by Ateneo researchers suggests that artificial intelligence could make tuberculosis screening less expensive in rural parts of the Philippines, where access to medical specialists remains limited. Its findings do not support immediate nationwide adoption, but point to targeted trials in communities that struggle to obtain timely radiology services.

The research, reported by medicalxpress.com, examined the economic value of using AI to interpret chest radiographs taken during tuberculosis screening. The findings were published in the August 2026 issue of BMC Health Services Research by Harold Henrison Chiu, Bryan Christopher Lao and Gloanne C. Adolor.

Tuberculosis is a major public health concern in the Philippines. The World Health Organization estimated that 739,000 people in the country developed the disease in 2024. That represented 6.8% of the 10.8 million tuberculosis cases recorded globally that year.

Early diagnosis can allow treatment to begin before the illness becomes severe or causes permanent damage. Yet patients in geographically isolated areas may face several obstacles even after reaching a facility capable of taking an X-ray. A radiologist or teleradiology service may not be available promptly, forcing patients to make another journey, spend more money or lose time at work. Delays can also interrupt the process of getting follow-up care.

What the model compared

The researchers built a decision model around a hypothetical group of 1,000 patients with suspected tuberculosis. Each patient was assumed to undergo chest radiography at a rural health unit, and the analysis followed projected costs and outcomes over five years.

The comparison involved two approaches: AI-assisted interpretation and conventional manual reading. The researchers included the price of AI software, operating costs, radiologist interpretation fees and confirmatory GeneXpert testing. The latter remains necessary because an AI-supported X-ray assessment does not by itself establish a definitive tuberculosis diagnosis.

Under the study’s main assumptions, the AI-supported approach would cost about Php 877,330 each year for the 1,000-person group. Manual interpretation was estimated at approximately Php 1.14 million. On a per-patient basis, the figures worked out to roughly Php 877 with AI assistance and Php 1,142 with manual reading.

The model therefore found the AI strategy to be the cheaper option. The researchers’ argument, however, extends beyond a comparison of software and professional fees. In places where specialist expertise is scarce, AI could provide a way to make radiological support available closer to patients rather than requiring every image to wait for a distant expert.

Why local conditions matter

The potential benefit depends heavily on how the technology is introduced. The researchers said AI could be incorporated into existing tuberculosis programmes through portable digital X-ray equipment and systems able to function with limited internet connectivity. Such an arrangement could bring screening to underserved communities and reduce the geographic barriers that make diagnosis more difficult.

The results were not uniformly favourable under every set of assumptions. When the analysis used a lower fee for manual or teleradiology interpretation, or applied diagnostic performance estimates from a Philippine setting, AI continued to be more effective but was not always the least expensive option.

That finding limits how broadly the headline cost comparison can be applied. The study relies on a theoretical patient cohort and assumptions about expenses and diagnostic accuracy. Actual results could vary between health units depending on staffing, connectivity, equipment, reading fees and the performance of the AI system in local conditions.

The researchers also stressed that AI would not remove the need for confirmatory testing. A positive or suspicious chest X-ray assessment would still need to be followed by GeneXpert testing, adding both expense and a further step to the diagnostic pathway.

A case for carefully tested pilots

Rather than recommending an immediate nationwide rollout, the study calls for pilot projects in selected underserved rural health units. Those trials could test whether the projected savings and diagnostic benefits appear in routine practice, while also identifying technical or operational problems that may not be visible in a theoretical model.

The proposed approach includes local validation, quality assurance, ongoing monitoring and an assessment of available budgets. These safeguards would help determine whether AI can work reliably within existing tuberculosis services instead of becoming a separate technology initiative with uncertain long-term support.

For the researchers, the central issue is not simply whether a machine can match or assist a radiologist. It is whether expert-level support can reach communities where specialist care is scarce, at a cost health services can sustain. In that sense, the study presents AI as a possible tool for reducing unequal access—but only if its deployment is shaped by local evidence and the needs of rural patients.