AI in the early detection of TB among people living with HIV

Overview

An AI-powered diagnostic tool developed in Tanzania is helping clinicians detect tuberculosis in HIV patients from chest X-rays with 92% accuracy potentially transforming TB diagnosis in some of the world's most under-resourced health settings.

TB remains one of the leading causes of death among people living with HIV, yet accurate diagnosis in high-burden settings is consistently hampered by limited specialist capacity and overloaded health systems. In Tanzania, as across much of sub-Saharan Africa, delays in TB detection translate directly into preventable deaths.

To address this, researchers developed a deep learning model using Convolutional Neural Networks (CNNs) trained on chest X-ray images from approximately 3,000 patients across public health clinics in Tanzania. The model identifies TB cases with 92% accuracy, supporting clinicians with faster, more consistent diagnostic decisions without requiring specialist radiologists on-site. To ensure the tool is trustworthy and explainable, a heat map visualisation was integrated to show clinicians exactly which parts of an X-ray influenced the model's output, building confidence in AI-assisted decisions.

With TB-HIV co-infection rates remaining critically high across the region, a scalable, accurate, and explainable diagnostic tool of this kind could meaningfully reduce diagnostic delays, improve treatment outcomes, and ease the burden on overstretched health workers offering a responsible AI model for clinical deployment across high-burden, low-resource settings.