Researchers at the Aga Khan University School of Nursing and Midwifery, East Africa, are leading the Kikohozi Classifier Project to develop an artificial intelligence (AI) system that can identify respiratory diseases such as tuberculosis (TB), asthma, and chronic obstructive pulmonary disease (COPD) through cough sounds. The project, led by Associate Professor Dr. Kahabi Isangula, aims to collect recordings from at least 30,000 adults, including 5,000 people with each of TB, asthma, and COPD, and 15,000 healthy controls.
The AI system is being trained to distinguish between cough sounds from people with and without respiratory diseases, and eventually, between different respiratory conditions. The project has already collected data from 12,936 participants, including 2,739 people with TB, 2,658 with asthma, 963 with COPD, and 6,465 healthy controls, generating 69,940 cough sound samples. This data is being collected across Dar es Salaam, Kilimanjaro, Iringa, Dodoma, and Shinyanga, with plans to expand to Mwanza.
Tanzania faces a significant burden of TB, with the World Health Organisation (WHO) estimating that about 118,000 people developed TB in 2024, resulting in 23,500 deaths. The country remains among those with a high TB burden, despite making substantial progress in reducing TB deaths since 2015. The Kikohozi project has the potential to be a valuable tool in addressing this issue, as well as the growing challenge of non-communicable respiratory diseases such as asthma and COPD.
Asthma and COPD are among the most common chronic respiratory diseases globally, affecting over half a billion people. COPD alone caused an estimated 3.4 million deaths in 2023. The Kikohozi Classifier project could provide a useful screening tool for these conditions, allowing for earlier identification and treatment. According to Dr. Isangula, the system could also help address TB stigma by making the first interaction a general respiratory-health screening rather than a test labelled specifically as a TB test.
Existing cough-AI systems developed elsewhere cannot be assumed to work equally well in Tanzania, as the sound of a cough can be affected by age, sex, environment, recording conditions, and other characteristics. Therefore, the project is building a large, Tanzania-specific database. This is consistent with global evidence, which highlights the importance of locally developed and context-specific solutions.
The Kikohozi Classifier is not intended to replace established clinical assessments such as GeneXpert, sputum testing, or spirometry. Instead, its value lies in providing earlier screening and decision support, potentially telling someone that they are more likely to have a respiratory condition and should seek further testing. A smartphone-based tool could eventually allow a person to record a cough without providing blood or sputum, reducing some of the barriers associated with conventional screening.
The project raises a bigger question: can locally developed artificial intelligence help take screening closer to people rather than requiring people to travel to health facilities first? The Kikohozi Classifier has the potential to increase access to screening and diagnosis, particularly in rural or underserved areas. As the project continues to collect data and develop the AI system, it could provide a valuable tool in the fight against respiratory diseases in Tanzania and beyond.
Key points
- The Kikohozi Classifier project aims to collect recordings from at least 30,000 adults to develop an AI system that can identify respiratory diseases through cough sounds.
- The project has already collected data from 12,936 participants, generating 69,940 cough sound samples.
- The AI system could provide earlier screening and decision support, potentially telling someone that they are more likely to have a respiratory condition and should seek further testing.
