Step into the innovative world of healthcare at the Kenya Medical Research Institute, where a dedicated team led by Dr. Videlis Nduba is pioneering a revolutionary mobile phone application empowered by AI to revolutionize the diagnosis of tuberculosis and other respiratory ailments.
In a serene, specially equipped room, Dr. Nduba and his team meticulously record coughs from individuals afflicted with respiratory illnesses, including tuberculosis, alongside those who are healthy. Their mission? To develop sophisticated software capable of distinguishing between these coughs, ultimately giving rise to a mobile app that can reliably identify a TB-related cough and other potentially life-threatening conditions.
Utilizing an array of microphones, ranging from affordable models to high-definition versions, even harnessing the power of smartphones, they capture both natural and induced coughs. These recordings are then dispatched to the University of Washington, where they undergo rigorous analysis via an advanced computer software system known as ResNet 18.
Dr. Nduba elaborates, "Our software harnesses the power of artificial intelligence to dissect and interpret these coughs, transforming them into mathematical representations known as cough spectral grams. This allows us to discern subtle differences between the coughs of individuals with TB and those without, paving the way for swift and accurate diagnoses."
With the potential to significantly reduce the time it takes for patients to receive vital diagnoses and treatment, Dr. Nduba envisions a future where the spread of tuberculosis is curtailed. "By expediting the diagnostic process, we can mitigate the transmission of TB within communities. Time is of the essence, as individuals become infectious upon developing symptoms. Rapid identification through our software could be a game-changer in the fight against TB," he emphasizes.
However, despite promising progress, the software has yet to meet the stringent criteria set forth by the World Health Organization (WHO). WHO guidelines stipulate that the application must achieve a minimum accuracy of 90% in detecting TB infections and 80% in ruling out non-infections.
While initial trials have yielded encouraging results, with an 80% accuracy rate in detecting TB and 70% accuracy in excluding TB, there is still work to be done to meet the WHO standards. Yet, Dr. Nduba and his team remain undeterred, driven by their unwavering commitment to harnessing technology for the betterment of global health.