Kenya: Africas, By Baher Tarek
At the Kenya Medical Research Institute, researchers are developing a mobile phone application that utilizes artificial intelligence to diagnose tuberculosis and other respiratory disorders.
In addition, Dr. Videlis Nduba and his team record coughs from both persons with respiratory disorders like tuberculosis and those who do not have the condition in a specially designated silent area.
The goal is to produce software that can distinguish between the two and a mobile phone application that can reliably identify a cough associated with tuberculosis and other dangerous diseases. In addition, natural or forced coughs are recorded using three microphones: a low-cost version, a high-definition version, and a smartphone microphone. The data are submitted to the University of Washington, where they are processed through an existing computer software system known as ResNet 18.
Nduba, the primary researcher, explains: “This software employs artificial intelligence to analyze coughs, which we refer to as cough spectral grams, and then regresses them. So, it is a mathematical method of modeling the cough image in order to assess whether there is a difference between someone with tuberculosis coughing and someone who does not cough. Furthermore, Nduba believes that if the software can be demonstrated to operate effectively in trials, it will reduce the time it takes for a patient to receive a diagnosis and treatment, so helping to reduce the spread of tuberculosis.
Nduba, says that the most significant achievement is the reduction in time to diagnosis. So, the typical duration between when someone develops TB symptoms and when a doctor confirms they have TB and need treatment can range from 3 to 2 months to a year. And when they are in the population, they are infectious and transmit tuberculosis. When they acquire a cough, exposing them to this software and determining whether it is TB will limit TB transmission in the community, and transmission is responsible for a large portion of TB cases.
However, the program is currently insufficiently accurate to meet the World Health Organization’s standards. On the other hand, according to the WHO, the application must be at least 90% accurate in recognizing a tuberculosis infection and at least 80% accurate in detecting the absence of an infection. So far, Nduba’s trials have yielded 80% accuracy in detecting tuberculosis and 70% in detecting the absence of tuberculosis.
