The scourge of tuberculosis (TB) may be largely a distant memory for most Americans and Europeans, but it killed roughly 1.25 million people last year around the world. A non-profit based in India, which accounts for more than a quarter of all cases, is developing AI tools that could boost efforts to eradicate the disease.
Roughly 10 million people a year fall ill with TB, making it one of the world’s most prevalent infectious diseases. In 2018, Indian Prime Minister Narendra Modi made an ambitious pledge to eliminate TB in India by 2025. With 2.5 million cases recorded in India last year, that goal clearly won’t be met; still, the country has invested hundreds of millions of dollars in a vast national TB program, and has reduced the disease’s incidence by about 18 percent between 2015 and 2023.
But diagnosing and treating the disease is a complex and lengthy process. TB is curable, but requires patients to undergo a strict six-month regimen of antibiotics. Any deviation from this program can quickly lead to drug resistance, so patients require constant supervision. And the gold standard approach to screening for the disease is chest X-Rays, which are difficult to conduct in the rural parts of India where TB is most common.
That’s why Indian non-profit Wadhwani AI has developed a suite of AI-powered tools to assist health workers detect undiagnosed cases, decide on treatment plans, and prevent people from dropping out of treatment. Working with the Indian government and the U.S. Agency for International Development, the organization is currently piloting these tools across the country. And Wadhwani’s director of solutions, Nakul Jain, says 2025 could see several incorporated into India’s national TB patient management system, Nikshay.
“We have seen some very encouraging results in the first few phases that we have launched and the government is supportive,” says Jain. “We are hoping that most of the solutions will now be integrated into the mainstream Nikshay application and they’ll be used across the country.”
Raghuram Rao, who works on TB programs at India’s Ministry of Health and Family Welfare, says that to meet the government’s ambitious goal of TB eradication, the ministry has embraced digital solutions. “Wadhwani AI’s innovative AI solutions, coupled with robust analytical support, have enhanced the program’s efforts to combat tuberculosis,” says Rao. “By enabling data-driven decision-making and aiding programmatic efficiency, Wadhwani AI has significantly supported us in the mission to end TB in India.”
How AI helps with diagnosing TB
One of the most promising tools Wadhwani has developed is an AI application that can detect potential TB infections from the sound of a patient’s cough. In India, health workers typically rely on symptom reporting to identify potential cases, and then send those people for a confirmatory X-ray, says Wadhwani’s chief AI scientist Alpan Raval. But the group wanted to see if they could use AI to identify cases before the patient even shows signs of the disease.
To do this, they had to collect a large dataset of cough sounds to train their models, and it was crucial that the data set included both symptomatic and asymptomatic patients, says Raval. With the help of The National Tuberculosis Elimination Programme, Wadhwani gathered 36,000 cough sounds from patients visiting X-Ray clinics for all manner of ailments, not just TB, which provided data from people with and without the disease. The researchers also took recordings from family members and close associates of TB patients who had yet to exhibit signs of infection. The sounds were collected through smartphones, says Raval, because the team knew that the solution wouldn’t scale if health workers had to rely on specialist equipment.
“We want to work backwards from usage, so we use the most basic smartphones to collect these sounds and we make sure there’s noise in the background that’s realistic,” he adds. “A lot of the work goes in building these models to make sure these solutions can be used robustly in the field.”
A health worker screens a patient for TB by collecting the sound of her cough.Wadhwani AI
The app works by converting the recorded audio into a visual frequency map, which is then processed by a computer-vision algorithm to predict if the patient has TB. In pilots across nine Indian states that started in early 2023, they found that the model detects 14 percent more cases than the existing screening approach, which suggests it is picking up asymptomatic cases. Their hope is to eventually make the model as accurate as an X-ray, which would drastically simplify screening efforts, though Raval says this will require much more data.
How AI helps with TB treatment
Once patients are diagnosed with TB, they’re typically given a test called a Line Probe Assay, which determines which drugs should be used to treat their particular strain of TB. The output of this test is a strip with a series of bands that indicate how the pathogen reacts to different medications. Normally, this test is interpreted by lab technicians and then uploaded to the Nikshay platform. But Wadwhani has created a tool that uses similar techniques to optical character recognition to automatically parse the test results, reducing errors and speeding up processing times.
Getting a patient into treatment is just the tip of the iceberg; ensuring that they stick to the lengthy treatment program is also a challenge, says Raval. Between 4 percent and 7 percent of patients drop out of the program before completing their course, he says, and even for those who stick with it, missing just one dose in ten increases the risk of developing drug-resistant TB by a factor of three.
Wadhwani AI’s Prediction of Adverse TB Outcomes tool shows a list of high-risk patients.Wadhwani AI
So Wadhwani has also developed an AI-powered system called Prediction of Adverse TB Outcomes (PATO), which predicts both the patients who will fall off the grid and those who will die from the disease. This can be used by health workers to identify people who need more intensive interventions, such as visiting their houses daily to make sure they’ve taken their medications.
Creating this tool was possible because of the scale of India’s TB problem, says Raval, which means the government has vast amounts of patient data. By law, each infection has to be tracked from diagnosis to the completion of treatment, and health workers collect detailed information including medical history, family situation, and economic status. Wadhwani used this data to train an AI model that uses the information collected at the outset of treatment for predictions. The tool has been piloted across 12 states since April 2023 and so far has reduced the number of adverse outcomes by 28 percent.
Using AI to decide who should be tested for disease and how they should be treated has inherent risks, Raval admits, but the organization ensures final decisions are always in the hands of humans. And in a country like India with a massive disease burden and an overstretched healthcare system, health workers need all the help they can get, he adds.
“We can’t get carried away by thinking that this is the Wild West and we can do anything we like,” says Raval. “But we need to compare against baselines. What’s best practice right now? And can AI make a significant difference [so] that the risks are worth it?”
From Your Site Articles
Related Articles Around the Web
