Updated: Jul 8, 2020
AI is not only making TB diagnosis accessible to people who were previously deprived of it, but by reducing radiologist effort required in reading X-rays, it is lowering healthcare costs and improving reporting times.
Tuberculosis is curable, and yet, it kills more people than any other infectious disease in the world. India has a quarter of the world’s TB population, which is more than any other country. Prime Minister Modi has promised to make India TB-free within the next five years. World Health Organization wants to eradicate the disease globally by 2030. Around USD 13 billion are pumped every year into this effort, and India has become a battleground for TB eradication. Despite all this, it is estimated that 27 lakh people in India were affected by TB in 2018. Although we are nowhere close to eradicating TB, we have now found a powerful ally that can help us win this fight: Artificial Intelligence.
To cure a person who has TB, you must first find this person, and that is where the challenge lies. TB does not cause mass epidemics. It silently affects individuals who remain unaware that they have the disease and go about their daily lives spreading it to others who come in contact with them. Instead of waiting for patients to show up at hospitals before being treated, the government has started active case-finding where health workers proactively screen vulnerable population groups for TB. The most accurate test to determine whether a person has TB is the GeneXpert test. Due to the high costs associated with this test, however, it is not affordable to use it for mass screening.
Chest X-rays have been commonly used in public screening drives for many decades. They were used in Japan for identifying TB patients in the years after the Second World War. Many countries including India have adopted mobile diagnostic vans equipped with X-ray equipment which drive around taking X-rays of hundreds of patients every day. There is however one major problem. Reading an X-ray requires qualified radiologists. Where are the radiologists to interpret these hundreds of X-rays?
In India, there is only 1 radiologist for 100,000 patients, and most of these radiologists are concentrated in the cities. Due to this acute shortage of radiologists, there are large and unacceptable delays in reading X-rays and reporting their findings. Patients often have to wait for weeks before their reports come in. This delay means that not only patients lose valuable days of life-saving treatment, but they continue to spread the infection during this period. If we want to eradicate TB, we must provide timely and accurate diagnosis to people of all demographics. This is where artificial intelligence is a game-changer.
Artificial intelligence has made astonishing progress in the past decade. AI models can now drive cars, predict stock prices, guess emotions, play board games, converse like humans, and compose poetry. Faster computers, availability of large medical datasets, and development of new algorithms has ushered in a revolution in using AI for medicine. AI has demonstrated expert-level performance in radiology tasks like diagnosing breast cancer from mammographs, segmenting brain tumours from MRI scans, and — you guessed right — in diagnosing tuberculosis from chest X-rays.
These AI systems are developed using neural networks, a technique inspired from biological neurons in human brains that enables computers to automatically learn to perform tasks. A neural network is fed a large dataset of X-ray images along with a label whether the person has TB or not. The neural network processes this dataset and extracts patterns in the images that allow it to predict the probability that a given patient has TB. Under certain conditions, AI models have demonstrated prediction performances that match or even exceed that of expert radiologists.
The Revised National Tuberculosis Control Program (RNTCP) — renamed as National Tuberculosis Elimination Program (NTEP) in December 2019 — was set up to provide universal access to TB diagnosis and treatment. Under this program, the government has begun conducting field tests for deploying AI solutions to counter the shortage of radiologists. A prominent example is Chennai where the Greater Chennai Corporation has collaborated with Clinton Health Access Initiative (CHAI) under the STOP TB Partnership to procure, maintain, and operate a fleet of mobile diagnostic vans. These vans capture thousands of X-rays every day. The X-ray images are uploaded to the cloud and processed by an AI-driven solution developed by DeepTek, a leader in smart solutions for medical imaging.
When asked if AI will replace radiologists, Dr. Amit Kharat, one of the co-founders of DeepTek, says, “Radiology is a vast field. Human radiologists are far better than AI at most tasks, while AI has shown exceptional performance at some tasks. We need to be talking more about combination or augmentation rather than substitution here. AI will assist human doctors in making critical life-saving decisions. And more importantly, AI has the potential to reach places where doctors are not present.”
Besides India, many countries in Africa and South East Asia have also adopted AI solutions to address the shortage of radiologists. AI is not only making TB diagnosis accessible to people who were previously deprived of it, but by reducing radiologist effort required in reading X-rays, it is lowering healthcare costs and improving reporting times.