These algorithms could end the scourge of tuberculosis

In some of the most remote and poorest corners of the world, where respiratory disease abounds and skilled healthcare professionals are afraid to walk, the diagnosis is increasingly being fueled by artificial intelligence and the internet.

In less than a minute, a new app on a phone or computer can scan an x-ray for signs of tuberculosis, Covid-19 and 27 other conditions.

Tuberculosis, the world’s deadliest infectious disease, killed nearly 1.4 million people last year. The app, called qXR, is one of many AI-based tools that have emerged in recent years for tuberculosis screening and diagnosis.

The tools offer hope in detecting the disease early and reducing the cost of unnecessary lab tests. Used on a large scale, they can also spot emerging clusters of diseases.

“Of all the AI ​​applications, I think digitally interpreting an image using an algorithm instead of a human radiologist is probably the most advanced,” said Madhukar Pai, director of the McGill International Tuberculosis Center. to Montreal.

Artificial intelligence cannot replace clinicians, have warned Dr Pai and other experts. But the combination of AI and clinical expertise is proving powerful.

“The machine plus the clinician is better than the clinician, and it’s also better than the machine alone,” said Dr. Eric Topol, director of the Scripps Research Translational Institute in San Diego and author of a book on the use. of AI in medicine.

In India, where about a quarter of the world’s TB cases occur, an app capable of reporting the disease in remote areas is urgently needed.

The Chinchpada Christian Hospital in Nandurbar, a small town in northwest India, serves members of the Bhil tribal community, some of whom travel up to 200 km to visit the center. The 50-bed hospital has eight doctors and only the most basic medical equipment.

Across the country, Simdega, one of India’s 20 poorest districts, is isolated from the nearest town, Rourkela, by nearly five hours of driving over bumpy roads. The tribal people of the district live in tiny hamlets surrounded by dense and evergreen forest. Simdega Medical Center, which has 60 beds and three doctors, is in a forest glade – “literally in the middle of nowhere,” said Dr George Mathew, the director.

The lean staff have to deal with everything that comes along, “from malaria and myocardial infarctions to seizures and head trauma,” said Dr Mathew. Over the years he has learned to read x-rays, and when he is perplexed he calls in radiologists among his distant friends and former colleagues.

Although Nandurbar and Simdega are over 800 miles apart, their populations are surprisingly similar. Malaria, sickle cell disease and tuberculosis are rife among them, compounded by poverty, the use of spiritual healers, and alcoholism – even among children.

“Tuberculosis tends to be overlooked and diagnosis is often delayed,” said Dr Ashita Singh, chief of medicine at Nandurbar Hospital. By the time people arrive at these medical centers, they are often “very, very sick and have never even been evaluated elsewhere,” she said.

But in some patients, x-rays carry signs that are too subtle for a non-expert to detect. “It is in this group of patients that AI technology can be of great benefit,” said Dr Singh.

The arrival of the coronavirus – and the lockdown that followed – cut off these remote hospitals from the nearest towns, as well as radiologists. It also delayed and complicated diagnoses of TB, as both diseases affect the lungs.

A few months ago, the two hospitals started using qXR, an app created by Indian company Qure.ai and funded by the Indian government. The app allows the user to scan an x-ray. If it finds evidence of tuberculosis, it assigns the patient a risk score. Doctors can then perform confirmatory testing on patients at highest risk.

At Nandurbar Hospital, the app diagnosed 20 patients with tuberculosis in October, Dr Singh said.

Applications like qXR may also be useful in settings where the prevalence of TB is low and for routine screening of people with HIV, who are at high risk of contracting TB, as well as those with other conditions. said experts.

“Most chest x-rays for people with suspected tuberculosis are read by people who are not remotely expert in their interpretation,” said Dr. Richard E. Chaisson, a tuberculosis expert at the University. Johns Hopkins. “If there was an AI software package that could read X-rays and CT scans for you in a remote emergency room, that would be a huge step forward.

qXR is among the most promising AI-based applications for the detection of tuberculosis. The company that created the app didn’t realize this potential until a doctor at an Indian hospital suggested it a few years ago.

In studies comparing different AI applications conducted by the Stop TB partnership, all AI applications outperformed experienced human readers, and qXR seemed to fare better.

The app identifies tuberculosis with 95% accuracy, according to Qure.ai CEO Prashant Warier. But this level of precision is not based on real world conditions, which Dr Topol called “a common problem” with AI-based applications. A tuberculosis control program may be less precise in the United States or Western Europe than in India, because the prevalence of the disease is lower in those regions, Dr Topol added.

The app has only been tested in adults, but is now used in children 6 years and older. Chest x-rays are particularly useful for pediatric tuberculosis because about 70% of cases in children cannot be confirmed by laboratory tests, said Dr Silvia S. Chiang, an expert in pediatric tuberculosis at Brown University.

“There is a huge shortage of trained professionals who feel comfortable interpreting pediatric chest x-rays,” she said, “the development and validation of computer-assisted reading technologies in children would help greatly. . “

Qure.ai said it is testing its application in children in Bangladesh and will release the data early next year. In the meantime, qXR and other apps will continue to improve as they learn as they go.

“The more you feed the beast with x-rays, the better it gets,” Dr. Pai said.

Experts were optimistic that AI-based applications could have a huge impact in the fight against TB, especially in countries like India that lack medical resources.

“I just dream of a time when something like this would be available to all the small public sector primary and secondary health care centers that are reluctant to do x-rays because they don’t have the confidence to do them. read, ”says Dr. Singh. “If this were to be made available to all radiology centers in rural India, I think we could beat TB.”