Replacing physicians has been cited as an aim of artificial-intelligence-based approaches to health care. Yet beyond the hype and hyperbole, there is a much more likely — and worthy — application of AI to medicine: infusing clinical expertise into regions where doctors are in short supply.

A key case in point: the global shortage of radiologists. The problem is particularly pronounced in low-resource settings, where both the diagnostic equipment and clinical experts to interpret imaging results are lacking. Consider that more radiologists work in the hospitals lining Longwood Avenue in Boston than all of West Africa. Yet wealthier nations are not immune to these challenges. For example, two analyses released in 2016 predict that European countries, including the U.K., will soon feel the squeeze of a shortage of radiologists and a growing tide of patients seeking CT and MRI scans.

Artificial intelligence (AI) can help fill this gap, and researchers are working on multiple fronts to develop AI-based applications for a range of important health conditions. Consider tuberculosis (TB), which is among the top ten causes of death worldwide. According to the World Health Organization, over 10 million people were sickened by TB in 2016 and nearly 2 million died. The vast majority of these deaths were in low- and middle-income countries.

Chest X-rays form a key linchpin of TB diagnosis. Although they are not sufficient to definitively diagnose the disease, they provide a cheap, rapid, and effective way of screening the lungs for TB-related abnormalities — particularly in areas where the disease is prevalent. Some progress has been made in improving global access to and affordability of X-ray machines, but many regions are still plagued by a lack of physicians with expertise in diagnostic radiology. That means patients often fail to get even the most basic screening tests, delaying TB diagnosis and treatment.

Now, various research teams, including ones in Texas and Pennsylvania, are harnessing deep learning methods to create automated tools for TB detection on chest X-rays. The accuracy of these models is quite high, approaching, and in some cases, even matching the performance of clinical experts. These efforts suggest that, by harnessing AI, it will soon become feasible to extend the reach of radiologists to places that currently lack care providers.

In a similar vein, investigators across the world are pursuing ways to transform the performance and interpretation of ultrasound technology. Many of these initiatives focus on the development of portable ultrasound devices that work in conjunction with smartphones or other handheld devices. This work is helping to improve portability and drive down the cost of ultrasound technology, which in turn, will help improve patient access, particularly in low-resource settings.

But price is not the only disruptor. Researchers in New York and Connecticut are developing a portable ultrasound device that harnesses semiconductor chips rather than piezoelectric crystals, which are the basis of conventional ultrasound machines. Moreover, the new device does not require multiple probes for imaging at different depths in the body — its single probe can carry out diverse functions, including fetal and obstetric exams, cardiac and peripheral blood vessel imaging, abdominal imaging, and musculoskeletal exams. This technology also incorporates a machine learning-based system to improve image acquisition as well as analysis. Together, these innovations could not only broaden the use of ultrasound — even expanding it to consumers — but also make it as routine and indispensable as stethoscopes and blood pressure cuffs are currently.

For more information about Dr. Kalpathy-Cramer’s research, please contact Partners HealthCare Innovation by clicking here.