Will Artificial Intelligence (AI) systems outsmart humanity and take over the world? It's not just a dystopian Hollywood fantasy anymore: some of the world's most serious technologists and thinkers—including Stephen Hawking, Bill Gates, and Elon Musk—have expressed concerns over the potential threat AI poses to our way of life.

Radiologists share these fears too, and many are concerned AI will replace their own expertise. There is a popular misconception that radiologists just read medical images, when in fact they are an integral part of cancer treatment and surgical teams, conduct patient-facing work such as biopsies, and can also treat patients directly. Radiology is one of the most diverse and important fields of medicine, but out of unwarranted fear and protectiveness, it's been hesitant to adopt AI.

AI will give radiologists more time to focus on other aspects of their work

These concerns are overblown, according to Reshma Suresh, head of operations for Qure.ai, an AI radiology and medical device company. Ms.Suresh emphasized that AI will improve radiologists' workflow efficiency by standardizing image-interpretation, allowing for a more accurate and faster diagnosis. Image reading and analysis can often be time consuming, particularly in low- and middle-income countries (LMICs) where there is a scarcity of radiologists and a heavy patient-load. And since the COVID-19 pandemic has taken off, the intensity of radiologists' workloads has only grown. By streamlining image-reading and assisting with other aspects of patient-care, AI will give radiologists more time to focus on other aspects of their work.

Companies like Qure.ai and Google Health's DeepMind support radiologists by automating radiological analysis. Their software uses machine learning to train algorithms to decipher computerized tomography (CT) scans, X-rays and magnetic resonance imaging (MRI) scans with the same, if not better, accuracy as a radiologist and at a much higher speed. In 2019, Google Health has launched a breast cancer AI based solution that has outperformed human radiologists by 11.5 percent in pre-identified data sets. According to Ms. Suresh, there is variability in radiological readings between readers. For instance, if two radiologists were provided with the same scan, their reading and ultimate diagnosis could be different in a few cases and, in the rare occasion, could miss an incidental finding It is exactly for this reason, she said, that AI systems will improve, not undermine or replace, the work of radiologists. These potential benefits are greatest LMICs, where a high burden of health often intersects with a lack of effective health-care systems and health agents. But the barriers to access these technologies are also higher in LMICs.

Boot used by medical staff are seen outside the coronavirus disease (COVID-19) ICU of Machakos Level 5 Hospital, in Machakos, Kenya October 28, 2020. — Boots used by medical staff are seen outside the coronavirus disease (COVID-19) ICU of Machakos Level 5 Hospital, in Machakos, Kenya October 28, 2020. REUTERS/Baz Ratner

Infrastructure Challenges

In low- and middle-income countries, where there are often limited resources and inadequate health-care infrastructures, new technologies—including AI in radiology—are slower to take off. Health-care providers in these places do not have the necessary clinical and technological expertise for operating these technologies, and there could also be a lack of the regulatory oversight and data privacy policies necessary to support the technology's adoption. For instance, several LMICs including Ethiopia and Indonesia have been slow to adopt telehealth during the COVID-19 pandemic.

And while it's challenging to implement AI in radiology in these settings, Qure.ai offers a compelling model of how it can be done. In Ethiopia, like many LMICs, health-care infrastructure is underdeveloped and is accompanied by opaque management and a lack of resources, stunting Ethiopia's ability to ensure quality health care. Qure.ai overcomes these hurdles by designing software that's compatible for most hardware systems, including outdated ones. This strategy allows Qure.ai to operate in a variety of health-care systems and facilitate radiologists' work across the globe.

Data privacy

Developed countries typically have strong privacy regulations—in the United States there's the Health Insurance Portability and Accountability Act (HIPAA) law, and in the European Union there's the General Data Protection Regulation— but many developing countries do not have strong oversight. This deficiency opens patients in low- and middle-income countries up to the risk of data exploitation, tracking, and other privacy violations. Health-care companies and nongovernmental organizations (NGOs) operating in these environments are out to prioritize data privacy, even if local regulations do not require them to do so.

Patients will not have to worry about the safety and integrity of their personal information getting compromised

Qure.ai protects data through region-specific regulations. "Patient data cannot leave the country," Ms.Suresh says unequivocally. Even though Qure.ai uses cloud systems to store demographic data, the company follows established regulations, like those outlined in HIPAA, to ensure personally identifiable patient information cannot be accessed outside of the local hospital network, similar to how current health-care systems operate. This means that patients will not have to worry about the safety and integrity of their personal information getting compromised.

The COVID-19 pandemic has made the need for AI-based advancements in radiology even more obvious to many experts. Over the past year, many health-care systems in low- and middle-income countries, such and Brazil, —as well as higher-income countries, such as the United States—have struggled to efficiently and effectively manage hospital crowding due to an overwhelming number of COVID-19 patients and a shortage of radiologists. Companies such as Qure.ai have started integrating their AI systems to help take on this challenge by helping radiologists quickly and effectively grade case-urgency and ensuring that cases are addressed in order of priority. In maximizing efficiency and clinical effectiveness by assisting with image-reading, AI allows radiologists to focus on patient-facing health interventions, treatments, and collaborating with health-care teams to guide medical procedures, allowing for a quicker turn-around between diagnosis and treatment and thereby improving health outcomes.

Despite concerns on the implications of AI in radiology, Qure.ai and Google Health have provided successful models for implementation and integration into global health delivery while navigating the infrastructure constraints and development barriers associated mainly with low- to middle-income countries, demonstrating that AI is nothing to fear.

A children's neurologist observes the x-ray of a baby's skull with microcephaly at the hospital Barao de Lucena in Recife, Brazil on January 26, 2016. REUTERS/Ueslei Marcelino

Newsletter

Think Global Health

Shuhab Elhag

Governance

Artificial Intelligence in Radiology: Hesitant Steps Forward

Governance