Winter 2019

To Boldly Go — Remote

The integration of artificial intelligence into telemedicine is enhancing the delivery of rural care and the training of specialists

Artificial Intelligence Issue

  • by Scott Edwards
  • 8 minute read

The medical tricorder in Star Trek: The Next Generation is one of many devices from science fiction that have posited ways in which medical data would be captured and communicated

The history of telemedicine—providing patient care from a distance using telecommunications technology—dates, somewhat surprisingly, to ancient times, when the Greeks and Romans used smoke signals and beacons to warn neighboring villages of disease outbreaks or to announce births and deaths.

With the development of radio and television, speculative articles on telemedicine bloomed. In the mid-1920s, a magazine called Science and Invention posited the notion that doctors would, in the future, use television and microphones and tend to patients via a “teledactyl” device, a tool with articulated appendages that would respond to remote manipulation by the physician, allowing the doctor to “feel his patient, as it were, at a distance.”

“The marriage between a physician and artificial intelligence provides an opportunity for substantial growth.”

From these humble yet futuristic beginnings, telemedicine today is on the brink of another revolution: the use of artificial intelligence, made possible with computers, algorithms, and technology, to analyze vast amounts of medical data to provide physicians and other caregivers with treatment guidance and recommendations, often remotely.

Service upgrade

The advent of the World Wide Web in the early 1990s brought about an information explosion that made possible the remote transfer of vast amounts of data, from basic vital signs and patient medical records to images from x-rays and MRIs among and between points of care. This virtual sharing capability has ushered in a new era in which people can use wearable devices to monitor personal health data and medical professionals can employ machine learning, pattern recognition, and computer algorithms—the basic elements of artificial intelligence—to diagnose, treat, and manage patients in clinics, hospitals, and at remote sites throughout the world.

“The marriage between a physician and artificial intelligence provides an opportunity for substantial growth,” says Paolo Silva, an HMS assistant professor of ophthalmology and the assistant chief of telemedicine at Joslin Diabetes Center’s Beetham Eye Institute. “By ‘training’ artificial intelligence with different skill sets, it can potentially lead to new medical discoveries.”

“Modern AI techniques can really affect the flow of decision-making information,” says Douglas Perrin, an HMS instructor in surgery and senior scientist at Boston Children’s Hospital who has spent the past fifteen years translating research advances in computer science into medicine. “It’s another tool in the physician’s tool kit.”

HMS physician-scientists are making inroads in using AI-based systems for tele-imaging and remote surgical education, training, and practice. Such initiatives could play a role in the future of how medicine is practiced remotely. And these systems may help fill a growing gap in health care, especially in rural and other underserved areas.

According to a 2016 report from the U.S. Centers for Disease Control and Prevention, data gathered between 1993 and 2015 indicate that nearly 20 percent of U.S. adults do not have regular access to health care. The National Rural Health Association reports that, as of October 2018, more than 120 rural hospitals have closed since 2005. Overall, the association says, more than one-third of rural hospitals are vulnerable, a number that jeopardizes access for an estimated 11.7 million patients.

In addition to the falling numbers of health care facilities, rural hospitals and clinics often do not have access to the array of specialists available in the more populated areas of the country. The use of AI may help reduce this dearth of rural specialists, according to a recent Forbes article on democratizing health care, by allowing specialists to remotely access a wealth of patient data and consult with local physicians on diagnosis and treatment determinations.

Pixelated care

At Joslin, Silva focuses his work in ocular telehealth for diabetic retinopathy, a condition that results when high blood-sugar levels damage the tiny blood vessels that supply the retina. At its most severe, it causes the retina to detach and leads to permanent vision loss.

Silva also directs telemedicine and retinal imaging research programs. One such program has led to collaborative efforts in the Philippines with Joslin and the federally funded Diabetes Retinopathy Research Network.

Based on this research, Silva has collaborated on the development of a fully automated computer algorithm to detect hemorrhages and microaneurysms, hallmarks of diabetic retinopathy, using ultrawide-field imaging. Unlike standard retinal imaging, ultrawide-field imaging captures up to 82 percent of the retinal surface in a single image, enabling clinicians to detect lesions on the retinal periphery.

Says Silva, “Our data show that when microaneurysms occur predominantly outside the fields that are imaged with traditional fundus photography, the eye is at high risk of worsening, increasing the risk of diabetic retinopathy progression over the long term.” Fundus photography captures the interior surface of the eye, including the retina.

In April 2018, the U.S. Food and Drug Administration approved the first medical device that uses artificial intelligence to detect greater than a mild level of diabetic retinopathy. In mild diabetic retinopathy, the retinal blood vessels weaken and, in small areas, balloon to form microaneurysms. Silva says this device uses traditional narrow-field fundus photography rather than ultrawide-field imaging.

portrait of Daniel Hashimoto
Daniel Hashimoto

Silva’s pilot study, presented in part at the 2014 annual meeting of the Association for Research in Vision and Ophthalmology, included 2,000 retinal images. To confirm the findings in a larger cohort of patients, he has since expanded the study to include more than 200,000 images. If successful, this algorithm may be able to identify lesions undetected by ophthalmologists, help clinicians predict the progression of the condition, and allow for the fine tuning of treatment plans. The need for such a tool is considerable.

In this country, nearly 4.2 million people have some form of diabetic retinopathy, according to the CDC, with more than 650,000 people at risk of losing their vision.

Part of Silva’s telemedicine practice at Joslin includes work within this vulnerable population. Through the Joslin Vision Network, created by Lloyd M. Aiello, a former HMS professor of ophthalmology, Joslin ophthalmologists are connected to more than ninety sites in twenty states. Many of the sites are within communities served by the Indian Health Service and the U.S. Departments of Defense and Veterans Affairs.

Through the program, Native Americans, a population with the greatest risk for diabetic retinopathy, can have retinal images taken annually. This effort has generated some 20,000 images. These are sent to the network’s reading and evaluation center in Boston, where clinical experts in diabetes eye care evaluate them for diabetic retinopathy and other ocular pathologies. Although the network doesn’t use specific AI applications, Silva hopes to apply his research with the Diabetes Retinopathy Research Network and “leverage artificial intelligence technology to provide care by reading retinal images at the point of care and decreasing the burden of evaluating images.”

Operating at a distance

As the surgical artificial intelligence and innovation fellow at Massachusetts General Hospital, Daniel Hashimoto spends much of his time investigating ways to improve the efficiency and quality of technical skills acquisition among surgeons. He also researches ways to integrate artificial intelligence into telemedicine in surgery. His work has been focused on the development of algorithms that can analyze video of surgeons conducting laparoscopic cholecystectomies, laparoscopic sleeve gastrectomies, and per oral endoscopic myotomies for ways to improve the safety and efficiency of their surgeries. In addition, through a process called telementoring, Hashimoto, who also is an HMS clinical fellow in surgery, has researched virtual and augmented reality platforms to remotely train surgeons.

Hashimoto is collaborating with scientists at the Computer Science and Artificial Intelligence Laboratory at MIT to analyze the millions of pixels in each of the twenty to thirty frames per second that are captured in videos of surgeries and create prediction models to help improve surgical procedures. Drawing on a library of data contained in videos that the researchers collected during the past four years from collaborating institutions, publicly available sources, and other databases, the algorithms the team has developed can predict the outcomes of techniques being used by the surgeons and, in turn, be used to guide them on ways to improve.

“Our hypothesis is that there is a sequence of events that can happen in an operation,” says Hashimoto. “Our work is focused on how we can use the data gleaned from the videos and make them actionable to help surgeons perform safer procedures for patients.”

Hashimoto’s efforts to remotely mentor surgeons as they acquire new skills involve virtual reality technology that trains them on the safe, effective use of surgical instruments. This virtual reality environment relies on algorithms to capture a variety of metrics, for example, how many millimeters the surgeon is moving an instrument or how many degrees they are rotating their wrist.

The effectiveness of this training approach has shown results; participating surgeons have improved their technical skills and have needed fewer hours to achieve proficiency benchmarks such as those set by the American Board of Surgery’s Flexible Endoscopy Curriculum. A study published in March in the journal Surgical Endoscopy by Hashimoto and his colleagues indicates that the team’s proficiency-based curriculum increased surgeons’ first-time pass rate for the Fundamentals of Endoscopic Surgery exam from 80 percent to 100 percent.

Silva, Hashimoto, and others are making advances in applying artificial intelligence to telemedicine and, they hope, to improving patient care in clinics far and near.

Scott Edwards is a Massachusetts-based science writer.

Images: John Soares