by Paul Govern
A machine learning algorithm developed by researchers at Vanderbilt University is just as powerful as humans at identifying skin lesions in clinical photos of people with monkeypox.
The team’s report appeared on September 15 in the Journal of Investigative Dermatology.
The number of skin lesions a person has is used by the World Health Organization as the primary measure of the severity of monkeypox in humans. Daily changes in the number of lesions could be used to assess patient responses in drug trials.
“If artificial intelligence based on photos could provide daily lesion counts in clinical trials of drugs being developed to treat monkeypox, it could make things a lot easier for everyone involved. While our proof-of-concept study uses limited data, our results clearly demonstrate that an AI solution to accelerate the intensive assessment of monkeypox severity is within reach,” said Eric Tkaczyk, MD, PhD, assistant professor of dermatology, Staff Physician at the Department of Veterans Affairs and Director of the Vanderbilt Dermatology Translational Research Clinic.
Monkeypox is endemic to the tropical rainforest regions of central and western Africa. From May, outbreaks of monkeypox were first reported from many regions outside of Africa. On September 13, the European Centers for Disease Prevention and Control reported a total of 19,379 cases since May in Europe, and on September 14, the US Centers for Disease Control and Prevention reported 22,774 cases since May in the United States, with 239 cases occurring in Tennessee. As of September 14, the CDC had put the global case count at 59,606.
First described in 1958, the disease is primarily transmitted through intimate contact, typically skin to skin. In addition to skin lesions, patients may experience fever, chills and other symptoms. The illness lasts two to four weeks, with most people infected in the current global outbreak recovering without requiring medical treatment. While deaths from the globally spreading variant have been rare, with only one death reported in the US and three deaths in Europe, higher death rates have been reported for people infected with a Central African variant in the past.
There is currently no specific drug treatment for monkeypox. According to the CDC website, “Antiviral drugs and vaccines designed to protect against smallpox can be used to prevent and treat monkeypox virus infection.” For more information on monkeypox, including prevention, visit the CDC website.
Tkaczyk led the machine learning study along with Andrew McNeil, PhD, a postdoctoral researcher in Electrical and Computer Engineering, Inga Saknite, Adjoint Assistant Professor of Dermatology, and Benoit Dawant, PhD, Cornelius Vanderbilt Professor of Engineering and Director of the Vanderbilt Institute for Surgery and Technology.
“This is a good example of what’s possible when researchers from the Vanderbilt Schools of Engineering and Medicine can pool their expertise,” said Dawant.
The project used 66 photos of 18 patients from an observational study in the Democratic Republic of the Congo. On each image, a team member marked borders around lesions and provided training data for machine learning. When testing photos retained by the training step, the lesion counts from the machine learning algorithm were on par with the counts from two human raters.
Others on VUMC’s study include David House and Ziche Chen. The team included researchers from the Institut National de Recherche Biomédicale in the Democratic Republic of the Congo, the Frederick National Laboratory for Cancer Research and the National Institutes of Health (NIH). The study was supported by the US Department of Veterans Affairs, the European Regional Development Fund and the NIH (CA090625, AR074589).
For more information on the use of the machine learning algorithm in drug testing, see the Vanderbilt University Research News story.
