Skin cancer is the most common cancer in the United States, and one in five Americans will develop skin cancer in their lifetime.1 The risk can be even higher for veterans; for example, a study of veterans exposed to Agent Orange during the Vietnam War found they had a skin cancer incidence of 51%.2 Early screening and treatment is critical to improve a patient’s prognosis.
With this challenge in mind, GDIT developed a tool that uses artificial intelligence (AI) to classify images of skin lesions, determine if they are indicative of a common skin disease, and if so, recommend immediate follow-up care.
AI Applied to Improve Veteran Health
The GDIT skin lesion classifier tool won third place in the VA National AI Tech Sprint 2020-2021, a competition organized by the Department of Veterans Affairs (VA) National Artificial Intelligence Institute (NAII) to match private sector talent with veterans, VA clinicians and other experts to brainstorm AI-based solutions that can improve veteran health and well-being. This is the second year the NAII AI Tech Sprint successfully developed new practices, tools, and products for VA and the veterans they serve.
“We are committed to helping our government customers meet their mission, and serving the veteran community is a core focus for GDIT,” said Kamal Narang, vice president and general manager for Federal Health. “Our experience, agility and commitment to innovation help us match powerful AI technology to tough customer challenges, and with this project, we aim to improve health outcomes for veterans.”
GDIT’s AI and data insights team uses artificial intelligence tools and methods to extract meaningful insights. “We’ve grown our expertise in machine learning with deep learning and transfer learning methods that help us build accurate computer vision and natural language processing solutions,” said Dave Vennergrund, vice president for AI and Data Insights. “The skin lesion image classifier we developed for the AI Tech Sprint is a great example of this expertise. We used open-source image classifiers trained on a collection of labeled images.”
We're using Machine Learning to help customers uncover insights from endless data.
In the tech sprint, the VA paired the GDIT team with Dr. Trilokraj Tejasvi, associate professor for at University of Michigan Health and Chief of Dermatology at the VA Ann Arbor Healthcare System. Tejasvi was particularly interested in image classifiers and how they might be applied to help diagnose skin lesions. He worked closely with the GDIT team, providing expertise and mentorship that was invaluable in building the solution.
Tejasvi is also the director of the University of Michigan Teledermatology Service and the teledermatology physician champion for the Veterans Integrated Service Network (VISN) 10. Many veterans do not live near a VA hospital, and prior to the COVID-19 pandemic, the VA already had the largest telehealth program in the country. By mid-2020, VA telehealth visits rose from 10,000 to approximately 120,000 per month – an increase of 1,000%.3
The need to develop virtual diagnosis tools suitable for telehealth delivery is paramount to reaching the veteran community, and the classifier can be accessed through any web browser. A physician will be able to simply pull a skin lesion image onto a web page and receive an instant classification and recommendation regarding follow-up care.
“When building an AI program, the output quality is highly dependent on the input,” said Tejasvi. “The GDIT team was very receptive to recommendations, incorporating them into building the image classifier and checking for image quality prior to transmittal. Each output report was easy to interpret and would help non-dermatology trained providers take appropriate steps – potentially saving veterans multiple trips to their provider’s office and promoting timely care. GDIT successfully created an ancillary tool to improve the teledermatology referral process.”
Given the high accuracy of the skin lesion classifier, GDIT was invited to submit a follow-on proposal to expand the classifier to cover a more diverse range of skin types, conduct field tests and verify accuracy through double-blind tests with VA oncologists and primary care providers.