Can Artificial Neural Network Help Predict Non-Melanoma Skin Cancer?
Posted: Monday, October 8, 2018
Based on personal health records alone, a multiparameterized artificial neural network, created by David Roffman, PhD, of Yale University, and colleagues, may be able to predict and stratify an individual’s risk of developing non-melanoma skin cancer. The study findings were published in Scientific Reports, and further clinical testing of this approach is ongoing.
“Our approach is easy-to-implement, non-invasive, and cost-effective while achieving comparable sensitivity and specificity to other approaches, which often require ultraviolet radiation exposure and family history data,” the investigators commented. However, they acknowledged, their model may be further improved with the inclusion of such data
Using the National Health Interview Survey adult data from the years 1997 through 2013 and 2015, the researchers identified 2,056 non-melanoma skin cancer cases and 460,574 non-cancer cases to train and validate the artificial neural network. They did further testing with the survey’s 2016 data set, totaling 28,058 respondents. Gender, age, body mass index, diabetic status, smoking status, history of stroke, among other parameters, were used for the network.
The researchers found that the network assessed cancer risk with a high sensitivity and a “decent” specificity (training sensitivity, 88.5%; specificity, 62.2%; validation sensitivity, 86.2%; specificity, 62.7%). And with the personal health data, they categorized patients into cancer risk groups: low, medium, and high.
“Moving forward, we envision that the developed [artificial neural network] could help direct primary care physicians in decision making on which patients are at highest risk for skin cancer, with subsequent referral to dermatology for total body skin examination,” concluded Dr. Roffman and colleagues.
