Machine learning method may predict skin cancer, study says

Sept. 12 (UPI) — A new non-imaging method has shown accuracy in spotting skin cancer, a new study says.

This machine learning technique can predict the development of non-melanoma with 89 percent sensitivity, according to research published this month in JAMA Dermatology.

“Most of the existing AI-medicine studies were about imaging, such as tumor detection, and computer-aided diagnosis for cancerous lesions,” Christine Wang, a researcher at Taipei Medical University, told UPI. “In our study, deep learning convolutional neural network was trained using non-imaging and sequential medical records to do the prediction.”

The researchers pulled data from the Taiwan National Health Insurance Research Database, collected between Jan. 1, 1999, and Dec. 31, 2013, for 1,829 patients who received their first diagnosis of nonmelanoma skin cancer and 7,665 without cancer.

They used a deep learning algorithm known as a convolutional neural network, training it to develop a risk prediction model to predict skin cancer based on analysis of medical data.

The researchers then used the model to assess 3-year clinical diagnostic data, medical records and temporal sequential information to forecast skin cancer risk within the following year using patient data collected from November 1, 2016 to October 31, 2018.

The researchers discovered when hypertension, chronic kidney disease and degenerative osteopathy were present, it increased the likelihood of a patient having skin cancer.

The researchers say some medications can reduce the skin cancer risk. Those include trazodone, acarbose, nonsteroidal anti-inflammatory drugs, systemic antifungal agents, statins and thiazide diuretics.

“Future research should not just focus on images, enough longitudinal cohort data could be of great research value,” Wang said. “Without looking at a patient in person, using electronic health records data-driven prediction model could help screen out high-risk populations and potentially enhance prevention measures or provide early diagnoses.”