Machine learning model reduces false positive lung cancer diagnoses

Machine learning model reduces false positive lung cancer diagnoses

The diagnosis of lung cancer is devastating news for patients—but what if the diagnosis is wrong? A low-dose CT scan is the standard diagnostic test for lung cancer for those at high risk, but it has a 96 percent false positive rate.

Nationwide, about a quarter of the scans reveal shadows indicating possibly cancerous nodules in the lung—but fewer than 4 percent of these patients actually have cancer. The University of Pittsburgh School of Medicine and UPMC Hillman Cancer Center researchers, including David Wilson, MD, MPH, associate professor of medicine in the Division of Pulmonary, Allergy, and Critical Care Medicine, and Panayiotis (Takis) Benos, PhD, professor of computational and systems biology, gathered CT scan data from 218 high-risk patients who either had lung cancer or benign nodules.

They created a machine learning model that calculates cancer probability by analyzing the number of blood vessels surrounding a nodule, the total number of nodules identified in the lung, and the number of years since a patient quit smoking. Comparing against the actual diagnoses of the 218 patients, the model’s assessment identified approximately 30 percent of patients’ benign nodules, all without missing a single positive case. Evaluating this technique in a larger population is the next step, the researchers indicate.