Could artificial intelligence detect early-stage lung cancer?
A small minority of pulmonary nodules can indicate early-stage cancers. As a result, detecting such pulmonary nodules in CT scans can represent a challenging opportunity for early lung cancer diagnosis. Grading the cancer-risks of nodules is a delicate and complex process; classifying a high-risk nodule as low-risk could hinder cancer diagnosis, while on the other hand, classifying a benign nodule as high-risk may lead to unnecessary invasive procedures and resulting complications.
A multi-institutional group of researchers have employed artificial intelligence to address this issue and assist the earlier diagnosis of lung cancer. Published in eBioMedicine, the study utilized CT scan data from approximately 500 patients with large lung nodules to design an artificial intelligence algorithm. The algorithm’s performance was then assessed. The algorithm analyzed each nodule and identified its risk of cancer with an AUC (a measure of a model’s performance, where 1 is 100% correct) of 0.87. The artificial intelligence algorithm also outperformed alternative commonly used clinical tests, such as the Brock score (AUC of 0.672) and Herder score (AUC of 0.83).