CT Scan Enhanced With Synthetic PET Can Predict Lung Cancer

Background: [ 18 F]Fluorodeoxyglucose PET and CT are commonly used to detect oncologic disease, including in potential lung cancer patients. One of the disadvantages of this technique is the radiation dose from the radiopharmaceutical and the expensive equipment required in this modality. An alternative method has been proposed using deep learning techniques to produce synthetic PET images from CT scans, which can simulate "true" PET images. Objective: To determine whether deep learning methods could produce synthetic PET images from CT scans, which could complement the information from the CT images and simulate the information from "true" PET images. Methods: The authors developed a conditional generative adversarial network, called cGAN, to produce the synthetic PET images from the CT scans. This research was performed using multicenter PET and CT scans. The deep learning training process paired the anatomical-to-metabolic features of the diagnostic CT and "true" PET scans. The sy more...