Abstract
Accurately predicting the prognosis of patients with lung cancer before or at the time of treatment would offer clinicians an opportunity to tailor management plans more precisely to individual patients. Considering that chest computed tomography (CT) scans are universally acquired in patients with lung cancer for clinical staging or response evaluation, fully extracting and utilizing the prognostic information embedded in this modality would be a reasonable approach. Herein, we review tumor-related prognostic factors that are extractable from CT scans, including the tumor dimensions, presence of ground-glass opacity (GGO), margin characteristics, tumor location, and deep learning-based features. Tumor dimensions include diameter and volume, which are among the most potent prognostic factors in lung cancer. In lung adenocarcinomas, the solid component size on CT scans as well as the total tumor size is associated with the prognosis. The areas of GGO indicate the lepidic component and are associated with better postoperative survival in early-stage lung adenocarcinomas. As for the margin characteristics, which represent the CT manifestation of fibrotic stroma or desmoplasia, tumor spiculation should be evaluated. The tumor location in the central lung is associated with occult nodal metastasis and is a worse prognostic factor per se. Last but not least, deep learning analysis enables prognostic feature extraction beyond the human eyes.