Sunday, November 29 | 1:00 p.m.-2:00 p.m. | S4-SSCH02-7 | Room E451A
A deep-learning model can predict mortality in heavy smokers from CT lung cancer screening exams, adding prognostic value beyond traditional risk factors and body composition measures, according to this presentation.
Changes in body composition, such as reduced muscle mass or excess fat, have been linked to cardiovascular and cancer-related deaths. Whether chest CT offers prognostic information beyond established risk factors and body composition has remained unclear, however, noted presenter Johannes Jahn, MD, of Universitätsklinikum Freiburg in Germany, and colleagues. To find out, the group developed CTchest-risk, a deep-learning algorithm that predicts mortality using low-dose CT (LDCT) lung cancer screening studies.
The researchers trained the model on data from more than 14,000 participants in the CT arm of the National Lung Screening Trial and tested it on nearly 9,000 others. They compared three versions: CT images only, CT combined with body composition segmentation masks, and masks only.
Over a median follow-up of 6.5 years, the CT-only version performed best. After adjustment for clinical factors and body composition, participants in its high-risk group had 60% higher mortality than those at intermediate risk.
“Automated mortality risk stratification from lung screening CT may enable opportunistic identification of high-risk individuals without additional imaging or manual [body composition] assessment,” the researchers wrote.
Check out this presentation on Sunday afternoon to learn more.
