Day 1 of RSNA 2025 is in the books, and new research into CT lung cancer screening dominated the scientific sessions at Chicago’s McCormick Place.
Lung cancer screening is drawing attention as screening programs go into effect internationally.
- In the U.S., lung screening is hampered by low completion rates (18-19%), but providers are finding that participation can be improved with aggressive identification and outreach to eligible patients.
Some highlights from Sunday (with handy session numbers to help you follow along) include…
- The ScreenLungNet AI model predicted three-year lung cancer risk from CT lung screening scans with AUCs from 0.93-0.94 (S4-SSCH02-1).
- In a study of 2.6k patients with lung cancer, only 36% met 2021 USPSTF lung cancer screening criteria, and just 5% actually got screened. Only 23% had data on their smoking history in the EMR (S4-SSCH02-2).
- Risk assessment scores were used to perform CT lung screening of lower-risk people every two years rather than annually, reducing screening’s harms without missing many cancers (S4-SSCH02-4).
- Compared to the landmark NLST study, a real-world CT lung screening program had fewer benign surgeries (12% vs. 18%), lower complication rates (24% vs. 32%), and better recurrence-free survival (HR = 0.60) (S4-SSCH02-5).
- CT radiation dose was reduced 51% and contrast iodine use 61% through a triple-optimized protocol that included 80-kVp scanning, GE HealthCare’s TrueFidelity deep learning reconstruction, and low-iodine adaptive contrast injection (S2-SSCA01-1).
- Using AI for automated patient positioning and scan range in CT exams cut positioning time 41% with 10-13% lower radiation dose and no discernible impact on image quality (S4-SSIN01-1).
- Measures of adiposity acquired opportunistically from coronary artery calcium CT scans using HeartLung Technologies’ AI-CVD algorithm predicted adults at risk of diabetes in a study of 2.9k people (S5-SSCA02-6).
- The Promedius AI algorithm for osteoporosis assessment of chest radiographs had an AUC of 0.84 in a study of 1k adults from three countries (M3-SSCH03-1).
- A real-world study of 2.1k patients found that DeepTek’s chest X-ray AI algorithm had an AUROC of 0.95 for detecting any of 13 clinically significant findings (M3-SSCH03-2).
- Researchers presented a feasibility study of a compression-free spectral DBT mammography system, finding spatial resolution close to state-of-the-art systems (S4-SSPH02-6).
- Researchers presented their protocol for MRI scanning of patients with cardiac implanted electronic devices. Over 10 years they scanned 7.3k patients with no major adverse events (S5-SSCA02-1).
- Adding MRI data to a multimodal transformer AI model improved its ability to predict five-year breast cancer risk in intermediate- and high-risk women (S2-SSBR01-6).
The Takeaway
RSNA 2025 is off to a great start. Be sure to check back with Thursday’s newsletter for more radiology news from Chicago, and follow along on our social media channels for ongoing video updates.
