RSNA 2018

Abstract Archives of the RSNA, 2018


Randomized Clinical Trial of CAD versus No CAD as First Reader of Lung Cancer Screening CT: Preliminary Report

Monday, Nov. 26 11:10AM - 11:20AM Room: E451A

Ren Yuan, MD,PhD, Vancouver, BC (Presenter) Nothing to Disclose
John R. Mayo, MD, Vancouver, BC (Abstract Co-Author) Speaker, Siemens AG
Renelle Myers, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Sukhinder Atkar-Khattra, BSC, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Isaac Streit, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
John Yee, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Kyle Grant, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Alexander Lee, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Anna L. Mcguire, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose
Colin Jacobs, PhD, Nijmegen, Netherlands (Abstract Co-Author) Research Grant, MeVis Medical Solutions AG
Bram Van Ginneken, PhD, Nijmegen, Netherlands (Abstract Co-Author) Stockholder, Thirona BV; Co-founder, Thirona BV; Research Grant, Varian Medical Systems, Inc; Research Grant, Canon Medical Systems Corporation
Martin Tammemagi, St. Catharines, ON (Abstract Co-Author) Nothing to Disclose
Stephen Lam, MD, Vancouver, BC (Abstract Co-Author) Nothing to Disclose

For information about this presentation, contact:


The accuracy of radiologists reading lung cancer screening CT in a previous study shows a false-negative rate (FN) of 3.5% to 8.1%. The purpose of this study was to assesses if CAD can reduce the FN and CT reading time.


We conducted a randomized trial in 148 smokers participating in our ongoing Lung Cancer Screening Project (75M:73F, 667yrs, 59 ex- vs. 89 current-smoker). Chest CTs were randomized into two arms. In the CAD and Technician first arm (CAD+Tech-1st), CAD findings were displayed first, a technician accepted or rejected CAD findings and added probable nodule(s), then a chest radiologist accepted or rejected the CAD +Tech findings adding additional nodule(s). In the RAD-first arm (RAD-1st) the same radiologist read the CT first with CAD marks hidden, then turned on CAD to accept true nodules including those only found by CAD and delete the non-nodule CAD findings. The number of true nodules and reading time were recorded.


The reading times were 6.2 3.4 min (range: 2-18) vs. 8.3 5.4 min (range: 3-30) for CAD+Tech-1st vs. RAD-1st arms (p=0.012) for CTs with >=1 nodule; and 4.41.5 min (range: 2-10) vs. 8.79.5 min (range: 3-30) for those without nodules (p=0.07). By the three detection methods, 212 true nodules were found in 97 CTs in the CAD+Tech-1st arm. CAD detected 82 and technician added 93 true nodules, giving a combined sensitivity of 83%. There were 37/212 nodules found only by the radiologist; 12/37 were the most important nodule, and 1/37 was the only nodule that drove follow-up. In the RAD-1st arm 71 true nodules were found in 51 CTs; 36/71 (51%) were found by both CAD and radiologist. The radiologist missed 2 true nodules in 2 participants (2/51, 4%) which were detected by CAD and altered their follow-up protocol. The radiologist's detection sensitivity slightly increased with CAD (97% to 100%). CAD missed 33/71(46%) true nodules found by the radiologist, 16/33 (48%) were key nodules and 11/16 were the only nodule, changing follow-up.


CAD+Tech speed up the radiologist's nodule detection on screening chest CT. CAD detected nodules in 4% subjects where no nodule was identified by the radiologist, changing imaging follow-up protocol.


While CAD+Tech as first reader cannot replace the radiologist, CAD could play an important role in lung cancer screening by saving radiologists' time, and importantly reduce their FN rate by 4%.