Sumiaki Matsumoto MD, PhD, Presenter: Research grant, Toshiba Corporation

Yoshiharu Ohno MD, PhD, Abstract Co-Author: Research grant, Toshiba Corporation Research grant, Koninklijke Philips Electronics NV Research grant, Bayer AG Research grant, DAIICHI SANKYO Group Research grant, Eisai Co, Ltd

Takatoshi Aoki MD, Abstract Co-Author: Nothing to Disclose

Hitoshi Yamagata PhD, Abstract Co-Author: Employee, Toshiba Corporation

Keiko Matsumoto MD, Abstract Co-Author: Nothing to Disclose

Munenobu Nogami MD, PhD, Abstract Co-Author: Nothing to Disclose

Yoshiko Yamnashita MD, Abstract Co-Author: Nothing to Disclose

Shunsuke Kinoshita, Abstract Co-Author: Nothing to Disclose

Yayoi Fukuda, Abstract Co-Author: Nothing to Disclose

Mika Nakabayashi, Abstract Co-Author: Nothing to Disclose

Kazuro Sugimura MD, Abstract Co-Author: Research grant, Toshiba Corporation Research grant, Koninklijke Philips Electronics NV Research grant, Mitsubishi Corporation Research grant, Bayer AG Research grant, Eisai Co, Ltd Research grant, DAIICHI SANKYO Group Research Consultant, Shionogi & Co, Ltd

00030490-DMT et al, Abstract Co-Author: Nothing to Disclose

To compare the reading time and the observer performance for actionable lung nodules when using CAD as concurrent reader (CAD-c) with those when using CAD as second reader (CAD-s).

Using a prototype CAD system, 3 radiologists (readers 1-3) and 3 radiology residents (readers 4-6) independently detected non-calcified nodules (≥ 4 mm) in 50 standard-dose 1-mm-thick CT datasets in two sessions, with the reading time being recorded. In this task, they rated each finding by assigning a score of diagnostic confidence on a 7-point scale ranging from 0 (definitely non-nodule) to 6 (definitely actionable nodule). In the 1st (then 4 weeks later, 2nd) session, a half of the datasets and the remaining half were read with CAD-c (then CAD-s) and with CAD-s (then CAD-c), respectively, in a randomized order. Later, 2 experienced chest radiologists reviewed all findings blinded to the source of each finding and determined the set of actionable nodules by consensus. The reading time of each reader with CAD-c and that with CAD-s were compared using paired t-test. The observer performance for actionable nodules with CAD-c and that with CAD-s were compared using jackknife free-response receiver operating characteristic (JAFROC) analysis with respect to all 6 readers as well as to every subgroup of them.

The set of actionable nodules consisted of 72 nodules. The reading times of readers 1-6 with CAD-c/CAD-s were 121±25 s/182±59 s, 118±59 s/258±110 s, 141±69 s/193±82 s, 157±80 s/221±138 s, 138±90 s/208±104 s, and 114±79 s/196±95 s, respectively, with the reading time of each reader with CAD-c being significantly shorter (P<0.01). No JAFROC analysis showed significant difference between the observer performance for actionable nodules with CAD-c and that with CAD-s (P>0.1; at the cutoff score of 4, the sensitivities for actionable nodules of readers 1-6 with CAD-c/CAD-s were 88%/86%, 89%/93%, 90%/90%, 88%/89%, 57%/68%, and 63%/78%, respectively).

The reading time with CAD as concurrent reader was significantly shorter than that with CAD as second reader, whereas the observer performance for actionable lung nodules with CAD as concurrent reader was not significantly different from that with CAD as second reader.

CAD might be used as concurrent reader instead of second reader to achieve relatively short reading time without compromising the observer performance for actionable lung nodules.

Matsumoto, S, Ohno, Y, Aoki, T, Yamagata, H, Matsumoto, K, Nogami, M, Yamnashita, Y, Kinoshita, S, Fukuda, Y, Nakabayashi, M, Sugimura, K, et al, 0, Computer-aided Detection (CAD) as Concurrent vs Second Reader for Lung Nodules on Multidetector CT: Comparison of Reading Time and Observer Performance for Actionable Nodules.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8005466.html