Takeshi Kubo MD, Abstract Co-Author: Nothing to Disclose

Gakuto Aoyama, Presenter: Nothing to Disclose

Koji Fujimoto MD, PhD, Abstract Co-Author: Nothing to Disclose

Masahiro Yakami MD, PhD, Abstract Co-Author: Nothing to Disclose

Masami Kawagishi, Abstract Co-Author: Nothing to Disclose

Kaori Togashi MD, PhD, Abstract Co-Author: Research Grant, Bayer AG Research Grant, DAIICHI SANKYO Group Research Grant, Eisai Co, Ltd Research Grant, FUJIFILM Holdings Corporation Research Grant, Nihon Medi-Physics Co, Ltd Research Grant, Shimadzu Corporation Research Grant, Toshiba Corporation Research Grant, Covidien AG

Yutaka Emoto MD, PhD, Abstract Co-Author: Nothing to Disclose

Ryo Sakamoto, Abstract Co-Author: Nothing to Disclose

Yoshio Iizuka, Abstract Co-Author: Nothing to Disclose

Hiroyuki Sekiguchi, Abstract Co-Author: Nothing to Disclose

Koji Sakai, Abstract Co-Author: Nothing to Disclose

Hiroyuki Yamamoto, Abstract Co-Author: Nothing to Disclose

A number of studies have demonstrated the improved accuracy of nodule diagnosis using CADx systems. The gain in the accuracy, however, may vary with nodule characteristics. Thus, to take full advantage of CADx, we evaluated the relationship between the performance of radiologists with a CADx system and the nodule characteristics.

In accordance with the IRB approval, we built a database of 491 lung nodules with clinical or pathological confirmation as a primary lung cancer, metastasis or benign nodule. The image findings were scored by consensus of two board-certified radiologists. We developed a CADx system (Bayesian network / Markov chain Monte Carlo method, 179 training data) that can provide the reasoning behind the suggested diagnosis. CADx which indicates the probability of diagnosis (ICAD) and probability of diagnosis with additional reasoning such as image findings and/or clinical information (RCAD) were used for this experiment. The reasoning was determined by the degree of influence on the highest possible diagnosis. 11 diagnostic radiologists with 5 years’ experience made the diagnoses for the 312 nodules with three different conditions (without CAD, with ICAD, with RCAD) with a more than 1 month interval. We focused on 61 clinically relevant nodule characteristics. For each characteristic, a group of nodules was defined so that the nodules in the group share that particular characteristic. With regard to 61 resultant groups, the mean accuracy of 11 radiologists was compared among three conditions (Wilcoxon signed rank test with Bonferroni correction).

For all groups, ICAD and RCAD significantly improved the diagnostic accuracy of radiologists (p<.05), except for groups of nodules with 1) irregular margin, 2) almost round shape, 3) lobulated shape, 4) calcification and 5) no contraction. For groups of nodules with 1) coarse speculation, 2) polygonal shape and 3) satellite nodules, accuracy was significantly better with RCAD than with ICAD (p<.05).

Effectiveness of CADx depended on nodule characteristics. Recognition of nodule characteristics that benefit from CADx support may lead to optimizing the CAD-assisted diagnostic process by the radiologists.

Kubo, T, Aoyama, G, Fujimoto, K, Yakami, M, Kawagishi, M, Togashi, K, Emoto, Y, Sakamoto, R, Iizuka, Y, Sekiguchi, H, Sakai, K, Yamamoto, H, Relationship between Characteristics of Pulmonary Nodules and Performance Improvement of Radiologists: Comparison between CADx with and without Reasoning.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13016982.html