RSNA 2013 

Abstract Archives of the RSNA, 2013


LL-INE3165-SUB

Does Computer-aided Diagnosis System which Presents the Reasoning for the Diagnosis Improve Radiologists’ Diagnostic Performance for Pulmonary Nodules on CT?

Education Exhibits

Presented on December 1, 2013
Presented as part of LL-INS-SUB: Informatics - Sunday Posters and Exhibits (1:00pm - 1:30pm)

Participants

Koji Fujimoto MD, PhD, Presenter: Nothing to Disclose
Masahiro Yakami MD, PhD, Abstract Co-Author: Nothing to Disclose
Takeshi Kubo MD, Abstract Co-Author: Nothing to Disclose
Ryo Sakamoto, Abstract Co-Author: Nothing to Disclose
Gakuto Aoyama, 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
Yoshio Iizuka, Abstract Co-Author: Nothing to Disclose
Masami Kawagishi, Abstract Co-Author: Nothing to Disclose
Hiroyuki Sekiguchi, Abstract Co-Author: Nothing to Disclose
Yutaka Emoto MD, PhD, Abstract Co-Author: Nothing to Disclose
Koji Sakai, Abstract Co-Author: Nothing to Disclose
Hiroyuki Yamamoto, Abstract Co-Author: Nothing to Disclose

BACKGROUND

Without reasoning, radiologists might not be able to determine whether the output of computer-aided diagnosis (CADx) is reliable or not. This may lead to difficulty in judgments of the CADx output by radiologists. The purpose of this study was to evaluate the effect of CADx software which presents reasoning for the diagnosis on radiologists’ performance. 

EVALUATION

With the approval of the IRB, we built a database of 491 lung nodules with clinical or pathological confirmation as primary lung cancer, metastatic, or benign. This database included thin-slice CT images, 49 nodule features interpreted by board-certified radiologists, laboratory data and patients’ past history. We developed a CADx that provide nodule features as reasoning for the suggesting diagnosis. An inference model with a Bayesian network was constructed using the Markov Chain Monte Carlo method with the 179 training data set. CADx with the inference of the Bayesian network (ICAD), with additional reasoning (RCAD) were evaluated. RCAD presented image findings and/or clinical information as reasoning according to the relevance with the presenting diagnosis. For evaluation, 11 radiologists interpreted 312 nodules under three different conditions; without CAD (NCAD), with ICAD, and with RCAD. Each radiologist inputted likelihoods of diagnosis (primary, metastatic or benign), which in total should be 100% on each nodule. The likelihood agreed with the confirmed diagnosis was regarded as the confidence in each interpretation. For each radiologist’s input, Shannon entropy was calculated using the likelihoods and was regarded as the uncertainty of the interpretation. Accuracy, AUC for each diagnosis, confidence, and uncertainty for 11 radiologists are compared for each condition (Wilcoxon signed rank test with Bonferroni correction).

DISCUSSION

Accuracy, AUC for primary lung cancer, and confidence were higher, and uncertainty was lower in the order of NCAD(0.71, 0.86, 59.2, 0.92, respectively), ICAD(0.76, 0.90 ,65.1, 0.80) and RCAD(0.77, 0.91, 66.3, 0.74). Significant difference was seen for NCAD vs ICAD, and NCAD vs RCAD.

CONCLUSION

RCAD improved accuracy and reduced uncertainty for their diagnosis, but significance was seen only with NCAD.

Cite This Abstract

Fujimoto, K, Yakami, M, Kubo, T, Sakamoto, R, Aoyama, G, Togashi, K, Iizuka, Y, Kawagishi, M, Sekiguchi, H, Emoto, Y, Sakai, K, Yamamoto, H, Does Computer-aided Diagnosis System which Presents the Reasoning for the Diagnosis Improve Radiologists’ Diagnostic Performance for Pulmonary Nodules on CT?.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13016986.html