Shoji Kido1, Hironobu Nakamura, Wataru Ito, Kazuo Shimura, Hisatoyo Kato. 1. Department of Computer Science and Systems Engineering, Faculty of Engineering, Yamaguchi University, 16-1, Tokiwadai 2-Chome, Ube, Yamaguchi 755-8611, Japan. kido@ai.csse.yamaguchi-u.ac.jp
Abstract
OBJECTIVE: To clarify the usefulness of computerized detection of pulmonary nodules (PNs) in single-exposure dual-energy subtraction computed radiography (CR) images of the chest. METHODS AND MATERIAL: Our scheme uses bone-subtracted CR (BS-CR) images, and consists of a contrast-adaptive filter for detection of the candidates for PNs and a vessel-enhancing filter for elimination of vessel opacities in the candidates for PNs. For the evaluation, 12 clinical cases with multiple PNs were used, and four radiologists participated in this study. RESULTS: The detectability of our computer-aided diagnosis (CAD) was compared with detectabilities of radiologists. The mean true-positive (TP) number of four radiologists was 1.60+/-1.03, and that of the CAD was 1.83+/-1.34. There was no significant difference in the nodule detectabilities between the radiologists and CAD (P=0.18). However, the false-positive (FP) rate of the CAD was 9.42+/-2.54 per image, whereas the mean FP rate of the radiologists was 1.40+/-0.64 per image. CONCLUSION: Computerized detection of PNs is considered to be useful in the determination of radiological diagnoses. However, reducing the number of FP findings remains an important problem to be solved.
OBJECTIVE: To clarify the usefulness of computerized detection of pulmonary nodules (PNs) in single-exposure dual-energy subtraction computed radiography (CR) images of the chest. METHODS AND MATERIAL: Our scheme uses bone-subtracted CR (BS-CR) images, and consists of a contrast-adaptive filter for detection of the candidates for PNs and a vessel-enhancing filter for elimination of vessel opacities in the candidates for PNs. For the evaluation, 12 clinical cases with multiple PNs were used, and four radiologists participated in this study. RESULTS: The detectability of our computer-aided diagnosis (CAD) was compared with detectabilities of radiologists. The mean true-positive (TP) number of four radiologists was 1.60+/-1.03, and that of the CAD was 1.83+/-1.34. There was no significant difference in the nodule detectabilities between the radiologists and CAD (P=0.18). However, the false-positive (FP) rate of the CAD was 9.42+/-2.54 per image, whereas the mean FP rate of the radiologists was 1.40+/-0.64 per image. CONCLUSION: Computerized detection of PNs is considered to be useful in the determination of radiological diagnoses. However, reducing the number of FP findings remains an important problem to be solved.