Incorporation of a computer-aided vessel-suppression system to detect lung nodules in CT images: effect on sensitivity a
- PDF / 814,546 Bytes
- 6 Pages / 595.276 x 790.866 pts Page_size
- 23 Downloads / 190 Views
ORIGINAL ARTICLE
Incorporation of a computer‑aided vessel‑suppression system to detect lung nodules in CT images: effect on sensitivity and reading time in routine clinical settings Taku Takaishi1 · Yoshiyuki Ozawa2 · Yuya Bando1 · Akiko Yamamoto1 · Sachiko Okochi1 · Hirochika Suzuki1 · Yuta Shibamoto2 Received: 30 July 2020 / Accepted: 10 September 2020 © Japan Radiological Society 2020
Abstract Purpose To evaluate whether a computer-aided vessel-suppression system improves lung nodule detection in routine clinical settings. Materials and methods We used computer software that automatically suppresses pulmonary vessels on chest CT while preserving pulmonary nodules. Sixty-one chest CT images were included in our study. Three radiologists independently read either standard CT images alone or both computer-aided CT and standard CT images randomly to detect a pulmonary nodule ≥ 4 mm in diameter. After an interval of at least 15 days to avoid recall bias, the three radiologists interpreted the counterpart images of the same patients. The reference standard was decided by an expert panel. The primary endpoint was sensitivity. The secondary endpoint was interpretation time. Results The average sensitivity improved with computer-aided CT (72% for standard CT vs. 84% for computer-aided CT, p = 0.02). There was no difference in the false-positive rate (21% for both standard CT and computer-aided CT, p = 0.98). Although the average reading time was 9.5% longer for computer-aided plus standard CT compared with standard CT alone, the difference was not significant (p = 0.11). Conclusion Vessel-suppressed CT images helped radiologists to improve the sensitivity of pulmonary nodule detection without compromising the false-positive rate. Keywords Tomography · Lung · Computer software · Radiologists · Workflow
Introduction Early detection of lung nodules in computed tomography (CT) images is essential to reduce the mortality of potential lung cancer patients [1]. To improve the detection of lung nodules, several computer-aided detection (CAD) systems have been developed. Previous studies demonstrated promising results of using CAD to detect nodules by reducing the reading time and increasing the overall detection rate [2, 3]. However, integrating a CAD system in a clinical workstation * Taku Takaishi [email protected] 1
Konan Kosei Hospital, Takayacho‑Omatsubara 137, Konan, Aichi, Japan
Department of Radiology, Nagoya City University Graduate School of Medical Sciences, Nagoya, Japan
2
has proven to be challenging and there have been few studies to judge the efficacy of CAD systems in routine clinical settings. Given the academic interest in machine learning and the demands of efficient CT image interpretation, we introduced a CAD system in our hospital and assessed its performance. This study was conducted in a municipal hospital and we adopted a commercially available CAD system that was approved for clinical use by an official agency in Japan. The CAD system automatically removes pulmonary vessels from che
