Virtual Library

Start Your Search

Eui Jin Hwang



Author of

  • +

    P2.11 - Screening and Early Detection (ID 178)

    • Event: WCLC 2019
    • Type: Poster Viewing in the Exhibit Hall
    • Track: Screening and Early Detection
    • Presentations: 1
    • Moderators:
    • Coordinates: 9/09/2019, 10:15 - 18:15, Exhibit Hall
    • +

      P2.11-16 - Variability in Reading Low-Dose Chest CT: Individual Readers vs. Central Review in a Nationwide Lung Cancer Screening Project (ID 2299)

      10:15 - 18:15  |  Presenting Author(s): Eui Jin Hwang

      • Abstract

      Background

      Variability in interpretation among readers or institutions is an important issue in the nationwide lung cancer screening. The potential of computer-aided detection and volumetry in reducing interobserver variability has rarely been investigated in the actual screening situation. This study aimed to evaluate the variability among experts in a nationwide lung cancer screening project.

      Method

      We evaluated 1647 consecutive baseline screening CT scans obtained during one month period of December 2017 from a nationwide lung cancer screening project in Korea (K-LUCAS) in which 14 institutions participated. One of 20 chest radiologists in each institution assessed CT scans using a thin-client system equipped with semi-automated nodule segmentation and computer-aided detection software based on Lung-RADS (institutional reading). One chest radiologist retrospectively reviewed all these CT scans while minimizing modification of segmentation results and rejection of tiny nodules (central review). Reading results between institutional reading and central review were compared. Per-case and per-nodule positive rates of central review using Lung-RADS and NELSON criteria were also compared.

      Result

      Per-case positive rate was significantly higher in central review (24.9% [410/1647; 11.1-32.7% across institutions] vs. 19.3% [319/1647; 5.6-30.0% across institutions]; P<.001), while per-nodule positive rate was significantly higher in institutional reading (10.2% [511/4497; 5.2-21.9%] vs. 19.9% [382/1920; 9.5-60.6%]; P<.001). The number of detected nodules was significantly larger in central review (3.04 vs. 1.17 nodule/case; P<.001), while the size of detected nodules was significantly smaller in central review (4.0 vs. 5.1 mm; P<.001). Variability in positive rates among institutions was significantly lower in central review (coefficient of variability, 21.9% vs. 40.2%; P=.044). Manual measurements after rejecting result of semi-automated segmentation occurred in 1.6% (80/5008) of nodules at central review and in 17.8% (342/1920) nodules at institutional reading. Positive rate with Lung-RADS was significantly higher compared with that of NELSON criteria (24.9% vs. 3.9%; P<.001), however, including indeterminate scan by NELSON criteria which requires additional scanning, positive rate was higher in NELSON criteria (24.9% vs. 33.4%; P<.001).그림1.jpg

      Conclusion

      Even in a situation where computerized tools are adopted, considerable variability was observed in interpretation of lung cancer screening CT even among expert chest radiologists. The variability was mainly caused by discarding tiny nodules and modifying or rejecting segmentations results, and reduced by central review. The NELSON criteria did not reduce the number of additional scanning in nodule management compared with Lung-RADS. The value of reducing variability by applying stricter rules should be further investigated.