Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster Display session

180P - Substantial radiologist workload reduction can be achieved if artificial intelligence is used as a first-read filter at baseline in lung cancer screening

Date

22 Mar 2024

Session

Poster Display session

Topics

Population Risk Factor

Tumour Site

Presenters

Harriet Lancaster

Citation

Annals of Oncology (2024) 9 (suppl_3): 1-6. 10.1016/esmoop/esmoop102576

Authors

H.L. Lancaster1, B. Jiang2, M. Silva3, J.W. Gratama4, D. Han5, J.K. Field6, G.H. de Bock7, M.A. Heuvelmans7, M. Oudkerk2

Author affiliations

  • 1 University of Groningen, University Medical Center Groningen, Groningen/NL
  • 2 The Institute for Diagnostic Accuracy, Groningen/NL
  • 3 Università di Parma, 43126 - Parma/IT
  • 4 Gelre Ziekenhuizen, Apeldoorn/NL
  • 5 i-DNA B.V., Groningen/NL
  • 6 NHS Liverpool Clinical Laboratories - Royal Liverpool University Hospital NHS Trust, Liverpool/GB
  • 7 UMCG - University Medical Center Groningen, Groningen/NL

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 180P

Background

Lung cancer screening is being implemented in multiple countries worldwide and consequently radiologists workload is ever increasing. Artificial intelligence (AI) could provide the solution to significantly reducing workload if used as a first read filter to accurately rule-out negative cases (none or lung nodules <100mm3). We aimed to validate an AI prototype in an external LDCT dataset.

Methods

Validation of the AI prototype was performed using 1254 LDCT baseline CT scans from the UKLS trial. Four manual readers and AI independently assessed all scans. A consensus read by two experienced radiologists, blinded from initial results, was performed in the case of discrepancies. All individual results were compared to the consensus read to determine a final classification. Individual cases were then classified as either; correct positives (CPs) or negatives (CNs)(≥100 or <100mm3, respectively), positive-misclassifications (PMs) (nodules classified by the reader/AI as ≥100mm3, but at consensus<100mm3) or negative-misclassifications(NMs) (nodules classified by the reader/AI as <100mm3, but at consensus≥100mm3).

Results

Of the 1254 cases, 816(65%) had no nodules or only nodules <100mm3 (negative cases). AI achieved a higher negative predictive value 91.7(89.8-91.4) than all manual readers [reader1; 79.0(77.5-82.0), reader 2; 80.1(81.3-85.4), reader 3; 77.5(76.0-79.0) and reader 4; 78.5(77.0-80.0)]. If AI was used as a first-read filter to rule out negative cases, workload reduction was calculated at 65% [(total scans;1254 - (CPs;370 + PMs;59)) / total scans;1254].

Conclusions

AI can achieve a better negative predictive performance than manual readers when used as a first-read filter at baseline LDCT lung cancer screening. If implemented in a lung cancer screening program, we estimate that only 35% of baseline cases with nodules >100mm3 would need to be assessed by a radiologist.

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.