Lung Cancer: AI Can Reduce False Positives on Low-Dose CT Screening Interview with:

Panayiotis (Takis) Benos, Ph.D.Professor and Vice Chair for Academic AffairsDepartment of Computational and Systems BiologyAssociate Director, Integrative Systems Biology ProgramDepartment of Computational and Systems Biology, SOM andDepartments of Biomedical Informatics and Computer ScienceUniversity of Pittsburgh

Dr. Benos

Panayiotis (Takis) Benos, Ph.D.
Professor and Vice Chair for Academic Affairs
Department of Computational and Systems Biology
Associate Director, Integrative Systems Biology Program
Department of Computational and Systems Biology, SOM and
Departments of Biomedical Informatics and Computer Science
University of Pittsburgh What is the background for this study? What are the main findings?

Response: Low-dose computed tomography (LDCT) scans is the main method used for early lung cancer diagnosis.  Early lung cancer diagnosis significantly reduces mortality.  LDCT scans identify nodules in the lungs of 24% of the people in the high-risk population, but 96% of these nodules are benign.  Currently there is no accurate way to discriminate benign from malignant nodules and hence all people with identified nodules are subjected to follow up screens or biopsies.  This increases healthcare costs and creates more anxiety for these individuals.  By analyzing a compendium of low-dose computed tomography scan data together with demographics and other clinical variables we were able to develop a predictor that offers a promising solution to this problem. What should readers take away from your report?

Response: Our analysis identified three variables that are highly predictive of nodule status (cancer/benign).

These are the total number of nodules, the number of vessels around the examined nodule and the years since the individual quit smoking prior to the low-dose computed tomography.  These three variables offer complementary information and are highly predictive of nodule status.  Our predictive model, LCCM or Lung Cancer Causal Model, provides a probability score on how likely is for a nodule to be malignant.  With proper thresholding, our results indicate that we can identify 28% of the benign nodules as such without risking any malignant nodule misclassification.  This can have a potentially big impact in the future on which people are asked to come for subsequent screenings. What recommendations do you have for future research as a result of this work?

Response: There are two lines of research we currently pursue. One is to see how generalizable our model is.  This involves examination of larger and more diverse cohorts. The second is to include additional clinical data and molecular biomarkers, such as those collected from easily accessible tissues like blood.  We hope that at the end a simple blood draw and a low-dose computed tomography scan can help to more accurately identify lung cancer cases early.

This study was conducted at the University of Pittsburgh School of Medicine and the University of Pittsburgh Medical Center.  The senior co-author of this study is David Wilson, MD.


Vineet K Raghu, Wei Zhao, Jiantao Pu, Joseph K Leader, Renwei Wang, James Herman, Jian-Min Yuan, Panayiotis V Benos, David O Wilson. Feasibility of lung cancer prediction from low-dose CT scan and smoking factors using causal models. Thorax, 2019; thoraxjnl-2018-212638 DOI: 10.1136/thoraxjnl-2018-212638 





The information on is provided for educational purposes only, and is in no way intended to diagnose, cure, or treat any medical or other condition. Always seek the advice of your physician or other qualified health and ask your doctor any questions you may have regarding a medical condition. In addition to all other limitations and disclaimers in this agreement, service provider and its third party providers disclaim any liability or loss in connection with the content provided on this website.


Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.