Developing a classifier of smoking history from chest x-ray images

Principal Investigator

Name
Hormuzd Katki

Degrees
Ph.D.

Institution
NCI/DCEG

Position Title
Senior Investigator

Email
katkih@mail.nih.gov

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCOI-514

Initial CDAS Request Approval
Aug 20, 2019

Title
Developing a classifier of smoking history from chest x-ray images

Summary
Smoking history is subjectively reported. To help provide an objective biomarker of smoking history, we propose to develop a classifier for smoking history from PLCO chest x-ray images. The classifier could have a stronger association with lung cancer than reported smoking history, in which case it may be useful in risk prediction models.

Aims

* Develop deep-learning-based classifier for aspects of smoking history (duration, intensity, quit-years, pack-years) by splitting data into training/validation/test sets
* Assess the accuracy of the predictions
* Include predictions as covariates into a risk-prediction tool, such as the Lung Cancer Risk Assessment Tool and see if it adds to, or replaces, reported smoking history

Collaborators

Martin Skarzynski, NCI/DCEG
Jonas de Almeida, NCI/DCEG
Anil Chaturvedi, NCI/DCEG
Li Cheung, NCI/DCEG
Rebecca Landy, NCI/DCEG