Pritam Mukherjee,
Postdoctoral Scholar,
BMIR, Stanford University

ABSTRACT:

The National Lung Screening Trial (NLST) screened high risk individuals, heavy smokers and between the age of 55 and 74 years, for lung cancer, using either low dose CT or X-ray imaging, over three rounds at one-year intervals. Overall, for the low-dose CT cohort, about 95% of the positive screening results were actually false positives. Our aim in this project is to develop a deep learning model that can reduce the false positive rate without sacrificing sensitivity. In this talk, we will discuss the main challenges of using the NLST dataset for our deep learning task, and ways to counter them. We will also describe our deep learning based approaches and models and present some of our recent results.