Deep Learning systems for detection of nodules and for simulation of new nodule training data.
Principal Investigator
Name
Keelin Murphy
Degrees
Ph.D.
Institution
Radboud University Medical Center
Position Title
Postdoctoral Researcher
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-812
Initial CDAS Request Approval
Jul 2, 2021
Title
Deep Learning systems for detection of nodules and for simulation of new nodule training data.
Summary
We are investigating the use of deep learning to detect nodules in CXR images. We would additionally like to investigate whether deep learning systems to simulate new training data are effective in improving the performance of detection systems. We will utilize all publicly available datasets with nodule bounding box annotations added by radiologists from our group where they are not available already.
Aims
- to determine the maximum performance of deep learning systems in the detection and localization of nodules in CXR images
- to determine whether performance can be improved by the use of training data with unique fake nodules, generated by a deep learning system trained for that task
- to compare the performance of deep learning systems for nodule detection in CXR (CT confirmed nodules) with the performance of multiple expert readers on the same images.
Collaborators
Professor Bram van Ginneken, Radboud University Medical Center
Ecem Sogancioglu, Radboud University Medical Center