Deep Learning Reconstruction of Chest CT Volumes from Single or Multi View Chest Radiographs
Principal Investigator
Name
Nadav Shapira
Degrees
Ph.D.
Institution
University of Pennsylvania
Position Title
Research Associate
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-697
Initial CDAS Request Approval
Jul 24, 2020
Title
Deep Learning Reconstruction of Chest CT Volumes from Single or Multi View Chest Radiographs
Summary
We want to explore possibilities of deep learning in the effective reconstruction of three-dimensional CT volumes from both single and multiple chest radiographs measured at varying projection angles. We intend to train two models with custom architectures: a 2D-to-3D encoder-decoder feature representation network and a generative adversarial network (GAN). We intend to use the NLST CT volumes to generate synthetic x-rays to create the training and validation datasets for these deep learning models.
Aims
- Use NLST CT volumes to generate synthetic chest radiographs at varying projection angles to create a training/validation dataset
- Use the training CT/x-ray data to train deep learning models for patient-specific chest CT reconstruction from single or multiple x-ray projections
- Compare generated CT volumes to actual NLST volumes to determine accuracies of developed models
Collaborators
Dr. Peter B Noel
Mr. Siddharth Bharthulwar