Synthesizing a complete tomographic study with nodules from multiple radiograph views via deep generative models

Chest radiography encodes a 3D anatomy into a complex 2D representation. This projection creates distinctive challenges even for the most experienced radiologists as many critical findings are superimposed, often resulting in error or further imaging. In particular, it is difficult to visualize the volume and density of lung nodules in chest radiographs. A deep generative model, commonly used to synthesize realistic images, can be used to perform 2D to 3D translations. In this thesis, we propose a generative model, optimized using pixel-wise error, that can synthesize a complete tomographic study containing nodules from frontal and lateral chest x-ray radiographs. Additionally, the generated studies maintain the proper chest cavity structure.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Chen, Andrew

Contributors dc:contributor

• Koyejo, Oluwasanmi

• radiography
• machine learning
• generative modeling
• synthesis

Statement dc:rights

• Copyright 2020 Andrew Chen

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108587

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108587