Generative adversarial networks and radiomics supervision for lung lesion synthesis

Shaoyan Pan, Jessica Flores, Chen Ting Lin, J. Webster Stayman, Grace J. Gang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Realistic lesion generation is a useful tool for system evaluation and optimization. Generated lesions can serve as realistic imaging tasks for task-base image quality assessment, as well as targets in virtual clinical trials. In this work, we investigate a data-driven approach for categorical lung lesion synthesis using public lung CT databases. We propose a generative adversarial network with a Wasserstein discriminator and gradient penalty to stabilize training. We further included conditional inputs such that the network can generate user-specified lesion categories. Novel to our network, we directly incorporated radiomic features in an intermediate supervision step to encourage similar textures between generated and real lesions. We calculated the network using lung lesions from the Lung Image Database Consortium (LIDC) database. Lesions are divided into two categories: solid vs. non-solid. We performed quantitative evaluation of network performance based on four criteria: 1) overfitting, in terms of structural and morphological similarity to the training data, 2) diversity of generated lesions, in terms of similarity to other generated data, 3) similarity to real lesions, in terms of distribution of example radiomics features, and 4) conditional consistency, in terms of classification accuracy using a classifier trained on the training lesions. We imposed a quantitative threshold for similarity based on visual inspection. The percentage of non-solid and solid lesions that satisfy low overfitting and high diversity is 87.1% and 70.2% of non-solid and solid lesions respectively. The distribution of example radiomics features are similar in the generated and real lesions indicated by low Kullback-Leibler divergence scores: 1.62 for non-solid lesions and 1.13 for solid lesions. Classification accuracy for the generated lesions are comparable with that for the real lesions. The proposed network presents a promising approach for generating realistic lesions with clinically relevant features crucial for the comprehensive assessment of novel medical imaging systems.

Original languageEnglish (US)
Title of host publicationMedical Imaging 2021
Subtitle of host publicationPhysics of Medical Imaging
EditorsHilde Bosmans, Wei Zhao, Lifeng Yu
PublisherSPIE
ISBN (Electronic)9781510640191
DOIs
StatePublished - 2021
EventMedical Imaging 2021: Physics of Medical Imaging - Virtual, Online, United States
Duration: Feb 15 2021Feb 19 2021

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume11595
ISSN (Print)1605-7422

Conference

ConferenceMedical Imaging 2021: Physics of Medical Imaging
CountryUnited States
CityVirtual, Online
Period2/15/212/19/21

Keywords

  • Deep learning
  • Generative adversarial network
  • Lesion generation
  • Virtual clinical trial

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Biomaterials
  • Atomic and Molecular Physics, and Optics
  • Radiology Nuclear Medicine and imaging

Fingerprint Dive into the research topics of 'Generative adversarial networks and radiomics supervision for lung lesion synthesis'. Together they form a unique fingerprint.

Cite this