Unpaired PET/CT image synthesis of liver region using CycleGAN

Gianmarco Santini (1) , Constance Fourcade (1) , Noémie Moreau (1, 2) , Caroline Rousseau (3, 4) , Ludovic Ferrer (3, 4) , Marie Lacombe (4, 3) , Vincent Fleury (5) , Mario Campone (6, 4) , Pascal Jézéquel (5) , Mathieu Rubeaux (1)

1 KMI - Keosys Medical Imaging [Saint-Herblain]
2 LS2N - équipe IPI - Image Perception Interaction
3 CRCINA-ÉQUIPE 13 - Nuclear Oncology
4 UNICANCER/ICO - Institut de Cancérologie de l'Ouest [Angers/Nantes]
5 ICO - Integrated Center for Oncology [Angers]
6 CRCINA-ÉQUIPE 8 - Stress Adaptation and Tumor Escape in Breast Cancer

Cross-modality synthesis represent nowadays a promising application in medical image processing to manage the problem of paired data scarcity. In this work we designed and trained a CycleGAN model to generate PET/CT data from 2D slices collected from the liver body region of twelve patients. The results obtained from the six test patients show how our model can outperform baseline CycleGAN framework and effectively be used for synthesizing artificial images to be used for data augmentation or dataset completion.

Domaines: Cancer

Gianmarco Santini, Constance Fourcade, Noémie Moreau, Caroline Rousseau, Ludovic Ferrer, et al.. Unpaired PET/CT image synthesis of liver region using CycleGAN. 16th International Symposium on Medical Information Processing and Analysis, Oct 2020, Lima, Peru. pp.6, ⟨10.1117/12.2576095⟩. ⟨inserm-03500317⟩