Project ID: P202208030001
Feasibility study of synthesis from CT to PET using CycleGAN network for head and neck cancer patients
Deep Learning, Medical Images
• Deep learning (Python) • Medical Image expert
The objective of image translation is to map an image onto another image with different domains. A growing number of applications in computer vision and graphics, including image synthesis, style transfer, and segmentation tasks, broadly benefit from this approach. The overall objective of image-to-image translation is to find a mapping T to convert an input image in domain A to a target image in domain B, preserving the source content and changing the irrelevant feature of the input image. While synthetic medical images have been more widely investigated in the case of other cancers, synthetic PET has not received much attention in the case of head and neck cancer patients with generative adversarial networks, including Cycle Generative Adversarial Network (CycleGAN). Hence, this study aims to study the clinical feasibility of employing synthesis from CT to PET using CycleGAN for head and neck cancer patients. The synthetic PET images are then compared with the real PET images based on the quantitative evaluation.