Generation and Evaluation of Different Modality of Medical Image Based on GAN
Chikato Yamasoba, Tetsuya Tozaki, Michio Senda · 2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC) · 2021
In recent years, computed tomography(CT) image, positron emission tomography(PET) image and anatomical image are often used for medical diagnosis. Each have different pros and cons. However, the diagnosis based on some modality images sometimes forces time and cost. Hence We aim to generate image which has different modality among CT image, PET image and anatomical image. Firstly, We apply Generative Adversarial Network for generate a psudo image. And generate image using DCGAN and CycleGAN. In the experiment about CycleGAN, We try to improve accuracy by changing the dataset. Finally, we evaluate generated image.