Unsupervised Generative Adversarial Networks for Bidirectional Brain MR-CT Lesion Segmentation
Medical image acquisition faces many challenges, such as cost, radiation dose, patient age, and limitation of images for certain types of brain disease. Medical image synthesis can assist in this endeavour by generating required medical images without real physical scan. A Generative Adversarial Network (GAN) model is considered an interesting model for CT-MR image synthesis. It is the next-generation artificial intelligent approach that has shown promising results in image generation and image synthesis. The purpose of this project is to generate realistic clinically significant bidirectional CT-MR images* for improving brain lesion segmentation in both modalities.
*Unpaired MR-CT Brain Image Dataset for Unsupervised Image Translation (Download)