Subject-Specific Lesion Generation and Pseudo-Healthy Synthesis for Multiple Sclerosis Brain Images

Understanding the intensity characteristics of brain lesions is key for defining image-based biomarkers in neurological studies and predicting disease burden outcome. In this work, we present a novel foreground-based generative method modelling local lesion that can both generate synthetic on healthy images synthesize subject-specific pseudo-healthy from pathological images. Furthermore, proposed be used as data augmentation module to training image segmentation networks. Experiments multiple sclerosis (MS) acquired magnetic resonance imaging (MRI) demonstrate highly realistic pseudo-pathological Data using improves performance compared traditional methods well recent lesion-aware technique, CarveMix. The code will released at https://github.com/dogabasaran/lesion-synthesis .