Non-Deep-Learning-Based Medical Image Synthesis Methods
Jing Wang, Xiaofeng Yang · 2023
Image synthesis is of increasing enthusiasm due to its capability of generating images for unavailable modalities from existing databases. Image synthesis, e.g., synthetic computed tomography (CT) from MR sequences, shows promising potential in clinical workflows such as MRI-only radiotherapy or positron emission tomography (PET)/magnetic resonance (MR) attenuation correction. This chapter will overview the non-deep-learning-based and traditional machine-learning-based medical image synthesis methods, including bulk density overriding, atlas-based synthesis, and traditional machine-learning-based techniques. Most of the work focused on the brain/head site and utilized atlas-based methods. The atlas and machine-learning-based techniques were able to achieve acceptable synthetic CT/MR images in most cases to potentially contribute to simplified care pathways.