PROJECT TITLE :

One-Shot Generative Adversarial Learning for MRI Segmentation of Craniomaxillofacial Bony Structures

ABSTRACT:

Magnetic resonance imaging (MRI) delineation of craniomaxillofacial (CMF) bone structures avoids dangerous radiation exposure as compared to computed tomography (CT). Bony borders are difficult to see in MRI, thus structural information from CT is used in the training process. Due to the rarity of linked MRI-CT data, this is a difficult task. A one-shot generative adversarial model for automated MRI segmentation of CMF bone structures is proposed in this study, which makes full use of unpaired data, which are often abundant, combined with a single pair of MRI-CT data. Cross-modal image synthesis and MRI segmentation are two of the sub-networks in our model, which learn the CT-to-MRI mapping. Both of these sub-networks are trained together as a whole. Also in the training phase, a neighbor-based anchoring method and a feature-matching-based semantic consistency constraint are presented to alleviate the ambiguity problem inherent in cross-modality synthesis. Our method outperforms state-of-the-art MRI segmentation algorithms both qualitatively and quantitatively, according to experimental data.