A Semantic Model For Medical Image Registration

Rapid development in medical imaging and constant evolution in computer industry has enabled doctors to choose from a variety of techniques. To fully exploit the potential advantage of diversity in modalities, it is often of great bene?t to incorporate information from diverse images, which is exactly what medical image registration(MIR) does. Therefore, research in this areas has always been very active.Inspired by text analysis, a novel classi?cation methodology is proposed and classify MIR as pixel based, feature based and semantic information based, which correspond to words, phrase and semantic information, three level in text analysis.Semantic models, such as bag of words and latent factor model, have been widely used and fully explored in the ?eld of text analysis. Similarly, semantic models have been developed and accommodated to the speci?c requirement of computer vision.Natural scene is full of objects and colors, but not the case in medical images. Therefore, directional bag of visual words is proposed to overcome the shortcomings of visual words and takes directional information into consideration, which makes it more effective in medical images processing.Also, a semantic model for medical image registration is proposed. This method needs human interaction, and employs directional visual words model and k-means++algorithm to locate key parts. Then special emphasis is put on the areas around the key parts. This method can take advantage of semantic information, as well as ensuring a better registration accuracy around interested region, which is of great beni?t in practice.