The feasibility of using feature-flow and label transfer system to segment medical images with deformed anatomy in orthopedic surgery

In computer-aided surgical systems, to obtain high fidelity three-dimensional models, we require accurate segmentation of medical images. State-of-art medical image segmentation methods have been used successfully in particular applications, but they have not been demonstrated to work well over a wide range of deformities. For this purpose, I studied and evaluated medical image segmentation using the feature-flow based Label Transfer System described by Liu and colleagues. This system has produced promising results in parsing images of natural scenes. Its ability to deal with variations in shapes of objects is desirable. In this paper, we altered this system and assessed its feasibility of automatic segmentation. Experiments showed that this system achieved better recognition rates than those in natural-scene parsing applications, but the high recognition rates were not consistent across different images. Although this system is not considered clinically practical, we may improve it and incorporate it with other medical segmentation tools.