Semi-automatic medical image segmentation

We present a system for semi-automatic medical image segmentation based on the livewire paradigm. Livewire is an image-feature driven method that finds the optimal path between user-selected image locations, thus reducing the need to manually define the complete boundary. Standard features used by the wire to find boundaries include gray values and gradients. We introduce an image feature based on local phase, which describes local edge symmetry independent of absolute gray value. Because phase is amplitude invariant, the measurements are robust with respect to smooth variations, such as bias field inhomogeneities present in all MR images. We have implemented both a traditional livewire system and one which utilizes the local phase feature. We have investigated the properties of local phase for segmenting medical images and evaluated the quality of segmentations of medical imagery performed manually and with both systems. Thesis Supervisor: W. Eric L. Grimson Title: Bernard Gordon Professor of Medical Engineering Thesis Supervisor: Carl-Fredrik Westin Title: Instructor in Radiology, Harvard Medical School