Integration of Interactive Corrections to Model-Based Segmentation Algorithms

3D deformable shape models have become a common approach for solving complex segmentation tasks in medical image processing. Nevertheless sometimes the segmentation fails due to low image resolution or contrast, structures lying closely together or an insufficient initialization of the model. Although the error is often obvious to physicians, they have no opportunity to improve the result. This paper presents 3D tools for the correction of erroneous segmentations and provides a method which allows the integration of these corrections to deformable model-based segmentation methods. The integration is accomplished by a user deformation energy which is defined in a way that allows efficient corrections without the need to segment the complete erroneous region manually. This new approach is illustrated on the segmentation of a vertebra and a femur-head.