Evaluation of a surface-to-volume registration method for prostate gland and its application to ultrasound image segmentation

This paper evaluated a surface-to-volume registration method for its ability of intra-operatively locating the prostate gland under the ultrasound (US) guidance, with the aid of pre-delineated prostate surface from the high-contrast magnetic resonance imaging (MRI) images of the same patient. The rigidity assumption and the registration credibility cross the modalities were both examined. When no deformation occurs, the prostate shape delineated from MRI images is consistent with the high gradients around the prostate boundary in the US images. The registration problem is therefore solved by looking for the transformation that maximizes the average image gradients along the surface normal. Self-registration experiments verified that the average accuracy is within 1 mm in translation and 3° in rotation. Cross registration shows that the prostate deforms between the modalities but the registration method is still effective when the deformation is small. And the registration accuracy deteriorates with the amount of deformation. To precisely locate and outline the prostate, segmentation can be applied based on the initialization from registration.