Multi-Modal Registration for Image-Guided Therapy

The real-time monitoring of non-stationary targets for image guided therapy is critical for accuracy in determining the relative locations of organs. Currently, methods for obtaining real-time monitoring using image guidance involve expensive intra-operative equipment, ionizing radiation, or are limited to surface imaging of areas accessible through videoendoscopic tools. Of all imaging modalities, ultrasound is the most cost-effective, portable, subsurface-capable, and non-ionizing. However, low contrast and high speckle noise content have prohibited its widespread use in image-guided therapy. We have sought solutions to two key elements involved in the registration of ultrasound with CT/MR: feature extraction in ultrasound and accurate multi-modal registration. A comparison is provided that illustrates substantial improvements in 2-D ultrasound edge detection using phase-based methods as opposed to traditional gradient-based methods. The elastic registration of 3-D CT/MR is accomplished by defining point correspondences used to compute a dense deformation field, which maps the atlas image (CT) from its coordinate system into the patient image (MR). We performed quantitative and qualitative error analysis to determine the accuracy of the registration. Also, a preliminary investigation into the Iterative Closest Point Algorithm (ICP) was begun and results are shown for rigid and non-rigid point patterns with the objective of extending the algorithm to register ultrasound edge points with CT/MR surface points. Thesis Supervisor: W.E.L. Grimson Title: Bernard Gordon Professor of Medical Engineering