Multimodal image-guidance for noninvasive surgery: registration, segmentation, and statistical imaging models

Summary form only given, as follows. Image-guided noninvasive and minimally invasive therapeutic procedures are becoming increasingly attractive in the practice of surgery in modern medicine. The last decade has witnessed significant efforts to develop therapeutic devices employing various forms of non-ionizing radiation to produce localized tissue necrosis and/or ablation to achieve a desired therapeutic end point. However, image guidance is one of the major challenges common to all noninvasive and minimally invasive procedures including biopsy, thermal ablation, endoscopy, and laparoscopy. Interactive image guidance paradigms increasingly utilize complimentary information from image data from two different modalities. In thermal ablation, for example, 3D MR patient data sets are utilized for treatment planning and target delineation while 2D real-time ultrasound is utilized for visualization of the ablated region. In this case, successful registration of images from the two modalities is key to the eventual success of this kind of noninvasive surgery in the future. The general area of multimodality image registration is currently receiving significant attention from the medical image processing community. Interactive image guidance requires intramodality as well as intermodality registration of time varying images of the region of interest. The time varying nature of the problem is due to tissue motion and deformation as well as changes in tissue properties due to the therapeutic agents, e.g., heat. Here, the current status of image registration in medical image processing is described. Examples of frame based, point landmark based, and voxel based image registration algorithms are given. Some of the special considerations for successful registration of time varying imagery undergoing motion and deformation are described. Optical flow techniques for motion analysis and target tracking are discussed. In addition, statistical imaging models for treatment monitoring and damage assessment are addressed and illustrated with examples. Signal processing aspects of the outstanding problems are highlighted