ABSTRACT

CONTENTS 9.1 Introduction 130 9.2 GPU vs CPU 131 9.3Image Registration Algorithms 133

9.3.1B-spline-Based Free-Form Deformation 133 9.3.2 Demons 135 9.3.3Optical Flow Method 136 9.3.4Evaluation of Image Registration 138

9.4Clinical Uses of Unimodal Image Registration 139 9.4.1Image Fusion 139 9.4.2 Segmentation 139 9.4.3Adaptive Radiation erapy 142

9.5 Conclusion 143 Acknowledgments 144 References 144

The efficacy of radiation therapy, beyond tumor biology, is deter-mined by two issues that are increasingly reliant on medical imaging: (1) accuracy in delineation of target tumor and neighboring organs at risk (OARs) and (2) accuracy of treatment delivery [1]. Accurate target contouring utilizes high contrast and spatial resolution imaging and can involve multimodal imaging. However, unimodal image registration remains prominent in clinical practice, as observed in strategies that use serial CT imaging to monitor treatment ecacy and re-treatment of patients where there is concern of excess dose to OARs. In these cases, the patient anatomy necessitates deformable image registration (DIR). While DIR is clearly necessary for extracranial imaging, it can also play a role in cranial imaging, where physiological changes can deform the anatomy [2]. us, fast computation strategies are needed for unimodal deformable image registration that is used in a variety of clinical settings.