UCLA

Medical image registration automatically brings two images into maximal spatial and anatomical correspondence. The registration of brain images taken at different modalities and time points helps in designing a treatment plan in neurology besides assisting in surgery. The registration of stained and unstained skin samples in dermatology used in virtual histological staining helps the dermatologists by removing the labor and cost that is involved in invasive techniques such as skin biopsies. The registration of histological skin images can be more challenging than the registration of brain images due to repetitious image textures, and the various expansions and morphological changes of tissue samples caused by the contrast agent used to stain the skin. For these reasons, the standard registration approaches may not adequately address this area of medical image registration. Advanced registration techniques have been developed to overcome such challenges. Recently, machine learning methods have gained popularity in multi-modal medical image registration. The goal of the thesis is to apply the standard, advanced, and machine learning-based multi-modal registration methods to the 3D microscopic histological dermatological data and to compare their performance.