Image Registration Assessment Performed with and without Landmarks Extracted using SIFT Technique

Abstract

In Image-Guided Radiation Therapy (IGRT), it is common to acquire several images of a patient and consequently perform image registration to compare the images. Therefore, both good registration and good quality assurance (QA) of the registration must be performed. The scope of this work is to assess an image registration when performed with and without landmarks. For this, Computed Tomography (CT) images of a radiation therapy patient were used to perform rigid and deformable registrations, with and without landmarks. The Scale Invariant Feature Transform (SIFT) technique was used to develop a code for the semi-automatic extraction of stable key points from images, that is, landmarks, both for registrations and the assessment of such registrations. Through the mean and maximum error and Mutual Information (MI) values ​​found, it was possible to verify a better alignment of the images when the registration was performed starting from the extracted landmarks, compared to the alignment performed without these landmarks. SIFT proved to be a great tool to perform both tasks and, when possible, the clinic professional should perform a good quantitative QA of image registration, considering landmarks distributed by the images.