Registration of Thoracic CT-CT Images Using Improved Demon Registration

Computed Tomography (CT) image is commonly used for medical diagnosis, to monitor disease progression as well in radiotherapy planning and treatment. In basis, image registration aims to accurately align two or more monomodal or multimodal images taken at different time or places. In order to accurately register two CT images, an accurate and reliable registration algorithm is required. This paper proposes an improved Demon registration technique that uses sum of conditional variance (SCV) and multi-modality independent neighborhood descriptive (MIND) similarity metrics instead of the conventional sum squared difference (SSD) demon method to register whole body CT (PET/CT) and thoracic CT images that are acquired separately. We tested our proposed method on 9 whole body CT (PET/CT) and CT images of Non-small Cell Lung Cancer (NSCLC). Apart from visual observation, the proposed method is compared with the free form deformation (FFD) and standard demon methods. The registration accuracy was justified by measuring the lung volumes overlap between the two images post registration in terms of the Jaccard and Dice coefficients. The quality of the registered images was measured using three image quality metrics; structural similarity index (SSIM), peak signal to noise ratio (PSNR) and correlation coefficient (CC). In overall, the performance of the proposed demon is double than FFD and is superior than the standard demon. The average Jaccard and Dice coefficients are 0.83 and 0.90 respectively. Results of SSIM, PSNR and CC metrics also indicate that the improved demon method is the best, followed by FFD and standard demon.