Mass Transportation for Deformable Image Registration with Application to Lung CT

Computed Tomography (CT) of the lungs play a key role in clinical investigation of thoracic malignancies, as well as having the potential to increase our knowledge about pulmonary diseases including cancer. It enables longitudinal trials to monitor lung disease progression, and to inform assessment of lung damage resulting from radiation therapy. We present a novel deformable image registration method that accommodates changes in the density of lung tissue depending on the amount of air present in the lungs inspiration/expiration state. We investigate the Monge-Kantorovich theory of optimal mass transportation to model the appearance of lung tissue and apply it in a method for registration. To validate the model, we apply our method to an inhale and exhale lung CT data set, and compare it against registration using the sum of squared differences (SSD) as a representative of the most popular similarity measures used in deformable image registration. The results show that the developed registration method has the potential to handle intensity distortions caused by air and tissue compression, and in addition it can provide accurate annotations of the lungs.