Three-Dimensional Morphometry and Non-Rigid Registration for Quantitative Analysis and Clinical Assessment in Radiology

The capacity to reliably track, model and characterize morphometric changes in anatomic structures and tumors from 3-D images sequences is extremely valuable in staging disease progression and assessing response to treatment. We have designed, developed and evaluated two approaches to facilitate clinical assessment in diagnostic radiology. The first is a tool for performing comparative morphologic analysis and the second is a registration strategy which can compensate for changes in shape that occur in deformable organs when assessing response to treatment across consecutive imaging studies. The first prototype system was used to characterize the morphology of ventricles from MR brain scans of patients who had been diagnosed with Bipolar Disorder or Asperger's Syndrome. Preliminary studies demonstrated that conventional volumetric measurements were insufficient for detecting and characterizing subtle changes in anatomic profiles. We have investigated the use of a double elliptic Fourier transform to discriminate among 3-D changes of anatomic structures. It was shown that characterization using low frequency elliptic Fourier descriptors provided an accurate representation of the anatomical structures while allowing for reliable group separation. The shape-based 3-D object representation of brain structures developed in this project may provide insight regarding the underlying mechanisms leading to the onset and progression of these disorders. As an extension of these studies, a deformable registration technique was evaluated for tracking tumor response to radiofrequency ablation of patients with liver malignancies. The method exploits the combined power of global and local alignment of pre- and post-treatment CT images. The distinguishing characteristics of the system is that it can infer volumetric deformation based upon surface displacements using a linearly elastic finite element model (FEM). Using both 2-D synthetic phantoms and 3-D beef liver data we performed the simulation of gold standard registration by measuring the accuracy of non-rigid deformation. The sub-voxel mean displacement error of deformation demonstrates that the technique provides valuable information for surgical interventions. This approach is general methodology for tracking deformable organs using non-rigid registration with respect to FEM simulations. It provides a basis for monitoring tissue response and therapy planning for a range of medical applications in the brain, breast or heart.