Image-based evaluation of pathologies over time is complicated by variability in image acquisition and patient positioning, deformations due to respiratory and cardiac motion, and the clinical standard in radiology of comparing unaligned images. This application proposes a computational method for longitudinal image analysis that captures, illustrates and quantifies pathology-related changes over time, by compensating for non-linear background motion. Although the method is general, we will use one of the most challenging yet routinely conducted longitudinal imaging studies as a driving clinical problem, namely lung cancer assessment using periodic CT imaging for evaluating patient-specific response to therapy and for restaging. In particular, we focus on subsolid nodules, also known as ground-glass opacities (GGOs), which must be carefully monitored for signs of malignancy but which are difficult to compare across scans because of the significant impact that global lung motion has on their appearance. We propose continued development and novel application of our """"""""geometric metamorphosis"""""""" deformable image registration method. Unlike previous approaches, geometric metamorphosis handles growing or contracting pathologies whose morphology is additionally impacted upon by non-rigid global background deformations. This is achieved by simultaneously estimating both the global background and lesion deformations. Clinical applications of such a registration method are to 1) ease data interpretation for reduced workload and increased throughput in manual radiologist review while simultaneously increasing the accuracy of longitudinal measurements, by displaying aligned scans that emphasize lesion change after non-rigid background motion has been eliminated;2) enable voxel- wise metrics of lesion response to treatment, which includes estimating a lesion's deformation magnitude, internal composition changes and infiltrative growth pattern over time. Geometric metamorphosis was originally developed for change detection in neuroimaging, and the first goal is to continue to demonstrate its wide applicability and utility by customizing the method for serial chest CT registration. Second, it is hypothesized that using physically-inspired heuristics to model the patient-specific, inhomogeneous elasticity of lung tissue will enable more accurate recovery of the jointly estimated lesion deformations. Geometric metamorphosis will be extended to integrate locally adaptive elastic regularization and test our hypotheses of improved target registration accuracy and improved accuracy of the recovered lesion deformations. Finally, the clinical utility of the registration software will be demonstrated within a retrospectve clinical evaluation. This will be done by comparing the image-based longitudinal measurements derived from both geometric metamorphosis methods with those from alternative image registration strategies, and correlating extracted voxel-wise lesion deformations and growth characteristics with patient diagnoses of benign versus cancerous lesions.
Clinicians use periodic medical imaging to monitor a patient's disease and evaluate whether a treatment is effective. However, quantitatively measuring a pathology's change in size or rate of growth is difficult because changes in image appearance make it difficult to accurately reproduce measurements on different scans. We are developing a computer algorithm that, given a patient's medical images, can recover a pathology's deformation over time while accounting for the impact of background deformations, which will give clinicians more accurate metrics for determining a patient's prognosis or deciding whether a patient should be switched to a new treatment protocol.
| Liu, Xiaoxiao; Niethammer, Marc; Kwitt, Roland et al. (2015) Low-Rank Atlas Image Analyses in the Presence of Pathologies. IEEE Trans Med Imaging 34:2583-91 |
| Liu, Xiaoxiao; Niethammer, Marc; Kwitt, Roland et al. (2014) Low-rank to the rescue - atlas-based analyses in the presence of pathologies. Med Image Comput Comput Assist Interv 17:97-104 |
