Divide to Conquer: A Field Decomposition Approach for Multi-Organ Whole-Body CT Image Registration
This addresses the limitation of existing organ-specific registration methods for clinical practice by improving generalizability across multiple organs, though it is incremental as it builds on prior registration techniques.
The study tackled the problem of multi-organ whole-body CT image registration by introducing a novel field decomposition approach, which outperformed baseline optimization and deep learning methods on a dataset of 691 patients.
Image registration is an essential technique for the analysis of Computed Tomography (CT) images in clinical practice. However, existing methodologies are predominantly tailored to a specific organ of interest and often exhibit lower performance on other organs, thus limiting their generalizability and applicability. Multi-organ registration addresses these limitations, but the simultaneous alignment of multiple organs with diverse shapes, sizes and locations requires a highly complex deformation field with a multi-layer composition of individual deformations. This study introduces a novel field decomposition approach to address the high complexity of deformations in multi-organ whole-body CT image registration. The proposed method is trained and evaluated on a longitudinal dataset of 691 patients, each with two CT images obtained at distinct time points. These scans fully encompass the thoracic, abdominal, and pelvic regions. Two baseline registration methods are selected for this study: one based on optimization techniques and another based on deep learning. Experimental results demonstrate that the proposed approach outperforms baseline methods in handling complex deformations in multi-organ whole-body CT image registration.