Purpose: A previously proposed reconstruction of difference (RoD) algorithm enables ultra-low-dose sequential imaging applications like lung cancer screening. The algorithm reconstructs the difference image between the prior and current anatomy to directly highlight changes in lesion morphology. However, deformable registration required in such methods may distort the lesion, representing anatomical change (e.g., growth/shrinkage) as deformation. This work proposes a modified registration method that accounts for anatomical misalignment while preserving the lesion change in the difference image.
Methods: Previous implementations of registration in RoD have relied on a diffeomorphic demons algorithm applied to the entire prior image. We modify the approach by imposing a locally-smooth requirement within a region-of-interest around the nodule. Specifically, we used an inpainting method to obtain a smoothly varying displacement field to account for deformation from surrounding soft tissue without fitting to changes in the nodule. The method was tested in patient data where a nodule shows visible shrinkage from the prior to the current image.
Results: Naive implementation of deformable registration without modification shows that a nodule in a registered prior image is deformed greatly in an attempt to match to the nodule morphology of the current image. By modifying the output displacement field in and around the nodule region of interest, the true morphology of the nodule in the prior image is recovered with no major changes in the alignment between the current and prior image, representing the true change in nodule shape and size.
Conclusion: The modified registration method preserves nodule size and shape while effectively aligning prior and current anatomy. Thus, this work has the potential to facilitate reliable clinical deployment of RoD in low-dose lung nodule surveillance.
Funding Support, Disclosures, and Conflict of Interest: This work was supported, in part, by NIH grant R21CA219608.
Not Applicable / None Entered.