In this manuscript, we propose a novel hybrid Landmark and Contour-Matching (LCM) image registration model to align image pairs. The proposed model uses image contour information to supplement missing edge information in between exact landmarks. We demonstrate that the model circumvent the drawbacks associated with an straightforward application of the Thin Plate Spline (TPS) registration technique.
The proposed model provides higher post-registration Dice similarity between the reference and registered template images by improving the image overlap away from major landmarks and visually reduces the appearance of the ''unnatural bending'' typically present in TPS-registered images. We also show that naively increasing the number of landmarks in a TPS model does not always guarantee an accurate registration result. We indicate how the proposed model using even less number of exact landmarks along with additional approximate contour information provided suitable results, as opposed to the TPS model. Lastly, the proposed model produces physically relevant registration results with improved Dice similarity indices even when landmark localization errors are present in data.
Overall, the proposed Landmark and Contour-Matching (LCM) model increases the flexibility of the TPS approach especially when only a few landmarks can be defined, when defining too many landmarks leads to high oscillations in the registration transformations, or when the identification of exact landmarks is susceptible to human error.