Distal Femur Allograft Selection Using a Spectral Shape Descriptor

The automatic selection of bone allografts in a virtual bone databank is an active line of research. This work presents a new approach to solve this problem, based on a recently published image descriptor, called Volumetric Heat Kernel Signature. This descriptor is used to compare the size and shape of three dimensional thresholded computed tomography volumes. This approach is compared to a published method that uses the Iterative Closest Points algorithm to register a segmented search template to different candidates present in the databank. Statistical testing of the agreement between the two methods show that both approaches render similar results in the relevant clinical context. The proposed method avoids incorrect registrations due to local minima and does not require lengthly manual image segmentation and positioning before its use. The new method is conceptually simple and its mathematical basis is sound