(Automatic) Identification of 3D landmarks – first steps

In spite of the increasing number of research projects using 3D landmark data, so far only a very limited number of software packages are available for manual landmark identification and, in general, they are not specifically designed for this purpose. Placing landmarks on 3D data is thus generally very time-consuming and can be imprecise, depending on the structure to be marked. Moreover, the analysis of 3D data would greatly benefit from the opportunity to use automatically generated landmarks in order to improve precision, comparability, and speed of data acquisition. We propose an automatic landmark identification system for 3D data, in which manually identified examples of each landmark are used to define a likelihood function based on differences in appearance and general assumptions regarding measurement noise. Parameter covariances can be computed from this function in order to quantify location accuracy. This supports an interactive system, in which poorly located landmarks (i.e. those with larger than expected errors or low match scores) can be referred to the user for manual mark-up, and then added to the database of examples. Here we present, as a first step, a tool for the manual registration of landmarks on 3D micro-CT images of rodent skulls, which allows multiple simultaneous views of the data, including a 3D volume rendering generated in real time using the shear-warp algorithm.