Camille Izard

Automatic Anatomical Landmark Detection

Anatomical landmarks, i.e. well defined points in the anatomy provide meaningful information on the local geometry. They are widely used to analyze shapes or as control points for many registration algorithms. However there detection, which remains manual, is a tedious and time-consuming task, even for specialists.

We propose a generic algorithm to detect landmarks in a new image. Using a training set of hand-labeled images, we learn the local geometry. In a new image, the location of the landmark(s) is given by likelihood maximization.

Examples of landmarks:

  • HoH : apex of the Head of the Hippocampus (1)

  • HT : tail  of the Hippocampus (2)

  • UA : posterior Apex of the hippocampal Uncus (3)

More details in this poster or the publications.