Image registration based on automatic detection of anatomical landmarks for bone age assessment

Abstract

Bone age assessment is a medical procedure to diagnose bone diseases, specifically, growth pathologies. As of today, it is carried out by visual inspection which, needless to say, is a tedious and time consuming action. Automated methods to carry out such a task are therefore desirable. In this paper we take a step in this direction by proposing the automatic detection of a set of anatomical landmarks in positions of interest in radiographs. The algorithm locates the landmarks by performing a rough segmentation to find the finger axes and then intensity profiles along the axes are analyzed to locate joints between finger bones. Such landmarks are then used to carry out a registration procedure to eventually come up with a bone age estimation. Two landmark-based registration methods are compared, the TPS algorithm and the articulated registration algorithm. Registered images are compared using automatic and manual -positioned by an expert- landmarks with both registration algorithms.