The exposure index is an important measure used in digital radiography to control the dose at the detector. This value should be computed in regions of interest that are adapted to each patient's anatomy and pose.

We propose to define automatically these regions based on anatomical landmarks in the main structures of interest (head, thoracic spine, lungs, lumbar spine, pelvis, femurs, knees, tibiae). This task is achieved by combining the global information on the size and the positions of the anatomical structures on the one hand, with local analysis on the other hand.

Experimental results, on a varied database of 82 full-body acquisitions, demonstrate the interest of the proposed approach, with less errors than existing approaches, in particular on frontal view acquisitions. The method is also robust to variations in patient's conditions and to the potential presence of metallic objects.

The approach proposed in this paper allows consistently estimating exposure index values associated with different X-ray acquisitions. This suggests that the application of the proposed method to clinical practice is promising.