Deterministic Automatic Segmentation in MRI, CT and CTA: A Robust Method Based on Anatomical Structures Modelling and Local Recursive Intensity Analysis

INTRODUCTION Automatic segmentation of human anatomical structures is a challenging problem in medical images analysis. At the state of the art some semiautomatic and few automatic procedures have been developed, based on different techniques as threshold, clusters, deformable models and probabilistic methods [1]. Depending on the acquisition technique and on the anatomical structure, these methods offer different results. In MRI, manual, semiautomatic or automatic software have been developed for brain segmentation in the research sector [1]. Automatic software have been developed for bone segmentation in CT, while semi-automatic or multiscans algorithms have been developed for heart and vessels segmentation in CTA [2]. We describe a robust automatic method for the deterministic identification and segmentation of anatomical structures in medical 3D images. The method is based on the modelling and the analysis of anatomical structures physical and geometrical features. The method is characterized by high precision and use of non subjective parameters. METHODS We analyzed data of single anatomical structures, including the brain, white matter, grey matter and skull, in a database of 250 MRI scans of the head [4]. We then analyzed data of single anatomical structures, including the bone, the heart, the aorta, urinary tracts, bladder and kidneys, in a public database of CT and MRI (http://pubimage.hcuge.ch:8080/Obelix). We used mathematical models to perform analysis of the tissues intensity distributions. Physical and geometrical properties of the systems determine intensity distributions. Therefore they can be modelled as the combination of a function describing the random variability of the tissue detection intensity and a function describing its geometrical properties. Assuming that the problem is characterized by spherical symmetry, the function describing the tissues intensity distribution is: