An Algorithm for the Pulmonary Border Extraction in PET Images

Abstract

Automatic segmentation methods are mandatory for most computer-aided diagnostic methods, particularly in Positron Emission Tomography (PET), being extremely relevant for quantification purposes, disease diagnosis and staging. These automated approaches speed up the clinical work-flow, eliminating the need of a tedious manual delineation by physicians and greatly improving the reproducibility of the delineation procedures.This paper presents a novel approach for the fully automatic extraction of the pulmonary boundaries for PET images based in the concept of “marker-driven watershed segmentation”. Additionally, an algorithm for the lung border extraction in CT images was developed in response to physicians’ requirements for a better understanding of each individuals’ specific anatomy. The accuracy of both algorithms was assessed, comparing the results of both approaches to their correspondent manually depicted contours by several physicians, taking into account several figures of merit.