Self-adapting threshold of pulmonary parenchyma

Abstract

Segmentation for pulmonary parenchyma is a crucial step for computer aided diagnosis (CAD) systems. The accuracy of pulmonary parenchyma segmentation can have a great impact on further steps of CAD systems, such as pulmonary nodule detection and feature extraction. Before segmentation, preprocessing should be done to remove references outside the thorax. After preprocessing, pulmonary parenchyma area threshold will be employed to realize segmentation. However, current segmentation approaches are mainly based on a fixed area threshold, which confronts problem of mis-segmentation and high segmentation failure rate. This article proposed a novel self-adapting threshold segmentation approach, which realized fully automatic segmentation and held considerable accuracy rate. First, fitting of polynomials based on the least square law is constructed to fit curves of pulmonary parenchyma areas. Secondly, a golden standard is created to represent change trend of pulmonary parenchyma for all patients. Finally, the golden standard is adjusted accordingly to realize full adaption and automation of segmentation. Experimental results indicated that the proposed approach achieved excellent accuracy rate and precise segmentation.