Ground Glass Nodule Segmentation Based on Regional Adaptive MRF Model

Abstract

Early lung cancer often manifests as lung nodules. Ground glass nodules (GGN) is most likely the early manifestation of lung cancer. The cancerous rate of GGN is much higher than that of solid nodules, and it is more difficult to identify. In this paper we propose a GGN segmentation method based on the regional adaptive Markov Random Field (MRF) model. First, threshold segmentation and morphological operations are used to segment the lung parenchyma. And the rolling ball and convex hull methods are used to repair the lung parenchyma contour. Secondly, fuzzy clustering algorithm (FCM) and original MRF model are used for coarse segmentation of GGN. After that, the Top Cap Transformation (TCT) is applied to remove blood vessels, reduce false positive and locate at GGN. So that statistical parameters can be updated. Finally, the regional adaptive MRF model is used for fine segmentation to obtain the GGN. This method can segment GGN more accurately and the average value of the overlap area ratio between the segmentation results of the regional adaptive MRF algorithm and the results of physician segmentation is 0.9144, while the existing methods can attain only 0.8663.