A New Stochastic Framework for Accurate Lung Segmentation Ayman El-Ba 1 , Georgy Gimel’farb 2 , Robert Falk 3 , Trevor Holland 1 , and Teresa Shaffer 1 1 Bioimaging Laboratory, Bioengineering Department, University of Louisville, Louisville, KY, USA 2 Department of Computer Science, University of Auckland, Auckland, New Zealand 3 Director, Medical Imaging Division, Jewish Hospital, Louisville, KY, USA Abstract. New techniques for more accurate unsupervised segmenta- tion of lung tissues from Low Dose Computed Tomography (LDCT) are proposed. In this paper we describe LDCT images and desired maps of regions (lung and the other chest tissues) by a joint Markov-Gibbs random field model (MGRF) of independent image signals and interde- pendent region labels but focus on most accurate model identification. To better specify region borders, each empirical distribution of signals is precisely approximated by a Linear Combination of Discrete Gaus- sians (LCDG) with positive and negative components. We modify a con- ventional Expectation-Maximization (EM) algorithm to deal with the LCDG and develop a sequential EM-based technique to get an initial LCDG-approximation for the modified EM algorithm. The initial seg- mentation based on the LCDG-models is then iteratively refined using a MGRF model with analytically estimated potentials. Experiments on real data sets confirm high accuracy of the proposed approach. 1 Introduction Lung Cancer remains the leading cause of cancer-related deaths in the US. In 2006, there were approximately 174,470 new cases of lung cancer and 162,460 related deaths [1]. Early diagnosis of cancer can improve the effectiveness of treatment and increase the patient’s chance of survival. Segmentation of the lung tissues is a crucial step for early detection and diagnosis of lung nodules. Accurate segmentation of lung tissues from LDCT images is a challenging problem because some lung tissues such as arteries, veins, bronchi, and bronchioles are very close to the chest tissues. Therefore, the segmentation cannot be based only on image signals but have to account also for spatial relationships between the region labels in order to preserve the details of the lung region. In the literature, there are many techniques developed for lung segmenta- tion in CT images. Sluimer et al. [2] presented a survey on computer analysis of the lungs in CT scans. This survey addressed segmentation of various pul- monary structures, registration of chest scans, and their applications. Hu et al. [3], proposed an optimal gray level thresholding technique which is used to se- lect a threshold value based on the unique characteristics of the data set. A D. Metaxas et al. (Eds.): MICCAI 2008, Part I, LNCS 5241, pp. 322–330, 2008. c  Springer-Verlag Berlin Heidelberg 2008