VARIATIONAL APPROACH FOR SEGMENTATION OF LUNG NODULES Amal A. Farag a , Hossam Abdelmunim ab , James Graham a , Aly A. Farag a , Salwa Elshazly a Sabry El-Mogy cd ,Mohamed El-Mogy c , Robert Falk f , Sahar Al-Jafary e , Hani Mahdi b , and Rebecca Milam g a Computer Vision and Image Processing Laboratory (CVIP Lab), University of Louisville, Louisville, KY 40292 b Computer & Systems Engineering Department, Faculty of Engineering, Ain Shams University, Cairo, Egypt c School of Medicine, Mansoura University Egypt d Mogy Scan, Mansoura, Egypt e School of Medicine, Ain Shams University, Cairo, Egypt f Jewish Hospital and 3DR, Louisville, Kentucky g University of Louisville, Department of Radiology ABSTRACT Lung nodules from low dose CT (LDCT) scans may be used for early detection of lung cancer. However, these nodules vary in size, shape, texture, location, and may suffer from occlusion within the tissue. This paper presents an approach for segmentation of lung nodules detected by a prior step. First, regions around the detected nodules are segmented; using automatic seed point placement levels sets. The outline of the nodule region is further improved using the curvature characteristics of the segmentation boundary. We illustrate the effectiveness of this method for automatic segmentation of the Juxta-pleural nodules. 1. INTRODUCTION Computer-Assisted Diagnosis (CAD) methods lend benefit to the radiologists in early detection and follow-up of doubtful nodules visible in the low dose CT. various machine learning methods have been used for automatic and semi-automatic nodule detection (e.g., [1-3]). Our proposed CAD framework for automatic detection and classification of lung nodules consists of four main steps: 1. Pre- processing of the scans (i.e. acquisition artifacts removal and noise filtering); 2. Segmentation to isolate the lung tissue from the rest of the chest region; 3. Nodule detection to isolate candidate nodules and 4. Nodule classification of the detected nodules into possible pathologies. In this paper we use the level set methods for modeling and segmentation of the lung nodules which appear in low dose computer tomography (LDCT) scans. A pulmonary nodule in radiology is defined as a mass in the lung usually spherical in shape; however, it can be distorted by surrounding anatomical structures such as the pleural surface. In our work, we use the classification by [4] which sorts the nodules into four categories: juxta-pleural where a significant portion of the nodule is connected to the pleural surface; vascularized where significant connection(s) to the neighboring vessels is seen from the nodule, while being located centrally in the lung tissue; well- circumscribed are nodules centrally located in the lung without vasculature connection; and pleural-tail which is located near the pleural surface but connected by a thin structure. In all of these types there is no limitation on size or distribution in the lung tissue. Each of these types of nodules possesses shape characteristics that can be quantified and used in the energy function that controls the propagation of deformable models used to automatically outline the spatial support of the detected nodules. This paper uses mainly the curvature measures, and will use the Juxta-pleural nodules type as a case study. The closest related works to our applications are the following: In [5], automatic detection of lung nodules was performed in the signed distance field of the CT images. The main steps in this work were, first, linearly interpolating the CT images along the axial direction to form an isotropic data set, then a segmentation approach was applied to smooth the lung boundaries, and finally detection of candidate lung nodules by computing the local maximas of signed distances in each subvolume. In [6], a 2D multiscale filter was used to detect candidate nodules, then a region growing approach was used to distinguish between nodules and non-nodules, followed by false positive reduction step. In [7], we described three different methodologies for modeling the lung nodules and obtained templates for each nodule type, which we deployed for nodule detection using normalized cross-correlation (NCC). In [8], we examined false positive reduction using geometric feature descriptors. This paper deals with segmentation of nodules after the false positive step. By segmentation, we mean extraction of the spatial support of the nodules, starting from the positions of candidate nodules. These candidates may be random in a given region of the slice or 3D volume. Variational approaches provide a powerful mechanism for accomplishing our purpose, given prior knowledge about the shape characteristics of candidate nodules. In particular, the curvature properties are different in the four nodule classes under consideration (see Fig. 1). Variational level set approaches for segmentation is preferred over statistical methods followed by morphological operations [9]. The level set segmentation specifies the boundary between lung and background tissues. This boundary has a curvature which is maximized at the nodule region. This technique will be used to minimize false positives and hence enhance the detection rate. 2011 18th IEEE International Conference on Image Processing 978-1-4577-1303-3/11/$26.00 ©2011 IEEE 2157