Labeling and clustering-based level set method for automated segmentation of lung tumor stages in CT images
- PDF / 1,776,241 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 102 Downloads / 212 Views
ORIGINAL RESEARCH
Labeling and clustering‑based level set method for automated segmentation of lung tumor stages in CT images K. Yamuna Devi1 · M. Sasikala1 Received: 21 April 2020 / Accepted: 10 July 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract An unconstrained growth of abnormal cells in the lung causes tumors. The aim of the work is the accurate diagnoses of tumor at its early stage through hybrid segmentation algorithm. The objective of this paper is to propose a novel labeled cluster active contour method to improve the performance of automated lung tumor segmentation from 2D CT slices. To the input 2D slice, an 8-connected component analysis is implemented to differentiate the image spatial information and obtain the RGB labeled data. The tumor location is foreseen using an unsupervised k-means clustering algorithm. Then, an automated level set algorithm is carried out to appropriately localize and extract the tumor region. The amount of initial level set curve evolution is controlled by the clustering efficiency of k-Means and labeling efficacy of connected component analysis. The quantitative evaluation of segmentation is carried out based on Shape features like area, perimeter, eccentricity, convex area, solidity, and roundness. The statistical object-based and distance-based metrics were used to find the similarity between manual and proposed methods. Also, performance metrics like accuracy, specificity, sensitivity, precision, and recall are used to validate the segmentation results of the proposed method. The proposed system is evaluated over 42 datasets from the commonly available large dataset LIDC (Lung Image Database Consortium). The accuracy, specificity, sensitivity and precision of the proposed hybrid method are 97.5%, 97.43%, 91.67%, and 98.79%, which exhibited the best competence as compared to traditional methods on the same dataset. The statistical and quantitative analysis shows the efficiency of the present work. Keywords Lung tumor · 8-connected component labeling (CCL) · k-means · Automated level set · Shape features · Statistical metrics
1 Introduction Across the world, a large proportion of cancer-related deaths occur mainly due to lung cancer (Xue et al. 2019). Lung cancer is produced by an abnormality in lung cell proliferation. Men are mostly incidence to lung cancer but woman are catching up, so the mortality rate of woman is greater than man. A benign tumor can be completely removed and thus can be stopped it’s spreading, whereas the malignant tumor grows aggressively and cannot be stopped spreading. The major two types of lung cancer are non-small cell * K. Yamuna Devi [email protected] M. Sasikala [email protected] 1
Centre for Medical Electronics, Department of Electronics and Communication Engineering, Anna University, Chennai, India
lung cancer (NSCLC) and small cell lung cancer (SCLC). The three subtypes of NSCLC are adenocarcinoma which starts in the outer part of the lungs, squamous cell carcinoma which starts in the middle
Data Loading...
