Quantitative evaluation of margin sharpness of pulmonary nodules in lung CT images. Issue 9 (1st September 2016)

• Record Type: Journal Article
• Authors: Dhara, Ashis Kumar; Mukhopadhyay, Sudipta; Chakrabarty, Satrajit; Garg, Mandeep; Khandelwal, Niranjan
• Is Part Of: IET image processing. Volume 10:Issue 9(2016)
• Page Start: 631
• Page End: 637
• Publication Date: 2016-09-01
• Subjects: computerised tomography -- lung -- cancer -- tumours -- medical image processing -- image retrieval -- image segmentation; margin sharpness -- pulmonary nodules -- lung CT images -- lung cancer -- malignancy -- content‐based image retrieval system -- computer‐aided diagnosis system -- histogram spread -- nodule segmentation -- radiologist -- mean similarity -- normalised discounted cumulative gain -- LIDC/IDRI public database -- computerised tomography; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/iet-ipr.2015.0784
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract : Pulmonary nodules are the potential manifestation of lung cancer. Margin sharpness of a nodule is an important imaging characteristic for estimation of its malignancy. Quantitative evaluation of margin sharpness could be helpful in the development of content‐based image retrieval system of pulmonary nodules and computer‐aided diagnosis system for lung cancer. In this article, margin sharpness of a nodule is represented using the combination of histogram spread of averaged gradient and acutance of nodule. The novelty of the proposed method lies in the parsimony of model representing the margin sharpness. The most recent technique on the computation of margin sharpness of pulmonary nodules considers a representative slice of the nodule and it depends on manual segmentation of nodules by a trained radiologist. The authors have automated the most recent technique for comparative study. The reproducibility of the proposed technique is evaluated on reference image database to evaluate response (RIDER) coffee break data set. The performance of the proposed and competing techniques across modalities are evaluated in terms of precision, mean similarity and normalised discounted cumulative gain on 891 nodules of LIDC/IDRI public database. The proposed method outperforms the competing techniques.