Triple novelty block detection and classification approach for lung tumor analysis. Issue 2 (23rd October 2020)

• Record Type: Journal Article
• Authors: Rani, K. Vijila; Dayana, C. Thinkal; Therese, P. Sujatha; Prince, M. Eugine
• Is Part Of: International journal of imaging systems and technology. Volume 31:Issue 2(2021)
• Page Start: 1034
• Page End: 1049
• Publication Date: 2020-10-23
• Subjects: bilateral filter -- CAD -- firefly search‐based Macqueen's K‐means clustering -- histogram‐based affine invariant detection -- lung CT -- nodule detection -- spatial constraint‐based new fuzzy clustering -- SVM with radial basis functional classification -- texture feature; Imaging systems -- Periodicals; Image processing -- Periodicals; 621.367
• DOI: 10.1002/ima.22509
• Languages: English
• ISSNs: 0899-9457

Abstract

Abstract: Recently the increased utilization of computer‐aided detection is more helpful to assist the radiologist, recognize the minute lung lesion. In this research article, we propose three novel block detection methods for lung lesion segmentation. First, an automatic algorithm for lung lesion detection and segmentation by using histogram‐based affine‐invariant detection (HAID) is proposed. HAID is a novel segmentation technique that applies enhanced maximally stable extremal region algorithm results to segment the lung lesion. The lung lesions are segmented with an accuracy of 91.5% and false positives per scan is 5.52. To reduce the false positives per scan of lung tumor segmentation, a spatial constraint‐based new fuzzy clustering (SCNFC) approach is then proposed. The experimental results proved the efficiency of the SCNFC technique in the successful segmentation of lung tumor tissues with high accuracy and reduced the false positive per scan (95.21% and 3.69 for a database of simulated computed tomography images). Finally, firefly search‐based Macqueen's K‐means clustering algorithm (FSMKC) is proposed. The performance of the FSMKC method is evaluated on series assessments using the Lung Image Database Consortium dataset (1505 images) and it achieved the accuracy of 97.18% and 96.28% of sensitivity with only 2.27 false positives per scan.