Automatic lung segmentation based on image decomposition and wavelet transform. (August 2020)

• Record Type: Journal Article
• Authors: Liu, Caixia; Pang, Mingyong
• Is Part Of: Biomedical signal processing and control. Volume 61(2020)
• Publication Date: 2020-08
• Subjects: 00-01 -- 99-00; Lung segmentation -- Image decomposition -- Wavelet transform -- Contour correction; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102032
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We introduce a new lung CT image filtering approach based on an image decomposition technique, which has the ability of denoising while keeping lung contours. We present a novel lung segmentation approach, which considers over image details extracted by wavelet transform and performs well on lung segmentation of pathologic CT images. We propose an efficient lung contour correction method, which is constructed on a fast corner detection technique and can efficiently correct and smooth segmented lung contours. Abstract: Accurately segmenting lungs from medical images is still a challenge due to some negative factors involved in the work, such as inhomogeneous intensities, juxta-pleural nodules, image noises and so on. To deal with the problem, in this paper, we present a novel algorithm to segment lungs from CT images in an accurate and automatical fashion. In our algorithm, an image decomposition based filtering strategy is first introduced to denoise lung CT images while preserving their lung contours. Lungs are then segmented from the CT images by wavelet transformation combining with a group of morphological operations. The segmentations are further refined by a contour correction approach, which is built on a fast corner detection technique, to correct and smooth the extracted lung contours. Experimental results show that our algorithm has better performance than a set of classical approaches, and it achieved an averaged Dice similarity coefficient of 98.04% and Jaccard's similarity index of 94.91% on lung CT image segmentation compared with ground truths. Our algorithm can correctly segment lung tissues from lung CT images and is helpful for radiologists' diagnosis of lung diseases.