Lung lesion identification using geometrical feature and optical flow method from computed tomography scan images

• Published in: Intelligent Multimedia Signal Processing for Smart Ecosystems
• Main Author: Abdullah M.F.; Sulaiman S.N.; Osman M.K.; Karim N.K.A.; Setumin S.; Ani A.I.C.
• Format: Book chapter
• Language: English
• Published: Springer International Publishing 2023
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85189779867&doi=10.1007%2f978-3-031-34873-0_7&partnerID=40&md5=7d2e3768b24a26e71262114f7424de89

Lung lesion identification is essential to an early lung cancer diagnosis. Detecting lung cancer early may aid physicians in treating patients. This chapter presents a geometric feature and optical flow technique for diagnosing lung lesions using computed tomography images. According to prior research, automating lung segmentation is incredibly challenging since fluctuations in pulmonary inflation combined with an elastic chest wall can result in a great deal of volume and margin variability. In addition, the attributes used to describe a lung lesion emphasize image aspects such as geometry, appearance, texture, and others. In this study, lung lesions in computed tomography images are segmented using an image processing technique that uses image segmentation algorithms. The optical flow approach has been designed to work with various computed tomography scan slices that could contain lesions. Collected data, image segmentation, optical flow, and performance evaluation are among the stages of the recommended method that call for image processing techniques. The Advanced Medical and Dental Institute, Universiti Sains Malaysia database was used to gather the computed tomography scan images. According to the study, lung slices with lesions have a standard deviation of 0% and 2.0% for the optical flow method, while slices without lesions have a standard deviation between 2.1% and 9.2%. These results can aid radiologists in more accurately diagnosing lung cancer by helping them immediately identify slices with lesions. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.