Segmentation and reconstruction of cervical muscles using knowledge-based grouping adaptation and new step-wise registration with discrete cosines. Issue 1 (2nd January 2019)

• Record Type: Journal Article
• Authors: Suman, Abdulla Al; Aktar, Mst. Nargis; Asikuzzaman, Md.; Webb, Alexandra Louise; Perriman, Diana M.; Pickering, Mark R.
• Is Part Of: Computer methods in biomechanics and biomedical engineering. Volume 7:Issue 1(2019)
• Page Start: 12
• Page End: 25
• Publication Date: 2019-01-02
• Subjects: Whiplash -- muscle segmentation -- alignment -- elastic motion model -- reconstruction -- dynamic distance force; Imaging systems in biology -- Periodicals; Imaging systems in medicine -- Periodicals; Biomechanics -- Data processing -- Periodicals; Biomedical engineering -- Periodicals; 616.0757
• DOI: 10.1080/21681163.2017.1356751
• Languages: English
• ISSNs: 2168-1163

Abstract

Abstract: Structural changes in the cervical muscles are the cause of most injurious and non-injurious neck pain for which surgery and therapy are used as medical interventions. In clinical practice, the correct diagnosis of disorders and the planning of treatments in the cervical region require high-precision 3-dimensional (3D) visualisation of the anatomy of patients' muscles, which necessitates the highly accurate delineation of neck muscles. However, segmenting cervical muscles is an extremely difficult task due to their identical complexions and the compactness in clinical imaging data. As far as we know, past endeavours did not focus on neck muscle segmentation. Therefore, this paper presents a novel and complete automatic delineation and 3D reformation from tomographic data of some of the specific neck muscles responsible for injurious neck pain. Our method uses linear and non-linear registration frameworks to amend inequalities between the training and testing tomographic data. It can handle posture variabilities among patients using an alignment plan and also exploits a cognition-based grouping adjustment to enhance segmentation accuracy. Our algorithm obtains promising results for real clinical data and offers an average dice similarity coefficient of 0.85 ± 0.02 .