Automatic segmentation of supraspinatus from MRI by internal shape fitting and autocorrection. (March 2017)

• Record Type: Journal Article
• Authors: Kim, Sunhee; Lee, Deukhee; Park, Sehyung; Oh, Kyung-Soo; Chung, Seok Won; Kim, Youngjun
• Is Part Of: Computer methods and programs in biomedicine. Volume 140(2017)
• Page Start: 165
• Page End: 174
• Publication Date: 2017-03
• Subjects: Active contour segmentation -- Level-sets approach -- Shape fitting -- Supraspinatus; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2016.12.008
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: We propose an automatic segmentation method of supraspinatus from MRI. This method uses a region-based segmentation and a new shape fitting technique. It can automatically and accurately extract reasonable shape. This method provides the regular 3D surface of supraspinatus. Abstract: Background and objectives: With significant increase in the number of people suffering from shoulder problems, the automatic image segmentation of the supraspinatus (one of the shoulder muscles) has become necessary for efficient and deliberate diagnosis and surgery. In this study, we developed an automatic segmentation method to extract the three-dimensional (3D) configuration of the supraspinatus, and we compared our segmentation results with reference segmentations obtained by experts. Methods: We developed a two-stage active contour segmentation method using the level sets approach to automatically extract the supraspinatus configuration. In the first stage, a trial segmentation based on intensity and an internal shape fitting technique were performed. In the second stage, the undesired image portions of the trial segmentation were automatically identified by comparing the trial segmentation with the fitted shape, and then corrected by forcing the contour to stop evolution in the over-segmented region and pass through undesired edges in the under-segmented region. Results: The proposed method was found to provide highly accurate results when compared with the reference segmentations. This comparison was made on the basis of four measurements: accuracy (0.995 ± 0.001), Dice similarity coefficients (0.951 ± 0.011), average distance (0.440 ± 0.086 mm), and maximal distance (3.045 ± 0.433 mm). The proposed method could generate regular surfaces of the 3D supraspinatus. Conclusions: The proposed automatic segmentation method provides a patient-specific tool to accurately extract the 3D configuration of the supraspinatus.