Translator Disclaimer
29 March 2013 Example-based segmentation for breast mass images
Author Affiliations +
A new example-based mass segmentation algorithm is proposed for breast mass images. The training examples used in the new algorithm are prepared by three medical imaging professionals who manually outlined mass contours of 45 sample breast mass images. These manually segmented mass images are then partitioned into small regular grid cells, which are used as reference samples by the algorithm. Each time when the algorithm is applied to segment a previously unseen breast mass image, it first detects grid cell regions in the image that likely overlap with the underlying mass region. Upon identifying such candidate regions, the algorithm then locates the exact mass contour through an example based segmentation procedure where the algorithm retrieves, transfers, and re-applies the human expert knowledge regarding mass segmentation as encoded in the reference samples. The key advantage of our approach lies in its adaptability in tailoring to the skills and preferences of multiple experts through simply switching to a different corpus of human segmentation samples. To explore the effectiveness of the new approach, we comparatively evaluated the accuracy of the algorithm for mass segmentation against segmentation results both manually produced by several medical imaging professionals and automatically by a state-of-the-art level set based method. The comparison results demonstrate that the new algorithm achieves a higher accuracy than the level set based peer method with statistical significance.2
© (2013) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Qingying Huang, Songhua Xu, and Xiaonan Luo "Example-based segmentation for breast mass images", Proc. SPIE 8674, Medical Imaging 2013: Advanced PACS-based Imaging Informatics and Therapeutic Applications, 86740K (29 March 2013);

Back to Top