Yi Wang1,2, Glen Morrell2, Allison Payne2, Dennis L. Parker1,2
1Bioengineering, University of Utah, Salt Lake City, UT, United States; 2Utah Center for Advanced Imaging Research, Salt Lake City, UT, United States
We compare two methods of breast tissue segmentation: 1) fuzzy c-mean clustering, an unsupervised learning method that classifies voxels into a specified number of clusters by iteratively minimizing intra-cluster variation, and 2) the support vector machine method, a supervised learning method that uses training data to construct hyper-planes to minimize the margin between classes. We also investigate the effect of varying the number of output clusters and the combinations of input image types. Our goal is to segment breast images into fibroglandular tissue, fat, lesions, and skin. Among other uses, segmentation aids magnetic resonance guided high-intensity focused ultrasound therapy by improving the accuracy of proton resonant frequency thermal mapping and improving the modeling of the simulated ultrasound beam patterns.