Spatial Image Segmentation for Breast Cancer Detection in Terahertz Imaging

Abstract

This paper proposes a new spatial image segmentation algorithm for breast cancer detection in terahertz (THz) images of freshly excised human tumors. Region classifications of fresh tissue with 3 or more regions, such as cancer, fat, and collagen, remain a challenge for cancer detection. We propose to tackle this problem by exploiting the spatial correlation among neighboring pixels in THz images, that is, pixels that are close to each other are more likely to belong to the same region. The spatial correlation among pixels is modeled by using Markov random fields (MRF). A Gaussian mixture model (GMM) with expectation maximization (EM) is then used to represent the statistical distributions of the THz images in both the frequency and spatial domain. Experiment results demonstrated that the proposed spatial image segmentation algorithm outperforms existing algorithms that do not consider spatial information.